ADVANCEMENTS IN IDENTITY VERIFICATION: GAIT RECOGNITION AND DENSE NET TRANSFER LEARNING FOR THE FUTURE

Abstract

Transfer Learning for the Future Gait recognition has emerged as a cutting-edge biometric recognition technology with significant implications for everyday life. This study introduces a novel gait recognition approach, which uses Densely connected neural networks as the foundation for transfer learning, known as DenseNet-based transfer learning. The method begins by incorporating spatial information of gait through Gait Energy Image (GEI) input, followed by feature extraction using DenseNet-based transfer learning. The K nearest neighbor classifier (KNN) is then employed for classification and identification purposes. The proposed method is first tested on the extensive public dataset CASIA-B for same-view gait recognition, yielding impressive results with an average recognition rate of 98.86%. The method also demonstrates strong robustness under varying conditions. When compared to the VGGNet network, the proposed method reduces the number of network model parameters by 448M, or approximately 84.85%. These findings indicate that the proposed approach significantly enhances the speed and quality of gait recognition transfer generated images