OpenCV compatible real time processor for background foreground identification

The background identification methods are used in many fields like video surveillance and traffic monitoring. In this paper we propose a hardware implementation of the Gaussian Mixture Model algorithm able to perform background identification on HD images. The proposed circuit is based on the OpenCV implementation, particularly suited to improve the initial background learning phase. Bit-width has been optimized in order to reduce hardware complexity and increase working speed. The proposed circuit processes 22 1920×1080 frames per second when implemented on Virtex 5 FPGA.