Realtime and Robust Motion Tracking by Matched Filter on CMOS+FPGA Vision System

This paper describes realtime and robust tracking of a planar motion target by matched filter implemented on the CMOS+FPGA vision system. It is required to obtain positional and angular signals around 1,000 Hz to control a mechanical system. A vision sensor must obtain visual features of a target object, synchronizing its sampling rate to the sampling rate of the control. The CMOS+FPGA vision system has been proposed to realize 1,000 Hz visual feedback. Matched filter can compute the position of a target robustly against occlusion, change of illumination, and background texture but requires much computation time since it includes 2D Fourier transform of images. Thus, matched filter algorithm is implemented on the system so that the matched filter can be performed in realtime. First we briefly introduce the CMOS+FPGA vision system. Second, we summarize the algorithm of matched filter to describe the design of the circuit performing matched filter on an FPGA. Finally, we show the experimental results of planar motion tracking by matched filter implemented on the system.