UNIVERSIDADE ESTADUAL DE MATO GROSSOFACULDADE DE CIÊNCIAS EXATASCAMPUS DE BARRA DO BUGRES

ROOF CONTOURS RECOGNITION USING LIDAR DATA AND MARKOV RANDOM FIELD

MODEL ON GRAPH THEORY

AuthorEdinéia Aparecida dos Santos Galvanin

Aluir Porfírio Dal Poz

LiDAR technology has become common in

recent years, allowing rapid and efficient

acquisition of the Digital Elevation Models

Motivation Methodology Results Conclusion

Object segmentation in urban areas, due to

scene complexity, requires the development of

specific methods that integrate the

neighborhood information and the domain

knowledge of characteristics of the interest

objects

Introduction Methodology Results Conclusion

Object extraction and generation of Digital

Terrain Model

Introduction Methodology Results Conclusion

This paper proposes a methodology for

automatic extraction of building roof contours

using a graph-based MRF.

Introduction Methodology Results Conclusion

Main advantage is to provide a general and

natural model for the interaction among

spatially related random variables in the

image.

The proposed methodology comprises the following preprocessing steps:

Introduction Methodology Results Conclusion

High regions

roof properties definition

Energy function

Minimization

Stability Roof contoursyes

not

Recursive splitting region segmentation, region merging , vectorization and polygonization

Introduction Methodology Results Conclusion

Introduction Methodology Results Conclusion

High regions

roof properties definition

Energy function

Minimization

Stability Roof contoursyes

not

Introduction Methodology Results Conclusion

Using the available contours, a region adjacency graph (RAG) is constructed

The neighbourhood ,regions neighboring is defined as,

iR

rR,Rdist|R ijjr,Ri

Introduction Methodology Results Conclusion

The construction of the energy function depends on a prior knowledge of the properties of the object ´roof´.

The features for the first order clique used is the area and rectangularity.

The area feature allows small object in relation to roofs, can be discarded.

senR

Introduction Methodology Results Conclusion

The third attribute allows the verification of parallelism or perpendicularity between objects

Because if either (objects with parallel main axes) or if (objects with perpendicular main axes).

, 0i j

, 90i j

( , ) (2 )i j ijR R sen

Introduction Methodology Results Conclusion

High regions

roof properties definition

Energy function

Minimization

Stability Roof contoursyes

not

Introduction Methodology Results Conclusion

weights that gives relative importance to each term of energy

1 1 1 ( , )

1

(1 )(1 ) (2 )

ln 1 ln 1

n n ni

i i j ijii i i j G

n

i i i ii

pU r p p sen

A

p p p p

rectangularity parameter of the object .

the area of object

angle between the main axes of objects

Introduction Methodology Results Conclusion

High regions

roof properties definition

Energy function

Minimization

Stability Roof contoursyes

not

Introduction Methodology Results Conclusion

three-dimensional visualization of the DEM used in the test

Introduction Methodology Results Conclusion

The extracted polygons are overlaid in red on the intensity image. This figure also shows the reference polygons (in blue) and a false negative (in green).  

Introduction Methodology Results Conclusion

The choice of test area took into account the complexity of the configurations of objects in the scene

In general, a good indication of robustness of the proposed methodology was the lack of false positives and the verification of few false negatives.The completeness parameters showed that the extracted polygons generally have high superposition with their reference polygons. 

REFERENCES

Dubes, R. C., Jain, A. K., 1989. Random Field Models in Image Analysis. Journal of applied Statistics, v. 16, n. 2, pp. 131–164. Haala, N., Brenner, C., 1999. Extraction of buildings and trees in urban environments. ISPRS Journal of Photogrammetry e Remote Sensing, v.54, pp.130-137. Jain, R. Kasturi, R & Schunck, B. G., 1995. Machine vision. MIT Press and McGraw-Hill, Inc New York. Kinderman, R., Snell, J. L. 1980. Markov Random Fields and their applications. Providence, R.I: American Mathematical Society. Kirkpatrick, S., Gelatt, C. D., Vecchi, M. P. 1983. Optimization by Simulated Annealing, Science, pp. 671–680. Kopparapu, S. K., Desai, U. B. 2001. Bayesian approach to image interpretation. 127p. 

THANKS FOR ATTENTION