An efficient pixel-wise method for moving object detection in complex scenes

Moving object detection is often one of the most basic and important stages in computer vision applications. In this paper, a novel background model is proposed to extract moving foreground objects from videos that may contain different kinds of disturbance such as illumination changes, camera parameter variations, noises and dynamic backgrounds, etc. For each frame, a local frequency response map is generated using short-term Fourier transformation (STFT) in local regions, and by extracting the relations among neighborhoods of the response map, a compact pixel feature is introduced as local frequency pattern. Then, an adaptive probabilistic estimation of pixel feature sequence modified from kernel density estimation is performed to estimate the probability of a pixel being background. Experimental evaluations on complex scenes of surveillance videos demonstrate that the proposed method has archived satisfactory results.