Sensor Auto-Calibration on Dynamic Platforms in 3D

J. Britt, D. Bevly

Abstract:	Sensor misalignment can be simply annoying causing fuzzy point clouds or catastrophic when obstacle avoidance sensors are mal-aligned. However, despite the necessity of calibration, the process can often be tedious and time consuming, often requiring external targets or apriori knowledge of the environment. I will demonstrate an auto-calibration scheme that will find the translation and rotation in thee dimensions between any two sensors capable of determining their respective change in position and attitude. This work is an extension of a similar method that has since been confined to only two dimensions. The major benefit to this method over other auto-calibration schemes, is that this method make no inherent assumptions about the environment, needs no calibration target, and is designed to be performed on a vehicle while it is undergoing dynamics. This is especially beneficial to applications where a sensor might be jarred during transit due to the environment, vehicle damage, or when sensors are often repositioned on vehicles. The algorithm in question operates by assuming that there are two sensors where each is capable of determining a local change in pose (position and attitude). The principle being that if the two sensors are perfectly calibrated, they should both experience the same change in pose. A Kalman filter is used to obtain the true translation and rotation between the sensors by attempting to align these changes in pose measurements so that both sensors are sensing the same change in pose. This also implies that should the sensors become misaligned during a mission, they could be re-aligned on the fly. In addition to the calibration method itself, I will analyze the observability of the method. In general the circumstances where the filter is unobservable is when the vehicle can undergo holonomic motion, only makes a single perfect circle, or one of the measurement axes goes unexcited. Typically these are not seen in practice. The algorithm has been validated in simulation and various levels of sensor error have been introduced to exam the robustness of the algorithm. The algorithm will also be tested using real-world data supplied by MIT’s grand challenge team, where the algorithm will attempt to align the Velodyne lidar with the onboard IMU. The change in pose of the lidar will be obtained through the ICP (iterative closest point) algorithm. The results of this algorithm dynamic calibration will be compared to the more controlled environment used in the supplied dataset.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	2195 - 2203
Cite this article:	Britt, J., Bevly, D., "Sensor Auto-Calibration on Dynamic Platforms in 3D," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 2195-2203.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In