Design and Implementation of Vision Based Safety Detection Algorithm for Personnel in Construction Site

The failure of construction workers to wear safety equipment as required is an important cause of safety accidents. In fortunate cases people are injured and severe cases can cause death. In order to prevent this from happening, this paper proposes an algorithm consisting of data augmentation for complex environments and object occlusion, more than 17,000 sample images and an advanced neural network model faster R-CNN. The experimental results show that the method can identify the personnel and helmets in the surveillance video in real time, and the average recognition accuracy is up to 90.3%, and the detection time is up to 0.037s per image. It satisfies the safety requirements of real-time detection of construction sites.