Particle Filter for Reliable Estimation of the Ground Plane from Depth Images in a Travel Aid for the Blind

Abstract

This paper presents a reliable method for the segmentation of image sequences of 3D scenes used in an electronic travel aid for the blind. We propose an implementation of the particle filtering (PF) algorithm to estimate and track the orientation and position of the ground plane free of obstacles. We explain how the state vector in the PF algorithm was defined, and verify the results on a large set of indoor and outdoor sequences, shot by a moving stereovision camera. The movement of the camera is not restricted, and is capable of six degrees of freedom (DoF). We show that ground plane orientation root mean square (RMS) error estimates do not exceed 2 or 3°, for the roll and pitch angles, respectively. The overlap between the mean values for the Jaccard similarity coefficient of the detected ground-truth ground plane regions and the ground plane regions was 0.94. Although the method was developed for use in an electronic travel aid for the blind, it could also find applications in the automatic navigation of autonomous vehicles and unmanned aerial vehicles (UAVs).