Juxtaposition of Inertial Navigation Sensor and Camera Egomotion Estimates of Ground Vehicle Trajectory: Results and Implementation Details
Abstract
Accurate knowledge of the pose (position and orientation) is essential to the unmanned ground vehicle's (UGV) ability to follow a prescribed path and to reach a specified location, which is the essence of autonomous mobility. An essential element of an autonomous UGV's sensor suite is the navigation sensor. Given appropriate processing, other UGV terrain sensors can also provide independent estimates of vehicle pose or derivatives thereof that can be fused with the navigation data to provide a more robust knowledge of pose and implicitly, of registration between the two sensors. Improved accuracy of UGV pose in turn enables improved accuracy of terrain geometry estimates. In this report, we describe a precursor to fusion, which is juxtaposition of consistent entities ( apples to apples ). Trajectories of pose estimates are recorded as a surrogate UGV traverses rough terrain, one trajectory from an inertial UGV navigation sensor, and its counterpart from egomotion extracted in post-processing from a forward looking video camera. The sensors and the relevant aspects of the data collection methodology are described, as are the streams of pose estimates. We explain the procedure adopted for temporally aligning the data streams and the assumptions concerning registration under which the streams are juxtaposed. A notation is developed to relate the myriad coordinate frames implicit in the application. The juxtaposition is presented graphically, with recommendations for subsequent studies with the intent of achieving fusion. A small related study evaluating the possibility of calculating the registration from pose trajectories is reported.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2003
- Accession Number
- ADA485430
Entities
People
- Gary A. Haas
- William F. Oberle
Organizations
- United States Army Research Laboratory