PFI-TT: Broadening Real-Time Continuous Traffic Analysis on the Roadside using AI-Powered Smart Cameras

The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project will allow democratization of advanced Artificial Intelligence (AI) and computer vision technologies to enable convenient and efficient traffic analysis for transportation system engineers and local departments of transportation. Cities face various traffic-related problems that need high-quality data to solve, such as congestion, accidents, and road overload. With the smart roadside camera device developed from this research, traffic operators can obtain real-time vehicle trajectory data at a much lower cost and conduct traffic studies or make prompt decisions to improve road safety and transportation efficiency. Moreover, the 21st century is the era of information, and "data is the new oil" of digital prosperity. The technologies generated from this proposed research can potentially spawn a set of new applications for road transportation that benefit the nation's economy, such as cheaper insurance with a fine-grained driving behavior analysis, safer streets with accident alerts from roadside cameras, and better accommodation of future automated vehicles with the insights from the data obtained through this technology. Additionally, this research will broaden the research participation in a primarily undergraduate institution and a Hispanic Serving Institution, as well as train future leaders in innovation and entrepreneurship. The proposed project focuses on novel technologies that allow traffic researchers to collect fine-grained vehicle trajectory data that existing road sensor systems cannot provide, analyze data in real-time, detect and classify different vehicle types, estimate vehicle speeds and trajectories, and predict potential collisions and accidents. Specifically, the project will develop three technologies: First, a novel 3D vehicle model-based solution for accurate vehicle tracking and localization using vehicle key point and Artificial Intelligence (AI); Second, a method to accelerate AI on low-power edge devices deployed on the roadside for real-time performance through AI model compression and efficient parallel computing on the graphical processing unit (GPU); Third, robustness improvement of machine vision in twilight by leveraging a novel type of camera and multiple views. To demonstrate the usability of these novel technologies, a set of road traffic data analysis applications will be developed, including real-time online traffic visualization, fine-grained traffic counting, road safety analysis with quantitative metrics, and traffic incident detection. The outcome of this project will be realized in a roadside smart camera device that can track and localize vehicles in the 3D space with decimeter-level accuracy, real-time efficiency, and robustness under low lighting conditions. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.