In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C. Tournaire O. (Eds), IAPRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
162 
Comparison with manual measurements. In 54 images we 
measured all vanishing points manually, by interactively identi 
fying two straight lines pointing towards a vanishing point. We 
compared these directions with the one determined by the system. 
The histogram of the 143 angular differences between the manual 
and the automatic measurements is shown in fig. 4. About 1/3 of 
the manually measured vanishing points show a difference larger 
than 6 degrees, a threshold taken from the histogram. The other 
95 differences indicate an average angular difference of 1.1 °. As 
a spatial direction has two degrees of freedom, this corresponds 
to a standard deviation in each angular direction of appr. 0.7 C . 
which is slighly larger than twice the internally estimated accu 
racy of 0.3°. 
60 
SO 
40 
30 
20 
10 
0 
0 20 40 60 80 100 
Figure 4: Histogram of angular differences between manually 
measured and automatically computed vanishing points 
Finally, we evaluated the the method on the 102 images of the 
York Urban data base [Denis et ah, 2008], with images of size 
580x 640 pixels. There the vanishing points were manually de 
termined using multiple edges per vanishing point. From the 306 
vanishing points 286 (93 %) were automatically detected. They 
showed an angular difference of l.7°, corresponding to a direc 
tional uncertainty of l.2°. The lower accuracy probably results 
from the lower resolution, thus the shorter line segments used. An 
example, where only two vanishing points were found is shown 
in fig. 5. 
Figure 5: Image with only two vanishing points found correctly. 
The black star is the estimated vanishing point, the white diamond 
the reference point. The green line segments in the image are 
identified as outliers before boosting. After boosting the distance 
to the reference point is diminuished by appr. 30 %. The blue 
vanishing point can be automatically be identified as outlier, as 
So = 126.7 2 is significanly larger than 1. 
6 CONCLUSIONS AND FUTURE WORK 
It up to now does not perform enough self diagnosis, in order to 
reliably determine the number of correct vanishing points. For a 
practical use a priori knowledge should be integratable in a flexi 
ble manner, e. g. constraints on the orientation or the prior knowl 
edge about the relevance of line segments. 
References 
Barnard, S. T., 1983. Interpreting perspective images. AI 21(4). pp. 435- 
462. 
Collins. R. T. and Weiss, R. S.. 1990. Vanishing Point Calculation as a 
Statistical Inference on the Unit Sphere. In: ICCV'90, IEEE Computer 
Society Inti. Conf. on Computer Vision, IEEE Computer Society. Los 
Alamitos, CA, pp. 400-403. 
Coughlan. J. M. and Yuille. A. L., 2003. Manhattan world: orientation 
and outlier detection by bayesian inference. Neural Comput. 15(5) 
pp.1063-1088. 
Denis, P.. Elder. J. H. and Estrada. F. J., 2008. Efficient edge-based meth 
ods for estimating manhattan frames in urban imagery. In: ECCV '08: 
Proceedings of the 10th European Conference on Computer Vision. 
Springer-Verlag. Berlin, Heidelberg, pp. 197-210. 
Deutseher. J.. Isard. M. and MacCormick, J., 2002. Automatic cam 
era calibration from a single manhattan image. In: ECCV '02: Pro 
ceedings of the 7th European Conference on Computer Vision-Part IV. 
Springer-Verlag, London, UK. pp. 175-205. 
Fischler. M. A. and Bolles, R. C., 1981. Random sample consensus: 
a paradigm for model fitting with applications to image analysis and 
automated cartography. Commun. ACM 24(6). pp. 381-395. 
Grammatikopoulos. L., Karras, G. and Pets. E.. 2007. An automatic ap 
proach for camera calibration from vanishing points. 62(1). pp. 64-76. 
Hartley. R. I. and Zisserman. A.. 2000. Multiple View Geometry in Com 
puter Vision. Cambridge University Press. 
Heuvel. F. A. V. D.. 1998. Vanishing point detection for architectural 
photogrammetry. In: Proc. ECCV '02, Springer, pp. 652-659. 
Kong. H., Audibert, J.-Y. and Ponce. J., 2009. Vanishing point detection 
for road detection. Computer Vision and Pattern Recognition. IEEE 
Computer Society Conference on 0, pp. 96-103. 
Korc, F. and Forstner, W., 2009. eTRIMS Image Database for interpreting 
images of man-made scenes. Technical Report TR-1GG-P-2009-01. 
McGlone. C. J.. Mikhail, E. M. and Bethel, J. S., 2004. Manual of Pho 
togrammetry. Am. Soc. of Photogrammetry and Remote Sensing. 
Meidow. J., Beder, C. and Forstner, W., 2009. Reasoning with uncertain 
points, straight lines, and suaight line segments in 2D. International 
Journal of Photogrammetry and Remote Sensing 64. pp. 125-139. 
Rother, C., 2000. A new approach for vanishing point detection in ar 
chitectural environments. In: In Proc. 11 th British Machine Vision 
Conference, pp. 382-391. 
Schindler. G. and Dellaert. F., 2004. Atlanta world: an expectation max 
imization framework for simultaneous low-level edge grouping and 
camera calibration in complex man-made environments, pp. I: 203- 
209. 
Schmirt, F. and Priese, L., 2009. Vanishing point detection with an in 
tersection point neighborhood. In: Discrete Geometry for Computer 
Imagery, Vol. 5810, Springer Berlin / Heidelberg, pp. 132-143. 
Straforini, M.. Coelho, C., Campani, M. and Torre. V., 1992. The recov 
ery and understanding of a line drawina from indoor scenes. 14(2), 
pp. 298-303. 
We presented a new method for vanishing point detection, which 
tracks the uncertainty from the image data to the final result and 
empirically evaluated the result on 292 images. The estimation 
was consistently performed with homogeneous coordinates, for 
which we proposed to use a minimal representation for the un 
certainty. 
The system easily can be extended to determine parameters for 
lens distortion, as proposed by Grammatikopoulos et al. [2007]. 
Tardif. J.-P., 2009. Non-Iterative Approach for Fast and Accurate Vanish 
ing Point Detection. In: Proceedings 12th International Conference on 
Computer Vision. 
Tuytelaars, T., Gool, L. V., Proesmans, M., Moons, T. and Mi, E., 1998. 
The cascaded hough transform as an aid in aerial image interpretation. 
In: In Proc. ICCV, b. pp. 67-72. 
Wildenauer. H. and Vincze. M., 2007. Vanishing point detection in com 
plex man-made worlds. In: ICIAP '07: Proceedings of the 14th Inter 
national Conference on Image Analysis and Processing. IEEE Com 
puter Society, Washington, DC. USA, pp. 615-622.