Knowledge-Based System Framework for Training Long Jump Athletes Using Action Recognition

Journal of Advances in Information Technology Vol. 6, No. 4, November 2015

T. Kamnardsiria 1, W. Janchaia 1, P. Khuwuthyakorna 1, P. Suwansrikhama 1, and J. Klaphajoneb 2 1 College of Arts, Media and Technology, Chiang Mai University 2 Department of Rehabilitation Medicine, Faculty of Medicine, Chiang Mai University Email: {teerawat.k, worawit.j, pattaraporn.k, parinya.sg, jakkrit.k}@cmu.ac.th

P. Suriyachanc Faculty of Education, Chiang Mai Rajabath University [email protected]

Abstract—The long jump is a part of track and field event. The achievement of long jumping is attempt to jump as Also, it has been a standard event in modern Olympic Games. far as possible. The long jump which comprise 4 phases Athletes have to use their strength, skills and effort to make are the approach run phase, the take-off on the wood board distance as far as possible from a jumping point. The long jump consists of 4 phases: run phase, take-off phase, flight phase, the flight phase and the landing phase. Furthermore, phase, and landing phase. The actions in each phase effect to in order to make the flight distance, the biomechanical the flight distance. If athletes perform right actions in each parameters of the long jump have been determined by phase, their performance will be significantly improved Take-off speed, Take-off angle, Take-off height and prior. In order to have right actions, they need coaching Aerodynamics [2]. Moreover, Linthorne (2008) addresses from experts. However, due to the lack of experts in the field, that biomechanical principles of the long jump are the coaching the right actions con not be proceeded widely. In this paper, we propose a framework of an expert system for run-up, take-off, flight, and landing phases. These training long jump athletes by combing computer vision principles are behind the successful for the long distance techniques and knowledge management theory. The expert long jumps [3]. Therefore, long jump performance is system will capture and learn the right actions of the long considerable for constructing the long jump distance. jump experts in each phase. Then it will be able to analyse Performance of a long jumping is depended on not only and coach learner/jumper based on knowledge captured  a fast horizontal velocity at the end of the run-up but also from the experts. take-off technique and landing in the sand pitch. These

Index Terms—action recognition, expert system, long jump, phases of the long jump are very importance and effect to computer vision, suggestion system, knowledge-based system the distance of the long jump, especially, the run-up phase which is the first phase of the long jump. Hey (1985) states that in order to make a fast run up, most long jumps I. INTRODUCTION athletes use 16-24 running strides (around 35-55 m.) The athlete reached about 95-99 per cent of their maximum The long jump was a section of the first Olympics in sprinting speed at the end of the run-up [4]. Besides, many ancient Greece, circa 708 B.C., It had first become in the long jump athletes use a check mark at 4-6 strides before Olympic sport since 1896. It has been included in track taking-off on the wood board. Therefore, their coach can and field of the athletic sports on the local, national and easily monitor the accumulated error and also solve run-up also international levels. Long jumpers generally practice in the first phase of the long jump. The take-off phase is following the long jump principle techniques such as also one of important phase for the long jump performance. approach running, taking-off, flighting and landing to The long jump athletes have to use a suitable take-off achieve in maximum performance. Until, the world record technique. For instance, the take-off leg angle at about of the long jump distance for man stands at 8.95 m. The 60-65 degree to the horizontal [5], [6]. While, the flight record has been broken by Mike Powell of the United and landing phases are a difficult action on the air. States. The world record set held in Tokyo (1991). Because, after the take-off the athlete is in free flight, so Furthermore, since 1988 the women record was held by the all angular momentum have to be considered, for Galina Chistyakova of the Soviet at 7.52 m [1]. example, the forward angular momentum (circling the arms and legs forward). Finally, the end of the flight phase  Manuscript received April 7, 2015; revised August 2, 2015. is the landing on the sand pitch. Long jumpers have to prepare their bodies for the landing by lifting their legs up

