Hazardous Behavior Identification Based on BIM and AutoML Applied to Prefabricated Construction

With the continuous development of China’s construction projects, safety hazards have emerged endlessly. Safe and standard construction practices are essential in construction projects. In recent years, construction safety accidents caused by construction workers not wearing helmets correctly and safety accidents caused by workers smoking and igniting flammable materials to start site fires have accounted for a large proportion of the safety accidents that have occurred on major construction sites. However, traditional manual supervision is time-consuming, inefficient, subjective, and sometimes challenged by conflicts of interest. Over the years, deep learning-based target object detection has achieved satisfactory results, but the complex deep learning setup is not easily mastered by construction personnel. Therefore, it is of great practical importance to prevent and reduce construction safety accidents by automatically detecting behavioral norms for personnel entering job sites. To this end, this paper presents an automated machine learning (AutoML) framework for helmet detection and smoking recognition. This paper aims to benchmark the AutoML algorithm for helmet detection and smoking. The final results of the experiments show that for helmets, AutoML could achieve 99.0% precision, 91.89% recall, and 99.24% mAP, which demonstrates the usability of AutoML in the construction datasets.