Using images from a single camera, the device has a simple structure: unlike sophisticated stereo camera systems it does not need camera calibration. (The information is obtained using a 'camera coordinate system,' so separate images do not need to be taken to calibrate the device). The length of a pedestrian crossing is measured by projective geometry: the camera makes an image of the white lines painted on the road, and then the actual distances are determined using the properties of geometric shapes as seen in the image. Experiments carried out by Shioyama and his colleagues showed that the crossing length could be measured to within an error of only 5 per cent of the full length - which is less than one step.
Shioyama and Uddin have now made a breakthrough in detecting the location of crossings in the first place and added this to their original camera. To do this they used a calculation called the "projective invariant" which takes the distance between the white lines (called the band width) and a set of linear points on the edges of the white lines, to give an accurate way of detecting what is or isn't a crossing in a given image.
They used this technique to analyse 196 images and it proved successful in detecting whether there was a crossing present in 194 of them. In the two images where the system made a mistake, it said there wasn't a crossing where there really was one.
Katherine Phipps, Accessible Environments spokesperson at the Royal National Institute of the Blind said: "Mobility is a serious issue for blind and partially sighted people
Contact: David Reid
Institute of Physics