Research On The Vehicle Logo Recognition Based On PHOG And ORB

With the increasing number of vehicles,the traffic problem in China is becoming more and more serious.The acquisition and identification of vehicle identity is becoming a research hotspot in the intelligent transportation system,and vehicle logos,as one of the most distinguishing marks of vehicle identity information except for license plates,are very important for vehicle identity.However,due to the small size and the high similarity among different vehicle logos,it is a great challenge for the research on vehicle logo recognition.Aiming at the problem of low robustness and poor real-time performance of existing technologies,this thesis proposes a new vehicle logo recognition algorithm,which includes vehicle logos location and vehicle logos identification.The main work and innovations are as follows:In the stage of vehicle logos location,this thesis proposes a vehicle logo location method based on prior knowledge and background texture suppression.The license plate is priori located based on Adaboost with LBP feature extraction.Then vehicle logo candidate area is obtained according to the location structure characteristics between the license plate and logo.Furthermore,this thesis uses texture suppression combined with the Sobel operator and Gabor filter to remove the background of the vehicle,and the precise location of the vehicle is obtained.In the stage of the vehicle logo recognition,a two-stage recognition method based on PHOG and ORB is proposed.In order to solve the problem of high computation and high feature dimension of traditional PHOG features,the overlapped block sliding strategy is adopted in the PHOG.By using the overlapped block PHOG features combined with the SVM classifier,the first-stage vehicle logo identification is completed;in order to increase the robustness of the proposed recognition algorithm,the classification confidence is calculated according to the probability estimate of the classifier in the first-stage recognition.When the confidence is lower than a threshold,the template matching method based on ORB is used for second-stage recognition.After two-stage recognition,the vehicle logo recognition can be obtained accurately.In this thesis,the algorithm is put forward by the analysis of the experiment result,the public database and the self built database are respectively used to test the algorithm respectively.The experimental results show that the proposed method has high accuracy and time efficiency,and a good robustness.