THE   PROBLEM

Definition

The problem stated in most simple way is something like - given two camera images, or for that matter more, reconstruct the 3-D scene. None to partial information maybe available about the multiple images.

Our Implementation

Given an image pair for an object,or a scene in general, find the 3-D characterstics through depth map. No information is available about the camera. Although we are using two images we would like not to call it stereographic vision project. This is because, firstly, in the stereo vision the dispariy between images is quite less, our algorithm on the other hand can handle sufficiently large disparity; secondly, the stereo vision uses inly two images whereas our algorithm can be upgraded to any number of images.

Eg.,consider the images -

    
The 'Tripod' image

The various steps involved in processing this pair are shown in later sections of the document

Justification of problem

Why are we going all about 3-D scenes, doing calculations, trying to find depth and all that. Well, here are few point that can justify the problem :

• The robot vision problem - a movable robot must have some sense of 3 dimensions. It must be able to identify as to what is lying near and what is far away. This is neccesary in obstacle avoidance and in easy mobility. To achieve this the robot can be fitted with two cameras placed slightly apart and the algorithm discussed in this presentation can be used.
• Virtual Reality - another fastly emerging area employed not only in entertainment but also in other serious applications. Here also we require 3d information but through large number of cameras.
• Industrial quality control - to maintain dimensional accuracy and shape of finished products.
• Animation