Quantitatively Evaluate Work Zone Driver Behavior Using 2D Imaging, 3D LiDAR, and Artificial Intelligence in Support of Congestion Mitigation Model Calibration and Validation

This project addressed the following: 1) a lack of a system to process raw traffic footage obtained from work zones to extract traffic and driver behavior information, especially systems that leverage recent advances in computer vision and artificial intelligence (machine learning), 2) a lack of understanding of the factors impacting the accuracy of the data extracted using an AI-system, and 3) a lack of quantitative analysis that presents the potential benefits of using the extracted real-world traffic and driver behavior information in the work zone traffic simulation models in comparison to default values. As a result, a preliminary version of an AI-based work zone traffic and driver behavior information extraction system using widely available 2D camera images, machine learning, and computer vision was developed to extract real-world traffic and driver behavior information, including vehicle count, vehicle classification, vehicle speed, time headway, and lane change location. A case study using the real-world videos collected on I-95 has demonstrated that the data extracted using the preliminary version of the developed AI-system is promising and can be analyzed to obtain accurate and refined real-world work zone traffic and driver information. It can provide valuable input that has previously not available to transportation agencies for developing appropriate traffic control strategies to manage and/or improve safety and mobility in work zones, and it can greatly help develop accurate and reliable traffic simulation models. This research focused on studying the feasibility of using AI technology to extract traffic information to enhance traffic simulation. A separate effort with a pilot study with a large diverse data set is recommended in the future to further validate, refine and implement the proposed method.