EXAMENSARBETE INOM DATATEKNIK, AVANCERAD NIVÅ, 30 HP , SVERIGE 2020

Establishing and Evaluating Cinematography Principles in a Third Person Computer Game

NATHAN BHAT

KTH SKOLAN FÖR ELEKTROTEKNIK OCH DATAVETENSKAP

Establishing and Evaluating Cinematography Principles in a Third Person Computer Game

Nathan Bhat

May 20, 2020

Swedish Title: Etablering och utv¨ardingav cinematografiska principer i datorspel med tredjepersonsper- spektiv Author: Nathan Bhat Master’s degree project in Computer Science Master of Science in Engineering in Computer Science program Supervisor: Bj¨ornThuresson Examiner: Tino Weinkauf Host Company:

1 Abstract

As the computer game industry continues to develop, there is a growing demand for games to look as aesthetically pleasing as possible. Because of this, use of cinematographic principles in computer games has become commonplace. This study aims to evaluate and compare cinematography principles regarding framing, distance from the player avatar, and moving the camera when the player avatar is obstructing the view on screen in a third person computer game. Important cinematographic principles were chosen for comparison based on previous research and suggestions by professional game developers. A qualitative user study was performed, where seven expert level participants tested, compared and discussed different camera methods that fulfilled different cinematographic principles in a test scenario. Findings showed that there was a preference towards framing that gave more focus to the surrounding scene rather than the player avatar, that using classic cinematography principles may make the player avatar appear too close, that an ellipsoidal camera orbit may give a stronger sense of stability than a spherical camera orbit, and that players prefer to be able to frame the camera manually to suit them rather than an automatic solution.

Sammanfattning

Med datorspelsbranchens fortsatta utveckling f¨oljer en v¨axande efterfr˚aganf¨or spel att se s˚aestetiskt tilltalande ut som m¨ojligt. I detta syfte har bruk av cinematografiska principer inom datorspel blivit vanligt. Den h¨ar studien ¨amnar utv¨ardera och j¨amf¨ora cinematografiska principer i f¨orh˚allandetill komposition, avst˚andtill spelarens avatar, och f¨orflyttning av kameran n¨ar spelarkarakt¨aren skymmer vyn i datorspel med tredjepersonsperspektiv. Viktiga cinematografiska principer att j¨amf¨ora valdes utifr˚antidigare forskning p˚a omr˚adetoch f¨orslag fr˚anyrkesprofessionella spelutvecklare. En kvalitativ anv¨andarstudie genomf¨ordes, d¨ar sju deltagare p˚aexpertniv˚atestade, j¨amf¨orde och diskuterade olika kamerametoder som uppfyllde olika cinematografiska principer i ett testscenario. Studien fann att det f¨orekom en preferens f¨or kompositioner som gav mer fokus ˚atomgivningen ¨an spelaravataren, att bruk av klassiska cinematografiska principer kan resultera i att spelaravataren upplevs f¨or n¨ara, att en ellipsoidisk kamerabana kan ge en h¨ogre k¨ansla av stabilitet ¨an en sf¨arisk kamerabana, och att spelare f¨oredrar att kunna styra kamerans komposition manuellt framf¨or en automatisk l¨osning.

2 Establishing and Evaluating Cinematography Principles in a Third Person Computer Game

Nathan Bhat May 20, 2020

1 Introduction

When developing games and interactive media there may be cinematography constraints that the designer would like to impose on the behavior of the camera. However, in games it is of the utmost importance that the player does not get disoriented by either having camera cuts or by losing sight of the player avatar and any external objectives, therefore cinematography in games often differs from cinematography in film. In games, player immersion is also of extreme impor- tance, therefore the camera must behave in ways that do not break that immersion. This leads to a prob- lem with no optimal solution where both technical and cinematographic constraints should be fulfilled.

Figure 1: This research investigates the use of principles of cinematography in third person computer games and 1.1 Definition of concepts evaluates them based on visual appeal and player experi- ence. The following is a set of definitions of relevant con- cepts for this research project.

1.1.1 The Rule of Thirds

A cinematography principle that generally creates pleasing composition regarding framing is that of the “Rule of Thirds”. As stated by [Heiderich, 2018], the rule of thirds is about placement of key objects or actors in a shot to achieve a balanced composition. If one were to split the screen horizontally into three

3 1.1.3 Offset from player

The term “offset” in this paper refers to the dis- tance the virtual camera object is from the player avatar. Sphere-based offset in this paper means that the offset is constant, so the camera object’s position is locked to being on the surface of a hypothetical sphere around the player avatar. This can also be seen as a spherical orbit around the player avatar. Figure 2: A picture showing a rule of thirds overlay and a Ellipsoid-based offset refers to using an ellipsoid in- phi-grid overlay on a 16:9 aspect ratio rectangle. The red solid lines denote the rule of thirds grid, whereas the blue stead of a sphere, which could be useful for example dashed lines denote the phi-grid. The points at which two if the desired camera should have a larger offset when same colored lines intersect define the crash points of the aimed at the avatar’s shoulder to maintain a sense of respective grids. distance from the avatar.

1.1.4 Camera flipping equal pieces, and then vertically in three equal pieces, one would result in a 3 × 3 grid with four intersec- If the player avatar is aligned with a certain side of tions between grid lines on screen as seen in Figure the screen according to the rule of thirds, there may 2. These four intersections are referred to as “crash be a situation where the player avatar is obstructing points”. To follow the rule of thirds one would align the view of some external object. To solve this the key objects or actors with any of the crash points method of “camera flipping” is introduced. In this on screen and have longer objects, such as an ac- paper camera flipping refers to moving the camera in tor’s body or the horizon, flow between crash points. a way that moves the player avatar from one side of Generally all intersections would not have to be used the screen to the opposite side of the screen in hori- at once, and by aligning the key objects with dif- zontal direction. This camera flipping behavior could ferent crash points different dramatic effects can be be manual in that the player uses an input device to achieved. decide when the camera should flip, or it could be automatic, where the system would try to determine when the player might want the camera to flip.

1.1.2 The Phi-grid 1.2 Problem definition

The objective of this study was to investigate tech- The Phi-grid, is a similar principle to the rule of niques for solving common challenges when creat- thirds, and the only difference has to do with the ing interactive third person camera systems that are placement of the vertical and horizontal grid lines. meant to be aesthetically pleasing and practical in As opposed to the rule of thirds, the placement of games. The biggest challenge this study aims to over- grid lines is defined using the golden ratio instead of come is that of framing, with questions such as where thirds, as presented by [Svobodova et al., 2014]. The should the player avatar be in the frame, and what comparison of the phi-grid to the rule of thirds on a distance should the camera be from the player avatar. 16:9 screen can be seen in Figure 2. Because cinematography is subjective the metric used

4 to evaluate this should be the visual appeal as well Together, answering these questions should give in- as the quality of experience from the player perspec- sight to designing an in game camera for a third per- tive. Besides this the technical solution for the cam- son game, and what methods provide a high visual era should not be performance heavy as it needs to appeal as well as a high quality of experience from be able to run in real time, and be stable over a long the player perspective. Using these insights should period of time. help solve the problem presented.

To research this topic an “on-foot” camera system The different methods selected for comparison were (for when the player avatar is standing, walking or based on the background and research section of this running) was developed. This camera system uses thesis, as well as suggestions given by professional certain camera movement methods, to fulfill the de- game developers. sign constraints of: framing, offset, frame coherence and accessibility, as well as fulfilling the technical re- quirement of being performance effective. 1.4 Keywords

Cinematography, games, cameras, third-person 1.3 Research Questions

How does the rule of thirds framing compare to the phi-grid framing in regards to visual appeal and qual- 1.5 Contributions ity of experience?

