Student's T Robust Bundle Adjustment Algorithm
Authors:
Aleksandr Y. Aravkin,
Michael Styer,
Zachary Moratto,
Ara Nefian,
Michael Broxton
Abstract:
Bundle adjustment (BA) is the problem of refining a visual reconstruction to produce better structure and viewing parameter estimates. This problem is often formulated as a nonlinear least squares problem, where data arises from interest point matching. Mismatched interest points cause serious problems in this approach, as a single mismatch will affect the entire reconstruction. In this paper, we…
▽ More
Bundle adjustment (BA) is the problem of refining a visual reconstruction to produce better structure and viewing parameter estimates. This problem is often formulated as a nonlinear least squares problem, where data arises from interest point matching. Mismatched interest points cause serious problems in this approach, as a single mismatch will affect the entire reconstruction. In this paper, we propose a novel robust Student's t BA algorithm (RST-BA). We model reprojection errors using the heavy tailed Student's t-distribution, and use an implicit trust region method to compute the maximum a posteriori (MAP) estimate of the camera and viewing parameters in this model. The resulting algorithm exploits the sparse structure essential for reconstructing multi-image scenarios, has the same time complexity as standard L2 bundle adjustment (L2-BA), and can be implemented with minimal changes to the standard least squares framework. We show that the RST-BA is more accurate than either L2-BA or L2-BA with a sigma-edit rule for outlier removal for a range of simulated error generation scenarios. The new method has also been used to reconstruct lunar topography using data from the NASA Apollo 15 orbiter, and we present visual and quantitative comparisons of RST-BA and L2-BA methods for this application. In particular, using the RST-BA algorithm we were able to reconstruct a DEM from unprocessed data with many outliers and no ground control points, which was not possible with the L2-BA method.
△ Less
Submitted 6 November, 2011;
originally announced November 2011.
Locating the LCROSS Impact Craters
Authors:
William Marshall,
Mark Shirley,
Zachary Moratto,
Anthony Colaprete,
Gregory Neumann,
David Smith,
Scott Hensley,
Barbara Wilson,
Martin Slade,
Brian Kennedy,
Eric Gurrola,
Leif Harcke
Abstract:
The Lunar CRater Observations and Sensing Satellite (LCROSS) mission impacted a spent Centaur rocket stage into a permanently shadowed region near the lunar south pole. The Sheperding Spacecraft (SSC) separated \sim9 hours before impact and performed a small braking maneuver in order to observe the Centaur impact plume, looking for evidence of water and other volatiles, before impacting itself. Th…
▽ More
The Lunar CRater Observations and Sensing Satellite (LCROSS) mission impacted a spent Centaur rocket stage into a permanently shadowed region near the lunar south pole. The Sheperding Spacecraft (SSC) separated \sim9 hours before impact and performed a small braking maneuver in order to observe the Centaur impact plume, looking for evidence of water and other volatiles, before impacting itself. This paper describes the registration of imagery of the LCROSS impact region from the mid- and near-infrared cameras onboard the SSC, as well as from the Goldstone radar. We compare the Centaur impact features, positively identified in the first two, and with a consistent feature in the third, which are interpreted as a 20 m diameter crater surrounded by a 160 m diameter ejecta region. The images are registered to Lunar Reconnaisance Orbiter (LRO) topographical data which allows determination of the impact location. This location is compared with the impact location derived from ground-based tracking and propagation of the spacecraft's trajectory and with locations derived from two hybrid imagery/trajectory methods. The four methods give a weighted average Centaur impact location of -84.6796\circ, -48.7093\circ, with a 1σ un- certainty of 115 m along latitude, and 44 m along longitude, just 146 m from the target impact site. Meanwhile, the trajectory-derived SSC impact location is -84.719\circ, -49.61\circ, with a 1σ uncertainty of 3 m along the Earth vector and 75 m orthogonal to that, 766 m from the target location and 2.803 km south-west of the Centaur impact. We also detail the Centaur impact angle and SSC instrument pointing errors. Six high-level LCROSS mission requirements are shown to be met by wide margins. We hope that these results facilitate further analyses of the LCROSS experiment data and follow-up observations of the impact region.
△ Less
Submitted 8 March, 2011;
originally announced March 2011.
