Three-dimensional object identification and registration (computer vision, recognition)
David B. Shu · 1984
During the past few years there have been great advances in the techniques of acquisition of range data from various sensing systems. The problem of speedily and reliably interpreting these range data for industrial object recognition is becoming critically important in the field of robotics and computer vision. This dissertation presents a methodology to meet such a challenge. It attempts to perform as much classification analysis as possible during the off-line learning process, while only a very small subset of the range data needs to be processed for the on-line recognition. It is assumed that there are only a small number of object types and the objects can be approximated by polyhedra. An adaptive matched filter is developed to extract an object's surface feature vectors using range measurement and surface normals. A sparse representation for each object category j is constructed for classification purpose in the form of a Rj-table of selected feature vectors. The approach, based on the concept of the generalized Hough Transform, classifies an object by examining the maximum votes which it receives from various Rj-tables. During the off-line learning phase an object tree is constructed for each prototype. The massive geometric data of a given prototype can be reordered through the tree traversal. An optimal selection rule is established for minimizing the misclassification probability. In this way the common feature vectors among object categories are removed as much as possible, while the distinctive feature vectors are selected in the Rj-table with statistical significance. Thus, it will result in a significant saving of the on-line processing time in comparison to the conventional three-dimensional scene segmentation approach. An experiment with simulated range images of nine categories demonstrated the success of the proposed methodology.