This research question is meant to give insight to how This study evaluates and compares cinematographic different framing methods perform from an aesthetic principles and their practical implementation in a and experience perspective in a third person game. computer game based on visual appeal and quality of The framing here refers to where on screen the player experience. Specifically this study researches meth- avatar is placed, more specifically the head of the ods of framing the player avatar, setting the offset player avatar model. from the player avatar, and moving the camera to flip the player avatar from one side of the screen to How does a sphere based offset compare to an ellipsoid the other, horizontally. This study contributes al- based offset in regards to visual appeal and quality of gorithms that can be used to fulfill these cinemato- experience? graphic principles. This study discusses potential ar- eas for future exploration within cinematography in This research question is to give a comparison of the games. two different offset methods and assess the feasibility of using different offset methods.

How does an automatic camera flipping method com- 1.6 Delimitation pare to a manual camera flipping method in regards to quality of experience? This study will not include investigating camera set This research question is meant to investigate dif- ups for cinematic scenes (commonly referred to as ferent possibilities for flipping the camera when the cut-scenes), or other cameras besides the in game framing becomes obstructive and evaluating them walking, running and sprinting cameras. This study based on quality of experience. only investigates a limited amount of cinemato-

5 graphic principles. This study only investigates a lim- ited amount of technical solutions. This study does b not investigate the effect of different field of view an- φ = gles. a + b b b = + a b b = + 1 a 1 (2) = + 1 2 Background φ 1 φ = + 1 φ In this section relevant theory and related work re- φ2 = 1 + φ garding cinematography and in game cameras will be 2 presented. φ − φ − 1 = 0

Using the quadratic formula, equation (2) can be solved explicitly, giving only one positive root. Since the ratio is a geometric descriptor it must be posi- 2.1 The Golden Ratio tive, therefore only the positive root is defined as the golden ratio.

The golden ratio is a mathematical constant describ- √ ing a geometric ratio, usually between lengths. This 1 + 5 φ = ≈ 1.62 (3) ratio has been the subject of study for many years, 2 as there seems to exist an aesthetic preference for it in many people, however no conclusive evidence of this has been found, according to [Svobodova et al., 2.2 Cinematography 2014]. The golden ratio also appears frequently in nature (such as the layout of petals on a flower) as well as in human design as shown in [Akhtaruzzaman Cinematography is the practice of recording film, as and Shafie, 2011]. stated by [Fairservice, 2019, p. 10]. Since its incep- tion, directors have developed practices, rules and [Svobodova et al., 2014] expresses the mathematical theories about how footage should be recorded to definition of the golden ratio as follows: maximize visual appeal and the conveying of infor- mation, often called “Principles of Cinematography” according to [Christianson et al., 1996].

a b These principles of cinematography are not objective = = φ (1) b a + b rules, but rather guidelines meant as a set of tools for a filmmaker. For example, as discussed by [Pro- feres, 2017, p. 44], the height of the camera should be generally at the height of the actor’s eyes, how- Where the Greek letter phi: φ, is the symbol rep- ever this can be broken to convey drama and emo- resenting the golden ratio. The ratio can then be tion. Because of the ambiguity of the principles of calculated explicitly using the above definition: cinematography, there is no exact science to creating

6 tal rule of thirds line was perceived as more beautiful than placing the horizon line along the lower horizon- tal rule of thirds line.

[Berdan, 2004] states that there are factors that go Figure 3: An image depicting the phi-grid division along into composition that can overrule both the rule of a line. If the entire horizontal dashed line represents the thirds and the phi-grid and that ultimately it is up to width or height of the screen, the dashed vertical lines the photographer or cinematographer to decide how represent the grid lines for the phi-grid. The ratio b = φ. a to build the composition for their purpose. This image was taken and reformatted from [Svobodova et al., 2014]. There are other principles of cinematography used for framing, such as the rule of fifths, where the camera shot is split into a 5 × 5 grid, creating more grid lines and crash points. [Gooch et al., 2001] states that the rule of fifths can be used in combination with the rule pleasing cinematography, rather, it is artistic expres- of thirds by splitting up the image in thirds along one sion that governs cinematography. The Rule of thirds axis and fifths along the other. Another principle is and phi-grid are some examples of these principles. using the golden spiral. Similar to the Phi-grid the The phi-grid is defined from the golden ratio in that golden spiral is based on the golden ratio and which if the first dividing line in the grid is one unit from can be used to determine how the “flow” of the image the edge of the screen, then the second dividing line should be composed, according to [Svobodova et al., should be φ units from the edge of the screen. This 2014]. definition is applied from all four directions of the [Arijon, 1976] discusses different heights at which the screen to maintain a symmetry. This division is de- frame should cut off for aesthetically pleasing shots picted in Figure 3. Compared to the rule of thirds 1 of humans. The mentioned cutting heights are: where the dividing lines are at 3 of the screen space 2 1 and 3 of the screen space, the phi-grid lines are at φ2 1 and at φ of the screen space. • under the arm pits According to [Svobodova et al., 2014] and [Metiu • under the chest et al., 2014], the rule of thirds is an approximation of the phi-grid, however while Svoboda et al. states that • under the waist this approximation is considered sufficient in prac- tice, Metiu et al. argues that this approximation is a detriment to the golden ratio. The stance taken by • under the crotch Svoboda et al., stems from their survey of 15 differ- ent research papers that had conflicting conclusions • under the knees regarding a preference for the golden ratio, however their conclusion states that a preference for placing motifs in the crash points was more preferable. A final cutting height is mentioned, the “Full shot” which Arijon states must include the subjects feet as Further [Svobodova et al., 2014] showed, using a user cutting the frame above the ankles is not believed to study, that the placement of the horizon line affects result in a pleasant composition. In the reference im- the perceived sense of beauty in landscape photogra- ages provided, the top of the subject’s head is placed phy. It was shown that placing the horizon line in the close to the top of the frame (never more than half a center of the photograph, or along the upper horizon- head’s length away).

7 2.3 Virtual Cinematography

As discussed in [Burelli, 2016], virtual cinematogra- phy differs from classic cinematography mainly in the fact that the objects in the scene, including the cam- era, do not need to abide by the laws of physics. As the name denotes, virtual cinematography takes place in the virtual world, in other words, in a com- puter graphical world. This means that the camera is an abstract representation meant to mimic the be- havior of a real world camera. Because of the abstrac- tion virtual cameras have a higher degree of freedom compared to physical cameras and can be used in ways that would be difficult or impossible for a phys- ical camera. Despite this cinematography has mainly Figure 4: An image depicting the axes of freedom a virtual been dominated by physical cameras, meaning that if camera has. The three positional vectors, the yaw (θ), the virtual cameras are used in unconventional ways they pitch (φ) and the roll (ψ), as well as the field of view (γ). can contrast to classic cinematography and by doing Image taken from [Christie et al., 2008]. so disorient the viewer.

