The application of classical progressive triangulation filter algorithm for airborne point cloud is very successful, however, there is a big difference between airborne point cloud and vehicle-borne laser point cloud in spatial distribution, density and other aspects. In this paper, a lot of experiments are carried out to improve the filter algorithm for vehicle-borne laser point cloud, which includes as follows: (1) Establish grid index, such as 0.1 meters, only retain the lowest points, which can greatly reduce the number of suspected ground points, and the filtering efficiency is improved significantly; (2) According to the vehicle-borne height and track line, the road face points can be roughly determined. Then the convolution operation is used to ensure the real road points, which are also the ground points. This method cannot have to relax the filter parameters (which will lead to more non-ground points) and ensure the integrity of the road boundary; (3) A method named as "get more and remove some" is proposed for solving the filtering faults at the tail of every points segment caused by the incline scanning face. After the three steps, the filtering is improved obviously on qualification and processing speed.

ABSTRACT:Vehicle-borne laser scanning (VBLS) is widely used to collect urban data for various mapping and modelling systems. This paper proposes a strategy of feature extraction and 3d model reconstruction for main body of overcrossing bridges based on VBLS point clouds. As the bridges usually have a large span, and the clouds data is often affected by obstacles, we have to use round-trip cloud data to avoid missing part. To begin with, pick out the cloud of the bridge body by an interactive clip-box, and group points by scanline, then sort the points by scanning angle on each scan line. Since the position under the vehicle have a fixed scan-angle, a virtual path can be obtained. Secondly, extract horizontal line segments perpendicular to the virtual path along adjacent scan-lines, and then cluster line segments into long line-strings, which represent the top and bottom edge. Finally, regularize the line-strings and build 3d surface model of the bridge body. Experimental studies have demonstrated its efficiency and accuracy in case of building bridge model. Modelling the stairs at the both end of the bridge will be the direction of the next step.

ABSTRACT:In this paper a method "slicing-iteration" is proposed to realize automatic extraction of vehicle-borne laser point cloud, the main steps are as follows: (1)Confirm ground reference surface; (2)Dividing points: taking points at certain height range above reference surface, divide them by certain cell size (such as 0.1m) and record points number of every cell; (3)Forming polyline: based on the number, for every cell, axially cluster is made, then find a neighbor with most similar number and form the two-point-line, further forming a polyline; (4)Iteration: dividing points and forming polyline can be conduct at different height range and by x/y axis, so as many features as possible can be extracted. The method can also be used in half-automatic boundary extraction. Experiments show that the method in the paper is easy to use, highly adaptive, with good extraction result. Corresponding author (392842876@qq.com)