Deep Learning Posture Estimation for Archery Consistency Measurement

Consistency in archer’s posture is crucial to retain high precision and high reproducibility in the shooting movement. The current practice in measuring postural consistency is often limited to human visual inspection with biases and subjectivity. First, a camera is used to capture body posture of archer during anchoring. A series of body posture images are subsequently processed with a deep learning pose estimation model with the extraction of keypoints. The selection of joints keypoints for archery are grouped for postural consistency study during anchoring phase. This paper presents a study between the postural consistency and scoring performance using an automated posture estimation mechanism. The study consists of nineteen elite archers performing ten shots in every session. The posture of each session is recorded to study their joint profile and consistency. It shows that the consistency measures are correlated with the joint profiles. The measure of joint profile and consistency can help coaches and athletes in assessing and improving performance with their training strategy.