Improved Signalized Intersection Performance Using Computer Vision and Artificial Intelligence

Details

• Creators:
  Moomen, Milhan ; Meng, Xiangyu ; Sarker, Tonmoy ; Essuman, Jones B. ; Addo, Collins
• Corporate Creators:
  Louisiana State University (Baton Rouge, La.) ; University of Louisiana at Lafayette ; Louisiana Transportation Research Center
• Corporate Contributors:
  Louisiana Department of Transportation
• Subject/TRT Terms:
  Artificial Intelligence Computer Vision Data Analysis Deep Learning Feasibility Analysis Intelligent Transportation Systems Multi-object Tracking Performance Measurement Signalized Intersections Tracking Systems
• Publication/ Report Number:
  24-4SS
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report; January 2024 – December 2025
• Corporate Publisher:
  Louisiana Transportation Research Center
• Abstract:
  Signalized intersections are critical points in urban transportation networks where congestion, delays, and safety risks are most prominent. Traditional approaches for performance evaluation rely on manual field counts, loop detectors, or expensive infrastructure-based systems, which are often limited in accuracy, scalability, and adaptability. This project explored the use of computer vision and artificial intelligence to develop automated tools for intersection performance analysis. Three primary methods were investigated: (1) a vehicle counting framework based on detection–tracking–counting pipelines; (2) a queue detection approach integrated with traffic light state recognition; and (3) pedestrian behavior and interaction. Vehicle counting was implemented using a virtual line-crossing strategy combined with advanced detection and tracking models, enabling accurate measurement of turn movements across multiple lanes. Queue detection, in turn, was achieved by associating detected vehicles within lane-specific regions of interest with real-time traffic light states, providing insights into demand and delay at intersections. The pedestrian-vehicle interaction is based on trajectory extraction at the intersection. All methods were tested on drone- and camera-based video datasets collected from Louisiana intersections. Results demonstrate that the proposed algorithms achieved high accuracy, robustness to environmental variations, and efficiency suitable for near-real-time applications. A user-friendly graphical interface was also developed to allow engineers to apply these methods to raw video data, facilitating data-driven decisions for signal timing, intersection design, and congestion mitigation. The study highlights the feasibility of AI-based computer vision systems as cost-effective, scalable, and reliable alternatives for traffic performance monitoring.
• Format:
  PDF
• Funding:
  DOTLT1000515
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:1c9dc371d1083590f5bdadb11cf5ce8e8b52ea5da15919e874b6f6441b83902239a7402493e7cf7b0355299077a08240a7ab3b3d4872d0cabaf4bb31b6330b5c
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/92000/dot_92000_DS1.pdf
• File Type:
  [PDF - 4.77 MB ]