Automatic recognition of accessible pedestrian signals

Abstract

Accessible pedestrian signals (APS) enhance accessibility in streets around the world. Recent attempts to extend the use of APS to people with visual and audible impairments have emerged from the area of audio signal processing. Even though few authors have studied the detection of APS by sound, comprehensive literature in Biology has been published to detect other simple sounds like birds and frogs calls. Since these calls exhibit the same periodic and modulated nature as APS, many of these approaches can be adapted for this purpose. We present an algorithm that follows this approach. The algorithm was evaluated using a collection of 79 recordings gathered from streets in San Jose, Costa Rica, where an APS system will be implemented. Three types of sounds were available: low-pitch chirps, high-pitch chirps and cuckoo-like sounds. The results showed 91% precision, 80% recall, 83% F-measure, and 90% specificity.