A real time simulated human vision system using connectionist models applied to target tracking

Summary form only given. An algorithm and an architecture of a simulated human vision system using connectionist models are discussed. An original approach to solve two major problems, occlusion and collision, is proposed for a multiple target tracking system. This approach uses the recall properties of various neural network models. The system has been built on a 64-node transputer system with 2 MB memory on each transputer. This system employs both image-intensity-based optical flow methods to compute motions of multiple targets and token-based methods to track a designated target