concl.tex

%!TEX root = main.tex
\chapter{Conclusion}\label{chap:concl}

Block matching is currently one of the fastest dense depth map algorithms
[TODO: source?]. Though the quality is far from the best, it is probably
usable in some applications. With GPUs and CPUs getting better and cheaper,
the OpenCL and CUDA frameworks maturing, a regular home computer has the
computing power of a small super-computer, depth estimation

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Summary}

In this section, you typically write about wyat you have done to address the
problem statement in section~\ref{sect:prob-statement}.



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Future work}

Paired with a silhouette extractor, the speed and quality can be further
improved. Silhouettes can be used to identify object borders and depth
discontinuities so they can be treated differently, with a smaller aggregation
window for example, or a different cost matching method, and also mask out a
region of interest, allowing the depth estimator to ignore background, if it
is of no interest.

A better understanding of signal processing and image sampling is needed to
improve on the Pyramid algorithm and Compute mask steps.