and extending them in front of the body [4]. computer vision techniques in order to help for analysing According to the long jump techniques, the existing human movement and also improving athlete skills. coaching methods of long jump are monitoring during the The review of literatures above has shown feasibility to jump, and then suggesting the right actions to the long apply action recognition techniques to long jump jumpers or as a slow motion video which is a popular biomechanics training jumpers. With respect to construct method for analysing motion. Nonetheless, they have to the high performance smart trainer of the long jump. In rely on many professional coaches for suggestion this paper of losing knowledge skills, we introduce a new techniques and also to correct the wrong actions of the framework of Knowledge-Based System for long jump long jump. For example, the optimisation performance of athlete by using computer vision techniques. The goal of the long jump such as the approach run [4],[5], the take-off this work is to support the coaching of the long jump [5], [6],[7],[8],[9], the flight and landing [10], [4], [11]. athletes and assist for improving their long jump These research have been studied in the sports science techniques. Also, the system is able to transfer the long research in order to improve the long jump techniques for jump knowledge to the new generation athletes. the athlete’s long jump. Therefore, the knowledge of long The major contributions of this article are the following: jump techniques are held in professional coaches and long (i) we propose a novel framework of knowledge-based jumpers. Thus, problems in transferring the skilled trainer for improving skills of the long jump athletes; (ii) knowledge from professional coaches to the new we present a system design of the application of the generation long jumpers are to be considered. In order to computer vision techniques for analysing the make robust suggestion techniques. Computer science biomechanics of the long jump. (iii) we suggest the technology is an alternative way for helping coaches and potential of the integrating system design for athletes to practice long jump techniques correctly. recommendation athletes in order to prove their skills with Nowadays, the development of digital world with the knowledge management systems. wide spaced digital equipment enable the use of The organization of this paper are the background and technology to produce huge amount of image and merges relate works presented in Section 2. In the flowing, the the technology into many automatic applications. In the conceptual frame work of the knowledge-based trainer for computer science research, computer vision techniques long jump athletes is described first in the Section 3. In the have been employed computer vision techniques for Section 4 Knowledge-Based system is designed. Section 5 analysing the human movement, called the action the preliminary experiment and results are showed at the recognition. The action recognition aims to automatically end of the paper. Finally, Section 6 the conclusion as well analyse and interpret continuity of human movements that as the future work are determined. made by a human agent during doing of a task into actions. The typical action recognition system consists of the II. BACKGROUND AND RELATED WORK feature extraction, the action learning, the action A. Long Jump Biomechanics classification, and the action segmentation [12]. For instance, in Thomas (2001) the long jump technique in the Long jump principle has been changed over time to the transition from the approach run to the take-off using modern athletics in the mid-nineteenth century. It consists of sprinting on the runway, taking-off on the wooden Time-Continuous kinematic data has been analysed. They board, flighting through the air and then landing in a sand classified the complex movement patterns by using the pit. An excellence long jumper must be a fast sprinter, single linkage algorithm. The single linkage algorithm is strong legs, a good complex take-off, flight and landing. based on minimising the distance between two objects About 6.5-7.5 m is the best female long jumper distances, [13]. Besides, Niebles (2010) presents the framework for while the best male distances could be reached around recognition complex human activities in the video by 8.0-9.0 m. [3]. using temporal model [14]. Hsu and his colleaques (2006) studied an automatically detect the motion during a standing long jump. First of all the silhouette of the jumper from video sequence were segmented from the background for all frames. Then a GA-based search was used to find the stick model. The stick model points out the essential joints of the person [15]. In addition, Costas et al. (2006), they studied an automatic human detection, tracking and action recognition based on real and dynamic environments of athlete finding. These sports such as pole vault, high jump, long jump, and triple jump were tested in this study [16]. While, in Robin et al. (2012), the use of image processing for measuring swimming biomechanics Figure 1. Biomechanics of the long jump. is demonstrated. They analysed techniques to enhance images for improving clarity and automating detection of Fig. 1 shows the long jump biomechanics that consists both arms and legs segments [17]. These research used of the approach run phase, the take-off phase, flight and

landing phase, and the fight distance can be derived from depend on the angular momentum of they take during all phases of the long jump. take-off as well as the available time on the air before 1) Approach run phase landing on the sand pitch. Many coaches recommend This phase is an important aspect of the long jumping the ’hang’ technique for the novice athletes because they that lead to a good performance by a fast and accurate produce a lower angular momentum and also have less run-up. The tasks of the athlete during run-up compose of available time on the air. Whereas, the ’hitch kick’ the acceleration to near-maximum speed, lower the body technique is appropriate for the better athlete. Long jump during almost final step and bring to take-off and take-off styles comprise hang style, stride jump and hitch kick are foot on the take-off board accurately. shown in Fig 2 [3]. Run-up velocity: Most long jumpers take about 16-24 4) Flight distance equation percent for striding over a distance of about 35-55 m. Fig 3 shows the jump distance that can be calculated Then the athlete reaches around 95-99 percent of their by the summation of the take-off distance, the flight maximum sprinting speed. Long jumpers avoid run-up all distance, and the landing distance [20]. 100 percent of sprinting speed [18]. Equation (1) is the total of jump distance. Wakai Run-up accuracy: The long jump run-up consist of two (2005) states that 90 percent of the total jump distance is phases which are the acceleration phase and the the flight distance. Hence, the biomechanical factors of zeroing-in phase. During the last few strides before the long jumpers flight distance are essential in long take-off on the board, the athletes are adjusting the length jumping. The effects of gravity are much more than of their strides in order to check how far they are from the aerodynamic forces during the flight phase of the long board. About five strides before the board have been used jumping. Therefore, the jumper may be determined as a for high performance long jumpers. In addition to that, projectile in feet [20]. The trajectory of the COM is long jumpers are able to adjust the stride with a small loss defined by the conditions at take-off as well as the flight of horizontal velocity. Novice athletes tend to have distance is given by mistaken their stride adjustment and high accumulated error later than highly skilled jumpers. A check mark 4-6 d jump  dtakeoff  dlanding (1) strides has been used for many long jumpers before the board [19]. Run-up to Take-off: Hay and Nohara (1990) states that skilled long jumper prepared about two to three strides before take-off with lower their centre of mass (COM). A low position into take-off is vital to provide a large vertical range of motion (ROM). Most long jumpers spend more time for practicing to lower their COM while, minimizing any reduction in run up velocity [19]. 2) Take-off phase In order to perform an appropriate take-off technique at the end of run-up and take-off foot well ahead of COM at touchdown to produce low position at the beginning of the take-off. Later a body of jumper pivots up and over the take-off foot, in at the same time the take-off leg quickly flexes as well as extends. Bridgett and Linthrne, (2006) states that a projectile event is principally for long jumping [5]. Additional, the long jumper desires to maximise the flight distance of the human projectile by optimumising take-off velocity and take-off angle. In 2000, Seyfarth and his colleagues revealed that in the long jumping, the optimum take-off technique will run up as fast as possible [6]. On the other hand, planting the Figure 2. Long jump style: (a) Hang style, (b) Stride jump, and (c) take-off leg should be approximately 60-65 to the Hitch kick. horizontal [5].

