Depth Versus Inertial Sensors in Real-Time Sports Analysis: A Case Study on Fencing

Automatic analysis of human motion in sports can provide useful feedback for athletes and coaches. Both depth and inertial sensors are commonly employed for this task. Each type of device has its strengths and weaknesses. Depth cameras allow for efficient tracking of joint positions, although they require a clear field of view. Inertial sensors are useful for determining acceleration and orientation but need to be mounted on the practicing person. In this paper, we perform a case study on fencing in order to investigate the usefulness of both data modalities in real-time qualitative motion analysis in sports. Qualitative analysis of motion goes beyond typical action recognition scenario as it requires the detection of actions in continuous movement as well as determining parameters of the performed motion. In this work a dedicated dataset of fencing footwork is introduced, including synchronized depth and inertial data. Methods for detection and qualitative analysis of the fencing lunge action are proposed and experiments are conducted in order to compare the accuracy and efficiency of analyzing fencing footwork with both depth and inertial sensors.