Navigation System for Mobile Robots Using PCA-Based Localization from Ceiling Depth Images: Experimental Validation

Abstract

This paper aims the experimental validation of a mobile robot navigation system, using self-localization based on principal component analysis (PCA) of ceiling depth images. In this approach, a roadmap based on generalized Voronoi diagram (GVD) is built from an occupancy grid, that is defined in the ceiling mapping to the PCA database. The system resorts to the Dijkstra algorithm to planning paths, using the GVD-based roadmap, from which a set of waypoints are extracted. During the mission, the robot is commanded by a controller based on dipolar navigation functions (DNF) along the waypoints, being self-located using only the information provided from ceiling depth images and other on-board sensors. The navigation system ensures that the robot reaches its destination, travelling along safety trajectories, while computing its pose with global stable estimates, from Kalman filters (KF). The navigation is achieved without the need to structure the environment, searching by specific features, and to linearize the model. The results are experimentally validated in an indoor environment, using a differential-drive mobile robot.

Ceiling-Vision-Based Mobile Object Self-Localization: A Composite Framework

“…For example, A. Ribacki et al use an upward facing camera to detect the ceiling boundaries and to estimate the ceiling space density from the current image [26]. Other authors, for example [5,6] use the ceiling depth images for robot localization. In these approaches self-localization is obtained from Principal Component Analysis of ceiling depth images.…”

An Innovative Monocular Mobile Object Self-localization Approach Based on Ceiling Vision

“…For example A. Ribacki et al use an upward facing camera to extract the ceiling boundaries for estimating the ceiling space density from the current image [11]. Other authors, for example [12,13] use the ceiling depth images for robot localization. In these approaches self-localization is obtained from Principal Component Analysis of ceiling depth images.…”