Automatic Archery Scoring System Using Deep Learning and Image Processing

This paper proposes an automatic archery scoring system that uses smartphone to read and record scores, addressing the limitations of existing manual and hardware-based scoring systems. Utilizing advancements in computer vision technology, the system consists of two main components: arrow localisation and arrow scoring. The localisation model uses a modified U-Net architecture to accurately identify the landing positions of arrows on a target, generating a saliency map that is converted into pixel coordinates essential for scoring. A custom dataset, which is released to public, was created to train this model. The scoring component identifies the target face using colour masking and applies image processing techniques for noise reduction and skew correction. It then matches the coordinates from the localisation model to the parameters of each scoring ring to calculate scores. While the model shows progressive improvement during training, further diversification of the dataset, exploration of alternate neural network architectures, and optimisation of hyperparameters can enhance accuracy. Future work includes improving front-end development for seamless functionality integration and adopting augmented reality (AR) for improved perspective correction.