Developing monocular visual pose estimation for arctic environments

Abstract

Arctic regions present one of the harshest environments on Earth for people or mobile robots, yet many important scientific studies, particularly those involving climate change, require measurements from these areas. For the successful deployment of mobile sensors in the Arctic, a high-quality localization system is required. Although a global positioning system can provide coarse positioning (within several meters), it cannot provide any orientation information. A single-camera-pose-estimation system is presented, based on visual odometry techniques, which is capable of operating in the feature-poor environments of the Arctic. To validate the system, a prototype rover was developed and fielded on a glacier in Alaska. The resulting pose estimates compare favorably to values obtained by hand registering the same video sequence. Although pose errors do accumulate over time, these errors are typical of a standard odometry system but obtained in an environment where standard odometry is not practical. C 2009 Wiley Periodicals, Inc.

Expanding the Limits of Vision‐based Localization for Long‐term Route‐following Autonomy

“…This makes vision‐based navigation in winter environments especially difficult as the elevation of the sun is perpetually low on the horizon, and snow rapidly accumulates, melts, and provides little contrast to the scene. Williams and Howard () improve VO in snowy environments by applying contrast‐limited adaptive histogram equalization (CLAHE) to increase keypoint matches in images with snowy foregrounds. They show an increase in keypoint match count by an order of magnitude.…”

Improving Visual Feature Extraction in Glacial Environments

“…Similar to Paton et al, Williams and Howard developed and tested a 3D orientation (pose) estimation algorithm on the Juneau Ice Field in Alaska. Williams and Howard wrote, "When dealing with arctic images, feature extraction is possibly the biggest challenge" [3]. They used contrast limited adaptive histogram equalization (CLAHE) post-processing to enhance contrast and make features stand out better.…”

Autonomous GPR Surveys using the Polar Rover Yeti

“…Navigation combines an inexpensive GPS receiver with single‐camera simultaneous localization and mapping (SLAM) to achieve ∼1 m localization accuracy (Williams, Parker, & Howard, ). The team has made impressive progress toward feature extraction on low‐contrast, snow‐covered terrain (Williams & Howard, ), but the effort reveals the difficulty of extracting identifiable features even under generally good visibility conditions. Nevertheless, plans to reconstruct glacial surfaces using tracked features provide an added, scientific impetus to develop SLAM for snow‐covered terrain (Williams, Parker, & Howard, ).…”

“…The difficulties encountered with Nomad stereo cameras and laser scanners (Moorehead et al., ) highlight the difficulties of obstacle detection in polar terrain. During cloudy conditions, a dedicated effort of the sort initiated for SnoMote (Williams & Howard, ) would be needed to achieve reliable vision‐based obstacle detection on low‐contrast snowfields. We felt that we could meet our primary objective, namely to demonstrate the operational value of autonomous GPR surveys, without dedicating the significant resources needed to establish reliable obstacle detection over polar terrain.…”