! 2004 
data 
DSM, 
  
utput 
FF, 
ation 
| and 
acity 
and 
the 
; are 
ures 
user 
they 
ence 
nost 
1ced 
ality 
rief 
ully 
ned 
ing. 
sing 
met 
COSS 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
to data in case of multi-user processing environments to make 
sure that actions of one operator do not interfere with actions of 
another operator on the same data set. 
Project status monitoring maintains a detailled overview of 
which processing steps have been applied to which parts of a 
data set. This includes a graphical processing status mask 
showing areas of a project have been completed (e.g. filtered or 
classified), which areas are pending (currently being worked 
on), and which areas are still open. It also alerts the project 
administrator of delays with respect to the project timeline. 
Resource management manages storage space on the 
designated storage devices, caches data locally in networked 
processing environments, to provide the user with fast access to 
the data he is working on, and maintains data integrity for the 
entire project. Automated data backup and mirroring may also 
be integrated into the resource management feature. 
Quality control verifies that the quality criteria set forth in the 
project specifications are met throughout the processing 
workflow. Specifically, at the end of each processing task 
quality checks are performed before the task is closed and the 
user can move on to the next step. The results of the quality 
controls are automatically logged. 
= During data import input data quality is verified regarding 
completeness of coverage within the block boundaries, raw 
point density, and a general data quality assessment is 
carried out through statistics generated during the import 
process (raw data noise and shot statistics, point 
distribution statistics), 
= after strip adjustment the quality of alignment is controlled 
by analyzing residual deviation and height noise statistics, 
= after model adjustment and transformation the statistics of 
residual planimetric and vertical errors at control 
surfaces/objects are checked against project requirements 
* after classification, statistics are generated, ground point 
density and ground point distribution are analyzed for 
irregularities and for compliance with project 
requirements. 
" For output generation the parameters used (e.g. tiling 
boundaries, grid sizes, file naming conventions, file size 
limitations, etc.) are compared with project requirements. 
Part of quality control is also the documentation of the 
processing workflow including used input files, performed 
tasks, associated names of operators, used parameters, product 
lists, and their crosscheck with the project specifications. All 
protocols generated by the individual tasks are collected in the 
quality control report that is generated automatically at the end 
of the project. 
Those steps of the quality control procedure that are interactive 
like the interpretation of point distribution statistics require the 
user to confirm that the quality criteria are met. His name, 
"signature", any remarks, and time and date are logged in the 
quality control protocol. 
3. PROCESSING 
While most of the lidar processing components in LasTools 
employ advanced approaches worthy of detailled discussion we 
will focus on classification in the scope of this paper. 
Un 
3.1 Classification Workflow 
Lidar measurements unqualified in that point locations are more 
or less evenly distributed over the reflecting surfaces, but no 
direct information about what type of surface was hit by each 
shot is available initially. In order to derive elevation models 
that only represent a certain type of surface (for example: 
terrain) the surfaces that reflected the laser beam must be 
distinguished. Classification is the process of distinguishing and 
assigning individual 3D measurement points to surface classes, 
so that in subsequent processing surface and object modelling 
may be based only on the points from relevant surfaces. 
As so far no approach exists that can guarantee a 100% correct 
classification of lidar points automatically, the classification 
process is done in three steps: 
1. Interactive parametrization 
2. Automatic classification 
3. Interactive refinement 
3.2 Interactive Parametrization 
The user selects one or several areas of a block that contain 
features typical for the surfaces and terrain shapes in that block. 
These areas are displayed simultaneously on the screen. The 
user then interactively varies the initial classification parameter 
settings such that the classification results in the displayed 
sample areas are optimized. Changes of parameters invoke an 
immediate reclassification and display update to show the effect 
of the current parameter settings. The set of parameters is 
associated with the block and saved. It may also be reloaded as 
a starting point for other blocks or later projects that have 
similar characteristics. 
3.3 Automatic Classification 
An automatic classification is performed for the entire data set 
(or selected parts) using the optimized parameters. 
In LasTools, we implemented a contour/segmentation-based 
object-oriented approach to point and surface classification. 
Segmentation is performed by “contouring”: contour lines are 
generated from the highest elevation in an area down to the 
lowest elevation in relatively small steps (for example 0.5 m). 
At each elevation level new closed contours are searched. The 
first occurrence of a closed contour seeds a new segment. A 
segment in this context is a coherent planimetric area delineated 
by a closed contour. On subsequent (lower) levels, segments 
grow until their contours merge with neighboring contours. A 
primitive object is constituted by all contours above this level, 
i.e. by all contours that contain at most one contour at the next 
higher level. Complex objects come into existence when 
multiple segments merge, i.e. a complex object contains more 
than one contour or object (both primitive and complex) at the 
next level. Complex objects are the “parents” of their complex 
or primitive "children". Obviously, at the lowest elevation 
(root) level a single complex object exists that contains all other 
objects and represents the entire block. 
These objects are so far only abstract entities representing 
hierachies of enclosed contours and should, of course, not be 
confused with actual surface objects as they may contain any 
type of surface objects (vegetation, buildings) as well as 
ground. This hierachical description of the data set, however, 
facillitates searching for “real” surface objects significantly.