Research On Vehicle Detection And Multi-dimensional Rocognition Technology Of Vehicle Information Based On YOLOVv3 And MSER

Real-time detection and extraction of vehicle information in complex natural scenes have always been one of the important contents in the field of computer vision.The breakthrough of problems in this field can not only bring effective promotion to the realization of auto driving technology,but also have important practical significance in the improvement of real-time parking monitoring system and automatic parking schedule algorithm in parking lot.Besides,in the process of vehicle information extraction and recognition after vehicle detection,the precision of vehicle plate text location is a key and fundamental step of the module’s performance.In this paper,the above key problems are studied,the main work as follow:1.In order to solve the problems of incomplete vehicle information detection area,low accuracy and inability to locate distant vehicles in real-time vehicle information detection,a new real-time vehicle detection and classification model—Vehicle-YOLO is proposed.Based on the latest yolov3 algorithm model,the model improve the accuracy and universality of vehicle real-time information detection by changing image parameters,enhancing features extraction ability of deep residual network,using 5 different scale convolutional feature maps to extract potential bounding boxes of vehicles.Then test and analyze the model performance with datasets,such as KITTI and PASCAL VOC.The experimental results indicate that Vehicle-YOLO model achieves 96% mean average accuracy and transmission speed about 40 fps on the dataset of KITTI,in other words,the model can still maintain a better real-time detection speed when detection accuracy is improved.In addition,the experimental results of Vehicle-YOLO on other datasets such as VOC also show different degrees improvement of accuracy,which demonstrate that the model has better universality and performance in common object positioning detection than traditional algorithms model of object detection.2.A new method of license plate location based on Maximally Stable Extremal Regions and Convolutional Neural Network is proposed in order to solve the problems of low accuracy,strong noise and interference factors in complex scenes.The method takes advantage of MSER to find more stable binary sub-images,and then,according to prior knowledge of license plate,some sub-images that do not conform to features of license plate characteristics obviously are filtered out,after that,makes use of heuristic knowledge and CNN to recognize true license plate character and find out their location in the complete image.Finally,sliding window of CNN isutilized to search for beginning and ending positions of a license plate.The experimental results indicate that the proposed algorithm has strong robustness,harder to impact by complex environment and improves accuracy of license plate text location.3.Vehicle detection and vehicle information recognition system is designed and implemented.First,vehicle in an image is located and cropped by Vehicle-YOLO model.Then,the vehicle license plate is obtained by using license plate recognition algorithm based on maximum stable extreme value area.Finally,color and brand of vehicle is recognized through CNN model.Experimental results demonstrate that the classification information of the vehicle obtained during the locating process can help the performance of the subsequent recognition module,and verify the practicability of the method given in this paper.