Research on Multi-Feature Gait Recognition Based on Deep Learning

In recent years, surveillance cameras have been widely used in public safety and criminal investigation fields. In order to improve recognition accuracy, many monitoring devices are equipped with intelligent systems that can capture and analyze visual data and extract information related to suspicious behavior and personnel identity. Although widely used, common biometric recognition technologies such as facial recognition, fingerprint recognition, and iris recognition also face certain security risks, such as criminals using photos to evade recognition or 3D printed masks to deceive cameras. Unlike this, gait recognition technology verifies identity by analyzing an individual's gait characteristics, which has a wider recognition range and stronger robustness. Gait recognition technology has shown advantages, especially in complex environments and long-distance scenes, and is unaffected by masks and occlusions. Gait data such as video images and 3D motion capture data can effectively extract unique gait features of the human body. Due to the large number of dimensions in gait data, accurate feature extraction is crucial for improving recognition accuracy. Currently, gait recognition technology is mostly based on machine learning and deep learning methods, continuously improving recognition models' performance and application potential through training and optimization on large-scale datasets. The gait recognition algorithm proposed in this article is based on the LSTM-DBN network and evaluated on the CASIA-B dataset. As an important benchmark dataset in cross-view gait recognition, the CASIA-B dataset covers gait sequences from different perspectives.