Awesome
VI-Stereo-DSO
Related Papers:
-
Direct Sparse Odometry, J. Engel, V. Koltun, D. Cremers, In arXiv:1607.02565, 2016
-
Stereo DSO:Large-Scale Direct Sparse Visual Odometry with Stereo Cameras, Rui Wang, Martin Schwörer, Daniel Cremers, 2017 IEEE International Conference on Computer Vision
-
Direct Sparse Visual-Inertial Odometry using Dynamic Marginalization, Lukas von Stumberg, Vladyslav Usenko, Daniel Cremers, 2018 IEEE International Conference on Robotics and Automation (ICRA)
1. Installation
Please follow https://github.com/JakobEngel/dso.
2. Usage
-
Modify the bash file
-
For stereo:
bash ./run.bash -
For mono:
bash ./run_mono.bash
- T_stereo file:
transform matrix from right camera to left camera (4x4).
- imu_info file:
transform matrix from left camera to imu (4x4).
gyroscope noise density [ rad / s / sqrt(Hz) ] ( gyro "white noise" )
accelerometer noise density [ m / s^2 / sqrt(Hz) ] ( accel "white noise" )
gyroscope random walk [ rad / s^2 / sqrt(Hz) ] ( gyro bias diffusion )
accelerometer random walk [ m / s^3 / sqrt(Hz) ]. ( accel bias diffusion )
Experiments
- 20190424
SE(3) Umeyama alignment：
| weight 6,0.6,0.5 | MH01 | MH02 | MH03 | MH04 | MH05 | V101 | V102 | V103 | V201 | V202 | V203 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ape rmse(m) | 0.0321 | 0.0288 | 0.0743 | 0.119 | 0.072 | 0.0599 | 0.105 | 0.168 | 0.0852 | 0.0667 | 0.211 |
No alignment（Initialization may have an impact on the results）:
| weight 6,0.6,0.5 | MH01 | MH02 | MH03 | MH04 | MH05 | V101 | V102 | V103 | V201 | V202 | V203 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ape rmse(m) | 0.0993 | 0.0557 | 0.234 | 0.194 | 0.111 | 0.232 | 0.202 | 0.2777 | 0.102 | 0.114 | 0.263 |
- 20190416
Apply First Estimates Jacobians to scale to maintain consistency.
EuRoC MH01:
- EuRoC V1_03:
- EuRoC V2_03:
- 20190409:
green line: groundtruth, redline: VI-Stereo-DSO
P.S.
This project aims at verifying the location performance of VI-Stero-DSO. The computational efficiency is to be improved. Actually in current code, many preintergration operations are redundant and can be simplified.