Abstract

In this paper an algorithm that performs the tracking of points in a calibrated and noisy image sequence is presented. The candidate points to be tracked must satisfy certain constraints that can be deduced from the multiple view geometry. The idea is to consider as noisy points those candidates which cannot be tracked in the sequence. The robustness of the algorithm has been verified on simulated data using different constraints. The methods are assessed in several cases where the number of noisy points and the noise in the measurement of the points to be tracked are varied. Using this study, it is possible to know the performance of the tracking method. An example that shows a perfect tracking of 8 points in a sequence of 10 images with 500 noisy points per image is shown. © Springer-Verlag 2004.