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

Moving object detection is an important and fundamental problem in video analysisapplications, including video surveillance, content-based video coding, and human-computerinteraction. As a classic and effective algorithm for motion detection, Gaussian MixtureModel algorithm has been widely applied. However, there are some deficiencies whenmoving objects are staying still in the scene for an uncertain time; at the same time, movingshadow which might be detected as the objects could affect the extraction of moving objects.A lot of shadow elimination algorithms have been proposed. However, these methods aregenerally limited by the application scene. The purpose of this paper is to improve the effectsof Gaussian Mixture Model algorithm in motion detection and propose new shadowelimination algorithm which could be applied in various scenes. The main tasks of this paperare as follows:Firstly, this paper proposed a new method based on improved Gaussian Mixture Model.Moving objects can be converted into background ones by Gaussian Mixture Model whenthey might be staying still in the scene for an uncertain time. Therefore, a new method basedon improved Gaussian Mixture Model is proposed in this paper. To obtain complete objects, abackground learning parameter is introduced to update the model according to the detectionresult in previous frame. Moreover, the information of8-adjacent connection area is utilizedto suppress noises and improve its stability in the complex environment. Several experimentswere implemented and the results demonstrate its effectiveness in detecting the movingobjects which stay briefly in the complicated condition.Secondly, as for the disadvantages of shadow removal methods utilizing texture, thispaper proposed a new algorithm based on Gaussian Mixture Model and YCbCr color space.Firstly, moving regions are detected using GMM. Secondly, GMM models the shadow pixel,which is obtained through analyzing the color statistics of the difference between theforeground and background of the moving regions in YCbCr color space. Lastly, the thresholdvalue of the shadow is obtained according to the Gaussian probability distribution in YCbCrcolor space. More than70percent of shadow pixels in sequence images of the experimentscould be detected by the algorithm accurately. Experimental results show that the proposed algorithm is efficient and robust in object extraction and shadow detection under differentscenes.