Video abstraction method based on dynamic and static feature fusion

The invention discloses a video abstraction method based on dynamic and static feature fusion, which is used for video content understanding and processing, and extracts an important part in an original video as a video abstraction, so that the abstraction video contains the most effective information for people, and the total length of the video is shortened. Therefore, the time for people to obtain effective information is saved. In order to obtain visual features with richer hierarchical information, the method comprises the steps of firstly extracting dynamic and static features from videoclips and corresponding video images, then fusing the dynamic and static features, and finally sending the fused features to a video clip importance score prediction network; selecting partial videoclips according to the scores and connecting the video clips into a final video abstract according to a time sequence. Experiments are completed by using two data sets of SumMe and TVSum, and five times of cross validation are