Real time human motion recognition via spiking neural network

2018 3rd Asia Conference on Power and Electrical Engineering

能源学;电工学

Yang, Jing^1 ; Wu, Qingyuan^1 ; Huang, Maiqi^1 ; Luo, Ting^1

Beijing Normal University, Zhuhai Campus, Zhuhai, Guangdong, China^1

关键词: Encoding schemes;    Gradient descent learning algorithm;    Human motion recognition;    Human motions;    Human-action recognition;    Real-time motion;    Spiking neural network(SNN);    Spiking neural networks;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/366/1/012042/pdf
DOI  :  10.1088/1757-899X/366/1/012042

来源: IOP

【 摘 要 】

Real time human action recognition is to recognize the human motion type based on skeleton movement in real time and is always a challenging task. In this paper, a novel method is proposed to accomplish the classification by using Spiking neural network (SNN) which is biology oriented neural network dealing with precise timing spikes. First, a new temporal encoding scheme is used to encode the real time motion capture data into a series of spikes and the according type of the motion is represented by a spike time. Second, a two-layered spiking neural network is initiated and trained through a gradient descent learning algorithm. The experimental results show that this method achieves a good learning precision and generalization.