Real-time Motion Detection with High-speed Vision and Local Parallel Processing

In order to achieve highly real-time performance for motion detection and control system, a vision system based on high-speed CMOS sensor and local FPGA processing was designed. In this system, the data of high-speed CMOS sensor were directly collected by a near-end FPGA, on which the image sequences are processed by the logic of pipeline structure and a fast method called projection average algorithm was applied for centroid detection. With the synchronous processing and parallel computing, highly real-time target recognition and centroid detection were realized. Experiments show that this system realizes real-time detection of 1146fps at 560×480 resolution, whose target positioning accuracy is 1 pixel and detection latency is only 593 clock cycles, about 6.59 us at 90 Mhz. This high-speed vision-based object detection system achieves highly real-time performance and can be applied to a variety of high-speed motion measurement and real-time control systems.