EFFICIENT FEATURE EXTRACTION FOR STEREO VISION APPLICATIONS
The first part of this paper outlines the practical implementation of the Marr-Hildreth edge detection technique which is based on observations and hypotheses on the early processing of biological vision systems and leads to perform edge detection on several different channels in conjunction. Different band-pass and derivative filters are required to perform separate edge detection on these channels. This efficient method requires big convolution kernels and consequently considerable computing times. The package developed and written in C language is running in reasonable times on computer class serial machines. Benchmarks have been run using 512x450 pixel, 256 gray-level images. Experimental results are discussed and reported. The second part outlines two projects in which the package is being used, both concerned with depth reconstruction via stereo pairs matching with different correspondence approaches and algorithms.
Bibliographic Reference: 7TH INTERNATIONAL CONFERENCE ON ROBOT VISION AND SENSORY, ZUERICH/CH, FEB. 2-4, 1988, AVAILABILITY: CEC-LUXEMBOURG, DG XIII/A2, BP 1907, MENTIONING PAPER ORA 33810 E
Availability: Can be ordered online
Record Number: 1989126066300 / Last updated on: 1989-05-01
Available languages: en