Calibration problem？

[cheneyfan]

in the Camera-IMU calibration，i get “T_cam_imu”， then put it in T_SC in the config file directly？

T_cam_imu and T_SC both mean transformation from Sensor to Camera？

[weblucas]

sensor = imu pose_insensorframe= T_SC pose_incameraframe vs pose_incameraframe= T_cam_imu pose_insensorframe

Are you using kalibr? you can use the exporter script for OKVIS https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/exporters/kalibr_aslam_config

[cheneyfan]

Yes, i’am using Kalibr. Directly replace T_SC with T_cam_imu？ If not， how to transform T_cam_imu into T_SC？

[weblucas]

No you have to invert the matrix. use this python script that it already prepare all the matrix to you https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/exporters/kalibr_aslam_config

[cheneyfan]

“T_cam_imu” in camchain-imucam-%BAGNAME%.yaml ， than convert it into “T_SC” in config_fpga_p2_euroc.yaml ？ can i just inverse “T_cam_imu”？

[aforster]

Yes. Invert it. S stands for Sensor which refers to the IMU. Please read the Readme.

Also check this for a good introduction to camera frame notation.

[weblucas]

yes, but to avoid numerical errors. do not invert the 4x4 matrix. use this approach http://www.cg.info.hiroshima-cu.ac.jp/~miyazaki/knowledge/image/eq53.gif,

On Fri, 26 May 2017 at 14:46 Andreas Forster notifications@github.com wrote:

Yes. Invert it. S stands for Sensor which refers to the IMU. Please read the Readme.

Also check this http://paulfurgale.info/news/2014/6/9/representing-robot-pose-the-good-the-bad-and-the-ugly for a good introduction to camera frame notation.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/ethz-asl/okvis/issues/38#issuecomment-304274021, or mute the thread https://github.com/notifications/unsubscribe-auth/AAxwtn1jf8_dvlTet4baJarIlmkqUqogks5r9sm3gaJpZM4NkkBp .

[cheneyfan]

OK, I have finished the calibration. The pose keeps unchanged when power on and it can run tracking normally within small and slight motion， but it will drift a lot with normal motion. How to optimize it？

[weblucas]

What is the range of the reprojection error of your calibration?

On Sat, 27 May 2017, 04:49 cheneyfan, notifications@github.com wrote:

OK, I have finished the calibration. The pose keeps unchanged when power on and it can run tracking normally within small and slight motion， but it will drift a lot with normal motion. How to optimize it？

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/ethz-asl/okvis/issues/38#issuecomment-304422101, or mute the thread https://github.com/notifications/unsubscribe-auth/AAxwtmkJ2PL_7-b_o9oBIB-4CfvwWa2gks5r948vgaJpZM4NkkBp .

[cheneyfan]

Calibration results

Reprojection error squarred (cam0): mean 0.155326923144, median 0.0510172933023, std: 0.624074143921 Reprojection error squarred (cam1): mean 0.15045563157, median 0.048621853134, std: 0.680495108651 Gyro error squarred (imu0): mean 198.677348797, median 7.02152733236, std: 749.008430121 Accelerometer error squarred (imu0): mean 5.01081140157, median 0.276420872561, std: 18.4145340448

Transformation (cam0):

T_ci: (imu to cam0): [m] [[ 0.0364117 -0.99857832 -0.03892978 -0.00774579] [-0.99917728 -0.03707442 0.01643904 -0.00258252] [-0.01785897 0.03829918 -0.99910672 0.00072436] [ 0. 0. 0. 1. ]]

T_ic: (cam0 to imu): [m] [[ 0.0364117 -0.99917728 -0.01785897 -0.00228543] [-0.99857832 -0.03707442 0.03829918 -0.00785827] [-0.03892978 0.01643904 -0.99910672 0.00046463] [ 0. 0. 0. 1. ]]

timeshift cam0 to imu0: [s] (t_imu = t_cam + shift) 0.0

Transformation (cam1):

T_ci: (imu to cam1): [m] [[ 0.02330732 -0.99947719 -0.02240795 -0.10869984] [-0.9995838 -0.02367924 0.01647798 -0.0027767 ] [-0.01699996 0.02201457 -0.99961311 -0.00136361] [ 0. 0. 0. 1. ]]

T_ic: (cam1 to imu): [m] [[ 0.02330732 -0.9995838 -0.01699996 -0.00026522] [-0.99947719 -0.02367924 0.02201457 -0.10867874] [-0.02240795 0.01647798 -0.99961311 -0.00375307] [ 0. 0. 0. 1. ]]

timeshift cam1 to imu0: [s] (t_imu = t_cam + shift) 0.0

Baselines:

Baseline (cam0 to cam1): [m] [[ 0.99977725 0.01339852 -0.01630746 -0.10090936] [-0.01339245 0.9999102 0.00048138 -0.00029849] [ 0.01631245 -0.00026288 0.99986691 -0.0019622 ] [ 0. 0. 0. 1. ]] baseline norm: 0.100928875648 [m]

Gravity vector in target coords: : [m/s^2] [ 5.27157227 -4.73801359 -6.7821717 ]

Calibration configuration

cam0

Camera model: pinhole Focal length: [419.7210910977609, 492.82337740835027] Principal point: [319.52734052614846, 235.52733517897832] Distortion model: equidistant Distortion coefficients: [-0.17748121501791742, 0.10571210592569351, -0.12395922727304214, 0.05614521113566777] Type: aprilgrid Tags: Rows: 6 Cols: 6 Size: 0.088 [m] Spacing 0.0264 [m]

cam1

Camera model: pinhole Focal length: [418.0268345400491, 490.8384144277561] Principal point: [317.8793551858768, 231.18805105856094] Distortion model: equidistant Distortion coefficients: [-0.16485391420830264, 0.06374731242574906, -0.06152118701051446, 0.024577910349410686] Type: aprilgrid Tags: Rows: 6 Cols: 6 Size: 0.088 [m] Spacing 0.0264 [m]

IMU configuration

Update rate: 200.0 Accelerometer: Noise density: 0.0025802 Noise density (discrete): 0.0364895383364 Random walk: 0.00048889 Gyroscope: Noise density: 0.00012901 Noise density (discrete): 0.00182447691682 Random walk: 6.23e-05

[lusj]

@cheneyfan have you got the propre value of "T_SC"?