We have applied the Hough Transform to detect kerbs in images and demonstrated its feasibility on some real road scenes. The uncertainty analysis of the ground plane parameters fit enables us to investigate kerb estimation uncertainty. The kerb detection module developed has been integrated with the GPOD system, which can now provide kerb information to the partially sighted and also detect small obstacles better.
Currently, we are experimenting with the kerb detection algorithm in various scenes to detect different types of kerbs, and extending it for steps/stairs.
The current kerb detection implementation, however, takes 3 seconds on the average to process a pair of 128x128 images on an Ultra-Sparc machine. So before it can be actually used by the partially sighted, we need to achieve at least near real-time speed. From Figure 5 , we can see that it can be optimised substantially by parallelisation: the left and right images can be processed individually, and the inside and outside regions can be processed separately.
Stephen Se