Evaluation of disparity map computed using local stereo parametric and Non-Parametric methods

Abstract

Disparity is inversely proportional to depth. Information about depth is a key factor in many real time applications like computer vision applications, medical diagnosis, model precision etc. Disparity is measured first in order to calculate the depth that suits the real world applications. There are two approaches viz., active and passive methods. Due to its cost effectiveness, passive approach is the most popular approach. In spite of this, the measures are limited by its occlusion, more number of objects and texture areas. So, effective and efficient stereo depth estimation algorithms have taken the toll on the researchers. The important goal of stereo vision algorithms is the disparity map calculation between two images clicked the same time. These pictures are taken using two cameras. We have implemented the non-parametric algorithms for stereo vision viz., Rank and Census transform in both single processor and multicore processors are implemented and the results showsits time efficient by 1500 times.

Underwater Target Detection and 3D Reconstruction System Based on Binocular Vision

“…In recent years, binocular stereo vision, which uses two cameras to get disparity map that can then be used to calculate the depth information and to achieve 3D reconstruction, has become popular in the field of computer vision [ 8 ]. Stereo matching, used for obtaining the disparity map, is a key issue and one of the most extensively studied problems in computer vision applications [ 9 , 10 ].…”

“…Stereo algorithms based on local matching could overcome the afore-mentioned drawbacks. They can work much faster because they estimate pixel correspondence only within a small window [ 8 , 28 ]. However, the matching costs of local matching are more susceptible to noise.…”