Archery score analysis system for outdoor environments

Abstract

Accurate scoring in archery is essential for objective performance evaluation and training feedback. However, existing systems are often limited by high cost, complex installation, or reliance on manual inspection. This study introduces a low-cost, image-based automatic scoring system optimized for outdoor environments, integrating conventional computer vision techniques with a deep learning–based heatmap regression framework. The system detects the target, applies circular geometric normalization, and employs a convolutional hourglass network to predict arrow impact points as heatmaps. A peak detection process then estimates scores based on the Euclidean distance from the target center. Localization error is measured in centimeters using a calibrated pixel-to-centimeter scale derived from a 10-point scoring zone diameter. Experimental results showed a mean localization error of 1.77 mm, with 67.83% of predictions within 2 mm, demonstrating reliable sub-half-centimeter accuracy under varying lighting conditions and partial occlusions. The system also provides directional feedback, supporting its application in both performance-monitoring and targeted-training enhancement. Overall, the approach offers a scalable, field-ready solution for archery that can be easily extended to other target-based sports.