Intelligent Traffic Signal Timing Optimization Method Based On Improved YOLOv5 Target Detection Model

Video surveillance and target recognition and tracking have become an important part of the current development of intelligent,informatized,and technological society.Research and application are more common in the construction of intelligent transportation systems.The rapid increase in the number of motorists and cars brought about by the steady improvement has made the congestion problems in traffic scenarios more and more serious,such as highspeed congestion on holidays,road congestion in the morning,middle and evening rush hours,and traffic congestion at intersections.Serious traffic jams not only affect people’s travel efficiency and the consumption of fossil resources,but also easily lead to traffic accidents,resulting in the loss of life and property of society and people.The optimal timing of traffic lights at intersections can improve the traffic capacity of vehicles at intersections,effectively help traffic departments solve the state of intersection congestion,and provide residents with better travel routes conveniently and effectively.Due to the complex environment of vehicles at intersections,the requirements for vehicle detection in different states are relatively high,and the requirements for real-time performance are very strict,so it brings arduous task challenges to vehicle detection and optimal timing of traffic lights at intersections.The traditional target detection and tracking algorithm uses a sliding window to detect the target object based on the area to be measured,which has low detection accuracy,single pertinence,high complexity,long operation cycle,and poor robustness.This paper proposes a real-time traffic flow detection method at intersections based on the improved YOLOv5 and DeepSort algorithm models.The improved YOLOv5 network model algorithm is used to achieve video small target vehicle detection,occlusion detection and complex environment detection.The deep learning multi-target tracking algorithm DeepSort algorithm is used to track the detected vehicles in real time,and inverse perspective transformation is performed on the delineated area of interest to better realize vehicle counting,and realize end-to-end real-time traffic flow detection at intersection monitoring.The experimental analysis and comparison show that this method has faster detection speed,better vehicle detection effect,higher accuracy and better real-time performance in the environment of complex environment,vehicle occlusion and high target density.The model fully meets the requirements of real-time detection of objects,can fully meet the effectiveness of vehicle detection at intersections,and meet the actual needs of use.Finally,the vehicle data of the four lanes obtained by the traffic flow detection is optimized by the improved adaptive genetic algorithm,and the effective green light is considered as the independent variable,and the vehicle average delay time function is the objective function.Next,a timing optimization model for traffic lights at intersections is constructed.Through the analysis and comparison of genetic algorithm,improved genetic algorithm and fixed duration,the effectiveness and feasibility of the algorithm for optimizing the timing of traffic lights at intersections are verified.