Programmable Vision Sensors with On-chip Real-time Image Processing

Abstract

In this contribution, the research activity on CMOS pro- grammable vision sensors, carried out at ITC-irst, is presented, with special attention on pixel-parallel architectures for low-level real- time processing tasks. Different pixel topologies are described for full-kernel spatio-temporal filtering, featuring a good tradeoff be- tween pixel-pitch, fill-factor and power consumption (1-4). Thanks to the switched capacitor pixel implementation, all the vision processing can be accomplished by simply changing the sensor timing diagram, which is the key idea of the presented vision sensors. Moreover, a novel approach for on-the-fly image processing over high dynamic- range images is described, adopting integration time sharing and oversampling techniques, which allow integration time and process- ing time to be overlapped, offering new ways of implementing on-chip real-time image processing.