ZebraRecognizer: Pedestrian crossing recognition for people with visual impairment or blindness. (December 2016)

• Record Type: Journal Article
• Authors: Mascetti, Sergio; Ahmetovic, Dragan; Gerino, Andrea; Bernareggi, Cristian
• Is Part Of: Pattern recognition. Volume 60(2016:Dec.)
• Page Start: 405
• Page End: 419
• Publication Date: 2016-12
• Subjects: Visual impairment -- Blindness -- Pedestrian crossing -- Mobile computing; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.05.002
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Independent mobility is a challenge for people with visual impairment or blindness. Groundbreaking innovation comes from mobile devices (e.g., smartphones) that are convenient platforms to provide assistive technologies in the form of mobile applications. This paper presents ZebraRecognizer, a software module that recognizes zebra crossings and that advances state-of-the-art along two directions. First, it removes projection distortion from the acquired image, hence improving the accuracy of the recognition and making it possible to compute the quantified relative position of the crossing with respect to the user, which is crucial to effectively guide the user. Second, ZebraRecognizer is efficient, as it adopts a customized version of the EDLines algorithm that is also implemented to run in parallel on the GPU. Experimental results show that ZebraRecognizer is accurate, efficient and it computes the crossings position precisely. Abstract : Highlights: ZebraRecognizer identifies zebra crossings from a mobile device camera. The use of accelerometer and gyroscopes help remove projection distortion. The most expensive operations are run on the mobile device GPU. ZebraRecognizer is reliable: precision 1, recall .93 and avg distance error 0.2 m. The system is efficient: it can process about 25 frames per second on iPhone 5S.