Research On The Vehicle Recognition Method Based On Sparsity Constraint

Vehicle recognition technology is widely used in the field of intelligent transportation and intelligent security,and it is also a hot topic in the research of computer vision,image processing and pattern recognition.Under this background,the research of vehicle recognition method is more and more and the corresponding vehicle recognition results are obtained.With the continuous improvement of vehicle recognition task,the new demand for vehicle recognition has been produced.In the practical application of vehicle recognition,bad weather conditions lead to vehicle target submerged,but the vehicle recognition task still need real-time imaging;incomplete body and nonuniform light illumination lead to serious noise existing in the vehicle target features;unique vehicle recognition has a strict requirements to the feature extraction.In order to improve the effect of vehicle recognition,this paper focuses on the related research work,and proposes a vehicle recognition method based on sparse constraint.The main research content and achievements of the thesis are as follows:(1)In order to solve the vehicle imaging problems due to severe environment,the vehicle detail enhancement based on sparse weighted filtering is proposed.Severe weather conditions such as fog and haze,rain and other extreme weather will cause the problem of vehicle target is difficult to distinguish from the screen,the problem directly affects the real-time imaging of the vehicle target.In order to obtain the stable real-time vehicle target imaging,the scene data of infrared thermal imaging needs enhancing adaptively to the real-time image.To this end,this paper proposes a sparse weighted filtering as an image processing tool to decompose the image into different components with corresponding frequency domain for adaptive enhancement.The detail enhancement process takes into account adaptive gain parameter control in the high frequency region and global contrast suppression in the low frequency in order to achieve the image detail,which is convenient for feature extraction and expression.Comparison with diverse methods on the real infrared data,the experimental results verify the proposed vehicle detail enhancement based on sparse weighted filtering have a reasonable compressing and enhancement,and outperform other enhancement methods;(2)In order to satisfy the vehicle image recognition requirement to the feature,sparse image feature with spatial pyramid structure is employed for vehicle image feature extraction.The running vehicle target in the real scene has diverse poses;furthermore,the vehicle image will degrade due to the weather and illumination etc.The factors above bring a huge challenge to the vehicle recognition task.In order to alleviate the impact on feature extraction,we extract local features and introduce sparse representation into the process of generating global characteristics using local features to refine the discriminant ability of vehicle target feature.At the same time,according to the various state distribution of the vehicle target in the real scene,the multi-level structure of pyramid is established to obtain the stable characteristic expression.For verifying the performance,the experiment selects the vehicle targets including sedan,taxi,van,and truck.The proposed method has an excellent performance on vehicle recognition compared with the traditional non-sparse methods;(3)In order to extract the unique feature of the vehicle target.A method on detecting the unique region of the vehicle target,vehicle window,is proposed.The information of vehicle outer contour or grid texture information of the vehicle air intake can help the extraction of discriminative features,but can not guarantee uniqueness of target characteristics of the vehicle.In order to extract the unique characteristics of vehicle targets,this paper,from the front of the vehicle target,selects the internal information of vehicle including vehicle interior decoration,drivers and other information as the only measure of the vehicle's target.To this end,the window region detection method is proposed.The vehicle window detected by the method benefits for the unique feature extraction.The feature set grouped by the local descriptors is more discriminant compared with other regions.The experimental results of the accurate vehicle target recognition verify the positive impact.(4)In order to solve the special problem of the uniqueness of vehicle recognition technology,it should guarantee that the vehicle target feature can lock the vehicle target for matching without the license plate information.This paper uses the window detection method to locate the discriminant region of the vehicle target,and employs the sparse feature of this region to regroup the structured feature and obtain the vehicle garget's feature with unique and discriminant ability.For showing the performance of the proposed method,the experiment randomly selects 120 groups of vehicle targets captured in the different time for testing with two different feature descriptors.According to the contrast experimental results on these vehicle data,it shows that the sparse feature is positive for matching and the vehicle window region is discriminative for feature extraction.