Video retrieval method, device and equipment and storage medium

Abstract

The invention provides a video retrieval method, device and equipment and a storage medium. The method comprises the following steps: acquiring a comparison video clip from a video library according to the time length of a to-be-tested video; determining the similarity between the to-be-tested video and the comparison video clip through a space-time neural network, wherein a space-time convolutionlayer of the space-time neural network is set to carry out two-dimensional convolution and time dimension information processing; and traversing the video library, and outputting a retrieval result according to the similarity. In the process of determining the similarity between the to-be-tested video and the comparison video clip, the space-time neural network is used for determining the similarity between the to-be-tested video and the comparison video clip; a space-time convolution layer of the space-time neural network firstly performs two-dimensional convolution, processes spatial features of a to-be-detected video and a comparison video clip, then processes time dimension information of the to-be-detected video and the comparison video clip, and realizes low-complexity space-time feature extraction in a mode of successively processing the spatial features and the time dimension information.

Description

technical field [0001] The present application relates to the field of video technology, in particular to a video retrieval method, device, equipment and storage medium. Background technique [0002] Video retrieval refers to finding the desired video or video clips from a large amount of video data. Space-based video retrieval methods generally extract the key frames of the video to be retrieved, compare the key frames with the videos in the video database, and find the desired video or video clips. The retrieval accuracy of this method is not high. Based on spatio-temporal video retrieval, the retrieval takes into account the temporal correlation between the frames of the video to be retrieved. This method has high retrieval accuracy, but has a large amount of computation and low retrieval efficiency. For example, if the video to be retrieved is a video of hands turning in circles, if the space-based video retrieval method is used, the retrieval results can be obtained ac...

AI Technical Summary

Problems solved by technology

The space-based video retrieval method generally extracts the key frames of the video to be retrieved, compares the key frames with the videos in the video database, and finds the required video or video clips. The retrieval accuracy of this method is not high.

The time-space-based video retrieval method takes into account the time correlation between the frames of the video to be retrieved. This method has high retrieval accuracy, but has a large amount of calculation and low retrieval efficiency.

For example, if the video to be retrieved is a video of hands turning in circles, if the space-based video retrieval method is used, the retrieval results can be obtained according to the position of the hand in each picture, but it is impossible to identify whether the hand is turning clockwise or counterclockwise. Therefore, the accuracy Not high; but with the video retrieval method based on time and space, the retrieval results can be obtained according to whether the hand is turned clockwise or counterclockwise, with high precision and low retrieval efficiency

Images