Robust global motion estimation

Abstract

A global motion estimation method based on robust statistics is presented in this paper. By using tracked feature points instead of whole image pixels to estimate parameters the process speeds up. To further speed up the process and avoid numerical instability, an alterative description of the problem is given, and three types of solution to the problem are compared. By using a two step process, the robustness of the estimator is also improved. Automatic initial value selection is an advantage of this method. The proposed approach is illustrated by a set of examples, which shows good results with high speed.

References

[1] Kumar S, Biswas M and Nguyen T Q., IEEE Proc. ICASSP., 2004, 333.

[2] Cohen I and Medioni G., Proc. Interpretation of Visual Motion Workshop, CVPR98. Santa Barbara, June 1998.

[3] Bergen J R, Anandan P, Hanna K, and Hingorani Rl., Proc. of 2nd European Conf. on Computer Vision. 1992, 237.

[4] IraniM., RoussoB., PelegS.. International Journal of Computer Vision, 1994, 12: 5

[5] ShiJ., TomasiC.. Proc. of CVPR. IEEE Computer Society, 1994, Washington, Seattle, 593

[6] OdobezJ. M., BouthemyP.. International Journal of Visual Communication and Image Representation, 1995, 6: 348

[7] ZhangZ.. Image and Vision Computing, 1997, 15: 5

[8] ReyW.J.J.. Introduction to Robust and Quasi-Robust Statistical Methods, 1983, Berlin, Springer

[9] Birchfield, S. KLT: An implementation of the Kanade-Lucas-Tomasi feature tracker.

[10] Tomasi. C and Kanade. T., Technical Report CMU-CS-91-132, Carnegie Mellon University, Apr. 1991.