Evaluation of stereo matching for mobile platforms with applications for assisting the visually impaired

Chidester, Benjamin Wilson

Permalink

https://hdl.handle.net/2142/46598 Copy

Description

Title

Evaluation of stereo matching for mobile platforms with applications for assisting the visually impaired

Author(s)

Chidester, Benjamin Wilson

Issue Date

2014-01-16T17:55:37Z

Director of Research (if dissertation) or Advisor (if thesis)

Do, Minh N.

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• stereo matching
• mobile devices
• visually impaired assistance

Abstract

The rising interest in immersive entertainment and enhanced image and video content, along with the development of stereo cameras for mobile platforms, motivates the presented evaluation of stereo matching algorithms for mobile devices. This work investigates this potential for stereo matching on a mobile device for real-time applications, in terms of computation time and quality of depth inference. Several algorithms are tested on an Android tablet housing a Tegra 3 processor using images captured from the on-board consumer-grade cameras. Despite distortions incurred by the lower quality cameras and the computational constraints of a tablet, results show that a simple block matching approach can perform reasonable inference at a rate of 10 frames-per-second. Other methods are shown to be too computationally demanding for real-time applications, as even the fastest alternative local method, using adaptive support weights, requires up to 20 seconds per frame on a 320x360 image. Results also show the impact of lower quality ``real-world'' images on inference performance on algorithms. Additionally, real-time stereo matching on a mobile device is applied to a novel application of assisting the visually impaired with navigation. A system is proposed using a simple block matching algorithm that infers the depth of the scene and communicates the presence of obstacles via sound to the user. The system is housed entirely within the mobile device, overcoming a primary hindrance of many users of assistive technology. The use of a mobile device also allows for an intuitive, interactive experience for the user with the depth information directly via the touch screen of the device. This system demonstrates the added functionality of real-time depth estimation on a mobile device and the potential for aiding the visually impaired with navigation.

Graduation Semester

2013-12

Permalink

http://hdl.handle.net/2142/46598

Copyright and License Information

Copyright 2013 Benjamin Wilson Chidester

Owning Collections