The Research On Vehicle Recognition Based On Invariant Feature And 2DPCA

License plate recognition,vehicle logo recognition,vehicle type recognition are important parts of the intelligent vehicle recognition system,They are widely used in the monitoring of vehicles on the road,parking lot management and vehicle violation records.Vehicle logo recognition is an important research field of intelligent transportation,and it actually means using the digital image processing and artificial intelligence method to recognize vehicle logo so that to get the information of the vehicle’s brand.Vehicle automatic recognition system has great practical significance.Vehicle logo recognition includes,in general,location and recognition of vehicle logo,and there is no universal method to the logo location.How to locate the vehicle logo from the image contains complex backgrounds,and accurately and efficiently identify the vehicle logo,do become important issues in vehicle identification system.This paper presents a two-stage logo positioning method based on scale invariant feature after analyzes some kinds of methods of vehicle logo location.Firstly,this method uses SURF feature locates the first character of the license plate,then the logo candidate region is determined according to the first character and then detects the vehicle logo precisely in the candidate regions by using Gabor features combine with BP neural network method.This method can avoid time-consuming caused by obtaining the license plate position and error localization caused by which relative formulas for vehicle license plate and logo are not accurate in the traditional vehicle logo location method based on a priori knowledge,it can also avoid the mathematical morphology method dealing with the limitations of scattered heat supply network around the logo.At the stage of recognition of the logo,this paper takes 2DPCA as the core,and a series of recognition rate and efficiency to PCA,PCA+LDA,KPCA+LDA,2DPCA,MBDPCA,2DKPCA,Mode 2DPCA,2DLDA,2DPCA+2DLDA of vehicle logo recognition method based on algebraic feature extraction have been compared and analyzed.Finally,the thesis also puts forward a vehicle logo recognition method based on deep learning of the deep belief network(DBN),and a comparative analysis on artificial feature selection method appeared in this paper has been presented.This paper established a non-constrained vehicles library which can reflect the environmental change to test the algorithm,the results proved the method of vehicle logo location based on scale invariant feature SURF presented by this paper has a better performance,and the vehicle logo recognition method based on 2DPCA+2DLDA has the advantages of in speed and recognition rate.Vehicle logo recognition method based on DBN in directly to the image pixels as input to obtain a satisfactory recognition rate and can meet the requirement of real-time.