3) Flight and landing phase The majority of long jumpers employ either ’hang’ position or execute a ’hitch-kick’ movement. The athlete actions are considered to control the forward rotation. It is informed by the body at take-off and also to reach an effective landing position [10]. Linthorne (2008) describes the athlete is free flight in the air, so the total angular momentum of athletes must be preserved. The forward angular momentum is produced by swinging the Figure 3. Diagram of the total jump distance. arms and legs. The long jumpers select their technique

And components of fitness consist of speed, power, balance,

1 flexibility, motion, and force. These components are 2   v sin 2   2gh  2  considerable factors in order to improve their long jump d flight  1 1  2g   v2 sin 2    skill. On the other hand, the long jump distance will be   increased. Each of the fitness components are described below: where v is the take-off velocity,  is the take-off angle, and Speed: Athlete sprints 50 metres, an excellent time is g is the acceleration due to gravity. h is the relative take-off height, which is given by: under 7.6 seconds for a male and under 8.1 seconds for a female. The sprint test is an essential component to execute the suitable jump at high speed. A good indicator h  htakeoff  hlanding of speed is ten stride test to monitor athletes’ ability and where htake-off is the take-off height and hlanding is the landing construct efficiently accelerate from a standing start height, the range of a projectile launched from the ground point. Running 6x20 m from a standing start point that level over a horizontal plane can be derived from works for the individual measures the time and distance covered in ten strides. v2 sin 2 d  Power: The standing long jump or vertical long jump flight g is the measurement of power. The outstanding result is over 2.5 metres for males, and over 2 metres for females from standing long jump test, while excellent vertical B. Long Jump Analysis jump test, is over 65 cm for males and over 58 cm for Long jump can be distinguished into two parts, females. Power is also important in the long jump. Hence, sprinting and jumping. The first part is the sprinting the competitors require dedicating a maximum power in which composes of two phases, the driving phases when the rapid time. the leg is in touching the ground and the recovery phases. Balance: The stork stand test is used for testing that Another part is the jumping movement of the long jump, excellent reading for both males and females is over 50 which very similar to sprinting and involves the hip knee seconds. Balance is extremely importance in the flight and ankle [21]. stage of the long jump. It aims to gaining the longest 1) The optimisation performance of physical fitness distance possibly from takeoff to landing by smoothly The long jump can be separated into four phases: the glide and in one ahead direction in the air. approach running, taking-off, flighting and landing. Each Muscular Strength: Leg Strength Test in which area of of these requires many fitness components. First of all, 25 metres is the measurement test. It is constructed by approach running is essential in order to execute a good cones in a straight line. The steps start jogging ten metres, jump that requires the speed fitness component. In this and then proceed to hop on the dominant leg until it phase it has been studied and found that the most reaches the end of the cone. influential factors of the jump distance is approach Flexibility: Sit and reach test is a measurement that running. The second phase, take-off with power, needs to excellent reading for males would be over 10 cm, and 15 be explosive and fast. The third phase, flight phase, cm for females. Flexibility is vital for helping athletes to requires the balance due to the good balance affect to protect their joints, connective tissue and muscle. flight actions in order to perform appropriate landing. Additionally, it is very importance in assisting the The last phase, landing, is very vital for avoiding athletes athletes to reach maximum speed, and power levels in the from muscle tissue and joints injury. Muscular strength long jump. Motion: In order to complete an excellent and flexibility are also essential. They help protecting the jump, the approach run is vital to construct the maximum body of athletes to control and change the direction speed that influences to the jump. Moreover, momentum shows in Fig 4 [21]. is also important especially angular momentum such as on arm and leg rotation. In addition, take-off and flight stages of the jump which are parts of momentum, are very necessary for a high jump as well as carry forward. The higher jump makes athletes stay longer in the air, the longer momentum can carry forward. The arms and legs in flight are also significant factors by swinging forward, as it generates further momentum. Force: is importance in the take-off stage of the long jump. Take-off on the board with the athlete body can be applied force in order to push off it and increase the height and length of jump. Moreover, the force can be Figure 4. Optimization performance of physical fitness in the long absorbed by bending knees, ankles and hips when the jump. body land [21]. 2) Methods of assessing the components of fitness C. Action Recognition Tanya (2010) addresses the methods for improving In the computer vision, an action recognition is become skill, as well as the performance of the long jump. The a very importance research topic including motion

recognition [22], [23], facial recognition [24], and training sequences that use to learn models. Finally, action movement behavior recognition [25]. classification is a reasonable comparison of dataset with a Deniel et al. (2011) addresses the action recognition different amount of training and testing, for instance; which is a sequence of human movements that constructed different subjects such as size, shape, speed, and style is by a human agent during the doing of a task. The task of shown in Fig. 5 [12]. Moreover, Ke et al. (2013) action recognition will be named actions, for instance, demonstrates human activity recognition on video data. define the action label for describing an action case when There are three aspects for human activity recognition acted by unlike agents under unlike perspective. The such as core technology, human activity recognition action recognition system will be done by sending system, and applications from low level to high level instructions to the actor then using simple action verb representation [26]. On the other hand, the Fig. 6 shows imperative and comparing with the recognized action typology of video-based human activity recognition names. It comprises the feature extraction, the action system. learning, the action classification, and the action segmentation. First of all, feature extraction is extracting postures and motions from video to distinguishable features between human actions. Secondly, action learning is the learning statistical model after features extraction is be done. This model is used to classify a new feature observation. Thirdly, action segmentation is cutting motion streams into a single action for example, initial Figure 5. Data flow of the long jump action recognition system.

