d07wals.x.g2db Page 1 of 7 Thursday, April 4, 2002 8:47:50 AM HEAD: Understanding Scene Graphs DEK Using graph-based Bio data structures to Aaron is chairman of Mantis Development, and teaches computer graphics and In- organize and manage ternet/Web application development at Boston College. He can be contacted at scene contents [email protected]. Byline NOTE TO AUTHORS: PLEASE CHECK ALL URLs AND SPELLINGS OF NAMES CAREFULLY. Aaron E. Walsh THANK YOU. Pullquotes Before scene-graph programming Scene-graph models, we usually nodes represent represented scene objects in a scene data and behavior procedurally d07wals.x.g2db Page 2 of 7 Thursday, April 4, 2002 8:47:50 AM 2.1 cene graphs are data structures used 2.68 At the scene level, you concern your- – to organize and manage the contents – self with what’s in the scene and any as- – of hierarchically oriented scene data. – sociated behavior or interaction among – Traditionally considered a high-level – objects therein. Underlying implementa- 2.5 S – data management facility for 3D content, tion and rendering details are abstracted – scene graphs are becoming popular as – out of the scene-graph programming mod- – general-purpose mechanisms for manag- – el. In this case, you can assume that your – ing a variety of media types. MPEG-4, for 2.75 VRML browser plug-in handles low-level – instance, uses the Virtual Reality Model- – concerns. 2.10 ing Language (VRML) scene-graph pro- – – gramming model for multimedia scene – Nodes and Arcs – composition, regardless of whether 3D – As Figure 2 depicts, scene graphs consist – data is part of such content. In this arti- 2.80 of nodes (that represent objects in a scene) – cle, I’ll examine what scene graphs are, – connected by arcs (edges that define re- 2.15 what problems they address, and scene- – lationships between nodes). Together, – graph programming models supported by – nodes and arcs produce a graph structure – VRML, Extensible 3D (X3D), MPEG-4, and – that organizes a collection of objects hi- – Java 3D. 2.85 erarchically, according to their spatial po- – – sition in a scene. 2.20 Scene Composition and Management – With the exception of the topmost root – Scene graphs address problems that gen- – node (which defines the entry point into – erally arise in scene composition and – the scene graph), every node in a scene – management. Popularized by SGI Open 2.90 has a parent. Nodes containing other – Inventor (the basis for VRML), scene- – nodes are parent nodes, while the nodes 2.25 graph programming shields you from the – they contain are the child nodes (children) – gory details of rendering, letting you fo- – of their parent. Nodes that can contain – cus on what to render rather than how to – children are grouping nodes; those that – render it. 2.95 cannot are leaf nodes. Subgraph structures – As Figure 1 illustrates, scene graphs of- – in Figure 2 let a specific grouping of nodes 2.30 fer a high-level alternative to low-level – exist as a discrete and independently ad- – graphics rendering APIs such as OpenGL – dressed unit of data within the main scene- – and Direct3D. In turn, they provide an ab- – graph structure. Operations on the scene – straction layer to graphics subsystems re- 2.100 can be performed on all nodes in the – sponsible for processing, eventually pre- – graph, or they may be restricted to a par- 2.35 senting scene data to users via monitors, – ticular subgraph (scenes can therefore be – stereoscopic goggles/glasses, projectors, – composed of individual nodes as well as – and the like. – entire subgraphs that may be attached or – Before scene-graph programming mod- 2.105 detached as needed). – els, we usually represented scene data and – Scene graphs in Figure 2 resemble tree 2.40 behavior procedurally. Consequently, code – data structures when depicted visually. – that defined the scene was often inter- – Not surprisingly, trees are often used for – spersed with code that defined the pro- – scene-graph programming. The directed – cedures that operated on it. The result was 2.110 acyclic graph (DAG) data structure (also – complex and inflexible code that was dif- – known as an “oriented acyclic graph”) is 2.45 ficult to create, modify, and maintain — – commonly used because it supports node – problems that scene graphs help resolve. – sharing at a high level in the graph (nodes – By separating the scene from the op- – in a DAG can have more than one par- – erations performed on it, the scene-graph 2.115 ent) although typically at the expense of – programming model establishes a clean – additional code complexity and memory 2.50 boundary between scene representation – consumption. In a DAG, all nodes in the – and rendering. Thus, scenes can be com- – graph have a directed parent-child rela- – posed and maintained independent of rou- – tionship in which no