The Research Of Vehicle Recognition System Based On Image Processing

With the development of Pattern Recognition and Computer Vision Technology,vehicle recognition technology based on image processing has received more and more attention and study.At present,there are mainly two methods for acquiring vehicle information through images.Among them,the one is to identify the vehicle indirectly through the recognition of the license plate and the vehicle logo.And the other one is to divide into large,medium and small vehicles through the acquisition of vehicle shape parameters.The former is easy to be affected by license plate and the vehicle logo recognition accuracy,while the latter is simply classified.None of them can achieve accurate identification.The vehicle-face,the area of above the bumper and under the hood,is the main information gathering place for 80% of the vehicle.Compared with the license plate and the vehicle logo,the vehicle-face has a large target,high information content,and relatively stable.So vehicle attributes can be effectively expressed by vehicle-face image.A vehicle’s type recognition method which is based vehicle-face image was proposed in this dissertation.Based on this,a vehicle’s type recognition system is developed.The main work is as follows:1.A vehicle-face image database was established including 20 types of vehicles for the training and testing.In order to acquire a vehicle-face sample image more quickly and efficiently,a vehicle-face localization method based on license plates used in this dissertation.Through experiments,a group of better relationship that the relative position and size ratio between license plates and vehice-faces was obtained.Then,the contrast tests of image feature extraction algorithms(Such as Histogram of Oriented Gradient,Local Binary Patterns,and Gabor feature extraction methods)were performed on the self-built vehicle-face image database.The feature extraction speed and cosine similarity were used as the measurement standards.The experimental results show that the Histogram of Oriented Gradient method is more suitable for vehicle identification systems.2.A BP neural network with good calculation efficiency and suitable for the vehicle recognition system based on vehicle-face image was obtained by carrying out structural design and parameter adjustment experiments.And then a test experiment was performed on the self-built vehicle-face image database.Experimental results verify the effectiveness of the algorithm.3.Based on Java CV that is an open source computer vision processing library,a vehicle recognition system based on image processing is designed and developed under the Eclipse integrated development environment.The system can quickly identify the vehicle’s brand and model according to the vehicle face image,and the experiment has verified that the vehicle face has better robustness and recognition effect under the conditions of light and scale changes.