Figure 6. Typology of video-based human activity recognition system.

questions (queries) submitted through a computer. D. Expert System Generally, expert systems consist of three parts. The first, The Oxford dictionaries (2010) give the meaning of the is the knowledge-based component which contains the expert system as “A piece of software which uses information of experts and as well as logical rules. The databases of expert knowledge to offer advice or make second part, is an inference engine that is an interpreting of decisions in such areas as medical diagnosis” [27]. the submitted problem against the rules and logic with the Besides, business dictionary (2015) describes that the knowledge-based. Finally, the last part is an interface that knowledge of an expert in a particular subject in an allows the user to show the problem in a human language artificial intelligence based system (expert system) is such as English language [28]. converted into a source code. This source code can be combined with other such source codes (based on the other E. Related Works experts knowledge) and also used for responding In the literature review, the human activity recognition

in the video with respect to the sports are considered to system will not lose its stability [32]. While, Robin and his study. In this section we present an overview of some colleagues used image processing for measuring related works as well as refer the readers to [26], [12], [29] swimming biomechanics. They analysed techniques to for the more completed survey. There are several enhance images for improving clarity and automating researchers have studied about analysing the movement of detection of both arms and legs segments. The Normalized the long jump. They proposed the long jump techniques in Cuts techniques were used in the segmentation. Then order to improve the performance of long jump athletes. Global Probability of Boundary was considered for the For instance, approach run techniques, take-off techniques, extracting feature. Moreover, a freestyle swimming video flight and landing techniques jump [18], [10], [5], [6], [8], footage has been analysed in this study. The results found [30], [7], [9]. that the Normalized Cut algorithm provides reasonable Furthermore, in the computer science research, some super pixel maps. And also the boundary detector can researchers employed computer vision techniques for precisely detect the body and draws an almost closed analysing the human movements. Thomas has analysed contour around swimmer [17]. On the other hand, some the long jump technique in the transition from the researchers employed the expert system in order to approach run to the take-off using Time-Continuous construct the application for keeping knowledge from kinematic data, in 2001. They classified the complex experts. Such as in 2011, Rueangsirarak used Motion movement patterns by using the single linkage algorithm Capture Technology to diagnose falling patterns in elderly with 57 trials (4.45-6.84 m) of time-continuous data. The people with a Knowledge-Based System in order to single linkage algorithm is based on minimising the determine serious falling risks for them and recommended distance between two objects. In the results, it allows to guidelines for medical treatment [33]. Besides, Oliveira identify structural changes of movement during a singular (2014) presents the framework of a knowledge-based and individual movement styles within the same system in order to support the creation of e-learning movement type, as well as remarkable specifically materials. It would be easily adapted for an effective concerning the flight phases [13]. On the other hand, generation of custom-made e-learning courses or Niebles (2010) presents the framework for recognition programmes [34]. In addition, Welter et al., (2011) complex human activities in the video by using a temporal developed the concept of image-based case retrieval for model. 3-D Harris corner detector was used for extracting radiological education using content-based. It is called the features [14]. In addition, the library for Support Content-based image retrieval (CBIR). This paradigm Vector Machines (LIBSVM) of Chang (2001) was incorporates a novel combination with aspects of employed. This library was used to classify the activities diagnostic learning in radiology: (i) Experiences in a with KTH Human action dataset, Weizmann action training environment combined with the radiologists database, and Olympic Sports dataset [31]. Besides, Hsu working context; (ii) The training cases is up-to date cause and his colleagues (2006) studied an automatic motion they are stored routine by electronic medical records; (iii) detection during a standing long jump. The silhouettes of Support adults learning that appropriate for the patient and the jumper from video sequence was segmented from the problem-oriented [35]. background for all frames. Pose estimation for an individual silhouette and Hue-Saturation-Value (HSV) III. THE CONCEPTUAL FRAME WORK OF space were also considered to extract the features. Then KNOWLEDGE-BASED SYSTEM FOR TRAINING LONG JUMP the GA-based search was used to find the stick model. The ATHLETES stick model points out the essential joints of the person. 20 The Fig. 7 illustrates the conceptual framework of the frames of a standing long jump video sequence have been introduced Knowledge-Based System for Training Long used for analysing data. The result shows that silhouettes Jump Athletes. The framework aims at capturing the and computer can be generated stick model of the second knowledge of the expert long jump athletes and also the and the third frames [15]. Moreover, Costas et al (2006) suggestion in order to correcting the right movement of presents an automatic human detection, tracking and novice long jumpers. Besides, it will improve their skills action recognition based on real and dynamic and performances of athletes. The system is separated into environments of athlete finding. They used the Temporal four sections as (i) Domain Experts, (ii) Action Signal for segmentation the athlete videos. Additional, the Recognition, (iii) Expert System, and (iv) Action Human Points Detection algorithm has been employed for Suggestion. the extracting features. Besides, these sports such as pole vault, high jump, long jump, and triple jump were A. Domain Experts Section classified by using Silhouette analysis algorithm. 39 video In this section, we firstly prepare for exploiting data set (12 pole vault, 9 high jump, 8 triple jumps and 10 information from long jump experts both the primary and long jumps) were used for analysing. The result shows that the secondary resources. The primary information is to be the correct classification rates were (100%) for the pole collected from experts such as trainers, coaches, and vault, (88.9 %) for the high jump, (87.5%) for the triple skilled long jumpers. Thereby, the capturing knowledge jump, and (80%) for the long jump. And also, if some either of interviews or captures their movements of long frames the silhouette estimation algorithm ran failed, the jump athletes. Besides, the secondary information is the

