Vision-Aided Inertial Navigation for Spacecraft Entry, Descent, and Landing

Author: Mourikis

Year: 2009

Notes:

EKF that fuse visual features to IMU measurements for attitude, velocity and position in a rt fashion
0.16 m/s velocity error and 6.4m error for landing position
3D information of landmarks on the planet known from satellite images mapped landmarks (ML), rotationnal and rotationnal velocity with 2D-2D tracking opportunistic features (OF)
experimentation during a sounding rocker entry test, system suited for FPGA hardware
Classic Harris KP and crosscorellation for feature matching (no need for scale and rotation invariance for this scenario)
OF only used for pose constraint between camera poses
DIMES: descent and landing system of MER mission: feature tracking without IMU fusion

ML 3D-2D:

ML enables global positionning, perform by matching templates between descent images and satellite images, but at a moment, these become hard to match and the system switches to OF tracking only
For ML, map matching if performed: a harris pt is selected, with current pose a template of the point a Homoegraphy between the template and the map is computed. Then the convolution of the template is performed in the frequency domain (after FFT) to reduce comutationnal cost (convolution turns into a product)
Match is marked if the correlation peak heigh and width passe a thresh

OF 2D-2D:

high scale change in the end of the landing => LKT fails
flat assumption of the scene => warping the initial image to successive images
template warping given the homography, then 2D spatial coorelation based matching
matching very stable, no RANSAC needed

Estimator:

state vector: $x_e = [q, p, v, b_a, b_g]$, N past poses of the camera (sliding window, N chosen as the maximum number of frames an OF can be tracked)
Measurement model = reprojection errors based on absolute position (ML) or triangulated position (OF)
petite subtilité pour les OF où il faut manipuler le résidu pour qu'il ne dépende pas de la détéction courante

Experiment: