Recognizing Human Actions by Using the Evolving Remote Supervised Method of Spiking Neural Networks

Xiurui Xie, Hong Qu, Guisong Liu, Lingshuang Liu

Abstract: This paper proposes a novel approach based on spiking neural networks to recognize human actions in videos. In our method, a star skeleton detector is designed to extract spatial features of input videos, and a classifier using evolving ReSuMe algorithm is proposed, with scale and shift invariance, to recognize input patterns. In learning algorithm, the remote supervised learning method(ReSuMe) is improved by the particle swarm optimization(PSO) algorithm. Experimental results on KTH and Weizmann dataset prove that our method achieves a significant improvement in performance compared with traditional ReSuMe and other method based on neural networks.