lidar_IMU_calib icon indicating copy to clipboard operation
lidar_IMU_calib copied to clipboard

Published 20 hours ago • verified publisher icon APRIL-ZJU

How to use my own data？

Open qian5683 opened this issue 4 years ago • 14 comments

Thanks for your good job. But I have some questions about how to use my own data. My lidar device is RoboSense Lidar-16. and there is a error：

Load dataset from /home/qian/rosbag/lidarimu0903.bag /rslidar_packets: 300 /imu/data: 5999 [li_calib_gui-2] process has died [pid 2424, exit code -11, cmd /home/qian/catkin_li_calib/devel/lib/li_calib/li_calib_gui __name:=li_calib_gui __log:=/home/qian/.ros/log/1caebfe4-ecf3-11ea-b9b2-244bfe93b8c4/li_calib_gui-2.log]. log file: /home/qian/.ros/log/1caebfe4-ecf3-11ea-b9b2-244bfe93b8c4/li_calib_gui-2*.log

qian5683 avatar Sep 01 '20 10:09 qian5683

@qian5683

  1. Currently, the LI-Calib toolkit only supports VLP-16.
  2. If you want to use the other LiDARs, there are two main functions that need to be modified :

icameling avatar Sep 03 '20 07:09 icameling

hi~, I've modified the code and used it for my own data. However, there is a problem. The translation I calibrated is very large and does not meet the expectation. I want to know:

  1. What maybe causes this error.
  2. Are there any requirements for recording rosbag. For example, what are the requirements for motion trajectory

qian5683 avatar Sep 09 '20 03:09 qian5683

Hi,

I think I've modified the two functions you mentioned above properly and the initialization step has been done without errors, But it looks like the toolkit only supports organized pointcloud (currently with a height of 16 using VLP-16), and the lidar I'm using only generates pointcloud data unorganized with a height of 1. So I was wondering if there are any functions I need to modify before I can use this toolkit? I've tried to modify those functions requiring two parameters (width and height) such that they take only one (width), but it doesn't seem to work very well. The data association function returns:

Plane type :0 0 0; Plane number: 0 [ WARN] [1599639840.190547592]: [DataAssociation] fails. Please let me know if there's anything I can do to make it work. Thanks!! XXD

Best regards Jay

jayjayhee avatar Sep 09 '20 09:09 jayjayhee

hi~, I've modified the code and used it for my own data. However, there is a problem. The translation I calibrated is very large and does not meet the expectation. I want to know:

  1. What maybe causes this error.
  2. Are there any requirements for recording rosbag. For example, what are the requirements for motion trajectory

hi,you use the toolkit on your own car(much bigger than the paper)?It can almost only perform plane motion,which cannot ensure imu have sufficient linear acceleration and rotational velocity in x,y,z axis?

noodlles avatar Sep 09 '20 09:09 noodlles

hi~, I've modified the code and used it for my own data. However, there is a problem. The translation I calibrated is very large and does not meet the expectation. I want to know:

  1. What maybe causes this error.
  2. Are there any requirements for recording rosbag. For example, what are the requirements for motion trajectory

@qian5683

Data requirements

  • First, it is important to have the measurement data synchronized between the systems, and we strongly recommend hardware time synchronization. Otherwise, some intrinsic properties of sensor networks such as limited resources of storage, computation, and bandwidth, combined with a potentially high density of nodes make the timestamps of measurements inaccurate.

    Form the provided datasets, we can see the timestamps of raw measurements distribute unevenly. 3

    And after synchronization and preprocessing, the timestamps of synchronized measurements arranged evenly. 2 （you can use rqt_bag to plot time-series of message）

  • Second, based on the observability of IMU-LiDAR extrinsic calibration[1,2], at least two rotations about different axes are necessary. Meanwhile, large accelerations and angular velocities make the estimation more accurate, but you should make sure the error of the LiDAR odometry is within a certain range under that severe motion.

