Indoor Localization for Visually Impaired Travelers Using Computer Vision on a Smartphone
Giovanni Fusco, James M. Coughlan · 2020 · Proceedings of the 17th International Web for All Conference (W4A) · doi:10.1145/3371300.3383345
Summary
This paper presents a computer vision-based indoor localization system that runs as a real-time app on a conventional iPhone, designed to help blind and visually impaired travelers navigate indoor spaces where GPS is unavailable. The system combines several technologies into a coherent pipeline: Visual-Inertial Odometry (VIO) from Apple's ARKit tracks the user's relative movement through the environment, while a computer vision algorithm detects Exit signs — which are legally mandated in virtually all commercial and public buildings — to provide absolute position fixes. A particle filter algorithm maintains multiple hypotheses about the user's location and orientation, using a 2D floor plan map that encodes wall positions, sign locations, and traversability constraints. As the user walks and more Exit signs are detected, the particle filter converges on the correct location. A critical design advantage is that the system requires no new physical infrastructure — no Bluetooth beacons, RFID tags, or special markers need to be installed. It relies entirely on existing informational signs already present in the building. The user can hold the smartphone or wear it on a lanyard with the camera facing forward; importantly, they do not need to aim the camera at specific signs, which would be impractical for someone with low or no vision. The app provides audio feedback including text-to-speech announcements when the user enters predefined regions of interest such as office doors, elevators, or stairwells, along with audio tones indicating whether the system has a confident location estimate or is still uncertain.
Key findings
Two user studies with a total of six blind participants (ages 27-72, using white canes and guide dogs) demonstrated the system's feasibility. In the first study, offline analysis showed median localization errors under 1 meter for all four phone-carrying modalities tested (handheld, lanyard, pocket, and strap), with 95% of estimates within 1.5 meters. All four modalities performed roughly equally, suggesting users can choose whichever carrying method they prefer. When the starting location was unknown, the algorithm converged to the correct location after a median walking distance of approximately 12 meters, typically occurring after the user turned at least one corner — corners are particularly useful because they eliminate many false hypotheses that would collide with walls. Median convergence time was about 19 seconds. In the second study using the real-time app, five blind participants navigated routes of 77 and 58 meters. The false positive rate was zero across all participants, and the false negative rate was 0.02 or lower for four of five participants, confirming reliable real-time performance. Battery consumption was significant but comparable to video games — dropping from 100% to 87% over 16 minutes of continuous use.
Relevance
Indoor wayfinding remains one of the most significant daily challenges for blind and visually impaired people, who cannot rely on visual landmarks, directional signs, or room number labels that sighted travelers use constantly. While outdoor navigation has been largely addressed by GPS-based apps like Nearby Explorer and Seeing Eye GPS, indoor spaces — offices, hospitals, airports, universities — remain difficult to navigate independently. This research is notable for its infrastructure-free approach: unlike Bluetooth beacon systems that require installation and maintenance, it leverages Exit signs that already exist in every building. For accessibility practitioners and organizations, this suggests a path toward indoor navigation solutions that can scale to any building with a digital floor plan, without requiring facility managers to install or maintain new hardware. The work also demonstrates thoughtful accessible design — the system does not require users to point their camera at signs, accommodating people with no usable vision. Future development aims to add turn-by-turn directions, spatialized 3D audio guidance, and augmented reality overlays for users with residual vision.
Tags: indoor navigation · wayfinding · computer vision · visual impairment · blindness · smartphone accessibility · assistive technology