All the Way There and Back: Inertial-Based, Phone-in-Pocket Indoor Wayfinding and Backtracking Apps for Blind Travelers

TL;DR

This work addresses indoor navigation for blind travelers in GPS-denied environments by delivering two inertial-based apps: Wayfinding, which uses a known floor plan, and Backtracking, which operates without map knowledge. Wayfinding combines two pedestrian dead-reckoning pipelines (Azimuth/Steps and RoNIN) with a Particle Filter to localize on a floor-plan graph, while Backtracking relies on magnetic-field signatures and turn sequences with sequence alignment (iDTW) to match a return path to the way-in. A user study with seven blind participants demonstrates that pocketed smartphones and smartwatch-based voice alerts can support independent navigation, achieving an average route speed of around 0.50 m/s and a SUS score of 80.36, indicating strong usability despite robustness challenges in open spaces and magnetic variability. The findings show that infrastructure-free inertial navigation can meaningfully assist blind travelers in corridor-laden buildings, with Backtracking offering a flexible, map-free option and room for future improvements, including hybrid localization and enhanced contextual cues.

Abstract

We introduce two iOS apps that have been designed to support wayfinding and backtracking for blind travelers navigating in indoor building environments. Wayfinding involves determining and following a route through the building's corridors to reach a destination, and assumes that the app has access to the floor plan of the building. Backtracking one's route, on the other hand, requires no map knowledge. Our apps only use the inertial and magnetic sensors of the smartphone, and thus require no infrastructure modification (e.g., installation and support of BLE beacons). Unlike systems that use the phone's camera, users of our apps can conveniently keep their phone tucked inside a pocket, while interacting with the apps using a smartwatch. Routing directions are given via speech. Both apps were tested in a user study with seven blind participants, who used them while navigating a campus building.

Figures (11)

• Figure 1: (a): Blue line: Path from participant P5 measured with A/S. The particle cloud is shown with colors ranging from green (high weight) to blue (low weight). Note that while the bulk of the particles follow the actual path, turning into a corridor, some particles enter nearby rooms through their openings. (b): An illustration of route segment assignment. Four route segments meet at a junction (red circle). The walker's path is shown with white circles, while its projection to the assigned route path is shown with grey circles. When the walker is within a circle with radius $T$ around the junction, no segment assignment is made.
• Figure 2: Examples of return path matching using projected sequence (a) and hybrid matching (b). The way-in path is shown with a thick purple line, ending at the black square. The length of each segment is given by the number of steps recorded, multiplied by the step length measured during calibration. The actual path of the participant during the return phase is shown by a gray line. Projected sequences are shown with black lines. In (b), reliable matches are shown as yellow circles. Note that in (a), the length of the initial segment appears to be longer than during the way-in, possibly because the walker took shorter steps, or took additional steps while looking for a place where to turn. In (b), the trajectory is corrected as soon as a new reliable match is found.
• Figure 3: Examples of successful (a) and unsuccessful (b) way-in path optimization. The original reconstructed way-in path is shown in the left panels with a thick purple line, ending at the black square, along with the approximate actual path taken by the walker (as measured from the video), shown with a gray line. The optimized way-in paths are shown in the right panels.
• Figure 4: (a): An hypothetical junction, with the walker's location (as estimated by the app) shown by either gray-filled circle. The larger circles represented the expected radius of location uncertainty. If the walker is located by the app at the junction center (top circle), their actual location can be anywhere in the dashed circle of uncertainty (including before or after the junction). If the app locates the walker at the lower circle, the actual location of the walker is certainly before the junction. (b): An example of notifications produced during route traversal (participant P5, route R1W). Directional notifications are shown as symbols filled in gray, while landmark notifications are shown as filled in orange. For the symbols shown as diamonds, an actual notification was produced. For the symbols shown as circles, a notification was not produced as it would have interrupted an ongoing notification. For a few selected notifications, the content of the speech produced is also shown.
• Figure 5: Pictures of our participants during the trials. (a): P2 in R2W. (b) P7 at the beginning of R3W. (c) P3 in R1W. (d) P6 in R2W. (e) P4 in R1W. (f) P6 in R2W.