[1] LIC-Fusion 2.0 https://arxiv.org/pdf/2008.07196.pdf (Section IV. OBSERVABILITY ANALYSIS) [2] Hesch, Joel. (2016). Consistency Analysis and Improvement for Vision-aided Inertial Navigation. Retrieved from the University of Minnesota Digital Conservancy, http://hdl.handle.net/11299/180208. (Section 2.6 IMU-Laser Scanner Extrinsic Calibration)

icameling avatar Sep 10 '20 13:09 icameling

Hi author, I don't have a hardware to do time synchronization, use rqt_bag it seems msgs distribute evenly. image

  • When Initialization finished, the initial value of rotation seems correct. My imu basically aligins with laser(rotation approximate identity)

[Initialization] Done. Euler_ItoL initial degree: 3.78477 0.405341 -1.24306

  • Association shows that we have got Plane type :65 1 55; Plane number: 121.
  • When Batch finished, the result is different from the initial value. It seems thant it converges to the wrong solution.

euler_ItoL : 57.4901 5.51252 -8.14225

  • When Refine finished, plane number decrease, not increase.

Plane type :16 8 11; Plane number: 25 ... refine refine ...

The NDT_target_map.pcd basically meets the environmental scene, but the refined_map.pcd is mess. image image

narutojxl avatar Sep 11 '20 11:09 narutojxl

@narutojxl

Please check out the surfels map after the first Association , and make sure there are enough surfels in the direction of x-axis, y-axis and z-axis. It's better to collect calibration sequence in a structured environment with big planes.

icameling avatar Sep 15 '20 02:09 icameling

@icameling, thanks for your advise, I will find a place meets this requirements to do experiment :)

narutojxl avatar Sep 15 '20 06:09 narutojxl

@icameling, hi author, Following your advice, I do the experiment in a small room surrounded by walls. I move imu and laser around and along X,Y, Z axis separately to sample data. After first Association finished, there are many planes fit the walls: data_association when first refine finished, the result is something wrong. The result has not improved, even if refine is continuously executed. My param is as following(time offset got from rqt_bag, by manual observation). The bag data is here,,

<arg name="topic_imu"           default="/imu/data" />
<arg name="bag_start"           default="1" />
<arg name="bag_durr"            default="100" />   <!--30-->
<arg name="scan4map"            default="40" />   <!--15-->
<arg name="lidar_model"         default="VLP_16" />
<arg name="ndtResolution"       default="0.5" /> <!-- 0.5 for indoor case and 1.0 for outdoor case -->

<arg name="time_offset_padding" default="0.001" /> <!--0.015-->
<arg name="show_ui"    default="true" />

The terminal result is: Note: Although the terminal display has converged, the result is completely wrong.

Load dataset from /home/jxl/Documents/lidar_imu_calib_result/C1_11F_lidar_imu_sampledata/2020-09-16-15-59-39_big_house.bag
/velodyne_packets: 910
/imu/data: 5000
Framebuffer with requested attributes not available. Using available framebuffer. You may see visual artifacts.



Ceres Solver Report: Iterations: 51, Initial cost: 6.906546e+05, Final cost: 3.015079e+02, Termination: NO_CONVERGENCE
[Initialization] Done. Euler_ItoL initial degree:  178.244  -177.28 -172.971



[Association] start ....
Plane type  :148 103 180; Plane number: 431
Surfel point number: 81558
[Association] 67809.2 ms

================ Iteration 0 ==================
Ceres Solver Report: Iterations: 121, Initial cost: 1.197994e+07, Final cost: 5.879937e+05, Termination: CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585;  0.134748  0.133536 0.0563047
[Accel] Error size, average: 4585;  0.0783432 0.0432497 0.0950025
[LiDAR] Error size, average: 81558; 0.184476

