Hi Patrick, great content ! At timestamp 25:48, you said we dont use the values of random walk and noise directly, but inflate thm by order of 10 times or by 20. Why is it so? Is there any mathematical or empirical significance to it? Or is it just that moving robot might incur more noise and bias for IMU than when calculated for steady IMU?
This is to take into account additional noises not fully modeled. Especially if you will mount this sensor on a mobile robot, there will be additional vibration and temperature affects.
Hi Patric, Thank you so much for your great video. At timestamp 37:34 you are mentioning the timeshift cam0 to imu0 that its not 0 and we have to take that into account. I have two questions: 1. By we should take that into account, we should use that as the time_offset in the kalibr_imu_chain.yaml file of our open-vins configuration? 2. If this data is bigger than 0.002 seconds, how it will effect our localization?
Yes, it should be taken into account at the point that the two sensor streams are combined (e.g. a visual-inertial navigation system). The magnitude shouldn't matter much but will cause the estimate to be delayed since we will need to buffer enough data until we have the sensor data from each sensor at the same timestep. Hope this helps.
Hi Patric, I am implementing openvins on d455. The Imu keeps drifting away even if the camera is in stationary. I recalibrate it and the values are very close to the once that already exists in rs_d455 config folder. Do you have any suggestions on how to rectify this?
Hi Patrick, how about your computer configuration ? and I have a Intel® Core™ i7-10700 CPU @ 2.90GHz × 16 computer and run one imu and two cameras calibration using kalibr for more than two hours, Could you give me some tips about how to accelarate?
Great video, but why multiply noise and random walk values by some constant value? Spend a lot of time to get accurate results and just multiply on random numbers it's strange, or there are something behind it?
We rather be a bit conservative. There are additional sources which the static motion doesn't take into account. For example when mounting on a robot which vibrates (or a computer with fans) there will be additional sources of noise which the static calibration can't capture. You can try not inflating of course!
Hi, I have question regarding when to assure that I have good calibration or not? I still have too much reprojection error, I recorded multiple times with 30 fps and 752x480 and the video contain some glitching, do u suggest to change any paramerts to enhance the calibration? shud the video be super smooth with no glitches?
Hi Patrick, great content ! At timestamp 25:48, you said we dont use the values of random walk and noise directly, but inflate thm by order of 10 times or by 20. Why is it so? Is there any mathematical or empirical significance to it? Or is it just that moving robot might incur more noise and bias for IMU than when calculated for steady IMU?
This is to take into account additional noises not fully modeled. Especially if you will mount this sensor on a mobile robot, there will be additional vibration and temperature affects.
Thank you for the amazing video.💮
Thanks, it helps a lot!
Hi Patric, Thank you so much for your great video. At timestamp 37:34 you are mentioning the timeshift cam0 to imu0 that its not 0 and we have to take that into account. I have two questions: 1. By we should take that into account, we should use that as the time_offset in the kalibr_imu_chain.yaml file of our open-vins configuration? 2. If this data is bigger than 0.002 seconds, how it will effect our localization?
Yes, it should be taken into account at the point that the two sensor streams are combined (e.g. a visual-inertial navigation system). The magnitude shouldn't matter much but will cause the estimate to be delayed since we will need to buffer enough data until we have the sensor data from each sensor at the same timestep. Hope this helps.
Hi Patric, I am implementing openvins on d455. The Imu keeps drifting away even if the camera is in stationary. I recalibrate it and the values are very close to the once that already exists in rs_d455 config folder. Do you have any suggestions on how to rectify this?
Hi Patrick, how about your computer configuration ? and I have a Intel® Core™ i7-10700 CPU @ 2.90GHz × 16 computer and run one imu and two cameras calibration using kalibr for more than two hours, Could you give me some tips about how to accelarate?
Ensure you build with release mode. Additionally, you can process the camera at a lower rate with the --bag-freq argument.
Thanks very much and that works @@patrickgeneva
Great video, but why multiply noise and random walk values by some constant value? Spend a lot of time to get accurate results and just multiply on random numbers it's strange, or there are something behind it?
We rather be a bit conservative. There are additional sources which the static motion doesn't take into account. For example when mounting on a robot which vibrates (or a computer with fans) there will be additional sources of noise which the static calibration can't capture. You can try not inflating of course!
I'm a begineer here. Is the procedure similar for d435i camera?
Hi, I have question regarding when to assure that I have good calibration or not? I still have too much reprojection error, I recorded multiple times with 30 fps and 752x480 and the video contain some glitching, do u suggest to change any paramerts to enhance the calibration? shud the video be super smooth with no glitches?
also, my image index scale is 0-10 only
Thank you!
Thank you for this video. but how do you generate : "+-- dataset-dir
+-- cam0
│ +-- 1385030208726607500.png
│ +-- ...
│ \-- 1385030212176607500.png
+-- cam1
│ +-- 1385030208726607500.png
│ +-- ...
│ \-- 1385030212176607500.png
\-- imu0.csv"?