goldbattle
commented
2 years ago This is likely the case that the IMU is too poor in quality, a long enough dataset was not collected, or some external factor is affecting the readings (eg. temp or drivers).
following tools to calculate the imu noise results and used them for Kalibr, but the acc convergence effect and reprojection error were not good in Kalibr
I tried to get better acc convergence results (reprojection error still high) when multiply factor with x10 for accelerometer_noise_density and accelerometer_random_walk
https://github.com/mintar/imu_utils https://github.com/ori-drs/allan_variance_ros
kalibr_result only vo
report-cam-our_nascar_datasetp50_imei7289calibr_728920220523_calibration20220523_calibration.pdf
allan_variance_ros result :
Accelerometer
accelerometer_noise_density: 0.0023093122966287663 accelerometer_random_walk: 9.754280328273595e-05
Gyroscope
gyroscope_noise_density: 8.556686388277563e-05 gyroscope_random_walk: 1.7024430789088757e-06
report-imucam-our_nascar_datasetp50_imei728920220523_calibration20220523_calibration.pdf
imu_utils results : Gyr: unit: "gyr_n: rad / sqrt(s), gyr_w: rad / s^2 / sqrt(Hz)" avg-axis: gyr_n: 6.4571634112763670e-05 gyr_w: 7.5463540571697402e-08 x-axis: gyr_n: 8.2809540837727724e-05 gyr_w: 2.2624825132915823e-07 y-axis: gyr_n: 7.0443502725776796e-05 gyr_w: 8.4495086640230569e-11 z-axis: gyr_n: 4.0461858774786483e-05 gyr_w: 5.7875299293741360e-11 Acc: unit: "acc_n: m / s^2 / sqrt(Hz), acc_w: m / s^3 / sqrt(Hz)" avg-axis: acc_n: 7.9859666683206910e-04 acc_w: 2.6832686051089446e-05 x-axis: acc_n: 9.1287741201428606e-04 acc_w: 5.1106059191407176e-05 y-axis: acc_n: 9.3866594974910584e-04 acc_w: 2.9383859860592782e-05 z-axis: acc_n: 5.4424663873281519e-04 acc_w: 8.1391012683762964e-09 report-imucam-our_nascar_datasetp50_imei728920220523_calibration20220523_calibration.pdf