Method of Safety Helmet Wearing Detection based on Key-Point Estimation Without Anchor

Safety Helmet wear detection is a very important task in the field of industrial applications which can greatly reduce the safety risk and provide a guarantee for better industrial production. Most of the popularly used detection methods realized by enumerating all the possible locations of detection objects, the classifier is then run to identify the final rectangular bounding box that wraps the target and the category to which it belongs. In this paper, we design a multi-scale key-point network to solve the size difference of the objects, coming up with a new loss function and training strategy to improve the accuracy of the result. Our method achieves 100 FPS 92% mAP on the SHWD dataset, which achieves the best trade-off between speed and accuracy.