An electronic 鈥渁rtificial eye鈥�, developed for people with impaired vision, has been shown to reliably identify pedestrian crossings, determine when it is safe to walk across and even measure the width of a road.
The system, created by Tadayoshi Shioyama and Mohammad Uddin, at the Kyoto Institute of Technology in Japan, consists of a single miniature camera that can be clipped onto a pair of glasses and a small wearable computer that analyses video images.
The artificial eye can identify Japanese pedestrian crossings by recognising the white stripes painted across the centre of a road. It can also tell when the signal is flashing to indicate that it is safe to cross. In testing, it successfully identified a crossing 196 times out of 198 and never 鈥渇ound鈥� a crossing where there was not one.
Advertisement
Furthermore, the system can measure the distance of a crossing to within the accuracy of a single step. The length of the crossing 鈥� the width of the road 鈥� is calculated using projective geometry.
Some, but not all, pedestrian crossing systems beep to let blind pedestrians know when to cross. This system 鈥渞elays information using a voice speech system and gives vocal commands and information through a small speaker placed near the ear,鈥� says Shioyama.
Cognitive approach
鈥淢obility is a serious issue for blind and partially-sighted people and new tools like this that may help people with sight problems get around safely are always welcome,鈥� says Katherine Phipps, spokesperson for the UK鈥檚 Royal National Institute of the Blind.
Hilary Buxton, an expert in computer vision at the University of Sussex, UK, says identifying even simple things can be extremely difficult for a computer.
鈥淭he challenge is that you鈥檙e not actually looking for one particular thing,鈥� she told New 杏吧原创. 鈥淭here are so many variations in colour, scale and lighting.鈥�
Buxton adds that a geometric approach to image analysis is good for spotting uniform structures, like a pedestrian crossing, but is ill suited to identifying more irregular, natural things. Many teams are now using machine-learning techniques to 鈥渢rain鈥� their software to recognise objects in a variety of settings, in a manner which mimics the way animals learn.
鈥淎s the set of variations has widened, researchers have moved to more 鈥榗ognitive鈥� [software] approaches,鈥� she says.
Journal reference: Measurement Science and Technology (DOI: 10.1088/0957-0233/15/12/008)