A PIPELINE ALGORITHM FOR BUILDING AND STRUCTURE SHAPES GENERATION USING DERIVED LAS DATASET: AN EFFICIENT ALTERNATIVE TO MANUAL DIGITIZATION FROM ORTHOPHOTOS IN FLOOD HAZARD FEATURE EXTRACTION

摘要

In this paper we propose and describe an implementation of a computationally efficient generation of building and structure shapes which dramatically improves the manual process in flood hazard feature extraction workflow without orthophotos. The impact cuts through not only on the cost in procuring very high-resolution true color RGB images but also ultimately to the reduction of average processing time from manual editing of 4.5 hours down to just over 40 minutes using the computational approach. This time saving scheme is the result from carrying out a prescribed pipeline operation on the LIDAR LAS dataset. The principle from which the generation of our wanted shapes rely uses the convex hull set that forms an outline from the LAS data points during program execution. To contrast when orthophotos with LAS point data they represent a 2-dimensional data matrix whose elements are light intensity RGB values in their horizontal and vertical pixels alone without any depth information and do not have elevation values of the spatial surfaces. In this paper we exploit the zth value representing the relative heights and elevation from ground in the LAS cloud point data that comprise of millions of laser point returns from its coordinates in 3D space. By incremental stepping through a number of height breaks, we are able to select structures in the floodplain areas that is otherwise could be difficult by visual acuity and experience in the polygon outlining by hand alone in 2D image. A test to quantify the difference between the manually derived polygons and those by machine, we used a 2-dimensional cross correlation to yield a number for goodness of alignment. Our experience indicated that an optimum for our test case was reached after a height break of 30m from an iteration of 3-30 meters and step size of 0.5-5.0 sq. meter. The new method prototype achieved practically a one to one alignment and an improved granularity in height discrimination.