e-Health video system for performance analysis in heart revalidation cycling

In this paper, an unobtrusive, fully-automated, model-based algorithm is proposed, which is able to estimate the pose of a cyclist and track it over time to extract important pose and movement parameters. Applied techniques include background subtraction, skin detection, principal component analysis and template matching. The proposed algorithm is robust against variations in human appearance such as body size and clothing. Besides pose estimation, we report on the ongoing development of a non-contact, vision-based heart rate estimation algorithm. The algorithm is able to detect and track the subtle temporal skin color changes caused by the flowing blood through the vessels and extract the corresponding cardiovascular Blood Volume Pulse (BVP) signal. Our proposed heart rate estimation algorithm includes (fore-)head tracking, independent component analysis and motion filtering, to eliminate frequencies caused by human motion. Our pose estimation and heart rate detection systems are successfully validated on experimental manually generated ground-truth data. The mean absolute body part orientation error is between 0.8 and 7.4 degrees for the pose estimation algorithm and between 1.9 and 3.9 beats per minute for the heart rate detection algorithm. By combining both systems, fully-automatic, non-contact cycling performance analysis can be performed based on video input only.