Application of deep learning in video target tracking of soccer players

Abstract

Football matches have a high degree of attention and the analysis technology used for video contents has important practical significance and good application prospects. However, due to the diversity of conditions, i.e., football venues, clothing colors, etc., there is no universal tracker that can perfectly adapt to all scenarios. Due to its excellent feature extraction capabilities, deep learning technology has been widely used in the field of computer vision in recent years. The main objective of this article is the extraction of player’s trajectory in a football game, i.e., the path tracking of the player’s goal. To achieve this objective, deep learning technology is used for automatic extraction of the characteristic features of the player’s target in the context of target detection and tracking. The target detection method employed in this study is based on deep learning by forming new multi-scale features and modifying the generation rules of anchor points of the captured videos, making it more suitable for small target detection tasks in football match scenes. The generation rules are based on a complex decision support system for target tracking. This decision support system uses the method of constructing a similarity matrix to transform the multi-target tracking problem into a data association problem that can be solved by the Hungarian algorithm. The proposed approach is compared against state-of-the-art techniques in terms of area under the curve (AUC) value, track set scene distribution, number of frames, and other parameters. Based on the experimental results, the proposed approach outperforms these existing techniques with much better results.