A Fast Foreground Object Detection Algorithm Using Kernel Density Estimation

a real-time foreground moving object detection algorithm based on Kernel Density Estimation in srgb color space is proposed in this paper, followed by an iterative noise reduction procedure, which removes dispersed noise and enhances the foreground contour in a binary image mask within a Bayesian framework. A simulated annealing strategy is applied in MRF-type decision making at the texture level to accelerate the convergence of the iterative noise reduction procedure. Experiments show that the proposed algorithm can resist undesirable effects of changing environmental illumination and shadow. Compared to several classical methods, better detection results are achieved on various datasets including both indoor and outdoor cases.