Crosswalk location, direction and pedestrian signal state extraction system for assisting the expedition of person with impaired vision

Object of this research is to test image processing algorithm under actual usage condition for assisting device that transmit time-varying spatial information of the pedestrian crosswalk to guide safe road crossing for person with impaired vision. Based on the opinion acquired from the hearing of six persons with impaired vision, the concept design of a pedestrian crosswalk guide was drawn as a wearable device with USB video camera, microcontroller, vibrator and tactile projector. In this paper, a crosswalk information extraction algorithm is tested on a test-bed laptop PC with a USB camera. The algorithm uses a template matching and normalized correlation method. The efficiency of the algorithm is evaluated under a field test where the image was gained at daytime on pedestrian road and crosswalk in front of two train stations. It was confirmed that the crosswalks were correctly detected in 86.7% of the total image frames. The algorithm was effected by disturbance from roadside tree shadow and non-targeted images which formed regular contrasting pattern.