collected data that have been analysed in empirical domain experts, the action recognition, knowledge research for example theory of long jump [3], long jump databases and also the action suggestion. In order to analysis [36],[21], optimisation of take-off [20], approach suggest good practice to athletes, the expert system will be run research [37], [38], long jump professionals articles train for the long jumping [42]. There are 4 sequences of [39], [40], textbooks [2], trainer programs [41], etc. Then this section are firstly, constructing the recognised pattern analysis both information primary and secondary in order (training image data set) and also text suggestion from to precisely suggestion, as well as recommendation the expert (training text), secondly, generating the rule of each right movement of the long jump. segmented objects with experts suggestion, thirdly, comparing the test movement with the trained pattern by B. Action Recognition Section using similarity function, and finally, suggestion text for This section includes about action recognition each action segmentation. techniques for recognising the long jump. We will employ the computer vision technique in order to analysis the D. Action Suggestion System Section athlete image sequence while acting long jump processes. In this section, an action suggestion system is to give It is called the action recognition [12]. There are 4 steps of recommendation or suggestion of correct actions for the action recognition. First of all, object segmentation will long jump athletes. The suggestion is an idea that will be use the static camera configuration with a camera for accepted or rejected. An action is taken after a decision is capturing the phase of the long jumping of taking-off and required [43]. Some research proposed successfully using flight phase. Secondly, the long jump feature extraction is of suggestion systems such as recognition of employees to extract characteristics of the segmented objects for for improving their jobs [44], [45], [46]. Thus, we will example, shape, silhouette, colours, and motion. Thirdly, apply the suggestion system to improve long jump skills. detection and classification algorithm will be used for Our suggestion system will be provided summary results recognising the long jump movements. It is based on the that show on the monitor. Moreover, training text for correcting and suggesting the long jump, for example, the represented features. Finally, if the data is accepted by approach run, the take-off, the flight, and the landing experts, it will be used to train the knowledge-based respectively will be shown as well. In addition, the system in order to make the master pattern of the long training programme of the long jump will be provided to jump. improve their skill and also long jump distance. Including C. Expert System Section the nutrition suggestion for the long jump athletes will be provided as well. In the expert system section, we aim to integrate the

Figure 7. Conceptual framework of a knowledge-based system for training long jump athletes.

Capturing device, (2) Preprocessing, (3) Segmentation of IV. KNOWLEDGE-BASED SYSTEM DESIGN the video to image sequences, (4) Feature extraction of the Fig. 8 shows the system design of Knowledge-Based captured video, (5) Training of the captured video sets, (6) System for the long jump. The system will be separated Discriminative of the Action correlation, (7) into two parts which are the Training process and the Configuration training text suggestion to the image Testing process. sequences. Finally, keeping to Knowledge Model. The The Training process consist of 8 processes: (1) Testing process will be provided the suggestion of the long

jump action. There are 7 processes of the testing process. similarity between the training image sequences from the There are (1) Capturing device, (2) Preprocessing, (3) Knowledge Model and testing image sequences will be Segmentation of video to image sequences, (4) Feature compared, after that, the results will be reported on the extraction, (5) Testing video sets, (6) Similarity monitor and also recommended the appropriate practice of measurement, and Finally, Suggestion Actions report. The the Long Jump Biomechanics.

Figure 8. Knowledge-based system design for the long jump.

Biomechanics. V. PRELIMINARY EXPERIMENT TABLE I. AGE, HEIGHT, AND WEIGHT OF THE PARTICIPANTS A. Methodology The purpose of this preliminary experiment was to Participant Age (years) Height (m) Weight (kg) Expert (E1) 39 75 collect pre-test data and evaluate factor that significant 1.80 Novice1 (N1) 19 1.60 55 affect the long jump performance. There are two sections Novice2 (N2) 21 1.74 70 including speed performance and long jump performance. Novice3 (N3) 18 1.79 69 To do this, we collected speed, long jump distance, Novice4 (N4) 22 1.75 64 take-off angle, and long jump action. We also recoded Novice5 (N5) 19 1.66 51 action for constructing the action recognition of system. Novice6 (N6) 19 1.80 70 1) Participants and jumping protocol Novice7 (N7) 23 1.73 68 Ten male students of Chiang Mai University and a long Novice8 (N8) 20 1.82 67 jump expert are participate in the study (Table I). They are Novice9 (N9) 18 1.74 56 novice long jumpers. The participants were informed Novice10 (N10) 27 1.68 58 about the procedures and the protocol prior to their 2) Procedures participation. All jumps were completed in the long jump Data collection: Data consist of 2 sections including sand pit. The take-off was performed from the synthetic speed performance and long jump performance. First running track, flat, dry. Besides, the sand landing area was section, speed performance were collected by 50 metres leveled with the take-off surface. Each participant running. While running, time were caught every 10 metres performed one time of 50 metres speed running, and also, (1-10, 10-20, 20-30, 30-40, and 40-50). Then, velocity three attempts of long jumps at his preferred. They jumped were calculated by distance/time. The maximum speed without any knowledge information of the long jump period of each athlete were collected in order to determine