Within virtual cinematography there are two types of There can be other parameters that affect the view of camera control; automated and interactive, according the camera such as an orthographic projection model, to [Christie et al., 2008]. In the prior the system however their connection to the camera itself is de- and developer are in full control of how the camera batable. will move and act, and if it is fully automated the user or viewer will have no influence on the camera whatsoever. Interactive camera control let’s the user control the camera’s behavior through different input 2.4 Game Cinematography means. This level of interaction can vary greatly, such as a user being able to switch between different Computer games are examples of applications that camera views at a button press to the user controlling use interactive virtual cinematography. According to the virtual camera through the motion of a virtual [Halper et al., 2001], the interactivity of games can reality headset [Wiedemann et al., 2016]. lead to “less-than-satisfactory camera position[s]” due to the player’s ability to move the subject of the Commonly cameras in virtual cinematography are shot freely. Because of the nature of games it is good expressed by seven values: x-position, y-position, z- practice to have camera movement bound by certain position, yaw, pitch, roll and field of view [Christie cinematographic constraints. For example “cutting” et al., 2008] and can be seen in Figure 4. However (instantly switching between camera positioning and the camera can also be expressed by three vectors angles) can be disorienting for a player who is con- and a value: the position vector, the normal of the trolling an avatar in the game [Oskam et al., 2009]. viewplane, the up vector (describes which direction is considered up for the camera) and the field of view Another constraint is having the camera avoid col- [Drucker and Zeltzer, 1994]. Generally the choice of lision with in game objects or avoiding occlusion representation is for the implementer of the system from in game objects. This occurs naturally in clas- to decide, and a combination of both can be used. sic cinematography where a camera operator simply

8 avoids running into things in the scene, however in games the camera operator is usually automated and must avoid collision using an algorithm [Halper et al., 2001].

As stated by [Christie et al., 2008], in games the cam- era is either directly controlled by the player through inputs that directly change the position and rotation of the camera, or indirectly in control of the camera by affecting its position and rotation through mov- ing a virtual avatar that is the subject for the cam- era. Two common methods for camera usage in 3D games is that of “First Person” and “Third Person”.

In First Person games, the camera is placed and con- Figure 5: A screenshot from the game Sekiro: Shadows trolled in a way that is representative of the perspec- Die Twice [FromSoftware, 2019]. Note how the player avatar is centered on the screen and that the camera is tive of the player avatar. This type of camera is gen- placed so that the avatar’s entire body is on screen. erally seen as less problematic due to the camera po- sition and rotation being trivial [Christie et al., 2008]. An example of a game that uses this type of camera is Doom Eternal [, 2020].

In Third Person games the camera follows the player avatar from a distance and must find a way to solve issues such as avoiding occlusion of the player avatar or a point of interest, or avoiding collision with in game geometry [Christie et al., 2008]. Due to this added complexity there are many different ways to frame the player avatar in a third person game. One common way of framing is to simply center the player avatar on the screen, as shown in for example The Legend of Zelda: Breath of the Wild [Nintendo EPD, 2017] and Sekiro: Shadows Die Twice [FromSoftware, 2019] which can be seen in Figure 5.

In third person games regarded as “cinematic” it is common to adhere to a non-centered shot, and try- Figure 6: A screenshot from the game God of War [SIE ing to fulfill more cinematographic principles. A cou- Santa Monica Studio, 2018] taken from [Nelva, 2018]. ple examples of this are the games The Last of Us Note how the player avatar is off to the side, and that [Naughty Dog, 2013] and God of War [SIE Santa the avatar’s head seems to line up with a rule of thirds Monica Studio, 2018] which can be seen in Figure crash point. Note also that the avatar’s entire body is not 6. on screen.

In all four of the third person games mentioned as ex- amples, the offset used is spherical [Nintendo EPD, 2017] [FromSoftware, 2019] [SIE Santa Monica Stu-

9 dio, 2018] [Naughty Dog, 2013]. While it is believed constraints at each frame. They evaluated different by the author of this thesis that a spherical offset solutions by testing participants with a demo game, method is the standard in the industry for third per- repeating it several times with varying heights, dis- son games, no source showing this could be found. tances and levels of frame coherence.

2.5 Related Work 3 Methodology [Bares et al., 1998] researches cinematography tech- niques being applied in real time to an interactive This section discusses and motivates the methodol- application using a constraint based solution. While ogy used in this thesis. this solution does contain a few cinematographic principles, it does not take frame coherence into ac- count and focuses mainly on interactive applications 3.1 Research Methodology rather than games.

[Halper et al., 2001] builds on the findings of Bares The main method of research for the background et. al, and extends the problem to a trade-off between of this thesis was based in modern papers, most of constraint satisfaction and frame coherence. The pa- which are from the past 10 or 20 years. The papers per researches methods to, in real time, show the on virtual cinematography and games come from re- visibility of important objects while avoiding camera spected publications and conferences, such as SIG- “jumps”. The paper focuses on computer games and GRAPH, and they are peer reviewed. Some of the defines certain desirable aspects of the camera such sources used for the cinematography background are as height and orientation towards objects in scene. from less known publications and their findings are Halper et. al’s paper also discusses “over the shoul- questionable. Despite this they are believed to be der” solutions, however does not go in depth about truthful and relevant as several of the sources reiter- placement of the actor, or player avatar, in frame. ate the information used for the background of this thesis. A paper that investigates and analyzes the use of cin- ematographic principles in computer games is that of There are a few examples in the background taken [Christie et al., 2008]. While this paper is a litera- from modern games. The games are all from the last ture and case study and does not have any experi- 7 years at the time of writing this report. They have ments, it reviews classic cinematographic principles all received high scores (over 90 out of 100) on the in games to a much greater degree than the two pre- website , and were therefore deemed as rel- viously mentioned papers. Christie et. al’s paper evant examples in this thesis. discusses the common problems that can occur when employing cinematography principles in a game and discusses different theory and methods to fulfill cer- 3.2 Evaluation Methodology tain constraints.

[Yannakakis et al., 2010] tests different camera se- In this study the main method of evaluation was a tups and camera control methods and measures par- semi-structured interview. The reason for this was ticipants psychophysiological features such as heart to be able to get indicative data by asking specific rate, but also perceived fun and frustration. The questions, but to also collect qualitative data regard- camera solver was based on a weighted Constraint ing how the camera is experienced by asking open Satisfaction Problem framework for satisfying view ended interview questions. This method of evalua-

10 After each of these gameplay segments the partic- ipant filled out a form with multiple choice ques- tions evaluating and comparing two camera meth- ods. They then were verbally asked to answer open ended questions and motivate their answers. These segments were intended to get a clear indication of the participants preferences and opinions about alter- native camera methods, and let them motivate and comment, ultimately collecting qualitative data.

For the multiple choice questions, this study opposed Figure 7: The player avatar as seen from above. The el- the choice of [Yannakakis et al., 2010] to use 4-AFC lipse around the avatar represents the offset of the camera and instead used a 7-point Likert scale. This deci- from the player avatar during the ellipsoidal offset setting. sion was ultimately based on the findings of [Watrin Note how the zeniths of the ellipse are placed to line up et al., 2019], where forced choice was compared to with the player avatar’s shoulders to maintain a sense of the Likert scale, and stated that while both 4-AFC distance from the avatar. The dotted circle represents the and 7-Point Likert showed an acceptable level of re- offset of the camera from the player avatar during the liability, the 7-Point Likert scale had a higher level spherical offset setting. of reliability compared to 4-AFC. 4-AFC was how- ever used once per segment; when the participant was asked to state which camera method they preferred when comparing two. This is because when showing tion was based off of the research done in [Ware and a preference between two methods it is easier to get a Osborne, 1990] with some minor modifications taken clear and non-nuanced indication of preference from from [Yannakakis et al., 2010]. 4-AFC than a Likert scale.

