An improved real-time soccer object detection and shooting method based on prior position information

Accurate detection and tracking of soccer balls in complex and dynamic environments such as soccer matches present a significant challenge due to the small size of the ball and its rapid movements. Existing detection models often struggle to balance accuracy with real-time performance, particularly in scenarios where the ball is partially hidden or blends with the background. This study introduces an improved method for soccer ball detection and shooting, leveraging an enhanced version of the YOLOv8 model optimized for soccer match scenes and enriched with prior position information. The initial phase of our method involved developing a specialized dataset for soccer ball detection, incorporating video footage from various professional soccer matches. The core innovation of this study is the integration of the Efficient Local Attention (ELA) mechanism into the YOLOv8 framework, supplemented by the use of prior position information to predict the ball's trajectory and enhance real-time shooting accuracy. The ELA mechanism, enhanced by prior knowledge of the ball's likely positions, focuses on critical local features essential for accurate soccer ball detection, such as its shape and size relative to its surroundings. This focus improves detection in scenarios where the ball is partially occluded or moving rapidly and diminishes the effects of distracting background elements, leading to more consistent and reliable detection performance. To assess the effectiveness of our proposed method, extensive experiments were conducted using the newly constructed soccer ball dataset. The results, compared to the baseline YOLOv8 model, demonstrated a significant improvement in detection accuracy, with the mean Average Precision (mAP) increasing from 0.548 to 0.581.