)04
79-
ata
hD
ge-
ent,
and
and
1 of
ing
sed
and
ton,
ing
ting
phic
)63-
OBJECT EXTRACTION AND RECOGNITION FROM LIDAR DATA BASED ON FUZZY
REASONING AND INFORMATION FUSION TECHNIQUES
F. Samadzadegan
Dept. of Surveying and Geomatics Engineering, Faculty of Engineering, University of Tehran, Tehran, Iran - samadz@ut.ac.ir
KEY WORDS: 3D Objects, Extraction, Recognition, Fuzzy logic, LIDAR, Region growing
ABSTRACT:
Three dimensional object extraction and recognition (OER) from LIDAR data has been an area of major interest in photogrammetry
for quite a long time. However, most of the existing methods for automatic object extraction and recognition from LIDAR data are
just based on the range information and employ parametric methods and object’s vagueness behaviour is basically neglected. Thus,
these methods do not take into account the extraction and recognition complexities and may fail to reach a satisfied reliability level
in complex situations. In this paper a novel approach based on the following strategies is formulated and implemented: (a) for a
more comprehensive definition of the objects, information fusion concept is utilized, i.e., object’s descriptive components such as
3D structural and textural (ST) information are automatically extracted from first/last rang and intensity information of LIDAR data
and simultaneously fed into the evaluation process, (b) for a more realistic expression of the objects and also for simultaneous fusion
of the extracted ST components, the fuzzy reasoning strategy is employed. The proposed automatic OER methodology is evaluated
for two different object classes of buildings and trees, using a portion of LIDAR data of an urban area. The visual inspection of the
recognized objects demonstrates promising results.
1. INTRODUCTION
Extraction: In this stage an inspection is carried out to locate
The idea of having a fully automatic three-dimensional OER and extract all 3D objects that exist in the entire area
(object extraction and recognition) system to replace the human irrespective of the objects identity. The morphological operators
operator has been one of the main aspirations and the final goal are applied to the range information of LIDAR data to delimit
for photogrammetry and computer vision investigators and isolate the individual 3D objects. In the next step each 3D
(Baltsavias and Stallmann, 1995; Brenner and Haala, 1998b; candidate region is mapped into the intensity images to
Brunn and Weidner, 1997; Collins and et. Al. 1995; Ebner and determine the corresponding region in the intensity space. The
et. Al. 1999; Fua and Haanson, 1988; Gruen and et. Al. 1997; final decision for each individual object's boundary is made by
Jaynes and et. Al. 1997 ; Ameri and Fritsch, 1999; Haala and à fuzzy-based region growing approach. Any modification of
Brenner, 1999; Lemmens, 1996; Maas, 1999). The existing the object boundaries as an outcome of the region growing
methods are mainly formulated using parametric approaches process will result a corresponding modification in the 3D
and just optimized for using single information. To exploit geometric information in the object space. At this stage the
more fully all available information that contribute to the system knows the presence and the location of the objects
extraction and recognition process and handling the object's without a definite knowledge about their identity.
vagueness behaviour, we propose an OER strategy which
makes use of the object information inherent both in range and Recognition and training: The geometric and radiometric
intensity information through a fuzzy reasoning strategy. contents of each segmented region are analyzed to derive the
double descriptive attributes that define the objects in an
2. PROPOSED OBJECT EXTRACTION AND integral manner, these are: structural and textural (ST)
RECOGNITION METHODOLOGY descriptive attributes. Simultaneous fusion of these parameters
yields the object’s identifying signature. Because of the
vagueness nature of the ST elements, the recognition engine is
designed based on a fuzzy reasoning strategy. The training
potentials are also embedded into the recognition engine to be
The overall strategy for our proposed operations may be
expressed, with reference to Figure. I, by the two interrelated
procedures, Extraction and Recognition.
EXTRACTION RECOGNITION used for unrecognized objects. Having stated the general
working principal of the proposed OER method, in the
Preliminarily 3D Regions BT Component following sections detailed treatments of the main individual
Morphological Operators Ha ysis modules that govern the OER process are presented.
Structural |
Recognition Engine 2.1 Object Extraction methodology
Information Fusion
Based on
Fuzzy Reasoning
m
3D Objects
| Fuzzy Based Region Growing
The proposed object extraction method is designed to perform
two sequential procedures, namely: (a) preliminary extraction
of all candidate objects of interest from first and last range
e
pulses LIDAR data, i.e., 3D candidate region extraction, and (b)
Figure 1. Extraction and recognition work flow
921
