The Research Of Motion Detection Based On Gaussian Mixture Model And Shadow Elimination Algorithm

Motion detection is one of important researches in digital image processing and is also the difficult problem which need to be solved in current motion vision research. As a classic algorithm of the motion detection, Adaptive Gaussian Mixture Model algorithm has been widely used, but its update strategy exists some deficiencies; In the same time, moving shadow would interfere the accurate extraction of moving targets, In order to reduce this interference, researchers has proposed a lot of shadow elimination method according to the features of shadow. However, these algorithms are generally limited by the application scenarios. The purpose of this paper is to improve the effects of motion detection based on Gaussian Mixture algorithm and extend the scenes to which shadow elimination algorithm applies, main tasks of this paper are as follows:Because traditional Gaussian Mixture Background Model algorithm uses uniform update rate and the Gaussian variance over convergence lead to the error detection. In this paper, we proposed a new update strategy. When updating the background model, the improved algorithm first divided the background into multi-modal regions, moving regions and escaping regions in dynamic, according to the model information and the result of prospect detection. Then, we assign different update rates for the different regions; Besides, when in the process of prospect detection we are introducing a threshold detection method, which could process error detection caused by illumination changes better.Shadow detection algorithm based on one kind of image feature is easily limited by the scene and illumination changes, which will leads to error detection. In this paper, a shadow detection algorithm based on HSV color feature and image Gradient feature is proposed. Firstly, the shadow is determined by the brightness differences between the shadow region and the background, the constancy of chroma and saturation. Then, the shadow is detected according to the Gradient invariance, Finally, the final results is gotten by combining these two results of shadow determination; Besides, we added a determine condition to the shadow decision rules based on HSV color features, which could solve the error detection caused by light intensity from weak to strong.To verify the effectiveness of the algorithm, with the programming tools of VS2005 and OpenCV, this paper achieved our algorithm and relevant comparison algorithm and applied these algorithms on the surveillance video material downloaded from the Computer Vision and Robotics Research Laboratory, University of California. The results show that:The Improved Gaussian Mixture algorithm could better handle the multi-modal regions and improve the precision of moving object segmentation; The fusion algorithm for shadow detection could achieve good inhibitory effect of the shadow and reduce the error detection caused by light mutation after introduction shadow elimination algorithm.