The study consisted of three different gameplay seg- Recordings of the demo sessions were reviewed and ments, one for framing, one for offset and one for relevant comments from the participant were pre- flipping. In each segment the participant was asked sented in the results section. The answers from the to complete a short test - consisting of a set of tasks - survey were used indicatively and were occasionally using one setting, then that setting was changed and presented in the results section when deemed rele- the test was repeated with the new setting. For the vant. framing test the two settings are the rule of thirds framing and the phi-grid framing, both placing the player avatar’s head in the upper left crash point. 3.2.1 Inclusion Criteria For the offset test the two settings are a spherical offset and an ellipsoidal offset, where the two zeniths of the ellipsoid are in line with the player avatar’s For this study it was important that the partici- shoulders as seen in Figure 7. For the camera flip- pants had a lot of experience playing third person ping test the two settings will be an automatic flip- 3D games. This was because expert opinions would ping method, that flips depending on what geometry be able to give a better assessment of the quality is around the player avatar (such as walls etc), and a of experience mentioned in the research questions as manual flipping method where the player can choose they can compare and draw similarities between other to flip the camera at a button press. These different games they have played. Participants who do not settings were meant to reflect the research questions have as much experience could cause a larger vari- presented in this thesis. ance in the results of the study, and potentially skew

11 valuable results, while more experienced participants ground in working with games and developing im- are likely to have stronger opinions. Therefore all provements. They generally self assessed positively, participants had to fulfill the following requirements: and none of the participants expressed a lack of back- ground knowledge or experience in the survey. The drawbacks of using a small set of participants from • The participant must have played at least 10 dif- one company is discussed in the critique of method- ferent third person 3D games. ology section in this thesis. • The participant must have played each of the 10 aforementioned games for at least 10 hours or to 3.2.2 Test Scenarios completion.

To evaluate the different features of the camera meth- A participant who fulfills these criteria should be seen ods, three different test scenarios were created. These as believable, as they have a wide variety of games were created with the intention of delimiting the ex- they are familiar with, they have played each game perience of the player to only what was necessary for a long period of time, meaning that they were able to create an understanding of the camera feature in to grow accustomed to whatever camera set up were question. The following list describes the three dif- in these games, and it is likely that the participants ferent scenarios: are relatively skillful at third person games. This should be evidence for the participants being reliable and their opinions being based on experience, rather 1. The player starts at point A and moves the than the tests feeling completely new to a participant avatar from point A to point B in a straight line. who will not be able to form a strong opinion in as The player then walks back to point A through short of a time. This inclusion criteria also allows the points C and D (which results in a path that is participant to discuss and draw comparisons to other not a straight line). games to a stronger degree. 2. The player starts at point A and looks at targets X, Y and Z without moving the avatar. In this study there were seven participants. These seven participants all either worked at, or were doing 3. The player, starting at point A, should move to- their theses at, Avalanche Studios, a large game stu- wards a target X and come to the edge of an dio that specializes in large 3D games. Their experi- in-scene wall (which should partially obscure X, ence and background as game developers varied. The triggering an automatic camera flip). The player average time they had spent professionally within the should then move the avatar towards target X game industry was 3.7 years. All participants self as- (so as to let the automatic camera flip back). sessed themselves to be either comfortable or very The player should then move the avatar so that comfortable playing third person games, and all ex- they can run alongside the wall, with the non- cept one self assessed as either skilled or very skilled dominant side of the screen, meaning the side at playing 3D third person games. The single excep- without the avatar, being on the side of the wall tion stated that they felt that they were neutral when (to trigger flipping while running along the wall) assessing their skill, however they have 13 years of ex- towards a target Y. This test has been depicted perience in the game industry making them reliable in Figure 8. as well.

These participants and their backgrounds were It is important to note that after each test the par- deemed as believable participants, since they not only ticipant was allowed to switch between the different fulfill the inclusion criteria, but also have a back- camera methods and control the avatar and camera

12 3.3 Implementation Methodology

The following section discusses the different methods implemented in this thesis, and presents the under- lying algorithms used for each method.

3.3.1 Framing Figure 8: A diagram of the player avatar’s movement around a wall in the third test scenario meant to eval- uate the camera flipping methods. In this test scenario To set the correct framing, the camera is moved so the right hand side of the screen is non-dominant. that a predefined focus point (in this case a point relative to the player avatar object that is roughly located by the player avatar’s head) lines up with a desired point on screen. The general method can be seen in Algorithm 1. freely as well. This was allowed since it helped the participants form their opinions and imagine what Algorithm 1: Framing Algorithm scenarios they would be using the camera set ups in. Data: distance, the distance the camera The first of these tests was to evaluate framing, as it object is from the focus point. fov, the required the player to simply move around the scene field of view of the camera. with the avatar using a certain framing. This test is aspectRatio, the aspect ratio of the performed for both the rule of thirds framing and the game window. horizontalF rac, the phi-grid framing. fraction of the screen the focus point should be horizontally. verticalF rac, The second scenario was also meant to test the offset the fraction of the screen the focus methods, spherical and ellipsoidal. The idea of this point should be vertically. was that the test scenario will go through the general Result: A vector describing the offset the different camera motions that would differ because of camera should be from the focus the offset method. The player experience could also point. be evaluated based on ease and comfort of moving the left := distance ∗ T an(fov/2) ∗ aspectRatio camera to look at different targets using the different right := −left offset methods. top := distance ∗ T an(fov/2) bottom := −top The third scenario was to to evaluate the camera flipping methods, automatic and manual. This test Vector res could demonstrate the different kinds of scenarios res.x ← Lerp(left, right, horizontalF rac) where a feature like this might be useful to the player, res.y ← Lerp(top, bottom, verticalF rac) and also evaluated how the methods performed in res.z ← distance those scenarios. return res Note: Tan(x) refers to the trigonometric The tests performed, the comparisons done and the tangent function executed on x. Lerp(a, b, t) questions asked in the user study were meant to get refers to linear interpolation from a to b expert opinions that could help answer and discuss using the interpolant t. the research questions presented in this thesis.

13 There are a few things that should be noted about Algorithm 1. Since the algorithm does not deal with the camera’s rotation the algorithm assumes that the camera has the correct orientation, that is to say that it is aimed at the focus point and uses a reasonable up vector before it is executed. This assumption is Algorithm 2: Offset Algorithm fulfilled automatically in the system worked with in Data: playerF or, the vector describing the this thesis, however it may need to be implemented player avatar’s forward vector. manually if recreated in another setting. Further, it cameraP os, the camera’s current should be noted that the camera’s rotation around position relative to the player avatar. the focus point is decided outside of this algorithm, dX, dY, dZ, the distances from the which is why the resulting vector does not deal at all focus point that the camera should be with rotation either. relative to the player avatar’s right, up and forward vectors (can also be seen After this algorithm is performed, the camera’s dis- as the three parameters defining the tance from the focus point may be larger than before. shape of the ellipsoid around the player This does not cause a problem in this application avatar). however, since the offset algorithm, Algorithm 2, is Result: The new distance the camera should performed directly after the Framing algorithm. This have from the focus point. ensures that the distance the camera is from the fo- cameraP osF lat := cus point is correct once all calculations have been cameraP os − (cameraP os · UP ) ∗ UP done. playerF orF lat := playerF or − (playerF or · UP ) ∗ UP azimuth := GetAngle(cameraP osF lat, playerF orF lat) polar := GetAngle(cameraP os, UP ) 3.3.2 Offset Vector ellipsoidV ec ellipsoidV ec.x ← dZ ∗ Sin(polar) ∗ Cos(azimuth) The offset method to bind the distance of the camera ellipsoidV ec.y ← to a predefined ellipsoid around the focus point is dX ∗ Sin(polar) ∗ Sin(azimuth) executed after the framing method. The algorithm ellipsoidV ec.z ← dY ∗ Cos(polar) for this method can be seen in Algorithm 2. return ellipsoidVec.length It may need to be clarified that the first two lines Note: UP refers to the vector (0, 1, 0). The in the algorithm are projecting the vectors on to the symbol · refers to the dot product operation. XZ-plane, so that the azimuth can be calculated. GetAngle(a, b) calculates the smallest angle between vectors a and b.Sin(x) and Cos(x) For the offset algorithm it should be noted that the refer to the trigonometric sine and cosine variable azimuth used is not the actual true value functions, respectively, executed on x. azimuth. Since it uses the smallest angle between v.length refers to the length of the vector v. two vectors it can never be larger than π radians. The algorithm maintains its correctness since only the ellipsoidV ec’s length is used, not the actual vec- tor itself. The length will remain correct despite the vector being mirrored.

