Abstract:

The growing proliferation of Artificial Intelligence (AI) has brought significant opportunities to make impacts on education. In this work, we explore how we might leverage AI to support fitness education. Specifically, we focus on teaching and learning two important exercises, namely squat and deadlift.

For many athletes, an essential aspect in learning and improving their skills comes from feedback. However, giving feedback on athletic movements, such as weight lifting or plyometrics, require analyses from either a professional coach, or an athletic testing kit, which is not accessible to most athletes especially when taking account for the affordability of these services. Currently, teaching a sports activity requires an expert coach to analyze specific movements, which is not only expensive but also resource dependent and time consuming. In addition, a coach can only cater to so many athletes at a time, resulting in a market demand for accessible, yet affordable coaching alternatives.

In our work, we aim to use AI, specifically skeleton detection models to support the teaching and learning of fitness activities. Specifically, our approach detects a learner’s posture during fitness activities and gives immediate and accurate feedback within seconds. Taking weightlifting as an example, KineticAI 1) allows a user to upload an image, 2) processes the image using a skeleton detection backend which captures the coordinates of the skeleton of the human in the image, 3) computes angles between different body sections, e.g., between shoulders and arms, and 4) give accurate assessment of the movement and feedback based on a set of heuristics based on the value of the angles. We are running a series of experiments with user-taken photos and online images of weightlifting athletes to examine the quality of feedback generated by KineticAI.

This research offers implications on the benefits of AI in enhancing fitness education through offering accurate and immediate feedback to athletes’ movements. This research also discusses the limitation of AI on human body skeleton detection in support of fitness education.

Keywords:

fitness training, skeleton detection, Applications of AI.