Deep Learning Based Automated Gait Recognition for Robust Person Reidentification

• Published in: 2023 6th International Conference on Engineering Technology and its Applications (IICETA) pp. 239 - 244
• Main Authors: Oleiwi, A. Sahib, Alkhafaij, Mahdi abdulkhudur, Ali, Eyhab, Saleem, Munqith, Habelalmateen, Mohammed I., Ali, Rabei Raad
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15-07-2023
• Subjects: Deep learning; Feature extraction; Gait recognition; Metaheuristic algorithms; Metaheuristics; Person identification; Support vector machines; Transfer learning
• ISSN: 2831-753X
• DOI: 10.1109/IICETA57613.2023.10351474

Generally, biometry-based control methods could not depend on separate expected performance or co-operation for operating properly. Conversely, such methods can be aware of malevolent processes to unauthorized access efforts. Certain works accessible in the literature propose the trouble with gait detection methods. These technique's purpose at recognizing humans with intrinsic perceptible features, despite dressed clothes or accessories. But this problem signifies a comparatively long-time challenge, one of the approaches established for handling the problem presents many disadvantages compared to feature extraction and minimal classifier rates, betwixt other problems. But deep learning (DL)-based techniques have recently been established as a powerful set of tools for controlling virtually several images and computer-vision-related issues, offering supreme outcomes for gait detection as well. In this aspect, this study develops hybrid metaheuristics with deep learning based automated gait recognition (HMDL-AGR) technique for person re-identification. The goal of the HMDL-AGR technique lies in the accurate detection of the different kinds of human gaits. In the presented HMDL-AGR technique, MobileNet model is applied as feature extractor with earthworm with crow search algorithm (EW-CSA). For gait recognition, pigeon optimization algorithm (POA) with support vector machine is used in this study. The experimental validation of the HMDL-AGR approach is examined and the outcomes show the supremacy of the HMDL-AGR approach over other current procedures.