Skeleton-Based Gait Recognition Using Modified Deep Convolutional Neural Networks and Long Short-Term Memory for Person Recognition

Vision-based person recognition is a method for identifying or recognizing a person based on gait features captured while walking with a camera. Despite the uniqueness of these gait features, their similarity among different individuals poses a significant challenge to the recognition process. This study contributes to the development of a classifier model that uses skeleton-based gait features to address these challenges. This novel approach incorporates deep convolutional neural networks and long short-term memory to significantly improve the performance of the model. By focusing on joint positions and angles of the hip, knee, shoulder, and wrist through human pose estimation, the study aims to enhance the accuracy of gait feature classification. The modified classifier model is based on AlexNet and includes LSTM and dropout layers. It was subsequently evaluated against other state-of-the-art deep learning classifiers to compare its accuracy. The results showed that the model achieved the highest accuracy score of 0.84, a precision score of 0.79, a recall score of 0.76, and an F1-score of 0.76. The performance trend indicated that this model was the most accurate for gait feature recognition.