P_LinI      :  0.0096321 -0.0583626   -1.15264
euler_LtoI  :  2.99366 0.706655  58.5238
P_IinL      : 0.0886249 0.0821533   1.14781
euler_ItoL  : 177.834 177.816 121.499
time offset : 0
gravity     :   -1.8447 -0.335032   9.60879
acce bias   :  0.250138 -0.192866 0.0372121
gyro bias   :   0.0103923 -0.00936599 -0.00503104
[BatchOptimization] 163356 ms





================ Iteration 1 ==================
[Association] start ....
Plane type  :5 1 7; Plane number: 13
Surfel point number: 1000
[Association] 18330.9 ms
Ceres Solver Report: Iterations: 201, Initial cost: 1.993042e+05, Final cost: 3.300348e+04, Termination: NO_CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585; 0.0514282 0.0520165 0.0321376
[Accel] Error size, average: 4585;  0.0208095 0.0155759 0.0465463
[LiDAR] Error size, average: 1000; 0.197449

P_LinI      : -2.32769 -3.49175   2.3051
euler_LtoI  :   177.47 -163.953 -168.067
P_IinL      : 2.34987 3.17279 -2.7084
euler_ItoL  :  5.96376  15.1328 -13.0828
time offset : 0
gravity     : -6.19972 -3.13915  6.89589
acce bias   :  3.84131  1.39964 0.670884
gyro bias   :  0.0160932 0.00051557 0.00186257
[Refinement] 44252.6 ms




================ Iteration 2 ==================
[Association] start ....
Plane type  :25 16 57; Plane number: 98
Surfel point number: 536
[Association] 18385.8 ms
Ceres Solver Report: Iterations: 123, Initial cost: 2.843387e+04, Final cost: 2.425103e+04, Termination: CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585; 0.0504337 0.0472536  0.029875
[Accel] Error size, average: 4585;   0.020195 0.0135573 0.0399376
[LiDAR] Error size, average: 536; 0.0970633

P_LinI      : -3.20017 -4.12333  1.94568
euler_LtoI  :  6.96212 -17.4698  10.3303
P_IinL      : 2.97674 3.37805 -3.2797
euler_ItoL  : 176.024 161.645 171.383
time offset : 0
gravity     : -6.33728 -2.79176  6.92019
acce bias   :  4.37851  1.51152 0.888586
gyro bias   :    0.0139098 -0.000735396   0.00314523
[Refinement] 21040.2 ms

================ Iteration 3 ==================
[Association] start ....
Plane type  : 13  15 122; Plane number: 150
Surfel point number: 800
[Association] 18593.7 ms
Ceres Solver Report: Iterations: 26, Initial cost: 2.345097e+04, Final cost: 2.334657e+04, Termination: CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585; 0.0502834 0.0469197  0.029966
[Accel] Error size, average: 4585;  0.0202703 0.0135586  0.039756
[LiDAR] Error size, average: 800; 0.0125079

P_LinI      : -3.22242 -4.10098  1.93203
euler_LtoI  :  6.99051 -17.1054  9.75836
P_IinL      :  2.98483  3.37833 -3.25768
euler_ItoL  : 175.752  162.04 171.966
time offset : 0
gravity     : -6.34575 -2.80172   6.9084
acce bias   :   4.3993  1.51087 0.897642
gyro bias   :     0.013666 -0.000854228   0.00302885
[Refinement] 4626.24 ms




================ Iteration 4 ==================
[Association] start ....
Plane type  : 16  17 133; Plane number: 166
Surfel point number: 884
[Association] 18639.6 ms
Ceres Solver Report: Iterations: 31, Initial cost: 2.334325e+04, Final cost: 2.329166e+04, Termination: CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585; 0.0501724 0.0468462 0.0299202
[Accel] Error size, average: 4585;  0.0203146 0.0135702 0.0396821
[LiDAR] Error size, average: 884; 0.012153

