Foreground detection in video streams in an FPGA without external memory

The paper describes a foreground detection algorithm for video streams and its implementation in an IP core that fits in a small FPGA without any off-chip memory. The algorithm uses down-scaling of the resolution of a processed image while minimizing loss of information through selecting a suitable pixel aggregation. The quality of the algorithm is compared to the MoG algorithm for standard image datasets. Implementation and performance results are shown for its implementation in Spartan-6 and Zynq devices. The algorithm performs quite well while using only minimum of FPGA resources.