Recently, object detection has become a popular research direction in the field of robot vision. Sports ball detection is one of the most challenging tasks because of high speed and tiny size, which is an essential part of a sports ball robot. In this paper, we focus on the detection of tennis balls. The tennis ball is tiny and flies very fast, which causes movement deformation. While some popular detectors, like YOLO family, still cannot effectively solve the challenge of detecting high-speed tiny objects. We conduct a lot of research and find that the anchor-based detector is not good at high-speed tiny object detection. Therefore, we propose an approach to detect tennis balls based on an anchor-free detector. The network architecture is altered to get better results. Moreover, a data augmentation method named "magnifier" is proposed for detecting extremely small tennis balls. Our approach is evaluated on the images of the tennis competition at 2017 Summer Universiade and the images including tennis in MS COCO dataset. The precision, recall, and f1-measure of our approach reach 91.77%, 93.43%, and 92.59%, respectively. Our approach outperforms YOLOv3 and achieves exceptional tennis detection performance.
展开▼
