Academy of Taiwan Information Systems Research (ATISR)
Abstract:
This study proposes an obstacle detection method based on depth information to aid
the visually impaired people in avoiding obstacles as they move in an unfamiliar
environment. Firstly, we have applied dilation of morphology and erosion of
morphology to remove the crushing noise of the depth image and have used the Least
Squares Method (LSM) in a quadratic polynomial to approximate floor curves and
determine the floor height threshold in the V-disparity. Secondly, we have searched
for dramatic changes depth value in accordance with the floor height threshold to find
out suspicious stair edge points. Thirdly, we have used the Hough Transform to find
out the location of the drop line. In order to strengthen the characteristics of the
different objects to overcome the drawbacks of the region growing method, we have
applied edge detection to remove the edge. Fourthly, we have used the floor height
threshold and features of the ground to remove ground plane. And then our system has
used the region growing method to label the tags on different objects. It has analyzed
each object to determine whether the object is a stair. Fifthly, if the result is neither up
stair nor down stair, we have used K-SVD algorithm to determine whether the object
is people. Finally, the system has assisted the users to determine the stairs direction
and obstacle distance through a voice prompt by Text To Speech (TTS). Experimental
results show that the proposed system has great robustness and convenience.