SEMIAUTOMATIC EXTRACTION OF 3D CURVES BASED ON SNAKES AND 
GENERALIZED POINT PHOTOGRAMMETRY FROM AERIAL IMAGERY 
Yongjun Zhang,Quanye Du 
School of Remote Sensing and Information Engineering, Wuhan University, 129 Luoyu Road,430079, China 
- zhangyj@whu.edu.cn -duquanye@yahoo.com.cn 
Commission III, ThS-7 
KEY WORDS: Photogrammetry, Feature, Extraction, Building, Edge, Aerial, Imagery 
ABSTRACT: 
The Snakes or active contour models of feature extraction algorithm integrates both photometric and geometric constraints. It derives 
the feature of interest by minimizing the total energy of Snakes with an initial location of the feature. Linear features can be directly 
processed with either x or y collinearity equation under the model of generalized point photogrammetry. In this paper, a new 
approach of extracting 3D curves based on Snakes and generalized point photogrammetry is proposed. Firstly, curve feature is 
extracted based on parametric B-spline approximation and Snakes on a single image. The seed points of curve feature on other 
images are determined by matching corresponding points. Then the corresponding curves are extracted by Snakes. Finally, the 3D 
curve model can be achieved by generalized point photogrammetry. Experimental results show that the proposed approach is feasible 
for 3D curve extraction. 
1. INTRODUCTION 
Linear feature extraction is one of important parts of image 
processing. At present, reconstructed buildings are usually 
regular objects. There are no available algorithms to realize 
automatic and mass reconstruction for which including curve 
edges. Curve extraction and expression on images have been 
studied many years in computer vision field, and many 
available algorithms have been developed. Meanwhile, in 
photogrammetry field, some extraction and reconstruction 
algorithms of roads, contour, coastline are developed, but they 
most applied to a single image, without considering the case of 
big overlap imagery. This paper presents an approach to extract 
and reconstruct building curve edges from digital aerial images. 
However, it is semiautomatic. The identification task is 
performed manually, and some few seed points as 
approximation of curve feature should be provided manually 
but coarsely on a single image. Subsequently, with these seed 
points, the curve feature will be extracted precisely and 
automatically by Snakes. Furthermore, the corresponding 
curves on other images are extracted automatically using 
corresponding point matching and Snakes, and 3D curve model 
can be acquired by generalized point photogrammetry from 
multiple images. 
2. SNAKES 
Snakes, or active contour models was introduced firstly by 
Michael Kass et al. (Kass, et al., 1988). It is used widely in 
many image processing areas, such as image segmentation, 
image tracking, 3D reconstruction, etc. In the recent twenty 
years, it has been researched, and developed Greedy algorithm 
Snakes(Williams, Shah, 1990) , Dynamic Program Snakes 
(Amini, et al., 1990), LSB-Snakes (Grun, Haihong Li, 1997), 
GVF Snakes (Chenyang Xu, Prince Jerry, 1998), FFA Snakes 
(Zhiqiang Hou, Chongzhao Han, 2005), etc. 
A traditional snake is a curve: 
v(s)=(x(s),.y(s)) (1) 
which is under the influence of image forces and external 
constraint forces, while energy is minimum, namely internal 
and external force balance, curve arrives object edges. Energy 
function is: 
Knak= jT TO te( V (9k fe = 
I f n,( v b))+£**«,(49)+e c 
Where E inl represent the internal energy of the spline due to 
bending, 
Ejmage gives rise to the image force, 
E con gives rise to the external constraint force. 
The internal energy can be: 
f =(44 v .(' s F +4 s K(4 2 ) /2 < 3 > 
Where a(s) and P(s) are coefficients, 
v s (s) = dv/ds, 
v ss(s) = d2 v/ds 2 . 
In fact, it becomes the following function during calculating. 
731