Study On Stereo Matching Technology Of Close-Range Photogrammetry Based On RBF Neural Network

With the further development of engineering construction in our country, the applications of tunnel construction monitoring are quite important. In recent years, photogrammetry technology has achieved great advancement. The photogrammetry, introduced in the safety monitoring of tunnel project construction, is a kind of promising technology.Firstly, according to the basic principles of the photogrammetry technology, the nonlinear imaging models and other related theories were preliminarily understood; the theory of stereo matching was introduced, in order to provide the theoretical basis for computational work in the experiments and modeling of Radical Basis Function(RBF) neural network.Secondly, based on smoothing, wiping off noise and enhancing techniques in the image processing, a method for target outstanding of image feature was presented. The images in different regions were processed to reflect validity of image processing through filtering. The methods of edge extracting and corner recognizing were adopted in the MATLAB software,which were the preparation jobs for camera calibration.Thirdly, the methods of digital camera calibration were introduced. The identified points and measured points were recognized in the experimental images, respectively. The intrinsic and extrinsic parameters of camera were obtained by the Camera Calibration Toolbox for Matlab(Calib Toolbox), for a basis to train stereo matching model of RBF neural network.Finally, after processing left and right images using stereo matching modeling of RBF neural network, there existed the deviation between the calculation results and the practical three-dimensional coordinates, not more than the permitted error. This fact ensures the accuracy and convenience of the stereo matching calculation with RBF neural network, whichprovides a new perspective and a new method for close-range photogrammetry technology.