טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentTomer Baum
SubjectComputational Aspects of Compound Eye Systems
DepartmentDepartment of Applied Mathematics
Supervisors Full Professor Rivlin Ehud
Dr. Katzir Gadi


Abstract

The compound eyes of invertebrates offer a concept of a visual system design different from common visual systems of vertebrates. Invertebrates perform vision based tasks such as hunting and navigating, providing us with unique system designs of proven efficiency. The compound eye comprises numerous small light sensing structures (ommatidia). Each eye can be considered as a surface of light sensors, directed in different directions and differing in characteristics such as size, aperture and color sensitivity. The shape of the surface influences the field of view and resolution. Thus the compound eye offers us a different approach to vision and the understanding of the observable space. A main objective of this research is the better understanding of the virtues of the compound eye and how they may be implemented in computer vision systems and provide new possibilities in producing vision systems that suit different tasks.

This work is in the research area of bio-inspired computer vision, a field that has two main directions of information sharing. The first direction, combines computer vision technologies in biological research aimed to improve the ability of obtaining and analyzing data. The second direction is the application of biological research and observations in systems and algorithms of computer vision and in mathematical models related to computer vision. Our research implements the two directions with a focus on the second:

In a work that is in the direction of information flow from biology to computer vision, we present a design for a stereo compound eye system that performs self-motion for depth estimation, based on the praying mantis' visual system. In a further study, we implement observations we had obtained in experiments with praying mantis in an algorithm for active tracking and pursuit of targets while overcoming occlusions. To the first direction of information flow, from computer vision to biology, belongs a work evaluating mantids' reactions to occlusions.