In view of the complex background environment (light changes, micro-animals, etc.) in real video images, moving target detection is not easy to extract complete moving targets, and an adaptive three-frame difference algorithm based on improved mean modeling is proposed. The algorithm uses the mean background model established by the previous k-frame as the intermediate frame of the three-frame difference method, and then uses the three-frame difference method, and selects the adaptive threshold to binarize the difference result. An AND operation is performed on the two detected targets, followed by morphological processing, filtering, etc., and then the true position of the moving target is obtained. Finally, the experimental results show that the proposed algorithm can adapt to the more complex background environment, is not susceptible to illumination changes or other minor changes, and can effectively overcome the phenomenon of void and edge loss, and has higher detection accuracy, suitable for unattended monitoring. surroundings.