Terrain Referenced Navigation Using SIFT Features in LiDAR Range-Based Data
Abstract
The use of GNSS in aiding navigation has become widespread in aircraft. The long term accuracy of INS are enhanced by frequent updates of the highly precise position estimations GNSS provide. Unfortunately, operational environments exist where constant signal or the requisite number of satellites are unavailable, significantly degraded, or intentionally denied. This thesis describes a novel algorithm that uses scanning LiDAR range data, computer vision features, and a reference database to generate aircraft position estimations to update drifting INS estimates. The algorithm uses a single calibrated scanning LiDAR to sample the range and angle to the ground as an aircraft flies, forming a point cloud. The point cloud is orthorectified into a coordinate system common to a previously recorded reference of the flyover region. The point cloud is then interpolated into a Digital Elevation Model (DEM) of the ground. Range-based SIFT features are then extracted from both the airborne and reference DEMs. Features common to both the collected and reference range images are selected using a SIFT descriptor search. Geometrically inconsistent features are filtered out using RANSAC outlier removal, and surviving features are projected back to their source coordinates in the original point cloud. The point cloud features are used to calculate a least squares correspondence transform that aligns the collected features to the reference features. Applying the correspondence that best aligns the ground features is then applied to the nominal aircraft position, creating a new position estimate. The algorithm was tested on legacy flight data and typically produces position estimates within 10 meters of truth using threshold conditions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 26, 2014
- Accession Number
- ADA615027
Entities
People
- Matthew T. Leines
Organizations
- Air Force Institute of Technology