Narrow Results

Model-Based 3D object recognition systems have a variety of potential applications, but widespread use of such systems has not occurred, due to a number of factors including the representational limitations of models. One historical limitation is the discriminatory representation of free-form objects. The system described in this paper recognizes free-form objects in dense range data acquired by a structured light rangefinder. Images and object models are represented as a network of salient segments which are then brought into correspondence until a reliable pose estimate is available. Experiments with a database of images and object models highlight the contributions of this system.

A recursive descent tree traversal algorithm is presented to find the point correspondences between two views. Given two sets of noisy 3-D feature points on multiple rigid objects at two time-sequential views, the points at the first view are constructed as a graph and then another graph is derived from those feature points at the second view so that a maximal matching point and minimum matching error is obtained. The correspondence of the vertices is then found. The algorithm can also he used when the number of points at two views are different. Therefore, after matching, the occluded points at either view or both views can be identified. The computation time of the proposed algorithm is large when the number of feature points is large. A data set splitting strategy for such cases, which can significantly reduce the computation time, is presented. Another algorithm presented is one in which motion parameters are estimated from a matched subgraph and are then used to guide the matching for the rest of the nodes. Computer simulations are performed to show the efficiency and accuracy of the algorithm.

This paper presents a new method of estimating principal curvatures of surfaces from range data. The method proposed is capable of using both scattered and gridded data. It is based on estimating curvatures of curves through sample points, unlike previously reported methods which rely on surface fitting. Results demonstrating that the method is applicable to a wide range of curvature values are presented.

A new approach for computing qualitative part-based descriptions of 3-D objects is presented. The object descriptions are obtained in two steps: Object segmentation into parts and part model identification. Beginning with single- or multi-view range data of a 3-D object, we simulate the charge density distribution over an object's surface which has been tessellated by a triangular mesh. We detect the deep surface concavities by tracing local charge density minima and then decompose the object into parts at these points. The individual parts are then modeled by parametric geons. The latter are seven qualitative shapes, each of which is formulated by a restricted globally deformed superellipsoid. Model recovery is performed by fitting all parametric geons to a part and selecting the best model for the part, based on the minimum fitting residual. A newly defined objective function and a fast global optimisation technique are employed to obtain robust model fitting results. Experiments demonstrate that this approach can successfully recover qualitative shape models from input data, especially when part shapes are not fully consistent with model shapes. The resultant object descriptions are well suited for symbolic reasoning and fast object recognition.

In the field of vision based robot actuation, in order to manipulate objects in an environment, background separation and object selection are fundamental tasks that should be carried out in a fast and efficient way. In this paper, we propose a method to segment possible object locations in the scene and rec-ognize them via local-point based representation. Exploiting the resulting 3D structure of the scene via a time-of-flight camera, background regions are elimi-nated with the assumption that the objects are placed on planar surfaces. Next, object, recognition is performed using scale invariant features in the captured high resolution images via standard camera. The preliminary experimental re-sults show that the proposed system gives promising results for background segmentation and object recognition, especially for the service robot environ-ments, which could also be utilized as a pre-processing step in path planning and 3D scene map generation.

Current robot vision systems consist of pseudo realtime feedback, where a scene is analysed while the robot is in a resting state, or otherwise occupied. For many applications a tight feedback loop with the robot servos being driven by sensor data is desirable. To this end, a lightweight, robot wrist mountable laser range finder has been developed at NRC. This sensor provides single scan lines of range data at a rate of approximately 13 calibrated profiles per second. Calibration of the range data is a nontrivial task. Linear interpolation is used on large samples collected from the range finder. Processing these calibrated scans to provide scene infonnation is a unique problem and considerably different from intensity image processing, and should also be recognized as distinct from processing entire 3-D range images. Although there is only a single raster scan of data, explicit cues to the content of the 3-D scene are available. This paper describes a method of calibrating range data, and a library of utilities written to extract information from the range profiles. Potential applications and future directions are discussed…