Method and device for rapid identification of video classification

Abstract

The invention is applicable to the technical field of computer vision, and provides a method and device for fast identification of video classification. The method comprises: obtaining the video to be tested, extracting the motion vector and the original RGB image from the video code stream of the video to be tested; converting the motion vector into a motion vector image; inputting the original RGB image to a predetermined convolution based on the RGB image In the neural network, to obtain the classification category confidence based on RGB; input the motion vector image into the predetermined enhanced convolutional neural network based on the motion vector, to obtain the classification category confidence based on the motion vector; the classification category based on RGB Confidence and the classification category confidence based on the motion vector are fused to obtain the classification category confidence of the video to be tested; the category corresponding to the highest score in the classification category confidence of the video to be tested is used as the content in the video to be tested corresponding category. The invention can effectively improve the efficiency of video classification and recognition.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a method and device for fast identification of video classification. Background technique [0002] Content recognition technology in video is widely used in video surveillance, retrieval, human-computer interaction and other fields. In recent years, feature extraction methods based on dense trajectories and methods based on dual neural networks (ie image deep network and optical flow deep network) have shown better action recognition capabilities. However, although the feature extraction method based on dense trajectories and the method based on double neural networks can effectively improve the accuracy of action recognition, since both the feature extraction method based on dense trajectories and the method based on double neural networks need to calculate optical flow (Optical flow) information, and the calculation of optical flow information is very time-...

AI Technical Summary

Problems solved by technology

[0003] In view of this, the embodiment of the present invention provides a method and device for fast identification of video classification to solve the problem of time-consuming identification of content in video in the prior art

Images