Traffic Light Detection for Portable Assistive Device to Aid Blind Pedestrians
thesisposted on 21.07.2015 by Andrea Manavella
In order to distinguish essays and pre-prints from academic theses, we have a separate category. These are often much longer text based documents than a paper.
Automated traffic light recognition is a key technology of interest in applications involving autonomous vehicles and safe driving. There are other important applications such asassistive technology for visually impaired pedestrians w he re traffic recognition is relevant. Crossing streets and navigating in crowded environments like cities can b e very hard for the significant segment of population that is blind or visually impaired. In the past few years, several research efforts were undertaken and papers published on regular traffic light detection. Some attempts have been made to detect traffic light signals but a comprehensive method still has to b e develop ed. The main goal of the work in this thesis was to develop technology to b e integrated in a simple wearable device for blind people to help them navigate outdoors to perform their everyday life activities and in particular to cross streets safely. The problem of automated traffic light in the absence of infrastructure is addressed. The proposed method examined various alternatives and an algorithm was devised to detect traffic lights by first selecting possible candidates by performing traffic light color extraction, pruning the large candidate set using traffic light properties, next carrying out recognition and classification of lights before finally making a decision on the traffic light signal. When tested on a set of image data, the algorithm achieved go o d results with the estimated correct detection rate of the prototype determined to b e above 90 %. The detection of pedestrian traffic signals indicating “walk and don't walk" was also considered and the algorithm devised for this problem also yielde d go o d results, again with an estimated correct detection rate better than 90 %