Detecting unseen anomalies in weight training exercises

Abstract

In weight training, correct exercise execution is crucial for maximizing its effectiveness and for preventing injuries. However, given the complexity of these movements, it is a challenge for trainees to know whether they are performing the exercise correctly. Considering the fact that wrong moves may result in life long injuries, it is important to design systems that can detect incorrect performances automatically. In this paper, we present a workflow to detect performance anomalies from only observations of the correct performance of an exercise by the trainee. We evaluated our algorithm on a benchmark data set for the biceps curl exercise, and also evaluated our system with a publicly available dataset, and showed that our method detects unseen anomalies in weight lifting exercises with 98 percent accuracy.