An Action Recognition Method Based on Deformable Convolution Network. (March 2020)

• Record Type: Journal Article
• Authors: Dong, Xu; Tan, Li; Zhou, Lina; Song, Yanyan
• Is Part Of: Journal of physics. Volume 1487(2020)
• Publication Date: 2020-03
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1487/1/012033
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: In order to solve the problem of behavior recognition in short video, this paper first proposes a key frame extraction algorithm based on mutual information entropy, which uses sliding window to preserve the timing information between frames. Based on the key frame extraction, a based on Deform-GoogLeNet, a two-stream CNN method for deformable convolutional networks, uses the two-stream network to extract the RGB features and optical Flow characteristics of the image separately, and uses the weighted average method to obtain the results of behavior recognition. On the public dataset Charades dataset, mAP is 22.9, which is higher than the similar fusion algorithm, which proves that the proposed algorithm has a good effect in short video behavior recognition.