Research And Application Of Vehicle Recognition Algorithm Based On The Combination Of Crowdsourcing And Deep Learning

In recent years,the number of motor vehicles have continued to grow,at the same time,the resulting shortage of traffic congestion and traffic accidents has become increasingly prominent.However,the emergence of intelligent transportation system become a significant way to solve traffic problems.Vehicle recognition technology is a critical technology of intelligent transportation system which has wide application in traffic management,such as pursuit of traffic offenders and tracking of vehicles with false and occluded license plate.At present,in the case that the license plate cannot be identified and the existing vehicle recognition technology can only realize the simple classification according to large,medium and small model,it can not meet the needs of traffic management such as pursuit and tracing.Therefore,how to quickly recognize and accurately classify the vehicle's brand and model information without the license information are focus and direction of study in current vehicle recognition methods,Besides its research has practical significance.Learning from the research results at home and abroad in the field of computer vision,this paper presents a model of vehicle recognition algorithm based on the combination of crowdsourcing and deep learning technology.The algorithm model uses crowdsourcing technology to obtain a large number of vehicle dataset labels and uses deep learning technology to achieve vehicle detection,recognition and classification.The research work of this paper is as follows:(1)Research on vehicle recognition algorithm based on deep learning.On the basis of the analysis of existing vehicle recognition methods,the vehicle recognition method based on deep neural network model is mainly studied.The main idea of the method is to use the deep learning target detection algorithm first to detect the target vehicle in the original image,achieving the distinction between the background and the target.Then,using the deep neural network model to accurately recognize and classify the extracted target vehicle type.The experimental results show that the ResNet model can obtain the vector representation of deeper vehicle features and has the highest accuracy in the vehicle recognition.(2)With the problem of slow training speed,a large amount of calculation and long training time in the established vehicle recognition network model,so an optimized ResNet model based on MobileNet theory is presented.The model mainly uses the MobileNet model structure to optimize the residual block of ResNet.The experimental results show that the optimized ResNet model has light-weighted structure,which not only ensures the accuracy of recognition,but also shortens the time of network training.(3)Research on vehicle recognition algorithm based on the combination of crowdsourcing and deep learning.In order to obtain more labeled vehicle dataset and solve the problem of time-consuming with high cost of marking in the data marking process,this paper adopts the crowdsourcing mode to label the vehicle dataset.Due to the high complexity of the existing crowdsourcing quality control algorithm and the poor ability to learn the correct label and labeler's capability level,a model of vehicle recognition algorithm based on the combination of crowdsourcing and deep learning is given.The experimental results show that the model can achieve higher recognition accuracy and verify the effectiveness of the model.(4)Implementation of the vehicle recognition platform.Based on the vehicle recognition model of crowdsourcing and deep learning,this paper designs a prototype of the vehicle recognition platform.Using the function of the platform can realize recognition and classification of vehicle as well as search and track specific vehicle information in the vehicle database.