A high-speed object recognition chip based on a biologically-realistic Hybrid Temporal Processing Element

Intelligent Reasoning Systems (IRS) is developing an integrated active vision system (AVS) based on design principles derived from experimental analysis of mammalian visual systems. The AVS design is based on a asynchronous analog encoding of location and motion data, and a custom VLSI Hybrid Temporal Processing Element (HTPE) developed by and available solely to IRS (patent pending). The HTPE can operate on analog data at frequencies in excess of 100 MHz, allowing rapid oversampling methods to be used for resolution enhancement, motion detection, and multiple-template matching. HTPEs have low device count and power dissipation, and can be fabricated in small layout areas. The AVS is intended for eventual on-board application in robots, intelligent machine tools, and other autonomous sensory-motor systems that require visible, IR, or similar input. Binocular fusion will be used to derive dipth information by comparing the primary-feature maps gnerated by feature detectors for two spatially-separated retinas. The goal of this Phase I proposal is to evaluate the feasibility of a binocular object recognition system for the AVS based on the HTPE.