Cascade RCNN with Hybrid Attention and Dual Pooling for Soccer Player Detection

Sports video analysis is of great significance to football players’ tactical deployment and daily training, and player detection is the basis of football intelligence analysis. However, in sports videos, there are problems such as the camera is difficult to capture fast athletes, and dense crowds causing occlusions. It is difficult for existing target detection algorithms to accurately identify athletes. To this end, based on Cascade RCNN, this paper proposes an anti-blurring and anti-occlusion football player detection model. First, we propose a novel Hybrid Attention Module (HAM) to refine the feature pyramid and enhance features for object occlusion and blurring. In addition, we introduce dual pooling (DA) to capture context information and enhance feature expression when providing candidate regions for the subsequent RPN. We verified the performance of our method on the football player detection data set. The experimental results show that our method is superior to other models, and can effectively solve the problems of missed detection and false detection of athletes caused by occlusion and blurring. It can be effectively applied in Sports video analysis systems.