Implementation of an automatic method to estimate the extrinsic calibration parameters of multiple sensors, explicitly handling the scale ambiguity arising from monocular cameras.
Authors: David Zuñiga-Noël, Jose-Raul Ruiz-Sarmiento, Ruben Gomez-Ojeda, Javier Gonzalez-Jimenez
Related Publications: Automatic Multi-Sensor Extrinsic Calibration for Mobile Robots
License: GPLv3
-
Boost (1.58.0.1ubuntu1)
sudo apt install libboost-all-dev -
Ceres (1.14.0-facb199)
See ceres documentation.
-
CMake (3.5.1-1ubuntu1)
sudo apt install cmake -
Eigen (3.3~beta1-2)
sudo apt install libeigen3-dev -
Gflags (2.1.2-3)
sudo apt install libgflags-dev -
Glog (0.3.4-0.1)
sudo apt install libgoogle-glog-dev
The input for the ground plane calibration are 3D point observations from the ground, in local sensor coordinates. The input is a csv text file as:
x,y,z
The input for the are synchronized per-sensor incremental 2D (coplanar) motions. For each sensor, incremental motions are grouped into a single text file, where each line contains a synchronized 2D incremental motion (angles in radians) in the following format:
x y yaw
Note: all sensors must have the same number of observations.
The multisensor calibration method can be invoked as:
calibrate [options] <motions1> <motions2> ...
Notes:
-
The method calibrates all sensors with respect to a reference one. The reference sensor is assumed to be the one observing the first incremetal motions.
-
Use the
--scale_ambiguous=n,m,...option to indicate which motions have scale ambiguity.