Abstract:To address the contradiction between peak-hour congestion and off-peak resource wastage in urban traffic efficiency, a study was conducted on an intelligent signal optimization system based on the collaboration of perception and planning. The system employs the YOLOv8 framework as the core for vehicle detection and multi-object tracking technology, enabling real-time and accurate statistics of traffic flow in each lane and generating visual analysis reports. By integrating multi-dimensional parameters such as lane vehicle density, start-up time, safe spacing, and intersection structure, a multi-phase dynamic timing algorithm is adopted, combined with a weighting model to calculate the optimal green light duration, achieving adaptive signal control. Key technological innovations include the construction of a fully automated closed loop from detection to decision-making, and the deployment of lightweight YOLOv8 models, vehicle tracking, lane detection, and signal timing algorithms on an edge FPGA platform, leveraging its parallel computing capability to achieve ultra-low latency control. Experimental tests show that the system improves time utilization by 16%–20% during light traffic periods and by 14%–16% during heavy traffic periods. It supports one-time passage for platoons of 26–32 vehicles, effectively eliminating secondary queuing phenomena. Practical applications demonstrate that the system meets the engineering requirements of smart city construction in terms of enhancing road network operational efficiency and travel experience.
