Sports Action Detection Based on Self-Supervised Feature Learning and Object Detection

Abstract

This paper presents an action detection method for sports videos based on self-supervised feature learning and object detection. Our previous method realizes action detection based on self-supervised feature learning. However, the self-supervised feature learning works well when a single person is on a video. Thus, we introduce object detection into our method to achieve action detection for multiple persons by tracking each person. The proposed method realizes action detection without a fine-grained annotation based on self-supervised feature learning. Experimental results using real-world sports dataset show the effectiveness of the proposed method.