Gait-CNN21: A Two-Phase Human Gait Recognition Technique for Frontal View Analysis

Security threats in different areas are the cause of main concern nowadays. Human gait recognition is one of the famous methods that address security threats by using the walking trait of a person for recognition. This is an efficient and robust technique but also has some factors, such as variations of clothing, carrying, and walking, that degrade the system's performance. To handle these factors, this paper proposes a new technique for human gait recognition called Gait-CNN21. The Gait-CNN21 is comprised of two phases that recognize a person using walking features. In Phase-1, a preprocessing technique is designed using four steps such as histogram equalization of the input image, normalization, augmentation, and Gaussian noise removal. Phase-2 is based on the design of a 21-layer CNN model and training the model using hyperparameters selected by extensive experiments. The Gait-CNN21 technique is evaluated by using frontal view data of the CASIA-B dataset and achieved the accuracy of 95.48%, 90.34%, and 89.44% on NM, BG, and CL, respectively. By comparing the results with existing techniques, it is evident that Gait-CNN21 obtained better results as compared to current schemes.