A methodology called the CIPS (cooperative independent parameter spaces) approach for reconstructing parametrized regular surfaces from range data is presented. The parametrizations are decomposed into subsets of parameters. The conjunction of the individual parameter detections in these subsets produces the full parametrization for a surface. The detections are accomplished using a multiwindow parameter estimation technique, multiresolution k-tree parameter space searching and voting, and a conflict resolution process that eliminates invalid parameter hypotheses and insures a single unique parametrization for each surface region. The overall decomposition of parameter detection spaces can be organized into a serial, parallel or hybrid architecture without problems of parameter crosstalk between spaces. Many of the major shortcomings of the Hough transform and other parameter space voting approaches are directly addressed by these methods. An implementation that detects spheres and cylinders in real, low-resolution range images is presented, and it is shown to be fast and accurate.>

A vision system to recognize 3-D objects is presented. A novel notion of generalized feature makes it possible to develop a homogeneous architecture to support recognition from simple partial features to complex feature assemblies and 3-D objects. Layered concurrent parameter transforms vote for feature hypotheses on the basis of image data and previously reconstructed features. Recognition networks, motivated by connectionist systems, collect votes, fuse evidence from various sources and ensure global consistency. In addition, they provide an integrated solution to the segmentation problem. The highly modular design allows fundamentally different types of features to interact in a coherent way, utilizing redundancies for more robust recognition. Within this paradigm, a system extracts planar patches, patches of quadrics of revolution, and the intersection curves of these surfaces (lines and conic sections in three-space) from a depth map. Reconstructed features index into a model database to form consistent object hypotheses. Experimental results detailing the recognition behavior or real depth maps are included.>

A generalized neighborhood concept is presented which extends the usual techniques for feature extraction using parameter transforms. Generalized neighborhoods allow operators to use the joint information contained in distant portions of the same feature; i.e. to utilize the long-distance correlation present in the image. The generalized neighborhood techniques, by correlating local information over different portions of the image, produce up to two orders of magnitude improvement in accuracy over conventional techniques. The response also becomes more complicated; false features may be detected due to a peculiar form of correlated noise. A general framework, motivated by connectionist networks, is presented which eliminates this behaviour by introducing competitive processes in the parameter spaces. A novel approach to the generation of lateral inhibition links in the networks is proposed which is consistent with generalized neighborhoods. Experiments are provided that show results on range data. Complex surfaces and 3-D surface-intersection curves are reconstructed from the data.>

A method for identifying unoccluded three-dimensional objects from arbitrary viewing angles is presented. The technique uses synthetically generated range data in a model-based feature vector classification scheme. Fourier descriptors and moments are used for feature vector generation from, respectively, contour imagery, and silhouette or range imagery. A method is developed for generating an exhaustive set of library views and worst-case test views that is based on a polyhedral approximation to a sphere. Analysis of the success of this approach is made with experiments on a six-airplane data set. A model of range data noise is developed, and results are presented for both ideal and noisy lower-resolution image-classification tests. The use of multiple views for object identification is discussed, and results for one-, two-, and three-view tests are presented.>

An object recognition system based on global features and nearest neighbor matching is extended and enhanced using classification quality assessment methodology. For quality assessment, a classification decision is processed at two levels. The first is to reject options that are not contained in the model database. The second is to identify the likelihood of error for classifications of known objects. Both stages are based on empirically determined thresholds of measures that are generated solely from the system's a priori knowledge, exploiting the known characteristics of both physical object space and feature space. Results are presented for a standardized object identification task, with a set of six similar known objects, and four unknown objects. It is shown that objects outside the model database are effectively rejected and that the accuracy of known object identifications is increased by rejecting views of low classification quality.>