Reseach On Three-dimensional Scene Information Acquisition And The Display Of Point Cloud Based On Binocular Vision

With the development of signal processing theory, electronic, computer and related technologies, machine vision has been developing rapidly. With its non-contact, real-time, visualization, automation and intelligence, machine vision are widely used in the national economy, scientific research, national defence construction and other fields. And machine vision is highly valued by all countries and industries. The binocular vision technology as an important branch of machine vision, is one of the important problems and key research direction of computer vision and non-contact measurement. Camera calibration, stereo calibration, stereo matching and point cloud display are four research points of binocular vision. In this dissertation, the above four key aspects are studied. The concrete research contents are as follows:First of all, we studied the model of binocular vision platform, using 2D checkerboard plane template to calibrate the camera. Stereo rectification technology is studied. And according to the binocular camera calibration data, images obtained on binocular camera platform are corrected;Secondly, stereo matching technology is studied. Modules and constraints of stereo matching algorithm are analysed. Based on analyzing the local stereo algorithm, a stereo matching algorithm based on guided filter is studied.Finally, the point cloud display technology is analysed. The reprojection of 3D point clouds is studied. Surface reconstruction of 3D point cloud has several key steps: filtering, segmentation, down-sampling, reconstruction. These steps are analysed.