Sports Assistant: Similarity Calculation for Sports Activity

Kun-Ru Wu, Yi-En Li, Yu-Yang Ding, Yu-Chee Tseng

Abstract: When participating in general sports activities, proper posture and movement are important to improve performance and prevent injury. In this paper, we propose a system to observe sports activity and calculate the difference between two people (for example, beginners and professional layers). The pose difference could help people improve their posture. The system consists of deep learning methods and similarity calculation. The deep learning method will detect human pose and output keyjoints. If the sport requires extra equipments, e.g. baseball bat, the system uses object detection to extract the keyjoints. Next, it calculates the pose similarity between two people. The developed system could help people know whether their posture is correct and which body parts could be improved.