Snap&Nav: Smartphone-based Indoor Navigation System For Blind People via Floor Map Analysis and Intersection Detection

Snap&Nav is a smartphone-based indoor navigation system for blind travellers that works in any building with a visible floor map, without requiring the building owner to pre-build a digital map, install BLE beacons, or deploy any other localisation infrastructure. The authors frame their contribution against two existing research lines: prebuilt-digital-map systems (accurate but rarely available in real buildings) and map-less systems that depend on sighted bystanders describing routes (free-form and often inaccurate). The core idea is to photograph a physical floor map at a building entrance, extract a graph representation of intersections and destinations from the image, and then use on-device intersection detection to localise the blind user on that graph as they walk. The system has two modules. The Map Analysis module runs on an iPhone 12 Pro: a sighted assistant (not the blind user) photographs the floor map, taps to annotate the user's starting position, and swipes to annotate their orientation; the image is then sent to a server where connected-component extraction, skeletonisation, Harris corner detection, and OCR produce a node map of intersections and destinations. The Navigation module uses the phone's LiDAR to build a local 2D occupancy grid and a YOLOv7 model to detect the shape of each intersection the user reaches, matching against the node map to advance the user's position and trigger turn-by-turn voice prompts with scale-estimated distances. The paper reports two studies: 20 sighted participants capturing floor maps, and 12 blind participants navigating three 100+ metre multi-destination routes.

In the sighted-assistant study, the map analysis algorithm achieved an Average Path Length Similarity (APLS) of 0.57 on first trial and 0.56 overall — close to the 0.5 threshold the authors use as the usability bar. Sighted participants captured floor-map images with a mean task completion time of 88.62s on first trial and 62.92s overall, recaptured only 0.21 times per trial, and annotated the user's position correctly in 99% of overall trials. Subjective usability ratings (7-point Likert SUS-derived items) were positive across the board, and all 20 participants said they would be willing to use the system if asked by a blind person. In the blind-participant study, the 12 totally blind users scored the system a median SUS of 92.5, and all six Likert items significantly favoured the system over cane-only navigation (confidence and cognitive load at intersections, near destinations, and overall; p < 0.05 on all items, p < 0.01 on four). Distance-to-destination error was generally below one metre in the system-aided condition versus several metres cane-only (B04 stopped 9.6m from the goal, B08 7.6m), and no participants needed to ask for route descriptions under the system-aided condition. Task completion time was not significantly different between conditions on most sub-routes, though the system added roughly 5 seconds of intersection scanning per intersection. Ten of 12 blind participants said the benefit of the system outweighed the inconvenience of asking a sighted assistant to capture the floor map; two objected to handing over their phone or relying on strangers.

For accessibility practitioners working on wayfinding in public buildings — shopping centres, museums, universities, hospitals, transit hubs — Snap&Nav is significant because it inverts the usual infrastructure burden. Most indoor-navigation accessibility solutions require the building owner to commission a digital map, install beacons, or maintain a mapping pipeline, which is why very few public buildings are actually accessible in this way despite a decade of research. Snap&Nav proposes that printed floor maps at building entrances — which are already legally required or de facto universal in many contexts — can act as the data source, and that a one-off interaction with a sighted person can substitute for permanent infrastructure. For practitioners advising building owners, this reframes the conversation: the question is no longer 'will you pay to digitise your building' but 'does your floor map signage meet a capture-able standard'. The paper also reinforces an interdependence-framed design pattern (sighted capture, blind autonomous navigation) rather than presenting full automation as the goal. Limitations are significant: the study is confined to one university building with simple 90-degree corridors and five floor maps, the map analysis algorithm does not handle complex intersections or open spaces, and the last-few-meters problem (locating a specific door within a destination area) is not addressed. Practitioners should also note that 2 of 12 blind participants opted out on privacy/ownership grounds, suggesting that 'hand the stranger my phone' interactions carry real social cost.