Traffic Target Detection Method Based On YOLO

With the rapid development of technologies such as big data,cloud computing,and mobile Internet,artificial intelligence has gradually matured,and unmanned driving has become possible.Autopilot technology not only reduces the burden on the driver,but also effectively reduces the occurrence of traffic accidents."Internet+Transportation" is the application of new Internet technologies to the field of intelligent transportation,enabling machines to identify a series of traffic targets such as vehicles,pedestrians,signal lights,traffic obstacles,etc.,thereby providing infrastructure for intelligent transportation and automatic driving.One of the most important technical points of autonomous driving technology is environmental perception.It uses image recognition technology to sense the surrounding environment and detect road vehicles,obstacles,traffic lights and traffic signs.In recent years,video image-based detection technology has begun to develop.This technology uses image acquisition device to perform real-time image acquisition on the detection target,and analyzes the collected data.However,how to improve vehicle safety performance,improve traffic target detection accuracy and detection speed is the key to achieving unmanned driving.In this paper,we study the characteristics of traffic targets,and realizes the reliability and real-time characteristics of autonomous driving.It analyzes and studies the specific traffic targets such as traffic lights and traffic obstacles.Firstly,we collect the traffic road images of multiple cities using the traffic road video captured by the on-board camera,and preprocess the images.Secondly,a total of seven types of traffic targets for traffic lights and traffic obstacles are manually labeled.By comparing the traditional traffic target detection method and image-based traffic target detection method,a YOLO real-time target detection method is used to realize the traffic target detection task in this paper.The latest two versions of YOLO algorithm are compared and analyzed,and for different network models of different structures,experiment with these traffic targets,train data,and obtain test models.Finally,the applicability of the two models to traffic detection tasks is analyzed from the aspects of precision,recall rate,mAP and detection speed.The experimental result shows that the use of YOLOv3 for traffic target detection can achieve a real-time detection effect,and there is a certain improvement in detection accuracy.