Automatic Rowing Kinematic Analysis Using OpenPose and Dynamic Time Warping

Abstract

In this study, we describe a system to automatically analyze rowing kinematic parameters using video capturing and processing and using a single RGB camera. Useful rowing 2D joint angles in the sagittal plane are estimated using the OpenPose API combined with an offline filter to overcome frame loss and oscillations. Rowing key moments are identified using a time series comparison between the joint angles curves from the movement performed and a manually labeled reference joint angles curves, representing a desired rowing stroke. The comparison is realized using the Dynamic Time Warping method. All the obtained data is displayed in a user-friendly interface to monitor the movement and provide offline feedback. The proposed approach enables automatic analysis of video-recorded training sessions.