RBFPDet: An anchor-free helmet wearing detection method

Wearing a safety helmet can reduce the accident rate in production and construction, and it is a necessary part of safety production management. At present, the effective supervision of helmet wearing still relies on manual on-site work, which is inefficient and wastes manpower and material resources. Therefore, the automatic supervision of helmet wearing detection is of great significance. Due to the detection difficulties such as small helmet target, complex background and variety of helmet shape with the posture of workers, helmet wearing detection has always been one of the most difficult tasks in the field of computer vision. To address this issue, we propose a novel object detection model based on anchor-free mechanism-Recurrent Bidirectional Feature Pyramid Detector (RBFPDet). Different from most other detection methods, we regard helmet wearing detection as a strong semantic feature points detection task. In order to prove the effectiveness of our method, we conduct control experiment and ablation study on two mainstream safety helmet wearing datasets. The experiment results show that our method significantly improves the accuracy of helmet wearing detection compared with other outstanding detection models in this field and our model can realize real-time detection under complex background. At the same time, we further intuitively illustrate the effectiveness of our method by means of feature map visualization.