| In the dissertation,I present a binocular visional navigation system for the Intelligent Vehicle(IV),including its theories,methods,and implementations．The vision navigation system should provide navigation information for the lateral control and longitudinal control．The road-marking detection,road-recognition,vehicle detection,and distinguishing the road-marking from vehicle or other obstacles are foundations of longitudinal control． With the development of the city and the prevalence of the vehicle,the problem about the traffic transportation become more and more serious．The autonomous vehicle can not only reduce the traffic accidents,but also can improve the security of vehicle and the efficiency of the highway,and can lighten the burden of the driver．The Intelligent Vehicle is a integrate system including the environment apperceive,layout decision,automatism drive and so on．As a way to gain the information of the outside,the vision system play an important role in the perfect Intelligent Vehicle．Through analyzing the domestic and international research of IV visual navigation system,and on the base of the analyzing the structure of human vision and the process of the vision formation,I make a research and analyzing of functions of the IV visional system,then according to the function analyzing , make a integrated analyzing and designing to the IV binocular visional system , decompose the visional system to several parts ,including the binocular stereo camera calibration and establishing the camera model ,image obtaining and image preprocessing ,image segmentation ,obstacle checking and tracking ,the road-marking checking and tracking , and the camera model parameters updating .At last , I make a farther analyzing and designing to the implementation algorithm of each function mode .The camera calibration need series datum marks, whose coordinates have been measured in the world reference frame and in the image reference frame , then according these datum marks to compute the parameters of the camera model . The parameters needed calibrated divide two kinds : internal parameters and external parameters . The external parameters are different to each frame image． But the internal parameters are same ．The parameters calibration is carried in the course of the transformation from objective scene to digital image ．The binocular stereo rig calibration need make a calibration to each camera firstly,which can gain its projection matrix , then get the results of the comparative orientation problem ,which is to use the projection of the scene point in the image plane to confirm the relation between the two camera reference frames．According to the projection matrixes, the relation of the binocular camera models can be gained．Then on the condition of the series of parameters and the transformation between the world reference frame ,camera reference frame and the image reference frame , I can establish the binocular rig model．In this dissertation , the model is enacted as a parallel binocular rig model , this means the light axis of the two camera are parallel．Take it into account that the image gained in the actual road environment will include noise, and in order to implement the lateral and longitudinal control exactly, the image must be preprocessed．In this dissertation , after analyzing all kinds of the image process algorithm, at last the Gaussian filter which is fast to remove noise is used in this dissertation．In order to divide the image, firstly the Helmhotz shear algorithm must be used to make a transformation to the two images, through which the disparity of objects above the ground is not zero and the disparity of the area belonging to the ground plane is zero．After this transformation , according to the disparity map I can divide the image into two regions, one is zero-disparity and another is nonzero-disparity． In the nonzero-disparity region, I check the obstacles and confirm those positions according to the 4 connectivity of pixel．In the algorit...
|
PDF Full Text Request