A computational model for motion detection and direction discrimination in humans

Abstract

Seeing biological motion is very important for both humans and computers. Psychophysics experiments show that the ability of our visual system for biological motion detection and direction discrimination is different from that for simple translation. The existing quantitative models of motion perception cannot explain these findings. We propose a computational model, which uses learning and statistical inference based on the joint probability density function (PDF) of the position and motion of the body, on stimuli similar to (Neri et al., 1998). Our results are consistent with the psychophysics indicating that our model is consistent with human motion perception, accounting for both biological motion and pure translation.