Self-Calibration and Visual SLAM with a Multi-Camera System on a Micro Aerial Vehicle

Author: Heng

Year: 2014

Notes:

config: two stereo pairs facing opposite directions with IMU
Self calib based on the generalized camera model (using many cameras as one)
Motion Estimation using Multiple Non-Overlapping Cameras for Small Unmanned Aerial Vehicles Kanade paper about non overlapping cams (without scale estimation)
M. Tribou: thesis about PTAM version with non overlapping cam, doesn't scale to large environments
Self Calib step:
- Estimate a consistent map for each stereo system with VSLAM
- Hand eye alignement to get a raw estimate of the inter stereo calib
- Merge of the two maps
- Perform full BA (intrinsics, extrinsics, poses and 3D points)
Robust pose graph optimization for loop closure (with relative pose given by stereo VO and loop closures)
Propose a pose estimation based on bearing vector and with gyro integration for rotation estimation