In the detection of moving targets in video, Gaussian mixture model can achieve better results, but it is easily affected by illumination mutation and environmental noise, and the integrated contour of moving targets is difficult to extract. This paper improves the Canny edge detection algorithm. The hybrid filter composed of the median filter and the bilateral filter replaces the inherent filter in the edge detection algorithm, and uses the Otsu algorithm instead of manually setting the double threshold to avoid losing the true edge, ensuring the integrity of the edge information, and processing with the frame. The Interframe processing four-frame difference method is applied to obtain the moving target region, and then the Gaussian mixture model is used to obtain the foreground image. Combining the foreground images of the two algorithms, a more integrated moving target contour can be obtained. According to the experimental results, contrast with the classic Gaussian mixture model, the presented algorithm can prevent the impact of certain illumination mutations, figure out the problem of hole in the target image and the loss of edge information caused by missed detection, and has stronger robustness.