search

ankitdhall / lidar_camera_calibration

ROS package to find a rigid-body transformation between a LiDAR and a camera for "LiDAR-Camera Calibration using 3D-3D Point correspondences"
http://arxiv.org/abs/1705.09785
GNU General Public License v3.0
1.49k stars 460 forks source link

met poor performace #13

Closed leica8244 closed 7 years ago

leica8244 commented 7 years ago

the data in poins.txt is as following: 8 -0.56811 -0.473328 1.79659 -0.213204 -0.134026 1.7816 -0.543555 0.218787 1.78348 -0.915342 -0.132713 1.8172 0.145415 -0.450564 1.79215 0.56126 -0.191263 1.78309 0.289001 0.229009 1.78225 -0.113636 -0.0443698 1.79478 -0.485469 -0.327932 1.47909 -0.144692 0.0051091 1.53701 -0.471128 0.350685 1.47055 -0.811904 0.0176443 1.41263 0.124684 -0.314695 1.50046 0.524897 -0.0503131 1.51868 0.260503 0.350273 1.51351 -0.139709 0.0858913 1.49529 it seems that the points' coordinates of the camera are much different from those of the lidar. at the same time, the result of my calibratioin seems not so correct. because my camera is just next to the lidar. ( Average transformation is: 0.996523 0.00304858 -0.0832566 0.176845 -0.00350322 0.99998 -0.0053152 0.124334 0.0832387 0.00558839 0.996514 -0.28119 0 0 0 1 ) can you help my figure out what may be wrong ?and how to make it more acurate?

ankitdhall commented 7 years ago

There are a few things you can check. [1] Check the values you have entered in the config_file.txt [2] marker_coordinates.txt has the correct measurements of the board on which the ArUco marker is stuck. [3] marker size in find_transform.launch is correctly entered You could probably re-read the README to make sure that you haven't missed anything.

leica8244 commented 7 years ago

thank you for that, it is the launch file which i forget to modify. it's now correctly calibrated.

leica8244 commented 7 years ago

Now we have gotten a good R&t matrix, and we try to fuse two point cloud. Even when we don't move the camera at all ,the final-fused point cloud seems not coincide so good. We put the lidar and camera together,and the t is 0.03,0.05,0.07, this seems good enough. And the final fused point cloud look like this . How can i get as a good result as yous? Thank you in advance! screenshot from 2017-08-17 13-26-56

ankitdhall commented 7 years ago

One of the things you could look at is ensuring that the intrinsic calibration of the cameras is good. Better the intrinsic calibration will lead to more precise fusion of the point clouds. Also, make sure that the cameras remain in the same place, even the slightest movements can cause discrepancy in the fused point cloud. A good idea is to fix them on a bar or some rigid object, so they do not change their relative positions.

wenjunj commented 5 years ago 
 @leica8244 你好，很高兴看到你是上海交通大学的，我也是上海交大的，想咨询你一点问题，我标定出来的结果RMSE很小，但是我实际配准激光和图片时效果不理想，请问大概是什么原因的，是否方便添加微信18818207380，非常感谢，希望有机会向你请教！

Num points is:8 Number of points: 8 0.559911 -0.074347 -0.00786292 -0.0554011 0.200673 -0.0119373 -0.0318813 0.00789918 0.000815446 Rotation matrix: 0.999208 -0.021123 0.0337249 0.0239192 0.996112 -0.0847847 -0.0318029 0.0855243 0.995828 Rotation in Euler angles: 1.37139 1.82261 4.90902 Translation: -0.185703 0.0968357 -0.0996245 RMSE: 0.0198393 Rigid-body transformation: 0.999208 -0.021123 0.0337249 -0.185703 0.0239192 0.996112 -0.0847847 0.0968357 -0.0318029 0.0855243 0.995828 -0.0996245 0 0 0 1

After 100 iterations

Average translation is: -0.176358 0.0460298 -0.0925272 Average rotation is: 0.999232 -0.0183505 0.0346194 0.0205445 0.997731 -0.064123 -0.0333642 0.064785 0.997341 Average transformation is: 0.999232 -0.0183505 0.0346194 -0.176358 0.0205445 0.997731 -0.064123 0.0460298 -0.0333642 0.064785 0.997341 -0.0925272 0 0 0 1 Final rotation is: 0.0354151 -0.999218 0.0175548 -0.0641067 -0.0198011 -0.997747 0.997314 0.0342099 -0.0647578 Final ypr is: 2.07551 -1.6441 -0.48601 Average RMSE is: 0.0216582 RMSE on average transformation is: 0.0366648