Fast and accurate ground plane detection for the visually impaired from 3D organized point clouds

Abstract

Detecting the ground plane is a prior stage to obstacle avoidance systems for the visually impaired. The ground plane is where the visually impaired can move on. This paper presents an algorithm that help the visually impaired navigate in a fast, safe, reliable and independent way. Using RGB-D scanners, enhanced RANdom SAmple Consensus (RANSAC) algorithm is proposed to eliminate the common RANSAC problems. The proposed algorithm is able to detect the ground plane and obstacles that face the visually impaired. The proposed algorithm consists of three main stages: data preprocessing, ground plane segmentation and object detection. Two sets of experiments have been made using two datasets and real world data. The results show accuracy (99.9%) and speed (21Hz) of our proposed algorithm.