Comparative Analysis of Distance Estimation Methods for Mobile Assistive Systems for Visually Impaired Users
Year: 2026
Type: conference-paper
Abstract: Distance estimation is a critical capability for mobile assistive applications designed to support the independent navigation of visually impaired users. This paper presents a systematic comparative analysis of six distance estimation methods applicable to monocular smartphone platforms under real-time constraints: (1) geometric boundingbox estimation based on the pinhole camera model, (2) monocular neural-network depth estimation covering Monodepth2, FastDepth, MiDaS, and Depth Anything V2, (3) stereo vision as a hardware accuracy reference, (4) Time-of-Flight and LiDAR sensors as a hardware performance ceiling, (5) optical flow and Structure from Motion depth, and (6) Google ARCore Depth API with Visual-Inertial Odometry. All software methods were benchmarked on 1,279 video frames using YOLOv8n as the object detector on an Android 12+ smartphone with a single RGB camera and no dedicated depth sensor. The study is conducted under the engineering constraints of the present research: total end-to-end latency not exceeding 250 ms, working range of 1 to 5 meters, acceptable relative distance error of 20 to 40 percent, and battery consumption below 40 percent per hour. The findings support ARCore as the preferred depth backend for production assistive applications, with the geometric method recommended as a lightweight fallback.
Institution: Novosibirsk State Technical University
Language: English
Topic: Advanced Vision and Imaging
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202289356