Motion Tracking with Surveillance Cameras

Abstract

Joint kinematics has been identified as a critical point of physical rehabilitation assessment. By tracking the gait patterns (i.e., stride length, step frequency, speed, knee flexion angle, and ankle flexion angle), walking abnormalities could easily be detected. Despite the substantial role of gait analysis, existing approaches are generally assisted by sophisticated equipment, making the popularization of gait analysis tough. In this project, an offline surveillance-camera-based motion tracking system is introduced. Joint points projected on video frames would be detected with the implementation of a deep-learning-oriented open source library named OpenPose. After the pre-processing of regression, an image distance detection method on account of trigonometry is implemented and calibrated. Fast Fourier Transform is then applied to extract the periodicity of the gait patterns. Besides, the system compares the calculated gait identities with healthy gait trends using time series similarity measurements. The project reduces the complexity of conducting gait analysis trails by limiting the required equipment to a single surveillance camera. By using a device as common as a smartphone, modern people could keep an eye on their walking patterns and follow routine physical rehabilitations easily.