Research On Key Techniques Of Building LiDAR Point Cloud Data Feature Detection And Registration

The feature detection of point cloud data and matching technology has currently been a area of research highlights. Concentrating on the feature detection of point cloud data and registration problems, this paper mainly investigates the feature points detection for building Li DAR data, the extraction of building horizontal and vertical edge feature lines, the issues of point cloud data registration. The main work is as follows:(1) Achieve point cloud data management using kd-tree, and study how to build a tree and improve the searching efficiency of k neighbor points. In order to establish the relationship between spatial topological adjacent points, this paper firstly establishes kd-tree using dichotomy, and implements the searching of the point cloud data of k neighborhood points, and then studies the time of building tree under different amounts of data and the searching efficiency of k neighbors. The research shows that with the increase of data volume, the time of building tree and the time of searching k neighbor points are both linear growth.(2) Put forward to a algorithm based on multi-structure data to extract the horizontal and vertical edges of buildings. On the basis of the least square plane cutting algorithm detecting feature points, history conditional sampling of model is perfomed using multi-structure estimation algorithm and is applied to search feature line equation. Using linear optimization algorithm, calculate the point count that the straight line distance is less than the threshold. The maximum point count of the line is the optimal line, which feature points located in are recored. In order to detect different targeted segment on the identical straight line, firstly rank by point in that line, and then search for the points which in the same line segment utilizing the comparison between point spacing and the threshold value. Finally, the intact extraction of the window edge is accomplished. The experiment shows that the multi-structure algorithm has a greater advantage of the speed and efficiency of searching the optimal linear, the capacity of the optimal linear containning inlier points than the traditional random sampling.(3) Implement the match of edge feature points of buildings. This article turns the registration of edge discrete feature point as the issue of probability density estimation.In accordance with the requirements of the model point set is greater than the data point, defined datum point cloud as the data set,waiting search point cloud set is defined as a model set. Building point cloud registration process, that is, to solve transformation parameters of turning data point set to model set. The gaussian mixture model likelihood function is designed as the objective function of matching, and the matching parameters of the three-dimensional point cloud data are obtained by applying the iterative EM algorithm in order to achieve the global matching of edge feature points of bulidings. On that basis, the effect of random noise on the matching process is studied, and the beneficial conclusions are received.