Multimodal Gait Recognition and Temporal Parameter Estimation Using IMU & Visual Sensors

This paper proposes a multimodal gait recognition system using wearable IMU sensors and video-based pose estimation to provide a robust framework for biometric applications. Temporal gait parameters such as stride time, swing time, and cadence are extracted from IMU signals and joint trajectories detected using OpenPose. These features help identify gait abnormalities and support health assessments. A Siamese Bi-LSTM network learns temporal gait representations from sensor and video data to verify identity for gait recognition. The system improves recognition accuracy and robustness across varied environments by fusing the outputs from both modalities. The integrated platform offers a web-based interface suitable for healthcare monitoring, rehabilitation tracking, and secure biometric authentication.