A single-user feasibility study demonstrating real-time raycast-based stair and elevation-change detection on consumer smartphone LiDAR hardware

For individuals who are blind or visually impaired (BVI), undetected stairways, curbs, and sudden elevation changes represent a leading cause of fall-related injury. Herein we present a preliminary, single-operator feasibility evaluation of a computationally efficient geometric algorithm for detecting stair and elevation changes in real time, implemented entirely on an LiDAR-equipped consumer smartphone (iPhone 17 Pro Max, iOS 26.2). The algorithm uses a set of eight downward- and forward-facing raycasts projected into an ARKit (Artificial Reality developer Kit)-reconstructed mesh and performs line-segment angular analysis at 60 Hz within a 16.7 ms per-frame budget, without requiring cloud processing, GPU inference, or custom hardware. A single trained operator conducted 120 trials spanning eight real-world hazard locations, five step heights (4.5 to 7.5 in; 11.4 to 19.1 cm), four stair grades (22 to 33 degrees), varied lighting conditions (indoor artificial, outdoor daytime, and outdoor nighttime), and both ascending and descending approaches, yielding an overall detection rate of 92.5% (95% CI 86.4–96.0%) and a mean alert distance of 2.31 ft (0.70 m; SD = 0.91 ft). No false-positive alerts were observed during the hazard-approach or traversal segments of these trials, although dedicated hazard-free control trials were not conducted, so this should not be read as a validated false-positive rate across general non-hazard environments. Detection rate and alert distance both tended to increase with step height, with the smallest steps and one specific indoor location accounting for most missed detections; however, pairwise differences across step-height, lighting, and approach-direction categories did not reach statistical significance at the sample sizes tested (Fisher’s exact p > 0.05 in all cases), and step geometry covaried with test location in this design, so these patterns should be treated as preliminary and hypothesis-generating rather than as an established reliability threshold. Descending approaches produced numerically lower detection rates and shorter warning distances than ascending approaches, consistent with a geometric limitation intrinsic to depth-based sensing of hazards that only become visible at close range. These results indicate that low-latency elevation-hazard detection is technically feasible on commodity mobile hardware under controlled, single-operator conditions; confirming this feasibility finding for real-world use by BVI individuals will require larger, multi-participant studies incorporating natural scanning behavior and dedicated negative-condition testing.