Detection of violations in construction site based on YOLO algorithm

Abstract

At present, the safety management of construction mainly supervises similar incidents based on the data of safety accidents that have occurred, but ignores the inherent uncertainty of safety accidents. As a result, the existing warning system is only suitable for a single project and does not have universality. In this paper, deep learning method based on YOLO is used to detect violations such as not wearing safety helmet and smoking in industrial sites. YOLO algorithm is improved by embedding convolutional block attention module (CBAM) and an adaptive spatial feature fusion module (ASFF) to enhance the detection effect of violations. CBAM can emphasize important features and inhibit general features. ASFF module can make full use of features of different scales. This paper combines the advantages of both modules with the fast performance of YOLO algorithm to achieve real-time detection of violations. The experimental results show that the improved algorithm has a higher detection accuracy than the traditional algorithm. The detection AP for smoking increased by 1.7^%, wearing a helmet of AP is improved by 1.0% and a 1.7^% improvement in the mAP.