Integrating computer vision and traffic modeling for near-real-time signal timing optimization of multiple intersections

Abstract Adaptive signal timing optimizations can improve the efficiency of road networks and reduce the emissions of pollutants, but most of the current studies still rely on simplified analytical methods to depict complex road transport systems and focus on optimizing traffic signals at an isolated intersection. A framework that integrates computer vision and traffic modeling is proposed to link the real-world transport systems and operable virtual traffic models for the signal timing optimization at multiple intersections. The integrative framework consists of six main steps, including configuring real-time video sources, conducting transfer-learning to develop the vehicle detector, comparing and selecting vehicle trackers, collecting traffic parameters by referring to the CV-TM ontology, establishing and running the traffic model, and operating simulation-based optimizations. The proposed integrative framework is demonstrated through a case study of the signal timing optimization at multi-intersections in a real-world road network. Three critical information items including the traffic volumes, vehicle compositions, and vehicles’ turning ratios are derived from real-time surveillance videos, and the extracted information is then automatically incorporated into TM to optimize the signal timings of interconnected intersections in a near-real-time manner. In comparison with the original signal scheme, the optimized one can reduce 14.2 % of average vehicle delays, 18.9 % of vehicle stops, 9.1 % of average travel time, and 2.3 % of pollutant emissions in this specific case. The results indicate that synchronously optimizing signal timings at multiple intersections increase not only the transportation efficiency but also the environmental friendliness of road transport systems. The proposed CV-TM integration framework is demonstrated to be a promising way for conducting near-real-time signal timing optimizations in intricate traffic scenes instead of at isolated intersections, helping decision-makers to promptly respond to the time-varying traffic conditions during various real-world events, and facilitating the transportation systems and cities to achieve sustainable development goals.