CS332 Visual Processing in Computer and Biological Vision Systems

Binocular Stereo Vision

Reading:

Palmer, S. E. (1999) Vision Science: Photons to Phenomenology, Cambridge: The MIT Press, pp. 206-216, 220-223.

Outline:

Our discussion of binocular stereo vision will begin with the computational aspects of stereo processing, then consider some important properties of human stereo vision and how these properties can be incorporated into a computer model. We will finish with some applications of stereo in computer vision systems. In particular, we will cover the following topics:

(1) Stereo geometry and the goals of stereo processing. We will examine the basic viewing geometry that underlies binocular stereo and the definition of stereo disparity. We will then introduce the Correspondence Problem as the main computational problem in stereo vision.

(2) Stereo correspondence algorithms. Algorithms for solving the correspondence problem can be classified according to a number of properties, including the choice of primitive features to be matched between the left and right images, the additional constraints used to perform a unique matching, and the specific strategy for incorporating these constraints into the matching process. We will first summarize the various primitives and constraints that are typically used in stereo correspondence algorithms, and describe simple algorithms that match regions of the left and right images using measures of similarity based on the sum of absolute differences or normalized correlation.

(3) Human stereo vision and the Marr-Poggio-Grimson multi-resolution stereo algorithm. We will review some important aspects of human stereo vision, including the use of multiple scales of processing, the role of eye movements in stereo vision, and properties of neural mechanisms that are selective for stereo disparity. With these observations in mind, we will explore a multiresolution stereo algorithm proposed by Marr, Poggio and Grimson that incorporates many known features of the human stereo system.

(4) Stereo applications. We will discuss some applications of computer stereo systems in areas such as robotics, automated vehicle guidance, construction of digital terrain maps from aerial photography, and image guided surgery.