Recognizing Primitives Using Aspect-Interpretation Model Matching
in A Both CAD and LP-based Measurement System
Zhou Guoging
Dept. of Computer Science and Technology, Tsinghua University,
Beijing, 100084, P. R. China
ABSTRACT
In order to realize measuring automatically CAD-based object using line photogrammtery[3,4,6],
the author in this paper presents recognizing primitive using model matching based on aspect-
interpretation. The idea is composed of aspect-interpretation, aspect-graph(also called attribute
relational graph) and model matching. The aspect-interpretation contains face-aspect code, face-
aspect mergence code and face-aspect split code. The aspect-graph includes attribute relational
graph, primitive attribute graph and attribute hypergraph. The model matching is to indicate
matching model aspect-graph with image aspect-graph. Finally, the measurement system, which is
an integration of CAD, Photogrammetry, Robot and CV, is briefly described. A group of
. experiments is tested and the results show the recognition scheme is feasible.
KEY WORDS: Recognition, Aspect-interpretation, Model matching, CAD, Line photogrammetry
these schemes only recognize an object from
1. INTRODUCTION
After the author presented the idea that CAD-
based industrial object can be measured using
line photogrammetry(LP) so as to built up a
bridge between CAD and CV system[3,4,6], a
set of soft-package of LP for measuring various
kind of primitives is developed. For the sake of
automatic measurement, it is necessary to
recognize automatically these primitives. In this
paper, the author describe how to recognize the
primitives in measurement system.
Since the model(data structure, strictly
speaking, abstract model) of object has been
stored in CAD system, recognizing primitives is
in fact CAD-based model recognition. Many
CAD-based recognition schemes have been
proposed in papers published. Woo[18] used
convex hull technique to describe the object.
Kyprianou[13] applied syntactic pattern
recognition methods to classify depression.
Henderson[8] and Kung[12] used logic
programming and expert system to extract shape
features. Joshi[9] developed attribute relational
graphs for a part. Lee and Fu[14] proposed an
algorithm for the exaction and unification of
relatively complex scene. So, in all of the
approaches above, they differ from recognizing
single primitives from an object. Comparatively
speaking, our recognition exists some characters
as follows.
e The primitives to be recognized are
numbered.
e Prior CAD information can be provided.
In this paper, the author presents a novelty
approach for recognizing different kind of
primitives. Briefly, the contributions of this
work can be summarized as follows.
e Aspect-interpretation was presented to
recognize some simple and regular primitives
that lost less information in projection, as well as
to assistantly construct aspect-graph.
e The constructing regulation of aspect-graph
based on aspect-interpretation was presented.
e The data structure of model matching was
studied
e A measurement system was developed.
2. ASPECT-INTERPRETATION
PRINCIPLE
some features from a CSG tree. Marefty[15] A 2-D image is perspective projection of 3-D
used a geometric reasoning for recognition object, which consists of a number of primitives
object. Most of these schemes adopted model by Boolean operator(unit, intersection, different)
matching, however, previous methods have in CAD system. For this 2-D image, a number of
limited success due to complex state of edges constructs a face in reasonable allocation
projection in intersection, difference and way, and a number of faces forms a volume-
emergence among primitives. Furthermore, primitive in reasonable allocation way, which
1018
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
pr
CO
fo
Ire
st:
th
-— FN e Fa o (YA 09 AN
