An Edge-based Inexact String Matching for Stereo Matching

Abstract

In the recent proposed stereo matching algorithm scholars have increasingly pursued accuracy Computation time has gradually been ignored which lead to the algorithm implemented cannot be applied to the real-time 3D computing and playback system The research of stereo matching in single frame has been gradually encountered a bottleneck Therefore this paper focuses on two key points: real-time computing as well as innovative methods This paper breaks the four stage stereo matching algorithm based on [1] as proposed and does not need to calculate matching cost which requires lots of computation time Instead of calculating the disparity of each pixel we use inexact string matching to find the corresponding edge pixel and then propagate the disparity to the other unmatched pixel In order to enhance the quality of the final result we also use vertical voting and post processing filters to refine the depth map Each step of the algorithm is designed with parallelism considerations such that the computations can be accelerated with CUDA implementations on GPU

Date of Award	2014 Jul 7
Original language	English
Supervisor	Pau-Choo Chung (Supervisor)