Real-Time Visal Motion Detection by Spatiotemporal Energy Model Implemented on GPU

The aim of this study is to develop a real-time visual motion detection system by using physiologically meaningful image processing algorithm. Spatiotemporal energy model has been recognized as the most plausible algorithm corresponding to the jobs in motion detection performed by simple and complex cells existing in area V1 of cats or macaque monkeys. Because of the parallelism of the brain, this algorithm inherently has high parallel performance. Together with the locality, spatiotemporal Gabor filtering and succeeding energy extraction process fit with the architecture of present GPU (Graphic Processing Unit). Enabling real-time motion detection at each pixel location over the entire input image is fundamental in many applications as for instances in robotics vision and carmounted camera. This system, moreover, is open for further expansion based on the physiological knowledge about mammalian visual system.