Deep Learning-based Surveillance System to Provide Secure Gait Signatures

Abstract

Recently, deep learning models have played a great role in various fields such as computer vision, speech recognition, and image recognition due to their ability to extract complex features from large data sets and learn automatically. Today, providing security has become one of the basic things to do to protect people and departments, so the need for a monitoring system has become necessary to achieve that safety in order to reduce security risks and breaches. Gait is the way a person walks while moving, and the movement of each individual is unique. Therefore, it can be used in biometrics to identify the person by using gait recognition technology for security purposes (healthcare, airports, crimes, etc.), which is distinguished by the fact that it does not require the cooperation of the individual. However, there are challenges represented by variations in viewpoint, clothing, carrying conditions, and so on. To address this issue and to increase the accuracy of identification, we investigate gait recognition using deep learning in this research and develop an innovative method based on convolutional long short-term memory (Conv-LSTM) and two other methods (Conv-AlexNet) and (Convl-ResNet150) to identify and recognize human walking. We implemented our approach based on two datasets: the CASIA B dataset and the local dataset. According to our experimental results, the results of the experiment show that the suggested approach achieved an excellent recognition rate of 95% when applied to the CASIA B dataset and 100% when applied to the local dataset.