Scene analysis of unstable video flows - using multiple retina channels and attentional methods

Anna Lázár, Karl Pauwels, Marc Van Hulle, Tamás Roska

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

The present work describes a bio-inspired approach designed to solve some tasks raised in the 'Bionic Eyeglass Project' [1], which aims to help the everyday life of blind or visually impaired people. The purpose of this approach is to provide a specific kind of information or to determine the region of interest ("ROI") in a low-resolution and unstable video flow recorded with a mobile phone by the visually impaired person. In the present paper we introduce a new stabilization method and three different tasks are resolved: Firstly, locating LED (light-emitting diode) indicators, secondly, locating traffic signs, and thirdly, deciding whether there is any switched-on lamp in a room. The test database has been made out of real-life scenes. We provide detailed evaluation results referring to the execution of those tasks. Our descriptive context refers to the recently modeled mammalian retina channel decomposition [2]. Using it we can avoid - at least partially - the classical difficulty that image processing algorithms nowadays face, namely that the intensity or color values of the same object largely depend on the actual lighting conditions. A further difficulty referring to the lamp-detection is that the solution has to be completely independent of the input's actual brightness. The method we introduce relies only on a single retina channel and achieves a very high accuracy: The ratio of the correct answers is around 99%. The other two tasks are to carry out an approximately real-time ROI-detection algorithm based solely on image information from an unstable low-resolution video-flow containing complex real-life scenes with unconstrained lighting conditions. The accuracy of the introduced methods is around 80%. We also make use of a stabilization algorithm designed especially for this project. In order to yield the desired information, we process channel-data as well as saliency maps. The presented method can be useful in a variety of other application areas.

Original languageEnglish
Title of host publicationIntegrated Circuits, Photodiodes and Organic Field Effect Transistors
PublisherNova Science Publishers, Inc.
Pages95-113
Number of pages19
ISBN (Electronic)9781617618680
ISBN (Print)9781606926604
Publication statusPublished - Jan 1 2009

    Fingerprint

Keywords

  • CNN
  • Cellular wave computing
  • Region of interest
  • Retina channel
  • Saliency map
  • Video flow stabilization

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Lázár, A., Pauwels, K., Van Hulle, M., & Roska, T. (2009). Scene analysis of unstable video flows - using multiple retina channels and attentional methods. In Integrated Circuits, Photodiodes and Organic Field Effect Transistors (pp. 95-113). Nova Science Publishers, Inc..