ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Computer vision as a navigation aid for visually impaired

Jan Markočič (2015) Computer vision as a navigation aid for visually impaired. EngD thesis.

Download (2854Kb)


    Accomplishing many of the every-day activities present a significant challenge to visually impaired people. One of such challenges is navigation in an unsupervised environment, which might be well addressed with technological solutions. The recent technological advances have made computational devices with significant processing power a widely affordable commodity. In parallel, the advances in computer-vision-based systems allow scene reconstruction in real time paving the way for realistic systems for obstacle detection. As the field of robotics is quickly advancing with the help of such technologies, we can use it to help visually impaired people. In this work we propose a computer-vision-based system for navigation of visually-impaired in unsupervised environment. The system is composed of environment analysis sub-system and the navigation sub-system. A direct monocular SLAM is used to estimate a model of the user's environment. A new algorithm for analysis of this model and detection of potential obstacles is proposed. Based on the obstacle detection result, the system selects the most appropriate direction of motion to minimize the probability of collision. This information is provided to the user by a sound-based interface -- a stereo headset. The direction of motion is provided to the user by placing sources of sounds in the virtual three-dimensional space that user perceives through the stereo headset. The system was evaluated in a real environment. The results have shown that the subjects wearing the system were six times less likely to collide with the obstacles in their environment than the subjects not wearing the system. Albeit basic, the experiments indicate a great potential of the proposed system for navigation in realistic environments.

    Item Type: Thesis (EngD thesis)
    Keywords: monocular camera, SLAM, depth estimation, obstacle avoidance, stereophonic sound
    Number of Pages: 57
    Language of Content: Slovenian
    Mentor / Comentors:
    Name and SurnameIDFunction
    doc. dr. Matej Kristan4053Mentor
    Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=51012&select=(ID=1536490691)
    Institution: University of Ljubljana
    Department: Faculty of Computer and Information Science
    Item ID: 3056
    Date Deposited: 08 Sep 2015 13:46
    Last Modified: 21 Sep 2015 13:36
    URI: http://eprints.fri.uni-lj.si/id/eprint/3056

    Actions (login required)

    View Item