dc.contributor.author |
Mamun, Md. Afif Al |
|
dc.contributor.author |
Kadir, Imamul |
|
dc.date.accessioned |
2020-11-21T10:22:02Z |
|
dc.date.available |
2020-11-21T10:22:02Z |
|
dc.date.issued |
2020-06-06 |
|
dc.identifier.uri |
http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/5116 |
|
dc.description.abstract |
Navigating from one place to another has been a problematic task for the blind. In
Bangladesh, the existing footpaths are mostly crowded or broken. Often, visually impaired
people get hurt while walking on a footpath as they do not have anything but a stick to help
them. Considering the problem scenario, we are proposing a smart solution to identify safe
footpath and detect obstacles in a footpath. The system will also be capable of estimating
the distance of the object as well as suggesting the safe pathway. To train the models we
built a dataset of footpath images of Dhaka containing 3,000 hand-annotated RGB images
for semantic segmentation and another dataset containing 500+ samples of real-world
distances of reference objects w.r.t to their pixel coordinates in an image for distance
estimation. We adopted and modified the U-Net architecture that is trained on our
segmentation dataset which is capable of inference safe footpath with 96% accuracy with
as low as 4.7 million parameters. The system utilizes YOLOv3 architecture for object
detection and a polynomial regression based novel approach to estimate object distance.
The distance measurement model obtains a score of 94%. |
en_US |
dc.language.iso |
en_US |
en_US |
dc.publisher |
Daffodil International University |
en_US |
dc.subject |
Computer Vision |
en_US |
dc.subject |
Visually Disabled Persons |
en_US |
dc.title |
an-Eye |
en_US |
dc.title.alternative |
Safe Navigation in Footpath for Visually Impaired Using Computer Vision Techniques |
en_US |
dc.type |
Other |
en_US |