The functional field of view for collision perception in real-time animation systems

C. O'Sullivan1, R. Radach2

1Image Synthesis Group, Computer Science Department, Trinity College Dublin, Dublin 2, Ireland (e-mail:carol.osullivan@cs.tcd.ie);
2Institute of Psychology, Technical University of Aachen, Jägerstr. 17, D-52056 Aachen, Germany

In recent years, the realisation has been growing within the computer graphics community of the advantages to be gained by using knowledge of human perception. We maintain that an extension of this approach to a study of human visual perception of dynamic scenes would be very beneficial in solving some of the problems of real-time animation systems. In interactive animation applications such as Virtual Reality (VR) or games, it cannot be predicted in advance how a user or the entities in a virtual world will behave, so the animation must be created in real-time. As the number of independently moving objects in the scene increases, the computational load also increases. Possible scenarios include crowd simulations or avalanches of rocks, where large numbers of homogeneous entities move around a virtual world in real-time. There are many bottlenecks in such systems, collision detection being a major one. A trade-off between detection accuracy and speed is necessary to achieve a high and constant frame-rate. The disadvantage of this approach is that objects may be perceived to be interacting in an unrealistic manner, repulsing each other at a distance, or interpenetrating. The analysis of visual perception of collision events enables a prioritisation of potential collisions to process within a given frame of an animation, hence reducing the negative impact of detection accuracy degradation.

To determine the functional field of view for the perception of collisions, we carried out 3 experiments with 12 participants. Filled white circles of 1 deg diameter presented on a black background served as stimuli. Subjects were asked to detect collisions (no gap between two circles) versus repulsions with minimal gap sizes of 0.1 or 0.4 deg. In experiment 1, we examined detection performance in a static display as a function of 3 levels of eccentricity, 4 directions of offset and 8 regions (up-left, right-down etc.). Experiments 2 and 3 used a dynamic situation with targets moving at a speed of 2.9 deg per second. In experiment 2 we added distracters that were different in colour from the colliding entities, and in experiment 3 we added distracters which were identical to the colliding entities (the "real-world" scenario).

In all three experiments, performance was found to be determined by eccentricity and separation of the colliding entities. In experiments 2 and 3, most subjects achieved maximum detection rates when collisions occurred at locations above the current fixation. The number of distracters significantly affected performance when the distracters were visually similar to the colliding entities (experiment 3), but not when the distracters were different (experiment 2).

Based on the results of these experiments, a plausible model of human visual perception of collisions was developed. In a real-time animation of large numbers of homogeneous three-dimensional objects, this model was used both to schedule collision processing, and as a metric to measure perceived collision inaccuracy. This system is currently being extended to include an eye movement contingent regime of prioritising collisions.