Vehicle Information Collection And Management System For Mobile Terminal

In July 2019,the Ministry of Public Security’s Traffic Management Bureau released data that in the first half of the year,the number of motor vehicles nationwide reached 340 million,of which 198 million were private cars.With the increasing number of vehicles,how to quickly complete vehicle collection and management anytime and anywhere has become the primary task of traffic management departments.Therefore,this paper designs a set of vehicle information collection and management system for mobile terminals,which is divided into two parts: cloud server subsystem and intelligent terminal subsystem.The cloud server subsystem is mainly responsible for vehicle information management tasks,including: vehicle information addition,single vehicle information query,multiple vehicle information query and vehicle information deletion,etc.The mobile terminal subsystem is responsible for the collection of vehicle-related information,in which the collected vehicle information includes: license plate number,vehicle geographic location information,time when the vehicle information is collected and ID of the vehicle information collector.Among them,license plate number collection based on intelligent terminal is the focus of this paper.The acquisition process of license plate number collection is: license plate detection,license plate correction,license plate character segmentation and license plate character recognition.In the vehicle information collection and management system for intelligent terminals,the main work completed in this paper is as follows:In the task of license plate detection,this paper adopts SSD-Mobilenet-v1 target detection algorithm based on lightweight convolution neural network,and adopts the idea of cascade detection,which effectively reduces the missed detection rate of license plate by adopting the cascade idea of first detecting vehicle and then detecting license plate.Because the license plate target is the target with a small proportion of the whole map,and SSD-Mobilenet-V1 has poor detection effect on small targets,the detection accuracy of the license plate is improved by expanding the size of the feature map generating anchor and increasing the proportion of the license plate in the feature map.Through the introduction of Mixup data enhancement strategy in the training phase of the algorithm,the detection accuracy of the algorithm is improved as a whole,and the mAP of the algorithm on COCO open source data set is improved by 0.3.In the stage of license plate character recognition,this paper designs a lightweight map classification network model LPR-net based on Lexnet-5.The minimum unit of network feature extraction adopts reverse residual structure,and the method of combining global pooling and point convolution is adopted to replace the full connection layer with large parameters.In order to improve the generalization ability of the model,Dropblock regularization is introduced into the network model to assist the convergence of the model.The classification accuracy of LPR-net in license plate character data set reaches 99.4%,which meets the application requirements in actual scenes.Through the demand analysis of vehicle information collection and management,the design and implementation of the system are completed.In the mobile terminal subsystem,the maximum time for license plate information collection is 1342 milliseconds and the shortest time is 520 milliseconds.Management functions such as deletion and query are implemented in the cloud server subsystem.Through experimental analysis,the system proposed in this paper effectively improves the efficiency of vehicle information collection and management and reduces the work intensity of vehicle management departments.