Personal tools

Difference between revisions of "Accelerator for Spatio-Temporal Video Filtering"

From iis-projects

Jump to: navigation, search
Line 1: Line 1:
−
[[File:SaliencyExample.jpg|thumb|500px|Example for a saliency map (image copyright belongs to the [[http://www.blender.org/foundation/ Blender Foundation]]).]]
+
[[File:spt_filtering.jpg|thumb|400px|Spatio temporal filtering example where sparse flow vectors are converted to a dense flow-field ([http://www.disneyresearch.com/wp-content/uploads/Practical-Temporal-Consistency-for-Image-Based-Graphics-Applications-Paper.pdf from]).]]
 
==Short Description==
 
==Short Description==
−
Saliency estimation is an important building block in video processing applications.  
+
There are many image and video processing algorithms (e.g. calculation of Optical-Flow, or Image Domain Warps) that are usually solved using large optimization problems. The solution of such
−
The goal of saliency estimation is to find visually important regions in the image,  
+
large optimization problems in real-time is difficult and sometimes even unfeasible. However, the mathematical structure of some of these problems allows us to approximate their solution by using
−
and this information can be used in further processing steps to adapt the video in a
+
non-linear filtering in the spatial and temporal domain. These filters scale better in terms of computational complexity than the corresponding optimization problems, and therefore would
−
content-aware fashion. Examples for applications that use saliency estimates are
+
allow to perform certain video processing steps more efficiently.  
−
"[[Real-time View Synthesis using Image Domain Warping]]" and "[[A Multiview Synthesis Core in 65 nm CMOS]]".
 
  
−
The goal of this project is to look into an advanced saliency estimation
+
In this project, we would like to implement the core parts of an efficient STEA filtering algorithm which has recently been developed in collaboration with [http://www.disneyresearch.com/research-labs/disney-research-zurich Disney Research Zurich].
−
algorithm, and to develop an efficient hardware architecture for it. Depending on
 
−
the progress, the implementation can also be prototyped on a FPGA.
 
  
 
===Status: Available ===
 
===Status: Available ===
 
: Scope: Semester or Master Thesis
 
: Scope: Semester or Master Thesis
 
: Looking for 1-2 Interested Students
 
: Looking for 1-2 Interested Students
−
: Supervisors: [[:User:kgf|Frank Gürkaynak]], [[:User:schaffner|Michael Schaffner]]
+
: Supervisors: [[:User:schaffner|Michael Schaffner]], [[:User:Lukasc|Lukas Cavigelli]]
  
 
===Prerequisites===
 
===Prerequisites===
 
: VLSI I
 
: VLSI I
−
: Introductory course in computer vision (recommended)
+
: Introductory course in computer vision (optional)
 
: Interest in computer graphics / computer vision
 
: Interest in computer graphics / computer vision
 
: Matlab, VHDL and C++
 
: Matlab, VHDL and C++
  
 
===Character===
 
===Character===
−
: 25% Theory & Literature Study
+
 
−
: 25% Matlab Evaluations
+
TBD
−
: 50% Hw Architecture & FPGA Implementation
 
  
 
===Professor===
 
===Professor===
Line 51: Line 47:
 
* '''[[Final Report]]'''
 
* '''[[Final Report]]'''
 
* '''[[Final Presentation]]'''
 
* '''[[Final Presentation]]'''
−
 
−
 
  
 
----
 
----
  
 
[[#top|↑ top]]
 
[[#top|↑ top]]
−
[[Category:Image and Video Processing]] [[Category:Digital]] [[Category:Research]] [[Category:Master Thesis]] [[Category:Semester Thesis]] [[Category:Available]] [[Category:FPGA]]
+
[[Category:Image and Video Processing]] [[Category:Digital]] [[Category:Research]] [[Category:Master Thesis]] [[Category:Semester Thesis]] [[Category:Hot]] [[Category:Completed]][[Category:ASIC]][[Category:FPGA]] [[Category:FPGA]] [[Category:2016]]

Revision as of 18:48, 14 April 2016

Spatio temporal filtering example where sparse flow vectors are converted to a dense flow-field (from).

Short Description

There are many image and video processing algorithms (e.g. calculation of Optical-Flow, or Image Domain Warps) that are usually solved using large optimization problems. The solution of such large optimization problems in real-time is difficult and sometimes even unfeasible. However, the mathematical structure of some of these problems allows us to approximate their solution by using non-linear filtering in the spatial and temporal domain. These filters scale better in terms of computational complexity than the corresponding optimization problems, and therefore would allow to perform certain video processing steps more efficiently.

In this project, we would like to implement the core parts of an efficient STEA filtering algorithm which has recently been developed in collaboration with Disney Research Zurich.

Status: Available

Scope: Semester or Master Thesis
Looking for 1-2 Interested Students
Supervisors: Michael Schaffner, Lukas Cavigelli

Prerequisites

VLSI I
Introductory course in computer vision (optional)
Interest in computer graphics / computer vision
Matlab, VHDL and C++

Character

TBD

Professor

Luca Benini

Partners

Disney Research Zurich

↑ top

Detailed Task Description

Goals

Practical Details

↑ top

Retrieved from "http://iis-projects.ee.ethz.ch/index.php?title=Accelerator_for_Spatio-Temporal_Video_Filtering&oldid=2131"