P_LinI      : -3.22564 -4.09614   1.9202
euler_LtoI  :  7.24535 -16.8125  9.45732
P_IinL      :  2.98282  3.37704 -3.25095
euler_ItoL  : 175.349 162.317 172.339
time offset : 0
gravity     : -6.34777 -2.80218  6.90635
acce bias   : 4.41002 1.51142 0.90247
gyro bias   :    0.0135861 -0.000874201   0.00295927
[Refinement] 5458.55 ms

================ Iteration 5 ==================
[Association] start ....
Plane type  : 18  12 142; Plane number: 172
Surfel point number: 901
[Association] 19696.2 ms
Ceres Solver Report: Iterations: 29, Initial cost: 2.328773e+04, Final cost: 2.326717e+04, Termination: CONVERGENCE

============== After optimization ================
[Gyro]  Error size, average: 4585; 0.0501242 0.0468216 0.0299117
[Accel] Error size, average: 4585;  0.0203191 0.0135803 0.0396618
[LiDAR] Error size, average: 901; 0.0112675

P_LinI      : -3.22544 -4.09933  1.91412
euler_LtoI  :  7.43774 -16.6043  9.36238
P_IinL      : 2.98595 3.37628 -3.2491
euler_ItoL  :  175.09 162.504  172.48
time offset : 0
gravity     : -6.34759 -2.80068  6.90713
acce bias   :  4.41377  1.51216 0.904278
gyro bias   :   0.0135546 -0.00086725  0.00291006
[Refinement] 5377.06 ms

Any help and suggestion will be very appreciated a lot ~

narutojxl avatar Sep 16 '20 09:09 narutojxl

Hello, the lidar I used is not vpl-16. The horizontal angle of lidar I used can only take 210 degree point cloud data, and the remaining 150 degree point cloud data is Nan. In this case, can I still use this program to calibrate?

In addition, since the radar used is not vpl-16, the two files mentioned above have been modified in the program.

I also want to ask if it can be used for calibration, do I need to modify other contents of the program?

zhangfan950304 avatar Nov 19 '20 09:11 zhangfan950304

Hi author, I don't have a hardware to do time synchronization, use rqt_bag it seems msgs distribute evenly. image

  • When Initialization finished, the initial value of rotation seems correct. My imu basically aligins with laser(rotation approximate identity)

[Initialization] Done. Euler_ItoL initial degree: 3.78477 0.405341 -1.24306

  • Association shows that we have got Plane type :65 1 55; Plane number: 121.
  • When Batch finished, the result is different from the initial value. It seems thant it converges to the wrong solution.

euler_ItoL : 57.4901 5.51252 -8.14225

  • When Refine finished, plane number decrease, not increase.

Plane type :16 8 11; Plane number: 25 ... refine refine ...

The NDT_target_map.pcd basically meets the environmental scene, but the refined_map.pcd is mess. image image

could you identify why tool behaves like this?

goktugyildirim avatar Mar 12 '21 14:03 goktugyildirim

@goktugyildirim Hello, have you Completed lidar-imu calibration? I have a try by this work,but I failed.Do you have any suggestions else?

guo-zixuan avatar Mar 29 '21 04:03 guo-zixuan

hi~, I've modified the code and used it for my own data. However, there is a problem. The translation I calibrated is very large and does not meet the expectation. I want to know:

  1. What maybe causes this error.
  2. Are there any requirements for recording rosbag. For example, what are the requirements for motion trajectory

@qian5683 , hello, i use plane motion imu data to do this calibration, it fail on init part with a NAN fault happend on ceres solver, i check my imu raw data, there is no nan value, and i am confuse and helpless. did you calib with plane motion data as metion below, and how is the result?

Chrislzy1993 avatar May 18 '21 07:05 Chrislzy1993

@goktugyildirim Hello, have you Completed lidar-imu calibration? I have a try by this work,but I failed.Do you have any suggestions else?

af-doom avatar Sep 19 '22 03:09 af-doom