DIDA - Dynamic Image Disparity Analysis.
Final rept. Jul 81-Dec 82,
MINNESOTA UNIV MINNEAPOLIS DEPT OF COMPUTER SCIENCE
Pagination or Media Count:
Disparity is a point by point characterization of the translational changes in an image sequence due to motion of the sensor or objects under view. Accurate estimates of disparity are required in order to understand scene dynamics and to determine a wide variety of spatial relationships. This report describes the results of the first phase of the Dynamic Image Disparity Analysis DIDA project. The DIDA project was initiated in order to develop methods for estimating motion induced disparity in real-time. The report outlines the types of approaches possible for estimating disparity. An extensive discussion on evaluating different approaches follows. Precise evaluation criteria are crucial if meaningful performance stands for DIDA are to be developed. The report argues that no single measure of accuracy is meaningful. Instead, performance should be characterized by a collection of measures. The evaluation of alternate disparity estimation techniques must consider this collection of measures together with a precise task analysis. In particular, different applications have different and often contradictory accuracy requirements.