Robust Human Gait Recognition with Convolutional Neural Network based on Gait Energy Image

Abstract

Purpose: Human gait recognition is one of the developments in artificial intelligence technology. Gait recognition is a biometric recognition technique that uses no direct interaction with an object, allowing for identification of individuals based on their gait. However, this recognition faces challenges, including varying camera angles (0⁰ - 180⁰), so this requires a more in-depth introduction.

Methods: Therefore, based on the references, this study proposes using the Gait Energy Image (GEI) and Convolutional Neural Network (CNN) features for in-depth extraction and recognition of each image in the Casia B Dataset, which is then compared with the results of previous studies.

Result: The results of this study, with the division of the Casia B Dataset 80% as training data and 20% as testing data and 11 camera angles between 0⁰ - 180⁰ produced an accuracy rate of 99.48%.

Novelty: So the accuracy achieved with this deep learning technique exceeds that of previous research using conventional methods and this gait pattern recognition technique can be used to be implemented in a biometric recognition system based on human gait patterns.