Research Of Moving Object Detection And Shadow Suppression Algorithm Based On GMM

Moving object detection of video images is to segment the moving objects from video sequences. It is the basis of the post-processing of moving objects classification, object tracking, behavior judging and scenarios understanding. It has been widely used in military and civilian fields.This paper described the common moving objects detecting algorithm and focused on researching the Gaussian mixture background model which based on Gaussian Mixture Model. Experimental results show that the model is good at modeling complex background, have excellent robustness and efficiency which compared with other methods.However, Gaussian mixture background model can’t restrain the shadows of moving objects. To resolve this problem of the shadow elimination, the Gaussian mixture shadow model was presented. The model learned the thinking which Gaussian mixture background model module well background. Firstly, the algorithm uses the Gaussian mixture background model to detect the moving objects from the video scenes. Then it made the suspected shadow model to judge shadow form the moving foreground detected. The pixels which were judged suspected shadows joined in the study and the update of parameters of Gaussian mixture shadow model. Finally, Gaussian mixture shadow model distinguished whether these pixels belong to the shadow.The experiment results show that the restrain performance of the algorithm is well. However, it had a disadvantage. Because parameters of model requires adequate shadow pixels participate in learning, but all the pixels in scene were covered in a low possibility with a short time. So, that led to the model parameters to study inadequate.The paper presents an effective and feasible improvement program for the algorithm to solve the problem. The paper used the color image segmentation technology to divide the video scenarios into different sub-regions which accorded to the color similarity. In the sub-region, the study and update of parameters of Gaussian mixture shadow model of non-suspected shadow pixel can use pixels information of suspected shadow. The experiment results showed that the improved algorithm has good performance to restrain and eliminate the moving shadows.Experimental results indicated that the improved algorithm can reduce greatly the adverse effects of the shadow of moving object detection. It eliminates shadow more effectively.