Font Size: a A A

The Vehicle Recognition Method Based On Multi-view Characteristics

Posted on:2015-01-18Degree:MasterType:Thesis
Country:ChinaCandidate:Y K AiFull Text:PDF
GTID:2252330425989054Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Vehicle recognition is one of the key technologies of intelligent transportation systems. Most of the researches on the vehicle recognition are based on one type of characteristic to classify, with its recognition accuracy rate stability under certain circumstances, and in order to obtain higher recognition accuracy, a large amounts of data analysis with much redundant information is need, which affects the real-time object recognition. Meanwhile, Adaptive optimization learning mechanism of the classification method is also need to be further improved.In response to these problems, a method of vehicle recognition based on multi-view characteristics is presented in this paper, designed to more quickly and accurately complete vehicle recognition. The main contents of this paper include:1、The establishment of a multi-view and multi-dimensional characteristic parameter system. Multi-view and multi-dimensional characteristics hybrid-tree structure system which includes the front-view, side-view and tail-view is established in this paper. contour model segmentation method based on adaptive significance level set is proposed, which the significant initial contour curve adaptive positioning algorithm based on two-dimensional convex hull is designed to obtain the initial position of evolution curve, while regularized P-M equation is used to replace the original Gaussian filter in Li model. On this basis, the region segmentation of the different view and the definition and quantifying of optimized characteristics parameter is completed.2、The research of characteristic parameters optimization method. Dimensionality reduction and optimization of characteristic parameter model based on improved kernel independent component analysis is researched in this paper. In order to achieve the purpose that multi-dimensional characteristics of the sample is projected onto a low-dimensional subspace-based independent, the independent primitives of images is obtained by KICA algorithm to construct an independent group subspace, while using2DPCA algorithm to complete the second order related to image and further reduce the dimension in the above method. Meanwhile, the parameter optimization effect evaluation method based on Amari error and average correlation degree is presented in this paper. Comparative simulation experiments show that the parameter optimization method can effectively reduce the dimension of multi-dimensional characteristic parameters. 3、The proposal of classification model based on improved SVM. An adaptive SVM classification model based on a combination of kernel function is presented in this paper, which studies the combined kernel function and loss of function in combination with the super combination. On this basis, the efficiency and the precision are controlled by the dual-angle constraint:on the one hand, based on the Mahalanobis distance and "ασ-principle", and combining with the weighted judgment, the sample data is sorted which is used to accelerate the training and testing speed of SVM and to improve the algorithm generalization efficiency; on the other hand, in the process of setting kernel function parameter, the optimal parameter iterative searching algorithm based on prior knowledge is designed to improve the classification accuracy of the classifier.Simulation experiments show that the accuracy of the contour model segmentation method based on adaptive significance level set is stable at95%or more; In the dimensionality reduction and optimization of characteristic parameter model based on improved KICA, the Amari error is less than6%, and the average correlation degree is stable at97%or more; the recognition rate for different vehicles of the adaptive SVM classification model based on a combination of kernel function is97.926%, its training time is1.9s and testing time is44.7ms. It proves that the improved model can meet the needs of vehicle recognition and classification, which have the advantages of faster recognition speed and higher accuracy. This has important theoretical and practical significance for the development of intelligent transportation systems and vehicle recognition system.
Keywords/Search Tags:Vehicle Recognition, Multi-view, Adaptive, Level Set, ImageSegmentation, Dimensionality Reduction, Support Vector Machine
PDF Full Text Request
Related items