Experts lift differently: Classification of weight-lifting athletes

The process of learning a novel body movement exposes a student to multiple difficulties. Understanding the range of motion is fundamental for learning to control the involved body parts. Theory and instructions need to be mapped to body movements: a student not only needs to mimic or copy the range of motion of individual body parts, but he also needs to trigger the motion fragments in the correct order. Not only correct order is important, but also precise timing. If the movements in questions are intensified by additional load, optimality of the motion patterns becomes crucial. Sub-optimal execution of an exercise reduces the performance or can even induce failure of completion. Correct execution is a subtle interplay between the correct forces at the right times. In this paper, we present a sensor system that is able to categorize movements into multiple quality classes and athletes into two experience classes. For this work we conducted a study involving 16 athletes performing squat-presses, a simple yet non-trivial exercise requiring barbells. We calculated various features out of raw accelerometer data acquired by two inertial measurement units attached to the athletes' bodies. We evaluated exercise performances of the participants ranging from beginners to experts. We introduce the biomechanical properties of the movement and show that our system can differentiate between four quality classes (poor, fair, good, perfect) with an accuracy above 93% and discriminate between a beginner athlete and an advanced athlete with an accuracy of more than 94%.