(Center Identification Number: 79060-02-C)
The long-term goal of the program is to develop a prototype device which will demonstrate the feasibility of the assistive technology to help blind pedestrians. In the proposed project we will develop technology to address two components of the technology needed for the long-term project: (a) Reliable traffic light detection using Google street view (GSW) and (b) Image Geolocalization using GSW.
Blind and visually impaired pedestrians are having increasing difficulty knowing when it is safe to cross streets because, with the advent of hybrid vehicles, cars are becoming quieter. The visually impaired used to be able to determine when it was safe to cross streets as they could hear cars approaching. Now there are increasing numbers of hybrid models being purchased by consumers, and this problem is becoming quite serious. The Department of Electrical and Computer Engineering became aware of this issue when one of its volunteers met with some visually impaired individuals at Friedman Place, a non-profit Supportive Living Community for blind and visually impaired adults, and learned they view these cars as the “silent killer.” In the long term our research group, led by Rashid Ansari, ECE Professor and Department head, plans to develop a small assistive device, possibly wearable, that will enable blind and visually disabled pedestrians to safely cross streets. In the proposed project we will develop technology to address two components of the overall technology needed for the long-term project:
- Reliable traffic light detection using Google street view (GSW) and
- Image Geo-localization using GSW
Reliable traffic light detection using Google street view (GSW): We will improve the performance of a real-time algorithm for traffic light detection building on our preliminary work coupled with the fact that blind or visually impaired pedestrians generally use a limited number of routes for their daily tasks. We will collect video data for typical route and investigate the performance of our algorithm
Image Geo-localization using GSW: We will work on improving the efficiency of a novel efficient algorithm for image analysis using Google street view and commonly used routes to create a pruned database and use image registration/retrieval methods to get accurate localization.
The concept underlying our approach to the assistive technology is to mimic the role of a sighted companion viewing a traffic scene and to communicate the information to a blind pedestrian. This will be done by using a camera to capture the visual street scene and then applying video processing and interpretation technology to understand the traffic scene and detect any approaching vehicles, followed by deciding when it is safe to cross the street and issuing synthesized sounds and/or machine-produced verbal instructions to the blind.
To be created are algorithms with improved performance for the tasks of reliable traffic light detection and image geo-localization described above.
July 1, 2014 to September 15, 2014
Total Project Cost $10,645