Modular Stereo Vision: Model and Implementation

Abstract

The two-frame stereo vision algorithm is typically conceived of and implemented as a single process. The standard practice is to categorize individual algorithms according to the ‘type’ of process used. Evaluation is done based on the quality of the depth map produced. In this chapter, we demonstrate that the stereo vision process is actually composed of a number of inter-linked processes. Stereo vision is shown to be modular in nature; algorithms implementing it typically implement distinct stages of the entire process. The modularity of stereo vision implies that the specific methods used in different algorithms can be combined to produce new algorithms. We present a model describing stereo vision in a modular manner. We also provide examples of the stereo vision process being implemented in a modular manner, with practical example code. The purpose of this chapter is to present this model and implementation for the use of researchers in the field of computational stereo vision.