Design of action detection system in wrestling match video based on 3D convolutional neural network. (24th April 2022)

• Record Type: Journal Article
• Authors: Liu, Yang; Mei, Qinyu; Gan, Xin; Zhu, Ya; Wang, Yongjie
• Is Part Of: International journal of wireless and mobile computing. Volume 22:Number 1(2022)
• Page Start: 29
• Page End: 37
• Publication Date: 2022-04-24
• Subjects: 3D-CNN -- three-dimensional convolutional neural network -- action detection -- wrestling video -- spatio-temporal features; Mobile computing -- Periodicals; Wireless communication systems -- Periodicals; 004.6
• Languages: English
• ISSNs: 1741-1084

Abstract

At present, there are some problems in motion detection in wrestling video at home and abroad, such as low detection accuracy and poor robustness. A motion detection system combining three-dimensional convolution neural network and recursive neural network is studied and designed. It uses three-dimensional convolution to obtain low-level feature code, then uses recursive memory module to obtain timing features, and finally completes motion detection according to timing features. Under the ratio of these three parameters, the accuracy of 3D-CNN convolutional neural network structure is higher than that of 2D-CNN. When the ratio of the influence factor of circular memory module P to that of circular memory module C is 1, the accuracy of 3D-CNN improves the fastest and the accuracy is close to 20%. The research results provide a new idea for the development of human motion detection and recognition technology.