14 Note that since the offset vector used for framing unflipped position if neither of the cameras are ob- is directly proportional to the distance the camera structed and the avatar is moving above a velocity is from the focus point, moving the camera along threshold is to prioritize a preference for placing the the axis from the focus point to the camera will not avatar to one side of the screen (a default state). change the framing (as it is equivalent to rescaling the offset vector), therefore the framing of the fo- The reason for this choice in method for automatic cus point is maintained after the offset algorithm has flipping was based in minimizing the amount of flips, been performed. so that they only occured when necessary, and also making sure the camera goes back to a default po- sition when the region around the player avatar and camera are relatively free from geometry. 3.3.3 Flipping Since flipping only changes the horizontal framing The flipping algorithm is rather trivial as it simply and not the rotation of the camera, the camera will be interpolates between two horizontal screen fractions aimed at the same point infinitely far away. That is to for the framing algorithm. However when this flip say if the player has aimed the camera at a point far occurs is non-trivial. In manual flipping this is obvi- away, and then a flip occurs, the camera will still be ously connected to a button press, however in auto- pointing relatively close to that original point. This matic flipping there must be some sort of analysis on difference however is larger and more noticeable if the the geometry around the camera and player avatar. camera is aimed at something close by. There are many different ways this analysis can be done, the method used in this thesis is described as follows:

1. Cast a ray a set distance from the unflipped and 3.3.4 Other Implementations flipped cameras in the direction they are aimed. Also cast a ray from the current camera (flipped The test scenarios were created using an existing level or unflipped) to the other camera. editor based in the Apex Game Engine, developed by Avalanche Studios, with basic block geometry. The 2. Flip if: method for switching between different camera meth- ods was also created for this thesis. The chosen set- (a) The current camera ray is obstructed and tings for the different camera set ups were chosen the other is not and the ray between the based off of general cinematography principles found two cameras is unobstructed. in [Arijon, 1976] and were guided by a professional (b) Or if none of the three rays are obstructed, game designer. These settings included the distance the camera is currently flipped and the the camera was from the player avatar, the speed at avatar is above a certain velocity threshold. which a camera flip happens, the sensitivity of input, and the field of view of the camera.

This method is meant to only cause flipping to oc- The speed at which the camera flips was different for cur when the current view is obstructed and the the manual and automatic camera methods. Manual other view is not obstructed. The algorithm must flipping had a quicker transition since it was meant also check that the path between the two cameras is to feel responsive, whereas automatic flipping had a free, so that the camera flip doesn’t cause clipping slower transition to reduce the feeling of the flip being through geometry. The reason the camera flips to an “jarring” or disorienting.

15 3.3.5 Implementation that already existed Participant Framing Offset Flipping Anthony ROT Both Manual Brian PG Ellipsoidal Both It is important to note that the system built for this Charles ROT Ellipsoidal Manual thesis built off of a pre-existing solution in the appli- David ROT Ellipsoidal Manual cation. This included implementation for handling Edward ROT Ellipsoidal Neither input and rotating the camera around the player Frank PG Both Manual avatar based on this input, as well as moving the George Both Ellipsoidal Both camera with the player avatar in the event that it moves. The player avatar and ability to control it was also an existing implementation. Table 1: This table shows the results per participant re- garding which camera method they preferred in each seg- ment. ROT refers to the Rule of Thirds and PG refers to Phi-grid. The questions were in 4-AFC form and the 4 Results participants could answer camera method A was preferred, camera method B was preferred, both were preferable or neither were preferable. In the following section results from the user studies will be presented. This includes the indicative re- sults from the survey and also thoughts and opinions expressed during the interviews. the rule of thirds framing put the avatar more out to the side, but that he did not like the fact that the This study consisted of seven different participants, avatar’s head was placed higher in screen space. for the sake of their anonymity they will be referred to with the following names (selected from popular boy’s The two participants who preferred the Phi-grid names): Anthony, Brian, Charles, David, Edward, framing, Brian and Frank, also both stated that the Frank and George. reason for their preference was that they felt they could see more. Brian stated that the reason he The different overall preference for each participant felt this was because he could see more of the scene of the different camera methods can be seen in Table above the avatar’s head. Frank stated that when the 1. framing was set to the rule of thirds that it felt like the avatar was obstructing the view. He had knowl- edge of the rule of thirds before the study and stated 4.1 Framing that when the player avatar’s head covered the crash point it gave the impression that the player avatar was blocking the view of more important objects in A sketch depicting what the two different framing scene. methods look like in game can be seen in Figure 9. Anthony, Brian and Frank thought that the differ- Most of the participants preferred the rule of thirds ences in framing were subtle, and because of this An- framing to the Phi-Grid framing. A reoccurring thony and Frank had only a mild preference for their theme for why this is that was stated by all four par- chosen framing method. Frank when asked about ticipants who showed preference towards the rule of purely visual appeal answered that he did not prefer thirds was that this framing gave more visibility of one over the other. George, while stating that both the scene due to the avatar being more off to the framing methods were just as preferable, showed pref- side. George who showed no preference between the erence towards the rule of thirds framing when asked two framing methods also stated that he liked how about visual appeal.

16 Participant Stance on noticeable difference Anthony Mildly agree Brian Agree Charles Mildly agree David Mildly agree Edward Disagree Frank Completely disagree George Mildly agree

Table 2: This table shows the results per participant re- garding their stance on the statement “I noticed a differ- ence between the two offset methods.” These results were gathered using a Likert scale question.

4.2 Offset

Figure 9: A sketch showing the difference in framing As seen in Table 2, most participants only noticed a in game between the two framing methods. The fram- minor difference between the two offset methods, one ing above uses the rule of thirds, and the framing below could see a difference and two participants could not uses the Phi-grid. The horizontal line indicates the hori- see a difference. While this difference felt very subtle zon, and the rectangles are representative of 16:9 screen to most of the participants, a clear preference seems size. Note that in the Phi-grid framing the player avatar to be found for the ellipsoidal offset method, as seen is lower and farther to the right on screen. in Table 1.

None of the participants showed strong dislike to- wards either offset method, and the survey results David, Frank and George felt that in both framing showed that they generally felt that both methods methods that the player avatar appeared too high up were acceptable solutions in a game. One exception in screen space, and Brian maintained that his reason to this however was Charles, who stated in the sur- for liking the Phi-grid framing better was because vey that he mildly disagreed to the statement “I feel the player avatar was lower on screen. Frank came I could play a game with the first offset method for with a suggestion regarding this; to move both the a long period of time.”, the first offset method be- player avatar and the aiming reticle down in screen ing the spherical offset. In regards to visual appeal, space. In the current solution the aiming reticle is both methods were found to be acceptable or even in the center of the screen, and Frank stated that aesthetically pleasing. Brian and David showed pref- usually that reticle will line up with the horizon line. erence for Ellipsoidal offset visually, however all other By lowering it, one could add more focus to what is participants showed no preference for either method above the horizon line, rather than below, according visually. to Frank. Both Frank and David stated that one must tilt the camera to see what is in front of the Anthony saw such a small difference that he stated player avatar, so Frank suggested that by moving the that he would have guessed they were the same. horizon line this could be remedied. David, Frank and George also all stated that the player avatar felt Brian immediately noticed a difference and showed too close to the camera and that it took up too much a clear preference for ellipsoidal offset. He stated he screen space. felt it was easier to aim using ellipsoidal offset.

