Abstract

GW4 is a real-time video segmentation algorithm for detecting moving objects in indoor and outdoor scenes. The platform for the final implementation is Field Programmable Gate Array (FPGA); a reconfigurable computing platform. The algorithm detects moving foreground objects against a multimodal background; it is motivated by two well-known adaptive background differencing algorithms: Grimson's algorithm and W4. The implementation is based on a single stationary camera transmitting RGB values at 25Hz. Background modelling at pixel level has been used in many applications, but may fail due to camouflage and foreground aperture problems. These common problems have been reduced in our approach with the use of pixel and frame level processing. To make the algorithm feasible and efficient for the final hardware platform, we avoid the use of floating point numbers and transcendental operations. The final implementation operates at real-time frame rates on 640x480 video streams. We present experimental results indicating processing speeds, and superior segmentation performance to Grimson's algorithm.