A low-power stereo vision system based on a custom CMOS imager with positional data coding

A new stereo vision system is presented, which is based on a low-power CMOS binary vision sensor, with embedded spatial contrast extraction and output data compression, through positional data coding. A novel disparity algorithm adapted to the output of the sensors has been developed. The algorithm has been simulated and tested on a real application. The presented system is targeted towards event-based applications, where information is only provided under significant changes in the scene. Due to the custom address-driven data coding of the sensors, a novel approach for image processing has been adopted, which is more efficient with respect to those adopted in standard raster-scan imagers. The system has been tested in a people counting application, installing the prototype on the ceiling, at around 3m height. In this way, the system is able to discriminate people heights with an accuracy of about 5cm.