An FPGA Implementation of the Mean-shift Algorithm for Object Tracking
Object tracking remains an important field of study within the broader discipline of Computer Vision. Over time, it has found application in a wide variety of areas, including industrial automation, user interfaces, navigation, object retreival, surveillance, and many more besides. A subset of these applications benefit from real-time or high-speed operation. This study attempts to implement the well known CAMSHIFT algorithm from its original specification in an FPGA. The inner loop operation to compute the mean shift vector is unrolled and vectorised to achieve real time operation. This allows the mean shift vector to be computed and the target to be localised within the frame acquisition time without the need for multiple clock domains.