INTEGRATION OF ARTIFICIAL INTELLIGENCE AND HUMAN FACTORS IN MOBILE WORK ZONES AND ROUNDABOUTS FOR SAFETY AND PERFORMANCE MONITORING

The transportation system is facing serious safety concerns at work zones and intersections, which are two major areas where accidents and fatalities occur. In addition, slow improvement in transportation industry workers’ performance is also a bottleneck to overall productivity. This dissertation aims to integrate artificial intelligence and human factors to improve the safety of mobile work zones and unsignalized intersections and monitor real-time worker’s performance.

To improve work zone safety, the Autonomous Truck Mounted Attenuator (ATMA) technology is explored with support from the Indiana Department of Transportation (INDOT). The ATMA can be driven automatically which removes drivers from the TMA truck to improve their safety. In this study, the ATMA system was tested under four mobile work zone operations, including trash pickup, crack sealing, Raised Pavement Marking (RPM) inspection, and drainage inspection with several roadway types, including interstate, trunk highway, and state road. During the testing, video, motion, and physiological data from the workers is collected. The data is used to develop models for transportation construction workers’ activity classification and physical fatigue level monitoring using various machine learning techniques. In addition, workers’ perception of the ATMA system is collected by a survey and the results found that more training or exposure to the ATMA system improved their evaluation of the system.

To improve unsignalized intersection safety, an in-vehicle warning system is developed and evaluated under various levels of aggressive vehicle behaviors across different warning conditions through a driving simulator study. A customized driving simulator is developed to support human driving experiment, which integrates SUMO and Webots. A real-world roundabout is built and calibrated in the simulator and both driving performance and eye movement data are collected from the experiments. The results indicate that advanced warnings can effectively influence vehicle speed, steering wheel control, and drivers’ attention on different areas of interests (AOIs). It is found that a proper warning time is critical to improve drivers’ safety and comfort. Gender differences are also identified from both types of data. Interestingly, although male drivers and female drivers demonstrate different driving behaviors, their safety performance in terms of minimum time to collision (TTC) is similar. Finally, to better facilitate the design of the advanced warning systems, two machine learning models are developed to predict minimum TTC and classify drivers’ perceived risk.

The contributions of this dissertation are summarized from the following four perspectives. First, this dissertation contributes to the body of knowledge by using a Deep Learning (DL)-based model for mobile work zone workers’ activity classification. The dissertation also innovatively integrates domain knowledge to refine the DL-based model’s performance. Second, this dissertation advances the application of feature-level data fusion in monitoring transportation construction workers. Specifically, the feature-level data fusion between kinematic and physiological data is found effective in improving model accuracy. Third, to improve mobile work zone safety, the ATMA system is tested with various road maintenance activities. This is the first ATMA test with a focus on mobile work zone operations with human workers working on the ground. The testing results are valuable for the future ATMA design and implementation. Fourth, this dissertation discloses the positive impacts of in-vehicle warning systems in roundabout merging scenarios. Furthermore, a customized driving simulator is developed to support human driving simulation experiments and is open-sourced for public use.