|M.Sc Student||Sasporta Aviva|
|Subject||Object Recognition Using Point Uncertainty Regions as Pose|
|Department||Department of Industrial Engineering and Management||Supervisor||Professor Ilan Shimshoni|
This work presents an efficient algorithm for geometric object recognition. The algorithm is based on matching geometric features. The model and image features are both represented in terms of point features.
In this algorithm we generate at each step a set of hypothesized matches between model and image points, estimate from them the pose of the object and store this pose in a lookup table. While a new hypothesis is being entered into entries in the table, it is compared to hypotheses already present in those entries. When two similar poses are found, the pose is suspected to be correct and the hypothesis is verified. Testing whether the two hypotheses are compatible is simply done by testing if several pairs of circles intersect. The “probabilistic peaking effect” is adapted when generating the hypotheses and improves considerably the running time.
The main contribution of this work is that poses and their uncertainty are represented by the uncertainty regions of the projections of several 3D points, whose uncertainty is due to the measurement uncertainty of the image features. When two poses are consistent, the pairs of uncertainty regions of the same 3D points will have a non-empty intersection. Therefore, instead of trying to compute the pose uncertainty region we can use the point uncertainty region to verify directly that poses are compatible. This is due to the fact that when given two compatible poses, the uncertainty regions of the projected model points will intersect in the image. The algorithm exploits the fact that these uncertainty regions can be computed easily and accurately.
The algorithm has been fully implemented and tested on real objects.