Real-time rendering of crowds consisting of high quality and distinct people

Project Details


Computer generated virtual models of cities alone are often empty and lifeless, the presence of a crowd helps bring them "to life". Currently real-time virtual models have only been able to show a small variety in the crowd, for example they often all walk the same way, even though the crowd itself is very large (thus containing many "clones" of each human). For example the realistic simulation and display of how people walk on uneven ground, step onto or off the curb without a foot floating in the air or disappearing into the ground has not been satisfactorily solved. The computer animation research community has not yet achieved this with sufficiently small computation cost to facilitate real-time crowd rendering. For example the different gaits people possess and the different ways that old, young, men, women and disabled people move is as yet uncharted in the virtual world of real-time crowd simulation. Our proposed work focuses on crowds of unique humans, in terms of how they look, how they move and how they interact with the world around them.

The challenges of our proposed research are split into three areas: firstly the behaviour engine for each crowd member has to work out what the individual's goal is and how it will be achieved. Once this is accomplished, the behaviour engine directs the animation engine to perform the required action. The human models are then deformed to fit the animation and implemented within the urban model using the rendering engine. The decisions cannot be pre-processed and must be based on the current situation at any point in time i.e. in real time.

Each area will bring features previously unseen in the crowd rendering field. The animation engines to date have a very limited number of actions they can perform, usually just playing an existing motion on loop. The real-time adaptation of existing animations (or generation of them if a suitable animation is not within the animation cache) will allow each crowd member to appear to perform unique movements, with the end goal of providing an extensive range of animations. This is a significant task and will require the development of algorithms capable of extreme scalability, from high precision and high cost close to the viewer, to low precision and low cost for the distant crowd (with many levels in-between).

Obviously if the animation engine is capable of constructing new and unique poses, the rendering engine must be able to represent these. A pre-selected number of positions, as is currently used, will not be sufficient. The rendering engine must be fully dynamic, abandoning traditionally static levels of detail, instead using, for example, animation caching, dynamic impostors and dynamic mesh simplification to achieve the desired speed without forfeiting the power of the system. With the rendering and animation engines providing this flexibility, it is clear the behaviour engine must be extended beyond those currently used in crowds in order to provide the information for the other parts to reach their full potential.
Effective start/end date1/02/0730/06/10


  • Engineering and Physical Sciences Research Council: £275,989.00