17 Charles stated that he felt ellipsoidal offset was more “hardcore” games (meaning challenging games, with responsive, however he was not sure why that was. a target audience of experienced game players) while automatic felt more cinematic. David preferred ellipsoidal offset because he noticed a larger distance between the camera and avatar in cer- Anthony stated that while he preferred the manual tain situations, and he stated that he had preferred flipping method more, he liked both, and said that the avatar to be farther away from the camera. he would like to have both methods in a game. An- thony mentioned that he felt it was easier to plan Edward did not initially notice a difference however using the manual flipping method, as you could flip after some testing he noticed the camera varied more to the side you wanted before the moment you need in the ellipsoidal offset method. After this he stated it. He also mentioned that while he understood that that he felt the avatar was more “stable” when using only the camera was flipping, having the camera flip the ellipsoidal offset method, which was his ultimate so that the player avatar looked like it was standing reasoning for showing preference towards it. behind a wall made it feel like the player was more protected from possible enemies. He stated that the Frank, though in the survey stated that they were current solution for automatic flipping felt like you, both as preferable, changed his mind after the offset as the player, needed to move the player avatar too segment and verbally stated that he preferred the el- close to a wall to trigger a flip. Anthony over all felt lipsoidal offset better. However he then went on to that both methods were convenient and improved the say that he felt it was more difficult to aim when the experience. ellipsoidal offset was turned on. Brian generally felt that both solutions improved the George could immediately see a difference, however experience, and that which solution was used de- found that difference to be negligible. Similarly to pended on the type of game. He stated that he also David, George noticed that the camera was farther thought that the actual flipping transition for man- from the avatar at certain angles and showed a pref- ual flipping could have been faster. While Brian ex- erence towards it because he felt he could see the pressed that he felt both solutions were as preferable, scene better. in the survey he answered the Likert scale questions in a way that showed discomfort, a feeling of disrup- tion of player movement and a feeling of frustration towards the manual flipping method.

4.3 Flipping Charles stated that while there may be benefits to an automatic camera flip, there is always a chance that it will flip when you as the player do not want As seen in Table 1, there was no participant who pre- it to, whereas manual flipping works every time. He ferred the automatic camera flipping method, how- thought that after some time manual flipping would ever 2 participants, Brian and George preferred both become second nature to the player. Though he pre- methods, and Edward preferred neither. ferred manual flipping he also said it might be inter- esting to see a hybrid solution. He noted that the A reoccurring theme, stated by Brian, Charles, Ed- change in reticle position due to the flip was accept- ward and George is that the flipping method depends able, and thought it was better to have a small change on the type of game. Brian stated that the automatic rather than rotating the camera to maintain a target. method could be more suitable in “casual” games (meaning not very challenging games, that do not assume high levels of experience in playing games), and Charles stated that manual felt more suitable in

18 David also showed a preference towards manual flip- bug where the camera could skip the transition and ping as he thought automatic flipping felt less pre- teleport. He also experienced another bug which was dictable. He stated that he saw the potential for au- that the manual flipping button could sometimes be tomatic flipping to be frustrating, however he also unresponsive, and admitted that the responsiveness stated that either flipping method was better than of the button was distracting despite knowing it was not having any flipping method. Because of the tran- not intentional. Ultimately he said that the auto- sition for automatic flipping was slower and the player matic flipping method did not feel accurate and was avatar passed through the reticle position, he felt that sometimes frustrating. might cause frustrating situations. The situations be- ing moments where the player is trying to aim at something, but having the player avatar obscure it due to a transition.

Edward was the only participant who did not like ei- 4.4 General Results ther method. He felt that the situations where he needed a camera flipping method were limited and While performing the user studies some participants far in-between, and that having a flipping method gave suggestions and feedback for improving the cam- when unnecessary simply complicates the game. He era system. George for example, stated that instead could not see himself using manual camera flipping of a camera flipping method one could incorporate often, and he stated that for either method it was a zoom out method, that moves the camera farther very important that the method did not feel in the away from the player avatar to let the player get a way. Despite this he answered the survey questions in better understanding of the surroundings. a way that stated that the flipping methods improved his observation of the scene and that there was some David stated that he preferred for the player avatar minor convenience to having them. He also expressed to be out of the way as much as possible, and Frank through the survey that the flipping methods felt dis- stated that it could be beneficial to push the player ruptive when it came to aiming. avatar even farther to the side of the screen when peeking around corners. George however stated that Frank expressed that he felt manual flipping was the farther you push the player avatar to one side of more responsive, and that the automatic flip should the screen, the less the player knows about what is on have a faster transition to make it more responsive. that side of the player avatar (this is logical as there He also felt that instead of just flipping from one side is less screen space on a side of the player avatar if it to the other, automatic flipping could be improved is placed on one side of the screen). by having it re-frame dynamically depending on the situation, and not just move to a set point. The camera methods did not show a decrease in sys- tem performance, and were deemed to be suitable in George felt there was only a minor convenience to- real-time applications. The algorithms for framing wards having a flipping method at all, and though and offset, while containing some heavier operations, he answered that both methods were just as prefer- do not contain any loops that can vary in size and able he showed a slight preference towards the manual are therefore efficient since they are only performed flipping method in the survey. However he felt that once per frame in the game. The algorithm for auto- he would seldom use the manual flipping method in a matic flipping technically includes checking ray in- game. He felt that the manual flip transition could be tersections with in-scene geometry, and are there- quicker, and that at certain times the automatic flip fore slightly more performance heavy, however, since felt too slow and at other times too fast; note that this there are few checks being done, the system did not opinion was stated after the participant witnessed a show a noticeable decrease in system performance.

19 5 Discussion the player avatar be placed so that it does not ob- struct the view of the scene and also so that it takes up less screen space than what was tested in this the- In this section the results of this study will be dis- sis. cussed in correlation to the research questions pre- sented. Further, certain topics will be discussed that Based on the user feedback, the view of the scene are not related to the specific research questions, how- is more important than view of the player avatar. ever are related to the problem statement. Therefore it could be important to prioritize cine- matographic rules that use the scene as the main subject rather than the player avatar, as it was in 5.1 Framing this thesis.

From a visual standpoint, only one participant pre- ferred the phi-grid framing, and one participant 5.2 Offset showed no preference. The remaining five partici- pants showed preference towards the rule of thirds While the difference between ellipsoidal and spherical framing, visually. This preference could be influ- offset was perceived as mild or not at all by most enced by the participant’s experience using the dif- participants, there is a trend in the results showing ferent camera systems, however more detail can be a preference for the ellipsoidal method. Despite this found when analysing what was said during the user preference, both methods were generally deemed by studies. the participants of the study to be acceptable.

