Vehicle Information Recognition Of Multi-angle Based On Improved YOLOv3

In recent years,the number of motor vehicles in China has been increasing year by year,and road traffic accidents have occurred frequently.Therefore,it is imperative to build an intelligent traffic monitoring system with powerful functions,efficient management and real-time information transmission.Vehicle is the key research target of intelligent traffic system,and vehicle information recognition is an important part of intelligent transportation systems.It has extremely important practical significance for traffic management departments to crack down on illegal activities involving license plates,ensure urban road safety,and improve road management.Considering the types of vehicles driving on actual urban roads,this paper is the first to detect and identify motorbikes and bikes.At present,at home and abroad are mainly the front face of the image data of vehicle models,vehicle colors,and vehicle brand recognition,due to the difference in the location of the road surveillance camera and the surrounding environment,the collected road vehicle images usually show multiple poses,multi-angle and multi-light distribution,which brings challenges to the detection and recognition of vehicle types,colors and brands.In view of the above problems,this paper based on the YOLOv3 detection and recognition algorithm,improved the YOLOv3 algorithm and designed the vehicle type,color and brand recognition network model,and built a multi-angle vehicle information recognition system on this basis.The system can capture and save video frames containing vehicle pictures in videos of road running vehicles,and detect and identify vehicle types,colors and brand information in vehicle pictures from different angles.The multi-angle vehicle information recognition system includes three parts: road vehicle video detection,image preprocessing and vehicle information recognition.Using the vehicle type,color and vehicle brand recognition network and other detection and recognition algorithms constructed in this paper,comparative training is performed on the self-built multi-angle vehicle dataset Vehicle8T＿10C＿157B to obtain the optimal vehicle information recognition network.The road vehicle video preprocessing module combines Gaussian mixed background modeling and background difference algorithm to realize the detection and storage of vehicle-containing video frames in vehicle video.The image preprocessing module uses image processing related algorithms to denoise,de-blacken and weaken the exposure of the vehicle pictures intercepted by the vehicle video detection to enhance the low-quality image,and then send it to the subsequent vehicle information recognition module to carry out the detection and identification of vehicle type,color and vehicle brand information.The experimental results show that the multi-angle vehicle information recognition system designed in this paper can effectively identify 8 vehicle types,10 vehicle colors and157 vehicle brands under eight angles.In the test experiment,the recognition rate of vehicle type can reach 97.3%,the recognition rate of vehicle color and vehicle brand is 94.2%,which proves the reliability and applicability of the system.