Research On The Recognition Of Vehicle Attributes Based On Deep Learning

Today,the car has become the preferred vehicle for people traveling.With the increasing number of vehicles,the recognition technology of vehicle attributes is being concerned by more and more scholars.At present,the cars of fake license and clone are rampant and crime cases involving motor vehicles are frequent.These had increased the pressure on traffic brigade and police.The traditional recognition technology of vehicle type and color had become overwhelmed when dealing with similar cases.So it is important to identify multiple attributes of a vehicle simultaneously.Therefore,new model is proposed in this paper and the main research as follows.(1)A model of vehicle type recognition based on deep learning is proposed.First,it overcomes the difficulties of high requirements for image preprocessing and feature extraction in the traditional method.Utilizing the machine learn from the sample data autonomously to obtain global features of low ambiguity based on convolution neural network which can achieve detailed vehicle classification.Second,we reduce the network parameters by improving the full connection layer of network which can shorten the test time of the target image and enhance the real time effectively.(2)A model of vehicle type and color recognition based on deep learning is proposed: we turn the single model single label model into the single model multi-label model,which can overcome the difficulty of only one vehicle attribute can be recognition of depth learning model.On the basis of vehicle type recognition,we added color recognition which can improve the accuracy of query for a specific vehicle.Therefore it can narrow the search effectively and lock the target vehicle information quickly in the criminal investigation.In this paper,we collect 215235 vehicle photos from the traffic jar to build the dataset.The experimental results show that the proposed methods have good recognition result and high recognition accuracy.In particular,the effect of large-scale vehicle type and color recognition of different year under the same vehicle sub-brand is better.