Research On Filtering Method Of Three-Dimensional Laser Scanning Point Cloud Data

Under the impetus of the applications in industrial design, three-dimensional mapping, space simulation, medical diagnosis, television and entertainment, three-dimensional data acquisition and processing technology of digital geometry are more and more concerned by scholars. In recent years, with the rapid development of the hardware and software of the three-dimensional model acquisition, three-dimensional laser scanning technology is used in various fields. Especially in the mapping field, it not only improves the efficiency, but also improves the accuracy. The point cloud data which is got through the three-dimensional laser scanning is not only simple in structure, easy operation, but also not stores the points and topological relations between points. Thus it is suitable to express the irregular object which has complexity geometric feature and characteristic surfaces. However, In the process of data acquisition, due to artificial factors, measurement of changes in the environment or defects in the equipment itself and many other factors, the data are polluted by the different levels of noise. Therefore, before using the data, we must filter the point cloud data. The goal is filtering the noise to reconstruct smooth surface, while maintaining the topological characteristics of the original sample surface and geometrical features.In this paper, point cloud data obtained by laser scanning as the research object, we study its filtering in-depth. The main content is as follows:1. Describes some of the classic scattered point cloud data filtering algorithm, in recent years, domestic and foreign literature has been more focused analysis of the algorithm, the principles they were described in detail, and summarizes the advantages of these algorithms Limitations on the measurement data after the filtering of laser focus are analyzed.2. Depth analyzes the noise of the point cloud model. Through establishing the mathematical model, summarizes the problems which must be solved in filtering process. Thus provides a theoretical basis for the filtering. 3. At the base of point cloud data filtering using Mean-Shift algorithm, this paper puts forward a number of improvements. Full consideration the normal vector and curvature the two geometrical to perform the characteristics of point cloud model, firstly it optimizes the k neighborhood using the similarity of normal vector, secondly filets the normal vector using Mean-Shift iterations to find a local model for each data point, thirdly filets the geometrical vector as the standard by normal vector. This method is effective not only for the point cloud filtering, but also for protection feature information and prevention the volume shrinkage.