High performance biomorphic image processing under tight space and power constraints

Image processing for space systems must be performed under tight space and power constraints while not compromising performance. Traditional computer vision approaches are not ideal because they are notoriously power hungry and physically large. We present two biologically inspired image processing systems that offer high performance on tight volume and power budgets. These systems follow a System-On-a-Chip (SOC) design methodology. They both exploit the promise of focal-plane computation offered by CMOS imaging technology. A general-purpose computational sensor has been fabricated in a standard 1.2 mu CMOS process, and its spatiotemporal filtering capabilities have been successfully tested. An array larger than 300 x 300 array will use only 0.5% of the chip area for the processing unit, while providing multiple spatiotemporally processed images in parallel. The 16 x 16 chip performs 1 GOPS/mW (5.5-bit scale-accumulate). An application specific sensor realizes a hybrid imaging system by combining a 120 x 36 low-noise active pixel sensor (APS) array with a 60 x 36 current mode motion detection and centroid localization array. The current mode array identifies pixels with time varying intensity. The centroid of all time varying pixels is computed. Clocked at greater than 60 fps, the chip consumes less than 2.5 mW.