Motion Capture System based on RGB Camera for Human Walking Recognition using Marker-based and Markerless for Kinematics of Gait

The motion capture system has the potential to perform kinematics of gait analysis. Gait analysis can be applied in human activity recognition (HAR) for human walking recognition technology. The walking recognition makes it challenging for researchers to develop using RGB camera with high accuracy. This paper compares the accuracy of vision-based motion capture based on an RGB Camera using marker-based and markerless methods. The evaluation to determine the accuracy of the proposed of both methods was compared with statistical analysis. The marker-based method uses the Kalman filter, and the markerless method uses MediaPipe to measure gait parameters. Development of motion capture that can detect joint leg positions and measure joint angles based on OpenCV. It is designed for joint trajectories and angles at the hip, knee, and ankle. The motion capture system is implemented by a Logitech C270 webcam, Intel core i5 2.1 GHz processor, 8 GB RAM, and processed by JupyterLab with Python programming. It has been tested on recorded video data containing the subject walking straight with three gait cycles: slow, fast, and zigzag. In the marker-based method, each movement’s average joint position detection errors are 22 pixels, 134 pixels, and 50 pixels. The angles of the hip and knee joints have an average angle difference with a reference of ±7°. In comparison, the markerless method has an average position error are 23 pixels, 65 pixels, and 49 pixels. And markerless has an average angle difference with a reference of ±5°.