Instrumentation & Measurement Magazine 23-4 - 27

volunteers for visual assistance; and iDentifi (iOS) recognizes
objects, colors, facial expressions, and handwritten text.
Another set of commercially available mobile apps is designed for navigation. For instance, Soundscape (iOS) uses
3D audio cues to enrich ambient awareness and provide new
ways to relate to the surrounding environment; Intersection
Explorer (Android) narrates the layout of the streets and intersections in neighborhoods as the user drags a finger over
the map; BlindSquare (iOS) describes the environment and
announces points of interest and street intersections while
travelling; Lazarillo GPS for the Blind (Android, iOS) provides
guidance by voice messages; Arianna Navigation (Android,
iOS) assists the user in indoor (e.g., airports, museums, hospitals, and shopping malls) or outdoor environments (stations,
sidewalks, etc.); and Lazzus identifies points of interest within
the visual field in real time. Recently, Google announced an intelligent Android app project, Lookout, dedicated to the needs
of the VI. It offers auditory clues to nearby objects, text, and
people. The app is designed to be used with a smartphone device hanging around a user's neck, with its camera pointing
away from the body.
The estimated growth of the VI population [1], the positive
user attitudes towards assistive technology [45], and advances
in computer vision technology form a solid foundation for research and development (R&D) projects in the field. Thus,
below we share some results of our findings in this research
area as well.
To get a systematic overview of the latest achievements in
the field, we followed the PRISMA methodology [9]. Three
major research databases (PubMed, IEEE Xplore, and ACM
DL) were queried using a combination of keywords: "visually impaired," "blind," "navigation," "video," "computer
vision," and "app." The exact query used was as follows: (('visually impaired' OR blind) AND navigation AND video) OR
(("visually impaired" OR blind) AND "computer vision") OR
(("visually impaired" OR blind) AND video AND app)). The
search was performed in May, 2018 and covered a publication
period of approximately five years (01 January 2013 to 05 May
2018). Older publications were not considered due to the rapid
development of computer vision algorithms at a speed that
can quickly render previously published solutions irrelevant.
The dataset was first cleaned to remove duplicates. This
resulted in 418 unique research publications included in our
literature screening process. Based on a screening of titles and
abstracts, 335 irrelevant papers were excluded from the review. Full-text analysis excluded an additional 68 papers. The
main reasons for excluding publications after full-text analysis were as follows:
◗◗ no details were provided on implementation, and only
a theoretical description of the proposed solution was
given (20 papers);
◗◗ did not meet our hardware assumptions, meaning the
solution was not smartphone-centric, did not use video
or images from the camera, or did not perform video
processing (for instance, if raw video was forwarded to a
human assistant for interpretation) (32 papers);
June 2020	

Fig. 5. PRISMA flow diagram for proper research papers selection.

◗◗ was focused on a very specific problem and did not
provide any generalizability (for instance, a pothole
detector) (16 papers).
Eventually, 15 research publications remained and were included in this systematic review. Our paper selection process is
visualized in the PRISMA flow diagram shown in Fig. 5.
Describing the surrounding environment through a set
of video-processing algorithms is a complex task, incorporating a high level of uncertainty. Lighting conditions,
movement, transparent and reflective objects, and regionspecific aspects (e.g., the appearance of a rather standard
object, such as a bus stop, can vary greatly in various
regions) are only a few of the challenges that must be addressed by such computer vision systems. On a more general
level, indoor conditions are different from those found in
outdoor scenarios; therefore, more than half of the papers
(eight papers) chose to focus on either indoor-only or outdoor-only tools, while the others (seven papers) aimed for
universal solutions.
Five papers selected an indoor scenario as the main focus
and addressed the challenges using various approaches. To estimate position and direction of movement, computer vision
functionality was supplemented by data from motion sensors
[47]-[49]. Such sensors were sufficient to track the location of
a VI user within a static map of a building, while providing
guidance towards the goal. However, motion sensors have
their limitations-they require maps and continuous tracking
of movements to localize the user. The use of fiducial markers
to improve localization was proposed in three papers (two of
which originated from the same project). Locations of interest were QR coded [47], [49], simplifying the computer vision
and localization tasks to the recognition of the QR codes. Pure
computer vision solutions for indoor environments were discussed in two publications, demonstrating the feasibility of
such systems [50], [51].

IEEE Instrumentation & Measurement Magazine	27
Instrumentation & Measurement Magazine 23-4

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-4
