Traditional Gaussian mixture modeling algorithm has the poor shadow suppression, and the problem of noise and being sensitive to illumination change also exist. In this paper, an improved Gaussian mixture modeling algorithm is used to extract the contour of moving objects. Gaussian mixture model is combined with Canny edge detection to extract the contour of objects with the help of the strong adaptability of Canny edge image to noise and illumination. However, this method is of high complexity and large amount of calculation, which can not meet the requirements of realtime video analysis. So this paper designs and implements the algorithm of contour extraction of moving objects based on CUDA platform by using GPU’s advantages of powerful computing power and parallel processing. Experimental results indicate that the algorithm improved the adaptability of noise and illumination, and effectively suppress the shadow in images. It can extract the contour of the moving objects in video sequences faster than in general environment under the precondition of the effect.