Tracking Objects with a Recognition Algorithm
Résumé
In this paper, we propose an efficient method for tracking 3D modelled objects in cluttered scenes. Rather than tracking objects in the image, our approach relies on the object recognition aspect of tracking. Possible matches between image and model features define volumes within a transformation space. The model is best aligned with the image in volumes satisfying the greatest number of correspondences. Object motion defines a trajectory in the transformation space. We propose an efficient algorithm to compute these transformations.