An Improved GMM-Based Moving Object Detection Method Under Sudden Illumination Change

Abstract

In traditional GMM-based moving object detection method, the foreground was segmented by comparing the current frame with the constructed Gaussian distributions, which might occur the detection failure under sudden illumination change because the Gaussian distributions were disobeyed in current frame. Therefore, an improved GMM-based moving object detection method under sudden illumination change is proposed in terms of the fact that the pixel intensity in the neighboring zone changes with the similar degree as illumination suddenly changes. The mean quadratic deviation of the gray values in four different directions are employed to build four independent Gaussian distributions instead of the pixel intensity. And a pixel is considered to be the foreground as all of mean quadratic deviations disobey the historical Gaussian distributions. Simulation results show that the proposed method can more exactly detect the foreground with less false pixels as illumination changes abruptly. Moreover, constructing the Gaussian model parallel with four threads reduces the total computation time, which meets the need of the video processing in practical engineering.