Friday, 1 July 2016

RECONSTRUCTION OF URBAN 3D MODELS FROM LIDAR DATA

1 Introduction
The needs for 3D city models are rapidly growing in various fields such as virtual city reality, town planning, microclimate investigation, transmitter placement in telecommunication, monitoring and protection of coastal zones, pipelines and corridor mapping and exhaust spreading in urban areas, Et cetera . There is a steady shift from 2D-GIS toward 3D-GIS now, therefore a great amount of accurate 3D city models have become essential to be produced in a short period of time and provided on the market [TAKASE et al]. Now a day, Light Detection and Ranging (LiDAR) is widely applied in urban 3D data analysis.

2 LiDAR Overview
Mounted on the aircraft, collecting the LiDAR data, is a Global Positioning System (GPS), a LiDAR sensor system and an Inertial Navigation System (INS). The GPS returns the longitude and latitude coordinates of the aircraft’s actual position. The LiDAR sensor system is a remote sensing instrument which consists of an emitter and a receiver. The INS tracks the altitude of the LiDAR sensor. The emitter sends out electromagnetic radiation (a pulse of light) into the atmosphere down to the target. The receiver (telescope) measures the intensity of the signal scattered back to the sensor. The time from sending to return of pulse is also recorded thus the distance of the sensor to the target can be calculated. In the case of a building surface, the laser will reflect off of the building’s surface and return to the sensor. Intensity and path of the electromagnetic radiation (laser beam) are distorted by the interaction between the beam and the atmosphere itself [NICHOLAS]. An example of LiDAR sensor system is shown in figure 1.


When two different heights of one point are received, this gives an indication of the presence of a penetrable object (tree). In contrast, for a non-penetrable object (building), data point has the same height for first and last returns. Figure 2 shows first and last return in case of a tree [ABDULLATIF].


2.1 Ground Surface
The point density is high around the scanner and gradually decreases with distance away from the scanner, the region of interest can be selected for example as an area 20 m x 20 m around the instrument to reduce the amount of no data areas [CHRISTOPH]. Figure 3 shows the representation of a manmade object (building) and a natural feature (tree) [VOLKER].


2.2 Tin Procedures
Triangulated Irregular Networks (TINs) are used for representing or interpolating point clouds by surfaces. TINs represents a meshing of points (xk, yk, zk) in the form of triangular surface in 3D with these points as vertices, constructed above a 2D triangulation of the surface (xk, yk) of the vertices in a base plane. Triangulations are usually constructed using the ‘Delaunay’ principle, which tells that circumcircles of the triangles do not contain the locations of vertices in their interior [CHRISTOPH]. Figure 4 shows the mesh points and TIN surface.


2.3 Greedy Insertion Triangulation
For all triangles, the distances between the triangles (planes) as shown in figure 5 and the points that they encompass (in x and y or longitude and latitude spacing) are calculated [NICHOLAS].


3 3D Reconstruction Algorithm
The following attributes should be a part of an ideal 3D reconstruction algorithm. The algorithm should be able to intelligently recognize complex (non planar) building structures besides simple ones. The algorithm should be generalized such that it has capability of analyzing raw (irregularly spaced) LiDAR data as well as where the LiDAR data is rasterized / grid-ed. The algorithm should reconstruct the buildings in the form of simple structures (roof planes, walls, etc.) as opposed to a collection of LiDAR points [NICHOLAS]. Figure 6 shows system block diagram.


3.1 Building Detection
The first step of building reconstruction algorithm is the detection of buildings. There have been many attempts to detect buildings using LiDAR data. The LiDAR points are classified according to terrain, buildings or others like vegetation. Morphological opening filters are have been used to determine a digital terrain model (DTM) which is subtracted from the digital surface model (DSM). An initial building mask is obtained by applying height thresholds to the normalized DSM [ROTTENSTEINER et al]. Figure 7 shows an example of a DSM [NORBERT et al].



3.2 Building Extraction
Figure 8 shows work flow for building extraction from LiDAR data [ROTTENSTEINER et al].


The algorithms for extraction of the building geometrical parameters will be performed after the complete edges of buildings have been detected. The LiDAR data interpretation incorporates following facts: (1) The buildings are higher than the surrounding topographic surface; (2) The laser penetrates into vegetation, thus giving echo from various heights, makes it possible to distinguish between man-made objects and vegetation [GUOQING et al].

3.3 Building Reconstruction
Object reconstruction or recognition presumes knowledge about the perceived objects by some kind of object model. Model that is used for building reconstruction should be able to describe buildings of various complexities and it should allow the representation of geometric constraints during the reconstruction. Object models can be visualized as abstractions of real world objects. The most important part in a model definition is the proper balance between the tractability and correctness, i.e. the outcome of the model must be adequate both in terms of the solution attained and the cost effectiveness [NORBERT et al]. Figure 9 shows an example of a refined reconstruction.


4 Uncertainty
Representing uncertainty can be a difficult problem. The uncertainty of all geometric entities like points, lines, planes and objects, can be represented derived by construction from those entities. But how certain is a segmentation result, for example, when multiple cues from different sources need to be combined? How that case is handled when a small (continuous) difference in geometry leads to a different roof topology (a discrete difference)? [CLAUS].

5 Conclusion
Recent development in 3D displays, computer graphics hardware and real-time texturing as well as the increasing availability of animation software tools have resulted in an increased demand for realistic three- dimensional virtual reality city models. This demand can only be satisfied by capturing urban scenes efficiently, which presumes the integrated use of various data sources [NORBERT et al]. In the near future, research will have to find a way to develop new specialized algorithms for the case, and devise methods to combine aerial and terrestrial sources efficiently [CLAUS]. Figure 10 shows an example of a virtual city model.



Written by,

























Share this



See All Articles

2 Responses to "RECONSTRUCTION OF URBAN 3D MODELS FROM LIDAR DATA"

  1. It's an awesome post about the LIDAR scanner Houston, Texas. You have describe around all the things that is needed to know. You can hire professional to do the job for you.

    ReplyDelete
  2. With the overabundance of information accessible with the organizations today, they must be careful not to get caught in an interminable examination work out. ai courses

    ReplyDelete