Abstract:In view of the huge amount of video information generated by the camera, the retrieval work needs a lot of manpower, material resources and time cost. The analysis shows that most of the traditional retrieval functions are based on text Keywords, and the coverage of video content is low and easy to rely on the subjectivity of the relevant staff. This paper proposes how to use traditional machine vision technology and deep learning technology to build an efficient video retrieval system. The innovation lies in fully exploring the information from the perspective of video frame image content. The process of information mining does not need manual intervention, thus improving the utilization rate of information.