the check mark stride. 2) The long jump performance The other section, the long jump performance were Due to the fact that some participants left from this collected from the participants, such as, start to take-off study after speed performance experiment, Eight of distance, time, long jump distance, and take-off angle participants (one expert and seven novices) cooperated for were noted in the form. In addition, long jump action were the long jump pretest. Table III lists the average values of captured by a digital camera. the start distance, velocity, jump distance, and take-off Data analysis: Descriptive statistics were used for angle. calculating the average as well as standard deviation (SD) 3) The relationship between velocity and take-off of start point to the take-off board, velocity, jump distance, angle effected the long jump distance and take-off angle. Fig. 10 illustrates the scatter plot of the relationship between velocity and take-off angle that effect to the long B. Results jump distance of all participants. The result was separated 1) Speed performance into three groups by the long jump distance which are Table II shows velocity of all participants with 10m per 3m-4.5m, 4.6m-5.5m, and 5.5m-6.5m respectively. It period of 50m distance. Besides, the Fig. 9 illustrates that shows the majority of the long jump distance is around almost all velocity of the participants increase slightly to 4.6m-5.5m. Among the plots, we noticed the similar the peak at the 30-40m period, then declines modestly to values of velocity and takeoff angle of 2 long jumpers, the end point. The participant N8 is the maximum velocity where the different long jump distances are significant. that rises to 11.90 (m/s). However, it drop rapidly to 8.93 For instance, two long jumpers tried to jump, the first one (m/s) at the end point. Therefore, this speed test shows that was velocity of 6.96 (m/s) with the take-off angle of all of participants should begin running at the point 21.49 and the other one was at velocity of 6.93 (m/s) with between 30m and 40m far from the take-off board, except the take-off angle of 22.28. Both of the long jumpers have for the participant N4, because, he has the peak at the similar values of velocity and the takeoff angle. The result 40-50m period. Hence, his start point should be farther revealed that the long jump distances these two around 40-50m from take-off board. participants are considerably different or around 1.18m (6.20m and 5.02m respectively). Therefore, we need to TABLE II. VELOCITY OF ALL PARTICIPANTS FROM START TO STOP know why the long jump distances are quite different. 50M (10M PER PERIOD) What alternative factor that influence the long jump Velocity of Start to end Distance 50m. (m/s) distance are performance. Participant 1-10 10-20 20-30 30-40 40-50 Expert (E1) 7.46 10.10 9.43 10.75 8.70 Novice1 (N1) 8.06 8.26 9.35 11.63 8.90 Novice2 (N2) 8.70 8.40 9.26 10.75 8.93 Novice3 (N3) 9.52 8.00 9.35 10.64 9.43 Novice4 (N4) 8.00 9.35 8.85 9.09 10.20 Novice5 (N5) 8.55 9.52 8.77 10.31 10.00 Novice6 (N6) 6.80 8.62 9.80 9.90 8.62 Novice7 (N7) 8.47 8.62 10.00 10.75 8.55 Novice8 (N8) 7.58 9.01 10.31 11.90 8.93 Novice9 (N9) 9.71 8.93 9.26 9.80 8.85 Novice10 (N10) 7.19 7.87 9.43 10.42 8.06

Figure 10. The relationship between velocity (m/s) and Take-off angle effected to long jump distance (m) of all participants

Figure 9. The relation between velocity of all participants and distance periods of 50m (10m per period)

or trainers thereby both interviewing and also capturing long jump motions using video capturing device. Besides, action recognition is to be implemented to analysing data then collect the knowledge dataset not only training image dataset for recognising action but also training text for suggestion actions. In addition, the expert system will be developed for training long jump athletes. One challenging extension is the capability of the system to provide the suggestion, such as approach run, mark points, long jump techniques, as well as training programmes for practices long jump performance.

Figure 11. The comparison of long jumpers with similar velocity and ACKNOWLEDGMENT take-off angle (a) The expert long jumper of (E1) and (b) The novice long jumper (N1) in Table III This study is supported by the Motion-Capture Laboratory, College of Arts, Media and Technology Fig. 11 shows the different movement of long jumping (CAMT), Chiang Mai University (CMU), Chiang Mai, of two long jumpers, while velocity and take-off angle Thailand in collaboration with the Department of are the similar. Nevertheless, it would be able to make different distance depend on the long jump Rehabilitation Medicine, Faculty of Medicine, Chiang action/movement in the air. Mai University Chiang Mai, and also, Department of physical education and recreation Physical, Rajabhat TABLE III. VELOCITY OF ALL PARTICIPANTS FROM START TO STOP Chiang Mai University, Chiang Mai, Thailand. Besides, 50M (10M PER PERIOD) the author would like to conveys thanks to Student

Participan Start distance Velocity Jump distance Take-off Development Division of Chiang Mai University as well t (m) (m/s) (m) angle () as Mr. Suriyo Thirach for their the long jump activity E1 26:80 ±2:86 6:79 ± 0:13 6:04 ± 0:20 21:89 ± 0:68 supports. N1 27:50 ±0:00 6:67 ± 0:40 5:30 ± 0:27 21:68 ± 3:44 N3 27:05 ±2:68 6:95 ± 0:24 4:91 ± 0:15 14:34 ± 3:57 REFERENCES 14:20 ± 1:27 N4 27:57 ±0:05 6:18 ± 0:75 5:38 ± 0:36 [1] IAAF. (2015). International association of athletics federations. N6 29:99 ±0:02 5:65 ± 0:28 4:80 ± 0:28 16:99 ± 1:12 [Online]. Available: http://www.iaaf.org/records/toplists/jumps/long-jump/outdoor/me N8 21:93 ±0:98 6:48 ± 0:16 4:80 ± 0:27 10:88 ± 2:22 n/senior N9 27:33 ±1:15 6:85 ± 0:02 5:37 ± 0:15 14:77 ± 3:60 [2] R. Bartlett, Introduction to Sports Biomechanics: Analyzing N10 29:63 ±1:50 6:40 ± 0:29 3:96 ± 0:67 11:88 ± 2:11 Human Movement Patterns, Routledge, 2007. [3] N. P. Linthorne, “Biomechanics of the long jump,” Routledge Handbook of Biomechanics and Human Movement Science, p. 340, VI. CONCLUSION AND FUTURE WORK 2008. [4] J. G. Hay, “The biomechanics of the long jump,” Exercise and We have reviewed the literature on long jump Sport Sciences Reviews, vol. 14, pp. 401–446, 1985. biomechanics, long jump analysis, action recognition [5] L. A. Bridgett and N. P. Linthorne, “Changes in long jump take-off technique with increasing run-up speed,” Journal of Sports system as well as the expert system for the Sciences, vol. 24, no. 8, pp. 889–897, 2006. knowledge-based system. We also proposed a new [6] A. Seyfarth, R. Blickhan, and J. V. Leeuwen, “Optimum take-off framework of the knowledge-based system for training techniques and muscle design for long jump,” Journal of Experimental Biology, vol. 203, no. 4, pp. 741–750, 2000. long jump athletes. The introduced comprises Domain [7] A. Lees, P. Graham-Smith, and N. Fowler, “A biomechanical Experts, Action Recognition, Expert System, and also analysis of the last stride, touchdown, and takeoff characteristics of Action Suggestion. Furthermore, we designed the the men’s long jump,” Journal of Applied Biomechanics, vol. 10, pp. 61–61, 1994. knowledge-based system for the long jump training for [8] A. Lees, N. Fowler, and D. Derby, “A biomechanical analysis of long jump athletes. The system consist of training, testing the last stride, touch-down and take-off characteristics of the process as well as knowledge model. Moreover, we women’s long jump,” Journal of Sports Sciences, vol. 11, no. 4, pp. 303-314, 1993. conducted a preliminary experiment in order to discover [9] N. P. Linthorne, M. S. Guzman, and L. A. Bridgett, “Optimum the speed and long jump performance of 11 long jump take-off angle in the long jump,” Journal of Sports Sciences, vol. 23, athletes. The results found that speed performance is quite no. 7. similar value around 9 (m/s) at the peak of velocity. [10] J. G. Hay, “Citius, altius, longius (faster, higher, longer): The biomechanics of jumping for distance,” Journal of Biomechanics, The result reveals the long jump performance is not vol. 26, pp. 7–21, 1993. only depend on velocity and take-off factors but also the [11] “Approach strategies in the long jump,” JAB, vol. 4, no. 2, 2010. movement on the air (flight) considerably influences [12] D. Weinland, R. Ronfard, and E. Boyer, “A survey of vision-based methods for action representation, segmentation and recognition,” improve the long jump distance. There are many future Computer Vision and Image Understanding, vol. 115, no. 2, pp. directions of our work. 224-241, 2011. For our future work, we will capture knowledge from [13] T. Jaitner, L. Mendoza, and W. Sch¨ollhorn, “Analysis of the long jump technique in the transition from approach to takeoff based on primary data such long jump experts, long jump coaches

