With the strong support of relevant national policies, this project designs a traffic signal intelligent control system based on the maximization of traffic efficiency. Through the real-time observation of the traffic flow at the intersection of two-way multi-lane and crosswalk, the contradiction between pedestrian violation and vehicle efficiency is comprehensively considered, and the concept of combining "manual" and "intelligence" is adopted, accurately design vehicle detection and pedestrian detection algorithms and special vehicle recognition in the deep learning architecture to minimize the degree of mutual interference between pedestrians and vehicles, maximize the efficiency of vehicle and pedestrian traffic at intersections, and ensure the quality of urban mixed traffic flow. The research shows that the intelligent signal light system based on adaptive fuzzy control and deep learning YOLO algorithm is superior to other fixed timing schemes and traditional pedestrian request signal light schemes in terms of pedestrian-vehicle conflict, vehicle queue length, waiting time and average speed. The project can reduce the total cost of the transportation system through public-private cooperation, and transform the existing traffic lights on the basis of large-scale rapid equipment. The rules are simple and easy for drivers and pedestrians to understand.