Robust Distance Estimation Using LiDAR-Guided Deep Learning for Assistive Mobility

Modern powered wheelchairs (PWs) can greatly improve mobility for individuals with severe motor impairments, but conventional control methods are not always suitable for users with advanced disabilities. This work presents a hybrid measurement framework that combines deep learning-based RGB image analysis with light detection and ranging (LiDAR) sensing to estimate, in real time, the distance between a wheelchair and its caregiver. A You Only Look Once (YOLO) object detection network identifies the caregiver's foot in RGB frames, guiding LiDAR measurements to the precise foot-ground contact point. A gradient-based refinement process further improves distance accuracy by reducing environmental noise. The system was evaluated through static and dynamic tests under realistic conditions, including variations in elevation and approach angle between the caregiver and the wheelchair user. Results demonstrate high measurement accuracy and repeatability, with mean relative errors below 2% in static tests and below 3% in dynamic scenarios. By focusing on the caregiver as a dynamic and reliable reference, the proposed method offers a practical and robust solution for safe, adaptive mobility in complex environments.