Aiming at the problems of light change, temporary occlusion and edge feature missing in moving object detection in real scenes, a mixed Gaussian model based on adaptive learning rate and an improved four frame difference target detection algorithm are proposed. The mixed Gaussian model based on adaptive learning rate is adopted to effectively solve the ghost and false detection phenomena in the traditional model; Secondly, the improved four frame difference method is used to overcome the influence of ambient light and motion occlusion, and the Canny algorithm is used to extract the edge of the target to supplement the edge information; Finally, the final detection result is obtained by "OR" operation of the three results. The experimental results show that the accuracy rate of the algorithm can reach 88.7%, the recall rate can reach 92.1%, and the harmonic average of the accuracy rate and recall rate can reach 90.4%. Compared with the traditional Gaussian mixture model, four frame difference method and the methods in the literature, the algorithm achieves the optimal results; The processing time of a single frame image is 24.3ms. On the basis of ensuring real-time, it can accurately extract foreground moving objects, and has certain anti-interference ability.