3-D object recognition using passively sensed range data
Model-based object recognition is typically addressed by first deriving structure from images, and then matching that structure with stored objects. While recognition should be facilitated through the derivar tion of as much structure as possible, most researchers have found that a compromise is necessary, as the processes for deriving that structure are not sufficiently robust. We present a technique for the extraction, and subsequent recognition, of 3-D object models from passively sensed images. Model extraction is performed using a depth from camera motion technique, followed by simple interpolation between the determined depth values. The resultant models are recognised using a new technique, implicit model matching, which was originally developed for use with models derived from actively sensed range data . The technique performs object recognition using secondary representations of the 3-D models, hence overcoming the problems frequently associated with deriving stable model primitives. This paper, then, describes a technique for deriving 3-D structure from passively sensed images, introduces a new approach to object recognition, tests the approach robustness of the approach, and hence demonstrates the potential for object recognition using 3-D structure derived from passively sensed data.
KeywordsObject Recognition Camera Motion Object Position Gaussian Image Directional Histogram
Unable to display preview. Download preview PDF.
- 1.Dawson, K. Vernon, D.: Model-Based 3-D Object Recognition Using Scalar Transform Descriptors. Proceedings of the conference on Model-Based Vision Development and Tools, Vol. 1609, SPIE — The International Society for Optical Engineering (November 1991)Google Scholar
- 2.Sandini, G., Tistarelli, M.: Active Tracking Strategy for Monocular Depth Inference Over Multiple Frames. IEEE PAMI, Vol.12, No.1 (January 1980) 13–27Google Scholar
- 3.Vernon, D., Tistarelli, M.: Using Camera Motion to Estimate Range for Robotic Part Manipulation. IEEE Robotics and Automation, Vol.6, No.5 (October 1990) 509–521Google Scholar
- 4.Vernon, D., Sandini, G. (editors): Parallel computer vision — The VIS a VIS System. Ellis Horwood (to appear)Google Scholar
- 5.Horn, B., Schunck, B.: Determining Optical Flow. Artificial Intelligence, Vol.17, No.1 (1981) 185–204Google Scholar
- 6.Grimson, W.: From Image to Surfaces: A Computational Study of the Human Early Visual System. MIT Press, Cambridge, Massachusetts (1981)Google Scholar
- 7.Faugeras, O., Herbert, M.: The representation, recognition and locating of 3-D objects. International Journal of Robotics Research, Vol. 5, No. 3 (Fall 1986) 27–52Google Scholar
- 8.Horn, B.: Extended Gaussian Images. Proceedings of the IEEE, Vol.72, No.12 (December 1984) 1671–1686Google Scholar
- 9.Brou, P.: Using the Gaussian Image to Find Orientation of Objects. The International Journal of Robotics Research, Vol.3, No.4 (Winter 1984) 89–125Google Scholar
- 10.Dawson, K.: Three-Dimensional Object Recognition through Implicit Model Matching. Ph.D. thesis, Dept. of Computer Science, Trinity College, Dublin 2, Ireland (1991)Google Scholar