PE&RS November 2015 - page 858

number of hypothesis-verification attempts for each frame
to obtain a winning candidate matching solution for each
approach. Our results suggest that LR-RANSAC considerably
outperforms the traditional RANSAC in both the outdoor and
indoor environments with approximately 1,000 less full-
verification attempts. Other sampling and consensus methods
reduce the number of iterations required by RANSAC. An exam-
ple of such an approach is BaySAC (Botterill et al., 2009) but
requires knowledge of prior probabilities of the inliers using
an additional pre-processing step.
Concluding Remarks and Future Works
We have presented a novel model-based vision method,
which automatically matches 2D image lines with 3D wire-
frame linear features by estimating focal length and angular
parameters of PTZ image sequences. The proposed approach
consists of: (a) an initialization of the camera parameters
using vanishing points, and (b) the Line-based Randomized
RANSAC (LR-RANSAC ). LR-RANSAC is an iterative algorithm com-
prising three stages: Hypothesis Generation, Pre-verification
Testing and Hypothesis Verification.
The method has been experimentally validated on test
video images with camera motion. Experiments with the
indoor dataset resulted in a 0.06 mm error for focal length,
while rotation errors were in the order of 0.18° to 0.24°. The
outdoor dataset showed a similar absolute mean error of 0.07
mm for focal length with rotational errors that range from
0.19° to 0.30°.
The LR-RANSAC approach has been compared to a sequen-
tial relative angular orientation of the PTZ video frames and
the results show that it is not prone to the parameter drifting
problem. Our results also suggest that LR-RANSAC requires
significantly fewer verification attempts and is therefore more
efficient than RANSAC for addressing the 3D/2D line-matching
problem. Since the efficiency of the LR-RANSAC has not been
fully evaluated using optimized programming, no conclusive
statements can be made about its effectiveness for real-time
applications. Assuming continuous and smooth camera mo-
tion can eliminate the need for time-consuming vanishing
point detection. In such a case, the camera parameters of the
previous frame can be used for initiating the LR-RANSAC pro-
cess in the current image frame. Vanishing points also limit
the algorithm to image scenes that conform to the Legoland
World assumption.
To handle images without this 3D local scene geometry,
sensors such as Inertial Measurement Units can be used in
conjunction with the camera to provide initial estimation of
rotation parameters. The nominal focal length value can be
used to initialize the system in such cases. LR-RANSAC can
then be applied for optimal, accurate matching of the ex-
tracted image lines with the back-projected CAD model lines.
In conclusion, our experimental outcomes show that the
developed method successfully estimates camera focal length
and orientation parameters in a fully automated process with
RMSE less than three pixels. Future work will include the es-
timation of the remaining elements of the interior orientation
(principal point and lens distortion parameters).
Acknowledgments
This work was financially supported by the Natural Sciences
and Engineering Research Council of Canada (NSERC) and
the Geomatics for Informed Decisions (GEOIDE) network. We
would like to sincerely thank Professor James Elder at Depart-
ment of Electrical Engineering and Computer Science (EECS),
York University, for providing York’s outdoor surveillance
video sequences. We also very much appreciate the comments
and suggestions provided by the anonymous reviewers, which
significantly improved the original version of the manuscript.
References
Adam, A., E. Rivlin, I. Shimshoni, and D. Reinitz, 2008. Robust
real-time unusual event detection using multiple fixed-location
monitors, IEEE Transactions on Pattern Analysis and Machine
Intelligence, 30(3):555–560.
Aider, O.A., P. Hoppenot, and E. Colle, 2005. A model-based method
for indoor mobile robot localization using monocular vision
and straight-line correspondences, Robotics and Autonomous
Systems, 52, pp. 229–246.
Baklouti, M., M. Chamfrault, M. Boufarguine, and V. Guitteny,
2009.Virtu4D: A dynamic audio-video virtual representation
for surveillance systems, Proceedings of the 3rd International
Conference on Signals Circuits and Systems, Vol. 6-8, pp. 1–6.
Botterill, T., S. Mills, and R.D. Green, 2009. New conditional
sampling strategies for speeded-up RANSAC, Proceedings of the
British Machine Vision Conference, pp. 1–11).
Canny, J.F., 1986. A computational approach to edge detection, IEEE
Transactions on Pattern Analysis and Machine Intelligence,
8(6):679–698.
Chum, O., and J. Matas, 2002. Randomized RANSAC with td, d test,
Proceedings of the British Machine Vision Conference (P. Rosin
and D. Marshall, editors), Vol. 2, pp. 448–457.
Coelho, C., M. Straforini, and M. Campani, 1990. Using geometrical
rules and a priori knowledge for understanding of indoor scenes,
Proceedings of the British Machine Vision Conference, pp. 229–234.
Denis, P., J. Elder, and F. Estrada, 2008. Efficient edge-based methods
for estimating Manhattan frames in urban imagery, Proceedings
of the European Conference on Computer Vision, pp. 197–210.
Fischler, M., and R. Bolles, 1981. Random sample consensus: A para-
digm for model fitting with applications to image analysis and
automated cartography, Readings in Computer Vision, ACM, 24(6).
Grimson, W.E.L., and T. Lozano-Perez, 1987. Localizing overlapping
parts by searching the interpretation tree, IEEE Transactions on
Pattern Analysis and Machine Intelligence, 9(4):469–481.
Habib, A.F., H.T. Lin, and M.F. Morgan, 2003. Line-based modified
iterated Hough transform for autonomous single-photo resection,
Photogrammetric Engineering & Remote Sensing, 69(12):1351–1357.
Hartley, R., and A. Zisserman, 2003. Multiple View Geometry in
Computer Vision, 2nd Edition, Cambridge University Press.
Hough, P.V.C., 1962. Method and Means of Recognizing Complex
Patterns, US patent 3069654 (18.12.1962).
Huertas, A., and R. Nevatia, 1998. Detecting changes in aerial views
of man-made structures, Proceedings of the 6th International
Conference on Computer Vision, Washington, D.C, pp. 73–80.
Jaw, J.J., and N.H. Perny, 2008. Line feature correspondence between
object space and image space, Photogrammetric Engineering &
Remote Sensing, 74(11):1521–1528.
Kim, K., S. Oh, J. Lee, and I. Essa, 2009. Augmenting aerial Earth
maps with dynamic information, Proceedings of the IEEE
International Symposium on Mixed and Augmented Reality
2009, Science and Technology Proceedings, pp. 19–22.
Kovesi, P.D., 2011. MATLAB and Octave Functions for Computer
Vision and Image Processing, Centre for Exploration Targeting,
School of Earth and Environment, The University of Western
Australia.
Leibe, B., K. Schindler, N. Cornelis, and L. Van Gool, 2008. Coupled
object detection and tracking from static cameras and moving
vehicles, IEEE Transactions on Pattern Analysis and Machine
Intelligence, 30(10):1683–1698.
Lowe, D.G., 2004. Distinctive image features from scale-invariant key-
points, International Journal of Computer Vision, 60(2):91–110.
Luhmann, T., S. Robson, S. Kyle, and I. Harley, 2006. Close Range
Photogrammetry: Principles, Methods and Applications,
Whittles, Caithness, 510 p.
858 November 2015 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING