A Novel Approach for Gait Recognition Based on CC-LSTM-CNN Method

Abstract

Gait recognition is a new biometric technology to identify different users through gait data. In this paper, a novel CC-LSTM-CNN gait recognition method is proposed based on the cross-correlation (CC) algorithm and the Long Short-Term Memory convolutional neural network (LSTM-CNN) hybrid classification model. The 3D cross-correlation features of X axis, Y axis and Z axis acceleration signals are calculated by cross-correlation algorithm, and the main features are extracted by Principal Component Analysis. Then LSTM-CNN hybrid model is designed to train the feature data series and get the CC-LSTM-CNN model. The results show that the accuracy for identifying different user reaches 99.3% thru gait data in 5 seconds. The result is significantly improved compared with the traditional machine learning classification algorithm.