Research On The Geometric Feature Extraction Of Vehicle Objects In Monocular Video And The Method Of Vehicle Recognition

With the development of the national economy,the intelligent transportation system(ITS)emerges because of the serious regulatory problems caused by the growing exponentially automobile.The research content of this paper is the sub task of the same temporal and spatial correlation retrieval for the vehicle in video.For the problem of vehicle identification can't be completed through the vehicle license in highway bayonet video,the geometric features of moving vehicles are extracted by digital image processing technology,the classification method of machine learning is used to identify the vehicle,and the vehicle identification system is developed based on the platform of Windows system,with VS2010,opencv2.4.3,QT5.5.The research content of this paper provides a method to quickly lock the target vehicle for analyzing on the same temporal and spatial correlation.The research contents of this paper are as follows:1.The establishment of the vehicle model standard feature template database that include vehicle complete parameter information.More comprehensive information provided by vehicle manufacturer are obtained through the network technology from the "home of trucks","car home" and other vehicle information sites.In this paper,an automatic climbing system of vehicle parameter information is developed for the vehicle information real-time renewal work.2.Moving object detection and camera calibration technology.In this paper,road background and vehicle objects are extracted with Gauss background modeling and background subtraction,and the edge of the moving vehicle is detected with Canny;the projection of the variable 3D model is based on the camera calibration,that the extraction of road background is calibration board for camera calibration.This calibration method is called Zhang Zhengyou calibration method.3.Parameter optimization algorithm.An improved Hausdorff distance is constructed as the vehicle object and model projection matching evaluation function.In view of the low efficiency of a large number of edge points matching in the matching process,this paper improves the efficiency by extracting a small number of key matching points as the two matching points to increase the matching speed and improve the efficiency.Using the quantum evolutionary algorithm to solve the optimal parameters.The geometric feature extraction of vehicle object is completed,which provides a stable and effective feature for vehicle identification.4.Based on machine learning models identification method research.In this paper,data mining analysis on the standard template data in the database is conducted by means of neural network method.The number of samples,the number of classification errors,and the recognition rate are compared and analyzed.Supervised Kohonen network can effectively accomplish the classification and recognition of the data provided in the database,through the geometric characteristic parameters of the vehicle,the average recognition rate of the vehicle classification using supervised Kohonen network is 91%.