All participants, in one way or another, justified their The preference for ellipsoidal offset for Dave and preferences by saying that the player avatar was more George was that it, in certain situations, gave a larger “out of the way” with their preferred method, regard- offset to the player avatar, allowing for the user to see less of whether it was the rule of thirds framing or more of the scene. This relates back to what was dis- the phi-grid framing. George also stated that what cussed regarding framing, that the user’s perception he liked about the rule of thirds framing was that the of the scene was deemed more important than the player avatar was more out of the way, however that view of the player avatar. the player avatar was also too high up in screen space. Brian, David, Frank and George all maintained that Brian, Charles and Edward all showed a preference the player should be lower down in screen space. This to ellipsoidal offset stating that it was easier to aim, preference could be because player’s felt that seeing felt more responsive and felt more stable, respec- the scene and area around the player avatar is more tively. The stability mentioned could be due to the important than seeing the player avatar. fact that the camera keeps a larger offset when aimed at the player avatar’s shoulders, which could be per- It seems that in general there is a preference for rule ceived as maintaining a more stable distance to the of thirds framing over phi-grid framing from both a player avatar. The opinions on aim and responsive- visual perspective and a quality of experience per- ness could be based in the fact that inputs that would spective, the main reason being that the player avatar move the camera have a smaller affect when the cam- is obstructing a less important part of the screen. De- era is aimed forward, due to the ellipsoid stretching spite this, it seems that the phi-grid is still a possible at that point. solution as there were participants who preferred it, even if their reasoning was mainly because the player Frank, who went on to change his mind and state avatar was lower in screen space. Based on the com- a preference for ellipsoidal offset noticed that when ments, future framing implementations should have aiming close to the player avatar’s shoulders that the

20 aiming felt more sensitive and difficult. This, simi- One thing that Anthony discussed was that the flip- larly to why the input is less sensitive close to the ping method gave a sense of security to the player. front of the player avatar, is most likely because of Anthony stated that the flip made it look like the the shape of the ellipsoid, and the zeniths of it being player avatar was more protected behind a wall. This oriented around the player avatar’s shoulders. could be important, especially given the manual flip as then the player could have an extra level of agency Visually, the preference for ellipsoidal offset was less on their own perceived security in game. This means clear, as only two participants preferred it visually, that the player might appreciate being able to flip the while all other participants remained neutral. This is camera at will to give themselves a sense of security most likely due to the fact that most participants only when hiding behind walls. noticed a minor difference between the two methods. Brian, David, Frank and George all stated that they As mentioned earlier, it is believed that the industry felt the automatic flip transition could have been standard is to use a spherical offset method. This faster. The speed of the automatic flip was chosen to means that the findings here, showing a clear pref- be slow so as to avoid it feeling jarring to the player, erence for ellipsoidal offset, goes against this indus- however player’s reacted to it negatively, feeling it try standard. This could be due to many factors, was not as responsive. George stated that he felt the however it is possible that the industry standard is manual flip could have been quicker as well. spherical offset due to simplicity. Charles and David both stated that while manual be- haves according to the player’s control, an automatic flipping solution always runs the possibility of assum- 5.3 Flipping ing incorrectly which side the player wants the cam- era on. Anthony, Frank and George expressed times when the automatic flip did not give a satisfactory The results showed that none of the participants had framing during the test, which further strengthens a preference towards automatic camera flipping, how- Charles’s and David’s argument for manual flipping. ever most participants thought both solutions im- proved the experience; the exception being Edward, Five of the participants thought a flipping method and to a much smaller degree George. Despite this improved the experience, however two did not. only four out of the seven participants showed a pref- George, one of the two who did not feel a flipping erence towards manual flipping. method was necessary, said there may be other meth- ods that are better suited for seeing the scene around Four participants stated that the suitable method for the player avatar, such as a temporary zoom out. flipping depends on what kind of game was being made. They stated that an automatic solution might work better in a cinematic game or a game that is more geared towards inexperienced players, while a manual flipping method would suit more experienced players. This could be due to more experienced play- 5.4 General Discussion ers having preference towards being in control of the player avatar and camera as much as possible. Most experienced players are able to plan and multitask, This thesis work has only a loose connection to sus- controlling things like camera and movement simulta- tainability, if any at all, in the sense that it con- neously, whereas more inexperienced players are usu- tributes to game development. Games have been ally not as capable. This has been informally shown shown to have cognitive, emotional and social ben- in [Razbuten, 2019]. efits [Fleming et al., 2017].

21 The contributions of this thesis rather lie in the game Users also chose their own method of remotely doing industry and real time computer graphics. By eval- the study, some participants were able to access the uating and identifying desired camera behavior for a demo on their work computers, while others used re- third person game using principles of cinematogra- mote controlling options. The choice of using mouse phy, this thesis’ findings help improve game develop- and keyboard as their main method of input, or a ment and interactive graphical applications from an gamepad as the main method of input, was up to the aesthetic stand-point. participant. The fact that repeating patterns could be shown despite this means that the results are de- fensible and can still be seen as valid. The repeating 5.5 Critique of Methodology patterns in the results are not bound to whether the participant used keyboard and mouse or gamepad ei- ther, further strengthening the results. This section is dedicated towards discussing critique of the methodology chosen in this thesis. To limit the length of the user study, to try and get more volunteering participants, only two methods per The participants of the user study came from a lim- category of framing, offset and flipping were tested. ited pool of people. All of the participants were men These categories were deemed to be the most impor- and were in some way affiliated with Avalanche Stu- tant, and were therefore prioritized, however based dios. Further, there were seven participants, which on the results, it seems that some alternative meth- was less than originally intended, this was due to the ods could have been worth exploring. outbreak of COVID-19 in the spring of 2020, which forced the study to be reworked to function remotely Further, the parameters for each method were chosen instead of in person. based on cinematography principles as well as a pro- fessional designer’s recommendations. This lead to Because of this limited demography of users, a con- similar looking parameters for each category, mak- clusion regarding the general public cannot be made. ing the difference between the framing methods and However, the participants were seen as experts, so the difference between offset methods seem more sub- their feedback and opinions can be regarded as valu- tle than if, for example, the player avatar had been able, and in some sense extendable to the general farther from the camera. Since several participants public. commented on the player avatar taking up too much screen space and being close to the camera, as well as The fact that the user study was performed remotely having difficulty noticing a difference between meth- could also have affected results, since there was more ods, it is likely that this affected the results. lag than usual involved. Observing the participant also became more difficult as this had to be done The flipping transition for manual and automatic flip- through a screen sharing application that, during cer- ping were different speeds, so as to make the flip feel tain of the participant’s tests, significantly dropped responsive when it was manual, and not too jarring in frame rate. when automatic. This could have affected the partici- pants ability to accurately compare the two methods, Multiple iterations of the camera system design were and the results showed a few participants discontent not performed due to the outbreak of COVID-19 lim- with the transition’s speed. iting the ability to continue development, and to do repeated user studies. There were some bugs in the implemented system that could have affected results, though the partici- Despite having to perform the thesis work under pants were made aware of them before the test and these conditions, the results showed repeating pat- were asked to ignore them. There were two bugs terns and can have conclusions drawn from them.

22 noted, one was that the manual flipping button could An area that could further be looked into is that sometimes not respond when clicked, meaning no flip of framing. Many participants discussed the player would occur. The other bug, relating to automatic avatar being too high up in screen space, therefore flipping, was that if too many flips were done in a moving the player avatar down to the lower crash short amount of time the camera would teleport to points or the center line could be explored. One could where it was supposed to be, instead of transitioning also explore the area of moving the aiming reticle smoothly. down in screen space, and by doing so moving the horizon line down. This goes against the findings of It is possible that the test scenarios did not let the [Svobodova et al., 2014] that states that the horizon user experience the camera methods to a full extent. should be on the center line or the upper rule of thirds Many participants either repeated the test before dis- line, despite this the feedback from users showed that cussing the methods, or moved about freely in the this may be an area worth researching. scene before forming an opinion. Specifically, sce- nario two, the scenario regarding the offset method, One could also investigate different distances between did not include moving the player avatar around, the player avatar and the camera. Many participants which could have lead to the method being more no- felt the player avatar was too close. Testing different ticeable. amounts of offset could therefore be useful. Different shapes of the ellipsoid offset method could also be When the participants were filling out the survey explored, by making the zeniths more extreme, for it was common for the participants to misinterpret example. One could also attempt to lower the sensi- which camera method was which. The participants tivity of input around the zeniths of the ellipsoid to sometimes noticed that they made a mistake and make it easier for players to aim around the shoulders went back to change their answers. This could have of the player avatar. been because the methods were referred to as the first method and the second method in the survey, Since it was found that the view of the scene is more while in the demo they were labeled as either “true” important than the view of the player avatar, re- or “false”. This could have affected the results of the searching cinematography principles regarding land- survey, however, since the users also got to motivate scapes and scenes could be explored, giving less em- their answers verbally this could be checked and no phasis on cinematography where the character is the conflicting statement regarding this was found. main subject of the shot.

