Abstract: A video retrieval method based on video fingerprints and spatio-temporal information was proposed to solve the problem caused by the whole drift of brightness, interference of abnormal point and video edit. First, the video fingerprint feature was extracted by frame segmentation. Then, an improved direct timing algorithm was used to eliminate the brightness drift interference. Finally, the concept of keyframes density was introduced to eliminate the time span problem caused by video editing and a comprehensive search strategy was presented. Experiment results show that the algorithm can effectively solve these problems with good robustness, low computing burden, and high recall rate and precision.