LiDAR-Based Obstacle Detection and Distance Estimation in Navigation Assistance for Visually Impaired

Abstract

People with visual impairments can face challenges with independent navigation and therefore may use traditional aids such as guide dogs, white canes, or a travel companion for navigation assistance. In recent years, researchers have been working on AI-based navigation assistance systems. Obstacle detection and distance estimation are two of the key challenges in such systems. In this paper, we describe a LiDAR-based obstacle detection and distance estimation technique. A lightweight deep learning-based model called EfficientDet-LiteV4 is used for obstacle detection, and a depth map from the LiDAR is used to estimate the distance to the obstacles. We have implemented and tested the approach with the LiDAR integrated into a Raspberry Pi4 board. The results show good accuracy in detecting the obstacles and in estimating distance.