Integration of Computer Vision, Microscopic Traffic Simulation and Heuristics for Optimizing Motorcycle-Dominated Signalized Intersections

Abstract

Traffic signal optimization in motorcycle-dominated environments remains a critical challenge in many developing cities, where the heterogeneity and high dynamics of traffic flows often limit the effectiveness of traditional control methods. This study introduces an approach that integrates computer vision, microscopic traffic simulation and heuristic optimization to design signals of motorcycle-dominated mixed traffic intersections. By leveraging visual data through modern object detection techniques, the proposed framework enables a more comprehensive and precisely of key traffic parameters including traffic volume and travel time - overcoming the limitations of conventional field surveys and loop detectors. These data are then utilized for developing and calibrating VISSIM models to accurately reflect reality. Rule-based and multi-start local search heuristics are implemented with VISSIM and Python to iteratively refine signal timing plans, aiming to minimize travel time and queue lengths at intersections. A case study conducted at a motorcycle-dominated intersection in Hanoi, Vietnam demonstrates the potential of this integration to improve both operational efficiency and adaptability of signal control systems. The chosen solution performs much better than the existing situation, with the average queue length and travel time reduced by approximately 52.5% and 16.3% correspondingly. The findings can prove the feasibility and accuracy of proposed integrated framework that traffic engineers and decision-makers might apply in motorcycle-dominated mixed traffic environments in practice.