Intelligent perception and personalized assessment of hazardous acts of construction workers

Abstract

Occupational injuries and illnesses among construction workers are not only related to the safety and health of workers but can also negatively affect the schedule and cost of a construction project. The immediate causes of occupational injuries and illnesses are hazardous acts. Real-time monitoring and assessment of hazardous acts with high accuracy not only improve workers' occupational safety and health (OSH) but also enhance project progress and cost management. However, previous hazardous acts monitoring practices have some inherent limitations such as insufficient accuracy and low stability on the complex construction sites. Few of the previous research have addressed the intelligent perception in hazardous acts monitoring in construction. Besides, previous OSH handbooks or international standards for hazardous acts assessment reflected only average acceptability levels of the workers' exposures to hazards. Such average acceptability levels might not be able to consider the individual differences. The primary aim of this thesis is to study and develop intelligent and personalized methods to monitor and assess hazardous acts of construction workers by addressing the problems of the existing methods. Specifically, literature reviews of previous related research are carried out. Then, four kinds of intelligent perception methods are respectively developed to convert on-site sensor or image data into real-time and accurate information of workers' hazardous acts. The first method is a wearable inertial measurement unit (WIMU)-based motion warning method to detect hazardous joint angle and holding time of body postures. The second method is a computer vision-based (CVB) view-invariant posture recognition method to monitor ergonomically hazardous postures and their frequency. The third method is an infrared sensor-based method to detect, locate, and warn non-hardhat-use (NHU) behaviors. The fourth method is a CVB view-invariant three-dimensional (3D) spatial proximity estimation method to monitor crowdedness of workers on the jobsite. Finally, a personalized hazardous acts assessment method is proposed with two components: one is a personalized motion warning method for recommending personalized acceptable holding time of working postures; the other is a personalized fuzzy inference-based hazardous acts assessment method for OSH performance appraisal and incentive among workers. Based on a series of experiments and field tests, the research findings from this study include: 1) The proposed WIMU-based motion warning method can assist workers in being timely aware of awkward working postures; 2) The proposed CVB posture recognition method outperforms the previous methods in terms of accuracy and view-invariance ability; 3)The proposed NHU detecting, locating, and warning method outperforms the previous methods in terms of NHU detecting accuracy, NHU locating accuracy, and the real time performance of NHU warning; 4) The proposed CVB crowdedness monitoring method enables the estimation of 3D spatial proximity within an error of 0.45 meters in a view-invariant manner; 5) With the assistance of the proposed personalized hazardous acts assessment method, the OSH performance of the recruited workers in the field test has been promoted. In summary, this research contributes to domain knowledge and methodologies for intelligent perception and personalized assessment of hazardous acts of construction workers by utilizing modern information technologies.