timecontinuous kinematic data,” European Journal of Sport athletes,” Journal of Sports Sciences, vol. 32, no. 2, pp. 149-156, Science, vol. 1, no. 5, pp. 1-12, 2001. 2014. [14] J. C. Niebles, C. W. Chen, and L. Fei-Fei, “Modeling temporal [38] T. A. Halgren, R. B. Murphy, R. A. Friesner, H. S. Beard, L. L. structure of decomposable motion segments for activity Frye, W. T. Pollard, and J. L. Banks, “Glide: A new approach for classification,” in Computer Vision–ECCV 2010. Springer, 2010, rapid, accurate docking and scoring. 2. enrichment factors in pp. 392–405. database screening,” Journal of Medicinal Chemistry, vol. 47, no. 7, [15] H. H. Hsu, S. W. Hsieh, W. C. Chen, C. J. Chen, and C. Y. Yang, pp. 1750-1759, 2004. “Motion analysis for the standing long jump,” in Proc. 26th IEEE [39] B. M. Rosenbaum. (2015). Mike powell offers advice and drills for International Conference on Distributed Computing Systems long jumpers. [Online]. Available: Workshops, 2006, pp. 47-47. http://trackandfield.about.com/od/longjump/p/powelljumpdrill.ht [16] C. Panagiotakis, E. Ramasso, G. Tziritas, M. Rombaut, and D. m Pellerin, “Shape-motion based athlete tracking for multilevel action [40] N. Newman. (2012). The Horizontal Jumps: Planning for Long recognition,” in Articulated Motion and Deformable Objects, Term Development. CreateSpace Independent Publishing Platform. Springer, 2006, pp. 385–394. [Online]. Available: [17] R. P. Dubois, D. V. Thiel, and D. A. James, “Using image https://books.google.co.th/books?id=RdBZLwEACAAJ processing for biomechanics measures in swimming,” Procedia [41] R. Stander. (2015). Long jump. [Online]. Available: http://www. Engineering, vol. 34, pp. 807-812, 2012. bolandathletics.com [18] J. G. Hay, J. A. Miller, and R. W. Canterna, “The techniques of elite [42] M. Negnevitsky. (2005). Artificial Intelligence: A Guide to male long jumpers,” Journal of Biomechanics, vol. 19, no. 10, pp. Intelligent Systems. Addison-Wesley. [Online]. Available: 855-866, 1986. http://books.google.co.th/books?id=1BxYQnrfv9MC [19] J. G. Hay and H. Nohara, “Techniques used by elite long jumpers in [43] J. Miller, “The suggestion system is no suggestion,” Modern preparation for takeoff,” Journal of Biomechanics, vol. 23, no. 3, pp. Management. (S. Parisa, Trans.) Tehran: Danshkar. Practicable 229-239, 1990. Ideas, Productivity in industry (Ministry of, Trans.) Tehran: sage [20] M. Wakai and N. P. Linthorne, “Optimum take-off angle in the entrepreneurs, 2003. standing long jump,” Human Movement Science, vol. 24, no. 1, pp. [44] P. Darragh-Jeromos, “A suggestion system that works for you,” 81–96, 2005. Supervision, vol. 57, pp. 6–7, 1996. [21] Tanya. (2010). Long jump analysis. [Online]. Available: [45] R. F. Bell, “Constructing an effective suggestion system-by http://pdhpecovenant.wordpress.com creating and implementing an effective employee suggestion [22] A. Erol, G. Bebis, M. Nicolescu, R. D. Boyle, and X. Twombly, system companies can tap into the reservoir of ideas and creative “Vision based hand pose estimation: A review,” Computer Vision thinking that all,” IIE Solutions, vol. 29, no. 2, pp. 22-25, 1997. and Image Understanding, vol. 108, no. 1, pp. 52–73, 2007. [46] C. Van Dijk and J. Van Den Ende, “Suggestion systems: [23] V. I. Pavlovic, R. Sharma, and T. S. Huang, “Visual interpretation Transferring employee creativity into practicable ideas,” R and D of hand gestures for human-computer interaction: A review,” IEEE Management, vol. 32, no. 5, pp. 387-395, 2002. Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 677–695, 1997. [24] W. Zhao, R. Chellappa, P. J. Phillips, and A. Rosenfeld, “Face recognition: A literature survey,” Acm Computing Surveys (CSUR), vol. 35, no. 4, pp. 399-458, 2003. [25] W. Hu, T. Tan, L. Wang, and S. Maybank, “A survey on visual Teerawat Kamnardsiri is currently pursuing surveillance of object motion and behaviors,” IEEE Transactions Ph.D. program in Knowledge Management at on Systems, Man, and Cybernetics, Part C: Applications and College of Arts, Media and Technology, Chiang Reviews, vol. 34, no. 3, pp. 334-352, 2004. Mai University, Thailand. He received his B.Sc. in [26] S. R. Ke, H. L. U. Thuc, Y. J. Lee, J. N. Hwang, J. H. Yoo, and K. H. Technical Education (Computer Technology) and Choi, “A review on video-based human activity recognition,” M.Sc. in Technical Education (Multimedia Computers, vol. 2, no. 2, pp. 88-131, 2013. Technology) at King Mongkut’s University of [27] Oxford Advanced Learner’s Dictionary of Current English, Oxford Technology North Bangkok, Thailand. He got a University Press, Incorporated, 2010. certificate of the Erasmus Mundus, E-tourism [28] Businessdictionary. (2015). Expert system. [Online]. Available: Project (Euro Asian project 2011-2014). His research areas are http://www.businessdictionary.com/definition/expert-system.html Knowledge Representation, Computer Graphics, Image Processing and [29] R. Poppe, “A survey on vision-based human action recognition,” Computer Vision. Image and Vision Computing, vol. 28, no. 6, pp. 976-990, 2010. [30] A. Lees and E. Fahmi, “Optimal drop heights for plyometric Worawit Janchai is a lecturer in the Knowledge training,” Ergonomics, vol. 37, no. 1, pp. 141–148, 1994. Management Department, College of Arts, Media [31] C. C. Chang and C. J. Lin, “Libsvm: A library for support vector and Technologies, Chiang Mai University, machines,” ACM Transactions on Intelligent Systems and Thailand. He holds PhD in Knowledge Technology, (TIST), vol. 2, no. 3, p. 27, 2011. Management from Chiang Mai University, Chiang [32] C. Cotsaces, N. Nikolaidis, and I. Pitas, “Video shot detection and Mai, Thailand in 2011, Currently, he is an condensed representation. a review,” IEEE on Signal Processing Assistant Professor in Knowledge Management Magazine, vol. 23, no. 2, pp. 28-37, 2006. and also the Head of Academic Office of College [33] W. Rueangsirarak, A. Atkins, B. Sharp, N. Chakpitak, K. of Arts, Media and Technology. His research interests are related to Meksamoot, and P. Pothongsunun, “Knowledge based system knowledge workers competency development and workplace learning framework of som and cbr techniques using motion capture especially in the learning organization context. technology in elderly falling risk for physiotherapist assessment and support,” International Journal of Research in IT Management and Engineering, vol. 1, no. 4, pp. 1-20, 2011. [34] P. O. Oliveira, et al., “A knowledge-based framework to facilitate Pattaraporn Khuwuthyakorn is a lecturer in the etraining implementation,” 2014. College of Arts, Media and Technologies, Chiang [35] P. Welter, T. M. Deserno, B. Fischer, R. W. G¨unther, and C. Mai University, Thailand. She received the B.E Spreckelsen, “Towards case-based medical learning in radiological (Electrical Engineering) from Chiang Mai decision making using content-based image retrieval,” BMC University (CMU), the M.E.M. and the M.S.E.E. Medical Informatics and Decision Making, vol. 11, no. 1, p. 68, from Queensland University of Technology (QUT) 2011. and the Ph.D. in Information Engineering from the [36] Z. Pan, “Analysis of mechanical model on factors influencing the Australian National Univesity (ANU). Her long jump result under the perfect condition,” 2013. research interests include computational method [37] F. N. Panteli, A. Theodorou, T. Pilianidis, and A. Smirniotou, development in computer vision, pattern recognition, machine leaning, “Locomotor control in the long jump approach run in young novice hyperspectral image classification and their applications.

