UltiVision: A Deep Learning Computer Vision Approach for Error Detection in Ultimate Frisbee Forehand Throws

Ultimate frisbee, despite its low barrier to entry, presents a challenge for beginners in mastering the forehand throw, a crucial skill for advancing the disc and scoring points. New players often rely on traditional coaching methods which require experienced players to guide them. However due to ultimate frisbee being a niche sport, these experienced players are hard to come by. Therefore, a computational approach is proposed to address this gap. A human activity recognition model (UltiVision) using computer vision and deep learning, is proposed to classify throws made by beginners into one of 4 classes, ‘good’, ‘no flick’, ‘no pivot’, and ‘both’, with both meaning a combination of no flick and no pivot. These mistakes are based on biomechanical analysis of the forehand throw. By identifying these errors, the model serves as a foundation for an automated coaching system, providing immediate feedback and accelerating the learning process for beginners. This can potentially lead to faster improvement, a larger talent pool, and a more vibrant ultimate frisbee scene. The model, with an accuracy of 92%, was able to accurately classify throws made by beginners into the 4 designated classes.