GPU parallel processing for fast robotic perception
Abstract/Contents
- Abstract
- In this thesis we detail how we built GPU-executed versions of two modern computer vision applications: one a visual servoing algorithm (to control a robotic arm with visual feedback) and the other a sliding-window object detector (to recognize instances of an object class in an image). Starting from slow reference solutions for both algorithms, our goal was to achieve as close to real-time speeds as possible so that these state-of-the-art algorithms could be incorporated onto a mobile robotics platform. We used general-purpose computing on graphics processing units (GPGPU) to accelerate the programs as this strategy has already been used to achieve large speed-ups (of factors of ten to a hundred) on other complex problems. We furthermore developed our applications with CUDA (Compute Unified Device Architecture), which is a programming interface developed by NVIDIA Corporation to perform GPGPU on their graphics hardware. In this thesis we first give a detailed introduction to CUDA, and then we describe in-depth how we built the GPGPU versions of the two computer vision algorithms. In both cases our goals of real-time execution were achieved, seeing end-to-end speed-ups of around fiftyfold.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Copyright date | 2010 |
| Publication date | 2009, c2010; 2009 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Baumstarck, Paul Gregory |
|---|---|
| Associated with | Stanford University, Department of Electrical Engineering |
| Advisor | Ng, Andrew Y, 1976- |
| Thesis advisor | Ng, Andrew Y, 1976- |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Paul Baumstarck. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Engineering Stanford University 2010 |
| Location | electronic resource |
Access conditions
- Copyright
- © 2010 by Paul Gregory Baumstarck
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...