Another possible factor that could have affected the Regarding camera flipping there are some varying ar- results is the placebo effect. This could have given eas of future research. Investigating differing transi- the user a false sense of seeing a difference between tion speeds, alternate automatic flipping algorithms methods, or given a false understanding of how the and even hybrid flipping where an automated system different camera methods affected the camera, ulti- can be overridden by a player could be areas of inter- mately skewing the results. No action was taken to est. One could also potentially research alternatives prevent the affects of placebo. to flipping in general, such as a zoom out function as suggested by George.

5.6 Future Work It seems, based on the feedback, that players have different needs and preferences, therefore including options to switch between different camera methods There are many opportunities for future work relating may be the best solution in a commercial game. to the work done in this thesis. The suggestions here are based on the feedback from the user studies as well as the results discussed.

23 This study was a qualitative study focusing mainly on Acknowlegements user feedback and statements, however a quantitative study showing a statistical significance of preference could be attempted to further strengthen the findings I would like to thank: Bj¨ornThuresson, Joacim Jon- of this thesis. sson, Fredrik L¨onn,Wade Konowalchuk, the anony- mous participants of the study, KTH University, Avalanche Studios and Jonas Iacobi, for their sup- port, guidance, help and hospitality during the course of this thesis work. 6 Conclusions

This study was a qualitative exploration of cine- References matography principles being practically implemented Md Akhtaruzzaman and Amir A Shafie. Geometri- and compared in a third person computer game. cal substantiation of phi, the golden ratio and the Regarding framing, both the rule of thirds and the baroque of nature, architecture, design and engi- phi-grid framing had benefits, however there was a neering. International Journal of Arts, 1(1):1–22, trend towards preferring the rule of thirds, with the 2011. reason being that the player avatar was obstructing a Daniel Arijon. Grammar of the Film Language. less important part of the screen. It was found that Silman-James Press, 1976. the player’s view of the scene was deemed as more important than the view of the player avatar. Using William H Bares, Jo¨el P Gr´egoire, and James C classic cinematographic principles for distance from Lester. Realtime constraint-based cinematography the player avatar could make the player avatar appear for complex interactive 3d worlds. In AAAI/IAAI, too close. pages 1101–1106, 1998.

It was found that while users did not see a strong dif- Robert Berdan. Composition and the elements of ference between spherical and ellipsoidal offset, that visual design. Photo composition articles, 2004. ellipsoidal offset was mostly preferred, due to a sense Paolo Burelli. Game cinematography: From camera of responsiveness and stability. However certain par- control to player emotions. In Emotion in Games, ticipants preferred the ellipsoidal offset because it pages 181–195. Springer, 2016. gave a larger distance from the player, which should have been regarded as unrelated to the offset method. David B Christianson, Sean E Anderson, Li-wei He, David H Salesin, Daniel S Weld, and Michael F Co- A manual flipping method was found to be preferable hen. Declarative camera control for automatic cin- to an automatic flipping method, as an automatic ematography. In AAAI/IAAI, Vol. 1, pages 148– flipping method can incorrectly assume the desire of 155, 1996. the player, and a manual flipping method gives more control to the player. A flipping method in general Marc Christie, Patrick Olivier, and Jean-Marie Nor- was deemed useful by some players, however some mand. Camera control in computer graphics. players did not believe a flipping method to be ben- In Computer Graphics Forum, volume 27, pages eficial. 2197–2218. Wiley Online Library, 2008.

Further research in game cinematography should be Steven M Drucker and David Zeltzer. Intelligent cam- performed to find more suitable camera methods in era control in a virtual environment. In Graphics the ever growing industry. Interface, pages 190–190. Citeseer, 1994.

24 Don Fairservice. Film editing: history, theory and Thomas Oskam, Robert W Sumner, Nils Thuerey, practice: looking at the invisible. Manchester Uni- and Markus Gross. Visibility transition planning versity Press, 2019. for dynamic camera control. In Proceedings of the 2009 ACM SIGGRAPH/Eurographics Symposium Theresa M Fleming, Lynda Bavin, Karolina Stasiak, on Computer Animation, pages 55–65, 2009. Eve Hermansson-Webb, Sally N Merry, Colleen Cheek, Mathijs Lucassen, Ho Ming Lau, Britta Nicholas T Proferes. Film Directing Fundamentals: Pollmuller, and Sarah Hetrick. Serious games and see your film before shooting. Routledge, 2017. gamification for mental health: current status and promising directions. Frontiers in psychiatry, 7: Razbuten. What games are like for someone who 215, 2017. doesn’t play games. https://www.youtube.com/ watch?v=ax7f3JZJHSw, 2019. Accessed: 2020-05- FromSoftware. Sekiro: Shadows die twice. [Computer 07. Game], , 2019. SIE Santa Monica Studio. God of war. [Computer Bruce Gooch, Erik Reinhard, Chris Moulding, and Game], Sony Interactive Entertainment, 2018. Peter Shirley. Artistic composition for image cre- ation. In Rendering Techniques 2001, pages 83–88. Kamila Svobodova, Petr Sklenicka, Kristina Mol- Springer, 2001. narova, and Jiri Vojar. Does the composition of landscape photographs affect visual preferences? Nicolas Halper, Ralf Helbing, and Thomas the rule of the golden section and the position of Strothotte. A camera engine for computer the horizon. Journal of Environmental Psychology, games: Managing the trade-off between constraint 38:143–152, 2014. satisfaction and frame coherence. In Computer Graphics Forum, volume 20, pages 174–183. Wiley Colin Ware and Steven Osborne. Exploration and Online Library, 2001. virtual camera control in virtual three dimensional environments. In Proceedings of the 1990 sympo- Timothy Heiderich. Cinematography techniques: sium on Interactive 3D graphics, pages 175–183, The different types of shots in film. Ontario Mining 1990. Assosiation., 2018. Luc Watrin, Mattis Geiger, Maik Spengler, and id Software. Doom eternal. [Computer Game], Oliver Wilhelm. Forced-choice versus likert re- , 2020. sponses on an occupational big five questionnaire. Diana Metiu et al. The golden ratio-from the oldest Journal of Individual Differences, pages 1–16, 03 to the newest arts. Cinematographic Art & Docu- 2019. doi: 10.1027/1614-0001/a000285. mentation, pages 52–56, 2014. Daniel PO Wiedemann, Peter J Passmore, and Mag- Naughty Dog. The last of us. [Computer Game], nus Moar. Virtual reality 3rd person camera be- Sony Computer Entertainment, 2013. havior modes. GAME-ON’2016, 2016. Giuseppe Nelva. God of war ps4 pro resolution vs Georgios N Yannakakis, H´ectorP Mart´ınez,and Ar- performance mode screenshot comparison: Simply nav Jhala. Towards affective camera control in beautiful. https://www.dualshockers.com/god- games. User Modeling and User-Adapted Interac- war-ps4-pro-resolution-vs-performance- tion, 20(4):313–340, 2010. mode-screenshot-comparison/, 2018. Accessed: 2020-04-16. Nintendo EPD. The legend of zelda: Breath of the wild. [Computer Game], Nintendo, 2017.

25 TRITA -EECS-EX-2020:182

www.kth.se