Human Gait Classification Using Deep Learning Approaches

Abstract

Human gait is one of the useful biometric traits which determines human’s identity by the manner of their walk in the video surveillance. It is considered as behavioral biometric cues which depend on walking pattern of the people rather than the look of the people. Various factors affect the performance of gait such as carrying condition changes, clothing condition changes, and viewing angle variations. We develop a model for gait classification using deep learning methods. Separate gait databases are used in our experiment. The gait signatures are extracted from gait energy image (GEI) using convolutional neural network (CNN). The classifiers we used for classification of human gait are support vector machine (SVM), random forest, and long short term memory (LSTM). These models are tested on standard gait dataset CASIA A and our gait database is created using Microsoft Kinect device. The experimental results are very promising on both the datasets.