Research On Vehicle Recognition Based On Optimal Transport Model

In recent years,with the rapid increase of the number of vehicles,a series of problems such as vehicle supervision difficulties and frequent traffic accidents have emerged.Therefore,it is of great significance to study a high-speed and effective vehicle identification system.Existing vehicle recognition technology can get great performance in the recognition of a coarse-grained tasks.But when there exist a variety of problems such as complex image background,different vehicle angles and various vehicle types,existing model still unable to obtain a good efficient.In order to solve these problems,this paper puts forward the model of vehicle recognition based on optimal transport model,the main work includes.Firstly,summarizing the research status of vehicle recognition technology at home and abroad,and analyzing the problems existing in the current vehicle recognition technology.Secondly,introducing the basic principle and the commonly solution method of optimal transport problem.By using the approximate solution scheme of Sliced Wasserstein distance,a group of optimal transport kernel functions are proposed,and an experiment is designed on the data set of bit-vehicles to compare the results of proposed kernel function and common kernel functions.Experimental results show that the proposed optimal transport kernel function can significantly improve the performance of vehicle recognition.Finally,based on the above theoretical research and experiments,a vehicle recognition model based on optimal transport model and multi-part feature fusion is designed to solve the problems existing in fine-grained vehicle recognition,which combining the optimal transport kernel and positioning network.The experimental results show that the accuracy rate of this model is 84.4%in the fine-grained vehicle data set bmw-10,which is greatly improved compared with other models.