Background Subtraction Algorithm Based On Gaussian Mixture Model And HSV

The background subtraction algorithm for sequence images obtain useful information by separating the moving objects and background. Analyzing the feature of the moving target is of great value in the field of target tracking, joint positioning, action analysis and 3D reconstruction.In this paper, we proposed the background subtraction algorithm based on the Gaussian mixture model, which is established after the researching of the distribution model of gray value in the background. There are two disadvantages of traditional EM algorithm, including a large amount of computation and local maximum value which is easily affected by the initial value. We proposed a new algorithm which incorporates the method based on image gray value. This algorithm can initialize the parameters of GMM very quickly and enhance the initialization speed and accuracy of the GMM parameters. In addition, we researched on the basic characteristics of the color space, then proposed the shadow detection algorithm, which is based on shadow brightness and chrominance features in HSV color space. And along with it is the elimination of shadow areas according to the result of Gaussian mixture model background subtraction.The algorithm proposed in this paper is based on Visual Studio2015 platform and OpenCV v3.2.0 library. The background subtraction experiment based on Gaussian mixture model and the shadow elimination in HSV color space were carried out respectively for the sequence images under different lighting conditions. And the algorithm was compared with the background subtraction based on kernel density estimation method and codebook method.Experiments show that the detection rate of the shadow elimination method proposed in this paper is higher than 80% averagely. Compared with the background subtraction algorithm based on the Gaussian mixture model alone,a more distinct shape of moving target can be obtained, and the shadow area is effectively reduced by this algorithm.