cycles are allowed — – tines that operate on them. In addition to 2.120 nodes cannot be their own parent. – making things easier, this lets you create – 2.55 sophisticated content using visual author- – Graph Traversal – ing tools without regard for how work is – Scene-graph nodes represent objects in a – processed. – scene. Scene graphs used for 3D content, – Listing One is VRML code for a scene 2.125 for instance, usually support nodes that – consisting of a sphere that, when touched, – represent 3D geometric primitives (prede- 2.60 appears yellow. As you can see, the ob- – fined boxes, cones, spheres, and so forth), – jects and their behavior are represented – arbitrarily complex polygonal shapes, lights, – at a high level. You don’t know (or care) – materials, audio, and more. On the other – how the sphere is rendered — just that it 2.130 hand, scene-graph programming models – is. Nor do you know or care about how – for other forms of media might support – the input device is handled by the un- – nodes for audio/video content, timing and – derlying run-time system to support the – synchronization, layers, media control, spe- 2.67 “touch” behavior. Ditto for the light. 2.134 cial effects, and other functionality for com- d07wals.x.g2db Page 3 of 7 Thursday, April 4, 2002 8:47:50 AM 3.1 posing multimedia. 3.68 because only one of each type can be – Scene-graph programming models sup- – bound, or affect the user’s experience, – port a variety of operations through tra- – at any instant in time. – versals of the graph data structure that typ- – 3.5 ically begin with the root node (root nodes – Every VRML node has an associated – are usually the entry point for scene ren- – type name that defines the formal name – dering traversals). Graph traversals are re- – for the node —Box, Fog, Shape, and so – quired for a number of operations, in- 3.75 forth. Each node may contain zero or – cluding rendering activities related to – more fields that define how nodes differ 3.10 transformations, clipping and culling (pre- – from other nodes of the same type (field – venting objects that fall outside of the – values are stored in the VRML file along – user’s view from being rendered), light- – with the nodes and encode the state of – ing, and interaction operations such as 3.80 the virtual world) in addition to a set of – collision detection and picking. – events, if any, that the node can send or 3.15 Nodes affected by a given operation are – receive. When a node receives an event, – visited during a corresponding traversal. – it reacts accordingly by changing its state, – Upon visitation, a node’s internal state – which might trigger additional events. – may be set or altered (if supported) so 3.85 Nodes can change the state of objects in – that it reflects the state of the operation – the scene by sending events. A node’s im- 3.20 at that point in time. Rendering traversals – plementation defines how it reacts to – occur almost constantly with interactive – events, when it may generate and send – and animated graphics because the state – events, and any visual or auditory ap- – of affairs change as often as the user’s 3.90 pearance it might have in the scene. – viewpoint, necessitating continual scene- – VRML supports a Script node that fa- 3.25 graph queries and updates in response to – cilitates dynamic behaviors written in pro- – an ever-changing perspective. To increase – gramming languages such as ECMAScript, – performance, effect caching can be used – JavaScript, and Java. Script nodes are typ- – so that commonly applied operations use 3.95 ically used to signify a change in the scene – cached results when possible. – or some form of user action, receive events 3.30 – from other nodes, encapsulate program – Virtual Reality – modules that perform computations, or – Modeling Language (VRML) – effect change elsewhere in the scene by – VRML is an international Standard for 3D 3.100 sending events. External programmatic – computer graphics developed by the – control over the VRML scene graph is pos- 3.35 Web3D Consortium (formerly the VRML – sible via the External Authoring Interface – Consortium) and standardized by ISO/IEC. – (EAI). Currently awaiting final ISO stan- – The complete specification for ISO/IEC – dardization as Part 2 of the VRML97 Stan- – 14772–1:1997 (VRML97) is available at 3.105 dard, EAI is a model and binding for the – http://web3d.org/. – interface between VRML worlds and ex- 3.40 An Internet and web-enabled outgrowth – ternal environments. – of Open Inventor technology developed – All in all, the VRML Standard defines – by SGI (http://www.sgi.com/), VRML stan- – semantics for 54 built-in nodes that im- – dardizes a DAG-based scene-graph pro- 3.110 plementers, such as VRML browser plug- – gramming model for describing interactive – ins, are obligated to provide.