Research On The Generation Algorithm Of Dense Point Cloud

Man is to know the world through the eye, the rate of the information acquired through the eye more than 80%.so for a long time, computer vision are hot topics in the study of human beings.people use the computer vision to better perceive the worldthrough the computer vision technology to all aspects of the study, a better service for human life, especially in recent years, as the unmanned aircraft, unmanned, robot, augmented reality (AR) and virtual reality (VR), and other aspects of the development of computer vision is reflected to a boom. And among them,3 d reconstruction based on forming dense point cloud is a hotspot of research in recent years, how to two-dimensional images were collected from the camera to restore the original 3 d information and object or scene three-dimensional dense point cloud has become one of the key parts. This paper puts forward the creative part of the two.1. In this paper, on the basis of SIFT was improved, puts forward a new kind of feature matching algorithm, first of all, through the method of image a moment more accurately calculate the feature points prototype area near the center of mass, seeks the center of mass are used to accurately draw the main direction of feature points, which can improve the efficiency of matching algorithm. By feature matching to get only the local information of the reconstruction, and can’t complete showed dense point cloud, need to be dense matching can generate dense point cloud. So this article on the basis of the improved SIFT operator is put forward based on regional growth strategy to implement the dense matching, and then generate dense point cloud.2. Parallax graphs generated by the improved algorithm to get the dense point cloud, firstly introducing the idea of segmentation, the mean shift algorithm for image segmentation, make the color similar to the part of the division to the same area, and then use local matching algorithm to get the initial parallax figure, through surface fitting, plane template generated parallax. Before then improved belief propagation algorithm, the belief propagation algorithm is based on the pixels of the image after segmentation, the belief propagation algorithm based on region segmentation which can reduce the computational complexity. And can be a very good sheltered situation of texture is not obvious and processing. In dense spread through the execution of the improved algorithm of parallax, generate dense point cloud.