Parinya Suwansrikham is lecturer of Software Administrative board committee, Royal College of Thai Physiatrists, Engineering at College of Arts, Media and Department of Rehabilitation Medicine, Faculty of Medicine, Chiang Technology, Chiang Mai University, Thailand. Mai University. His research interests include Rehabilitation medicine, He received his B.Eng. in Electrical Engineering Music therapy, Musculoskeletal medicine, Sports science and medicine, from Chiang Mai University and M.Eng in Electrodiagnosis, Prosthetics & Orthotics. Computer Engineering from King Mongkuts University of Technology Thonburi. His research areas are image processing, computer vision and information retrieval. Permsak Suriyachan is currently leader of department of physical education and recreation Jakkrit Klaphajone is a lecturer in the Physical, Rajabhat Chiang Mai University, Chiang Department of Rehabilitation Medicine, Faculty of Mai, Thailand. He received his B. Ed. (Physical Medicine, Chiang Mai University, Chiang Mai Education) and M. Ed. (Physical Education) at 50200, Thailand. He received the M.D. from Kasetsart University, Thailand. He has been a Faculty of Medicine, Chiang Mai University, 1993, guest speaker of the International Association of the certificate of Proficiency in PM & R from Thai Athletics Federations (IAAF). He is an expert at Medical Council, 1996 and the fellowship in physical fitness of Sport Authority of Thailand. Sports medicine and science, University of His research areas are Physical Education, physical fitness, Athlete Aberdeen, UK, 2000. He is currently positioning at performance practices.