Automated classification of construction site hazard zones by crowd-sourced integrated density maps

Current onsite safety management always relies on time-consuming predefinitions of hazardous zones based on the managers' personal capabilities. However, in a typical labor-intensive industry such as construction, the workers themselves can provide a wealth of information for hazard identification. Historical accident-free working locations on site provide a valuable means of recognizing safe workplaces. This paper presents an approach to the automated classification of construction site zones derived from the location tracks of workers collected from a real-time location system (RTLS). Through data mining, filtering and analysis, the location tracks are transformed into grid density maps and continuous density maps. These illustrate the characteristics of spatial-temporal activities onsite as well as providing a visual representation of the distribution of safe and hazardous individual workplaces. A personnel hazard map is generated automatically based on historical accident-free location tracks from a field project using the proposed approach. Compared with the actual workplaces in terms of accuracy, precision, sensitivity and specificity, the evaluation result reveals that the hazardous areas on a construction site can be automatically classified to improve the workplace management of individual workers. The contributions of this research include an automated zone classification algorithm and an evaluation framework consisting of four indicators for hazard awareness onsite.