Doctoral Dissertations

Date of Award

8-1991

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Electrical Engineering

Major Professor

Mohan M. Trivedi

Committee Members

D. Brzakovic, D. W. Bouldin, M. O. Pace, K. Soni

Abstract

The ability to produce precise depth measurements over a wide range of distances and the passivity of the approach make binocular stereo an attractive tool for range sensing. Yet, high computational costs and rigid input requirements have prevented stereo approaches from being widely adopted for practical applications. This dissertation develops and demonstrates a new computational framework for an accurate, robust, and efficient stereo approach. Most of the deficiencies prevailing in current computational models of stereo can be attributed to their use of a single, typically edge element based, primitive for stereo analysis as well as to their use of a single level control strategy. To alleviate these deficiencies, this dissertation develops the multi-primitive hierarchical computational model. In this model, stereo analysis is performed in multiple stages, incorporating multiple primitives and utilizing a hierarchical control strategy. The multi-primitive hierarchical system consists of three integrated subsystems corresponding to three primitives: region, linear edge segment, and edgel. Higher levels of the hierarchical system are based on primitives containing more semantic information. Thus, matching at higher levels produces fewer mismatches. Results of stereo analysis at higher levels of the hierarchy are used for guidance at the lower levels. The control and guidance is provided by generalized hierarchical constraints. This system will be able to address the limitations of purely edge based or region based approaches and is superior to a single level, single primitive system. To systematically evaluate the validity and performance of the multi-primitive hierarchical framework, extensive experimentation is carried out on a variety of scenes from two application domains (industrial automation and aerial image understanding). The images are representative of diverse imaging geometries, resolutions and scene content. These experiments validate the multi-primitive hierarchical model. Results show the utility of region as a primitive of stereo matching. Hierarchical constraints are shown to improve the matching accuracy. The complementary nature of the multiple primitives is also demonstrated. The performance of the stereo system demonstrates the accuracy, robustness, and the efficiency of the multi-primitive hierarchical model. The multi-primitive hierarchical stereo system is able to consistently analyze a variety of images with a high degree of accuracy and reliability. Compared with many existing stereo algorithms the multi-primitive hierarchical system produced better results in significantly less time.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS