Derivative edge detection techniques : state of art and a practical implementation for stereopsis
The Joint Research Centre Ispra is currently engaged in the development of stereo-image processing techniques for the input of 3D site geometry to a computer. A prototype system has been developed which demonstrates the complete link from stereo image processing to a simulation of the principal energy flows across a building envelope. The aim of the work is to improve the performances of the stereo reconstruction prototype system and decrease its dependence upon human intervention. The report describes the research done both to enhance the input to the stereo matcher phase, i.e. the preliminary segmentation and features extraction from the segmented image, and to test a different stereo matcher, based on dynamic programming. The preliminary segmentation was achieved using the Marr-Hildreth approach, which belongs to the class of derivative edge detection techniques. The theoretical part of derivative edge detection is described through an extended literature review, an exposition of the chosen approach and the performance comparison problem. The theoretical and practical problems faced in the implementation of the Marr-Hildreth edge detection scheme and the results obtained are described. The application of the developed implementation to a stereopsis module based on dynamic programming (Groppello87a) is presented together with the conclusions drawn.
Bibliographic Reference: EUR 12410 EN (1989) MF, 108 pp., ECU 8, blow-up copy ECU 13.75
Record Number: 198911168 / Last updated on: 1994-12-01
Original language: en
Available languages: en