Gait Recognition from Occluded to Reconstructed Gait Cycle Using Deep Learning

Gait recognition plays a crucial role in video-based behaviour analysis, particularly in surveillance and security applications. Recognizing individuals based on their walking patterns can provide valuable insights for tracking and identifying people in crowded and occluded environments. This research presents an innovative approach using deep learning techniques for gait recognition from occluded to reconstructed gait cycles. We propose a pre-trained bidirectional gait reconstruction network combined with a CNN-RNN model to address the challenges posed by occlusion in gait recognition. The reconstructed gait cycles are obtained by detecting and reconstructing occluded frames using the spatiotemporal data from the gait sequence. We conduct experiments on the CASIA B dataset and achieve a trained accuracy of 97.14%. Performance evaluation on entire, occluded, and reconstructed gait cycles demonstrates that the reconstructed gait cycle outperforms the occluded gait cycle, with accuracy levels almost comparable to the entire gait cycle. These results highlight the effectiveness of our proposed approach in overcoming occlusion challenges and improving gait recognition accuracy. Our findings contribute to the advancement of deep learning-based gait recognition methods and offer insights for enhancing video surveillance systems in complex real-world scenarios.