UAV Low-Altitude Remote Sensing Inspection System Using a Small Target Detection Network for Helmet Wear Detection

Abstract

Automated construction site supervision systems are critical for reducing accident risks. We propose a helmet detection system with low-altitude remote sensing by UAVs in this paper to automate the detection of helmet-wearing workers to overcome the limitations of most detection efforts that rely on ground surveillance cameras and improve the efficiency of safety supervision. The proposed system has the following key aspects. (1) We proposed an approach for speedy automatic helmet detection at construction sites regularly leveraging the flexibility and mobility of UAVs. (2) A single-stage high-precision attention-weighted fusion network is proposed, allowing the detection AP of small-sized targets to be enhanced to 88.7%, considerably improving the network’s detection performance for small-sized targets. (3) Our proposed method can accurately categorize helmets based on whether they are worn or not and the type of helmet color, with an mAP of 92.87% and maximum detection accuracy in each category.