Abstract
This thesis examines computer vision applications for traffic monitoring and safety analysis. The focus is comparing proprietary computer vision technology and open-source computer vision code we developed. Two case studies are completed using proprietary sensors purchased from a company called Numina. The first case study demonstrates unique opportunities for using computer vision to monitor bicycle travel during snow events. The case study reveals that bicyclists do not use separated bike lanes if they are not adequately cleared of snow. The second case study, using proprietary software, further shows how computer vision can create pedestrian and bicyclist demand models. Three case studies were completed using our computer vision code. The code was written in Python and uses a detection model called YOLOv8. The first case study demonstrates how user counts can be obtained from video feeds and provides examples of insights that can be drawn from these counts. The second case study uses computer vision to create visualizations of user movements at intersections. The third case study develops and demonstrates the application of a new surrogate safety measure for pedestrian and bicyclist safety. Advantages and disadvantages of proprietary and open-source systems are discussed. Shortcomings and future opportunities of computer vision are discussed.