A Deep Neural Network Approach for Model-based Gait Recognition

Cholwich Nattee, Nirattaya Khamsemanan

Abstract

Since the declaration of  war on terrorism, human identification has gained more popularity  throughout various research communities. One of the sought-after  techniques that is used in the identification process is Gait  recognition. Gait recognition is a biometric recognition technique  that uses body measurements and movements during walks. This technique  is non-invasive and can be done without an awareness of a subject. In  this work, we develop identification models for gait information  extracted from sequences of walking by Microsoft Kinect, from  Andersson and Araujo, using a Deep Neural Networks (DNN) combined with  Majority Vote.  The results show our proposed model yields accuracy of  95.0%, which outperforms Support Vector Machine, k-Nearest  Neighbor,  Multi-Layer Perceptron and solely Deep Neural Networks  techniques.