Gait period detection method based on convolutional neural networks

Gait period detection directly affects the computational cost and accuracy of gait recognition. On the basis of deep convolutional neural networks, this paper models the gait cycle by classifying the gait sequence according to periodicity and fitting the gait sequence to sine functions to detect the gait cycle. The key is to abstract gait period detection as a classification problem or a sine function based on the regularity of the gait period. Each frame in a gait video corresponds to a category or a function value that can represent its periodic characteristics. Convolutional neural networks are used to extract periodic features of the gait frame, locate the position of the frame in the cycle, obtain the classification or return result, and finally realize gait cycle detection. In the CASIA-B dataset, we test a variety of network structures in terms of different views to verify the effect of periodic detection. The experimental results are compared with those of other state-of-the-art works and show that the proposed method achieves good accuracy and robustness for gait period detection. Copyright ©2021 Journal of Harbin Engineering University.