WCSE

Research on Real-time Behavior Recognition Method Based on Deep Learning

Abstract— With the advent of the era of big data, machine vision is growing rapidly and behavior recognition technology has a wide range of applications in our lives. As far as the current trend of behavior recognition technology is concerned, most of them have a series of problems such as slow calculation speed, low recognition accuracy and delay. In this paper, PoseNet deep neural network algorithm based on tensorflow.js is adopted to process the acquired image, train on the data set and extract the posture confidence and key point information of the human body. Through relevant algorithms, the behavior recognition of the target human body is completed, which has a broad application prospect in the future.

Index Terms— Tensorflow.js, Deep Neural Networks, Key points, Behavior recognition

Yuanjun Ding, Qingqing Yang, Haoyang Yu, Hongjie Wang, Xiaocong Chen, Haibo Pu
College of Information Engineering, Sichuan Agricultural University, CHINA
Haibo Pu
Key Laboratory of Agricultural Information Engineering of Sichuan Province, CHINA

[Download]

Cite: Yuanjun Ding, Qingqing Yang, Haoyang Yu, Hongjie Wang, Xiaocong Chen, Haibo Pu, "Research on Real-time Behavior Recognition Method Based on Deep Learning," Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering, pp. 307-311, Hong Kong, 15-17 June, 2019.