Design And Implementation Of GMM Algorithm Based On Gradient

The commonly used moving object detection algorithms with their advantagesand disadvantages are introduced base on the analysis of these their implementation.We proposed a robust gradient-based GMM algorithm. First, by studying the GaussianMixture Model in background modeling method, the problem that GMM can’t adaptto rapid changes in light illumination which cause background regions to be detectedas foreground regions is analyzed from three aspects as the model initialized,parameters update and foreground segmentation. Then, we proposed a region levelgradient based GMM algorithm rather than the traditional pixel level GMM algorithm.The GMM algorithm is extended based on image gradient information. Twobackground models are modeled based on color information and gradient information.By combine the result of these two models we achieved the foreground regionswithout false detections caused by the illumination area.The Algorithm is developed using Visual Studio2008and OpenCV on Windowsplatform. By Comparing the result of traditional GMM to proposed method on a size320×240video sequences, the results show that the proposed algorithm retains theflexibility and the accuracy of the GMM while overcoming the problems ofillumination.