Attention Guided YOLOv3 for Wearing Safety Helmet Detection

Abstract

Wearing safety helmet is of great importance to ensure the personal safety in power substations. The task of safety helmet detection still requires manual effort which is time-consuming and laborious. To address this problem, this paper introduces a computer vision method for automatic safety helmet detection based on YOLOv3 algorithm. We propose an attention mechanism which contains spatial-wise and channel-wise attention modules, to separately enhance the low-level and high-level features of deep convolutional encoder. Equipping YOLOv3 with our attention mechanism, the detector can improve its ability of detecting small objects, which is suitable for real-world safety helmet detection. Experiments on a publicly available helmet wearing detection dataset show that the proposed method is able to achieve good performance while running at a high speed.