search

drwnz / camera_lidar_pandey

A modified version of the camera/LiDAR calibration method (originally by by G. Pandey).
GNU Lesser General Public License v3.0
4 stars 2 forks source link

Issue for KITTTI Odometry Seq-00 Calibration Result #1

Open zhanghao-py opened 4 years ago

zhanghao-py commented 4 years ago

Hello drwnz,

First of all, thanks a lot for your contribution to modify this program to support the KITTI datasets. I have already run this executable program with sequence-00 (image_0) KITTI Odometry datasets. I found the experimental results as follows and can't obtain the correct calibration result as ground truth described.

  1. Ground Truth as KITTI calib file provided.

kitti_seq00_gt

  1. Optimization with Gradient Descent (calib.gradient_descent_search)

kitti_seq00_err_1

kitti_seq00_err_2

Initial guess: X0 = [0.0, 0.0, 0.0, -180, -90, -90];

After many experiments, I still didn't calculate the correct calibration result through Gradient Descent optimizer.

By the way, I unusually get the core-dump issue if the initial guess was selected poorly well because of bad histogram in mi_cost function.

  1. Optimization with GSL.

No matter gsl_minimizer (calib.gsl_minimizer in Calibration_deprecated.cpp) or gsl_minimizer_nmsimplex(calib.gsl_minimizer_nmsimplex in Calibration_deprecated.cpp), these optimizes' procedure are ended in failed status and can't also calculate the correct calibration result.

Could you please provide me for your experimental process and results? What's more, would you get the precise (LiDAR to Camera) calibration results? Thanks for your help.

drwnz commented 4 years ago

Hi, I was able to get calibration results with Gradient Descent. However, as you can see, on KITTI the optimization is badly affected by local minima, so initialization is very important. I'm currently working on other methods to overcome this issue.