Evaluation of the effect of camera viewing angles on the quality of human pose estimation in river surfing

Abstract

Capturing human poses with monocular RGB cameras seems like an uncomplicated compared to extensive motion capturing systems or calibrated multi camera setups. In our application scenario we are capturing the biomechanics of surfers on standing waves in rivers with GoPro cameras. Thereby we identified detection quality issues of the joints on the backside of the surfers’ bodies in certain situations.

In this paper we are evaluating different camera positions and angles with regard to the recognition quality of these critical body parts. We are describing the application scenario and the resulting requirements for human pose estimation and hardware setup.

Our evaluation results are identifying patterns of quality issues depending on surfers’ stance, skills and motions on the one hand and camera positions and angles on the other hand. We are concluding our paper with further approaches to improve detection quality.

Mehr zum Titel