Research on Gait Feature Fusion Method of OpenPose and GEI Based on Convolutional Network

Abstract

Gait recognition has the advantages of low resolution, easy acquisition but not easy to imitate, and difficult to camouflage, etc. It is one of the biometric recognition technologies with the most development potential at long distances. However, factors such as viewing angle, clothing, and walking with weights in gait data can make gait recognition performance poor. Convolutional neural network is one of the best deep learning techniques, which has the ability to fit features and extract typical features. Therefore, this paper proposes a gait recognition method based on a multi-layer convolutional neural network. This method is based on the OpenPose module to obtain the skeletal key point set of the human body in the gait cycle and the gait energy image (GEI) corresponding to the gait cycle, autonomously learn the characteristics of the target in a gait cycle, and use the triplet loss function to optimize the network. The recognition accuracy of the method above is 92.40% in a single view and 90.25% in a cross view after testing on typical gait dataset CASIA-B, which is robust to several uncertain factors.