Efficient visual recognition using the Hausdorff distance (Q2563915)

The book presents a computer vision system for searching an image for a specified model object. The author describes a own very reliable method that uses the minimum Hausdorff distance to visually locate an object in an image. The visual recognition system locates instances of a model in the image by determining transformations of the model which bring a large number of model features close to image features. The Hausdorff distance is a measure defined between two point sets representing a model and an image. The minimum Hausdorff distance is used to find a transformation of the model which brings it into closest correspondence with the image. Some lower bounds show that finding the exact transform that minimizes the Hausdorff distance may be expensive. An efficient rasterized approach to the search is developed. The search technique used is transformation space subdivision which works in a tree-like fashion. Some implementations of this approach are discussed, applications to motion tracking and mobile robot navigation are studied. The system is fast enough for real world applications and tolerates imperfections. The book closes with some interesting open questions for future work.