Elsevier

Image and Vision Computing

Volume 30, Issue 2, February 2012, Pages 109-121
Image and Vision Computing

A fast robot homing approach using sparse image waypoints

https://doi.org/10.1016/j.imavis.2012.02.006Get rights and content

Abstract

This paper proposes a fast image sequence-based navigation approach for a flat route represented in sparse waypoints. Instead of purely optimizing the length of the path, this paper aims to speed up the navigation by lengthening the distance between consecutive waypoints. When local visual homing in a variable velocity is applied for robot navigation between two waypoints, the robot's speed changes according to the distance between waypoints. Because long distance implies large scale difference between the robot's view and the waypoint image, log-polar transform is introduced to find a correspondence between images and infer a less accurate motion vector. In order to maintain the navigation accuracy, our prior work on local visual homing with SIFT feature matching is adopted when the robot is relatively close to the waypoint. Experiments support the proposed navigation approach in a multiple-waypoint route. Compared to other prior work on visual homing with SIFT feature matching, the proposed navigation approach requires fewer waypoints and the navigation speed is improved without compromising the accuracy in navigation.

Highlights

► We propose a fast image sequence-based navigation which represents a route in sparse waypoints. ► Images are registered by using log-polar transform in order to lengthen the distance between consecutive waypoints. ► The robot can depart from previous waypoint at a higher speed due to the lengthened distance. ► The navigation accuracy is maintained by adopting prior work on local visual homing based on SIFT and epipolar geometry.

Introduction

Autonomous robot navigation has been extensively studied over the years. Among various robot navigation approaches, navigating by a series of waypoint images is of particular interest. Image sequence-based navigation often adopts the two-stage t teaching–replay strategy, where the robot studies the environment by taking snapshots of the paths in the teaching stage and travels in the environment by tracing these paths by comparing its current view with the waypoint images in the replay stage.

Homing is a special type of navigation, where the focus is on guiding the robot back to its base station from arbitrary position. Robust homing is important to an autonomous robot, especially when it requires routine maintenance such as recharging and function calibration. In the case of local visual homing, numerous studies have emphasized on the accuracy of guiding a robot between consecutive image clues. In general, two classes of approaches are developed: holistic methods [1], [2] and correspondence-based methods [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22].

In a holistic homing method, robot's view is warped in order to generate the most similar image to the waypoint image [1], [2] because no correspondence is detected between two images. In [1], the robot's view is warped with different hypothesis of relative pose between the robot and the waypoint. A motion vector inferred from the best hypothesis minimizes image distance between the robot's view and the waypoint image. In [2], a correlation value for image similarity is maximized by rotating the robot's omnidirectional view. The direction of the motion vector is proportional to the angel of rotation.

A correspondence-based homing method detects correspondences of image features and computes a motion vector to minimize the displacement of correspondences. More discussions on the correspondence-based approaches will be provided in Section 2.1.

We propose an effective approach to reduce the runtime of the image-based navigation along a specified path in an indoor flat environment. Compared to other SIFT-based approaches, our method requires less intermediate waypoints. In order to initiate visual homing from a farther pose, log-polar transform is used to determine the correspondence of a circular template for its better tolerance of scale differences than the SIFT matching. To overcome the insufficient accuracy of the inferred motion vector, our prior work on local visual homing is adopted when the robot is relatively close to the waypoint.

We provide numerical results demonstrating that employing SIFT in the early stage of navigation does not contribute much to the final homing accuracy. Experiments show that our approach requires much less waypoints and effectively reduces the homing time, while not posing negative impacts on the final homing accuracy. In general, a robot is able to depart at a higher speed from a waypoint due to the lengthened distance between waypoints and decelerate less frequently because of the reduced number of waypoints.

Section 2 introduces related work on correspondence-based local visual homing and some techniques on reducing the runtime of navigation; Section 3 describes the procedure for deciding the waypoints and capturing the waypoint images; Section 4 details the proposed navigation approach; Section 5 provides experiments to validate the proposed approach and the performance; Section 6 concludes this paper.

Section snippets

Related work on local visual homing

This section reviews some related work on correspondence-based local visual homing.

Waypoint selection

This section explains how a route is represented by sparse waypoints in the teaching stage. In an environment, three types of waypoints are defined according to their functions: service waypoints, junction waypoints, and intermediate waypoints. Service waypoints stand for poses where the robot provides services. Junction waypoints represent road intersections where a robot may change its orientation largely. Intermediate waypoints are required when the distance between two waypoints is beyond

Local visual homing

This section introduces image registration by log-polar transform and the proposed local visual homing.

Robot platform and parameter settings for experiments

Our experiments are performed with U-BOT [29], [30]. U-BOT is a four-wheeled robot with two-wheeled drive in the front. The maximal velocity and acceleration of U-BOT are 1200 mm/s and 1000 mm/s2, respectively. Actuator commands are transmitted through serial communication to U-BOT. A USB webcam, QuickCam Ultra Vision [31], is mounted on the top of U-BOT. The navigation approach is performed by a laptop with Intel Core 2 Duo P7450 (2.13 GHz). The graphics card is NVIDIA GeForce G 105M so that SIFT

Conclusion and future work

This paper proposes a fast image sequence-based homing approach along a flat route represented as waypoints, where images are captured in advance. Given a robot with fixed specification, the proposed approach aims to speed up navigation without compromising navigation accuracy. Compared to prior work on local visual homing with SIFT feature matching, the average distance between consecutive waypoints can be lengthened and the robot is allowed to depart at a higher speed from each waypoint. To

References (31)

  • G.L. Mariottini et al.

    Image-based visual servoing for nonholonomic mobile robots using epipolar geometry

    IEEE Trans. Robot.

    (2007)
  • O. Booji et al.

    Navigation using an appearance-based topological map

  • S. Segvic et al.

    Large scale vision-based navigation without an accurate global reconstruction

  • F. Fraundorfer et al.

    Topological mapping, localization and navigation using image collections

  • T. Goedeme et al.

    Omnidirectional vision based topological navigation

    Int. J. Comput. Vis.

    (2007)
  • Cited by (0)

    This paper has been recommended for acceptance by Henrik Iskov Christensen.

    View full text