Vision-based Terrain Relative Navigation for Planetary Landing
Open Access
- Author:
- Sutterlin, Graeme
- Area of Honors:
- Aerospace Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Roshan Thomas Eapen, Thesis Supervisor
Sven Schmitz, Thesis Honors Advisor - Keywords:
- Computer Vision
TRN
Terrain Relative Navigation
Point Clouds
DEMs
Pose Estimation
Feature Matching - Abstract:
- ose estimation is an integral part of any navigation pipeline. Its goal is to estimate the location and orientation of a 6 degree-of-freedom sensor in a 3D scene. A crucial stage in this process is identifying the correspondence between the data obtained from a sensor and the 3D world model, after which algorithms like PnP are used to construct the camera posture depending on the correspondence. This research work aims to determine pose (orientation and translation) of a camera sensor based on the finding the correspondence between features in a image and a 3D points in the terrain frame using nonlinear optimization techniques. The 2D to 3D correspondence problem is reduced to a 2D to 2D correspondence problem (feature matching and tracking) by rapidly rendering a perspective projection of the available 3D point cloud. The use of Gazebo as a physics simulator to produce an exact lighting environment and simulate the descent of a camera-equipped spacecraft on a parabolic approach to the Rheasilvia crater is discussed in detail. The challenges posed by sensor limitations and camera noise are also evaluated in this work. It is shown that the nonlinear least-squares is able to accurately estimate the pose up to machine precision in the case without any noise, and up to a reasonable tolerance in the presence of noise. Monte-Carlo simulations are performed to validate these results.