Human Identification Using Gait Recognition

Abstract

Gait recognition is a way used to recognize individuals based on their walking style. Unlike other identification techniques, gait recognition can be performed from a distance using video footage without requiring the person's cooperation. It has found applications in various areas such as access control, surveillance, and forensics. There are three main approaches to gait recognition based on the setup namely, machine vision, floor sensor, and wearable sensor methods. In this project machine vision method is implemented. One or more cameras capture video footage, which is then processed to extract gait features. Gait recognition is subject to variations caused by factors such as the viewing angle, clothing, walking speed, footwear, and whether the person is carrying any items. The proposed system utilizes a Convolutional Neural Network (CNN) to extract high-level gait features from low-quality video. The process involves preprocessing the video to obtain input data, extracting features, and comparing them with a pre-trained neural network for recognition. We trained the model on Casia Gait dataset consists of twenty objects and walking patterns from three angles. We pre-processed each video of the object, then propagated it to the fully connected layer. We used the Mask R-CNN algorithm for silhouette extraction. Gait recognition serves as a biometric technology that can be used to identify individuals without their knowledge. The values obtained from the algorithm after training and testing were tabulated. We attained an accuracy of 91% hence it can be applied to the real world scenario.