weisongwen / UrbanLoco

UrbanLoco: A Full Sensor Suite Dataset for Mapping and Localization in Urban Scenes
https://advdataset2019.wixsite.com/urbanloco
394 stars 37 forks source link

question about the extrinsic #10

Open huanhexiao opened 3 years ago

huanhexiao commented 3 years ago

I am confused with CA data extrinsic paras, or the coordinate axis. 1) Could u please tell me the axis directions of each sensor? 2) What does each " [to LIDAR]" mean, senor_T_Lidar or Lidar_T_sensor ?

For instance,I guess the axis directions of Lidar seems like Right-Forward-Up(alongVehicle) according to prius.png, and Xsens's is Forward-Left-Up, as Xsens's extr is [0., -1., 0., 0., 1., 0., 0., 0., 0., 0., 1., 7.62e-02] According to pose convention a_T_b, the z-axis translation 7.62e-02 means the b is above a, so imu_T_Lidar is right, rather than Lidar_T_imu. Based on this supporse, each camera's extr means camera_T_Lidar, and each y-axis translation should be negative value since camera's y-axis defines the vertical direction, right? But the Cam4 and Cam5 is not.

My discription may seem complicated, but i think i'm talking about some simple issues. So can u do me a favour?

weisongwen commented 3 years ago

@huanhexiao thanks for the interest. " [to LIDAR]" means sensor_T_LiDAR. In other words, all the sensor measurements are calibrated to the LiDAR sensor on the roof of the vehicle.

Regarding the 7.62e-02, this means that the z-axis is almost the same with respect to the LiDAR and IMU. I am very sure that our calibration accuracy extrinsic is less than 7.62e-02.

Hope this helps. BTW, the recent papers using our datasets seem to be ok with the provided extrinsic. You can also take a look at those papers.

Hope this helps,

huanhexiao commented 3 years ago

@huanhexiao thanks for the interest. " [to LIDAR]" means sensor_T_LiDAR. In other words, all the sensor measurements are calibrated to the LiDAR sensor on the roof of the vehicle.

Regarding the 7.62e-02, this means that the z-axis is almost the same with respect to the LiDAR and IMU. I am very sure that our calibration accuracy extrinsic is less than 7.62e-02.

Hope this helps. BTW, the recent papers using our datasets seem to be ok with the provided extrinsic. You can also take a look at those papers.

Hope this helps,

Thanks for your reply, i've tried other translations, and quite sure about sensor_T_LiDAR.

However i still feel confused with the rotation, Could u please tell me the axis directions of the sensors? Here is my guess of the sensors' XYZ(along vehicle): Cam0: RDF (Right-Down-Forward) Lidar: RFU Xsens: FLU I tried to recover the euler angles from some rotation matrix, such as imu_R_Lidar, Cam0_R_Lidar, the values do not correspond with my guess.

weisongwen commented 3 years ago

@huanhexiao haha your questions are quite a navigation style. We directly calibrate all the sensors to the LiDAR on the roof and we do not provide coordinate information like "RDF (Right-Down-Forward)". But I think you can recover this information from the extrinsic. e.g. you can take a look at the LiDAR clouds data. I believe you can figure out this via existing data.

BTW, @yzhou377 may help but he is pretty busy. Please let us know if you still get problems and show us your algorithm details. about how you use the data.

huanhexiao commented 3 years ago

@weisongwen Thanks for patient and time. In addition to my navigation style, i want to estimate the extrinsic paras of IMU-Cam in a recent research, so i need an accurate one as a reference. i will make some tests first. Looking forward @yzhou377 your answer!

Kailthen commented 2 years ago

I'm ploting point cloud point to cam0(front image), the result seems not corrent. Any suggestion ? image

weisongwen commented 2 years ago

Hi @Kailthen , sorry for the delayed reply. You are right, the extrinsic parameters between the camera and the LiDAR are not accurate. During the dataset preparation, the extrinsic parameters are calibrated using the Autoware. You are suggested to use some existing open-sourced work for LiDAR-camera calibration to refine the extrinsic parameters. We will follow this up but It is hard to provide the exact timeline.

Thank you, Weisong,

wjf1997 commented 8 months ago

I am confused with CA data extrinsic paras, or the coordinate axis.

  1. Could u please tell me the axis directions of each sensor?
  2. What does each " [to LIDAR]" mean, senor_T_Lidar or Lidar_T_sensor ?

For instance,I guess the axis directions of Lidar seems like Right-Forward-Up(alongVehicle) according to prius.png, and Xsens's is Forward-Left-Up, as Xsens's extr is [0., -1., 0., 0., 1., 0., 0., 0., 0., 0., 1., 7.62e-02] According to pose convention a_T_b, the z-axis translation 7.62e-02 means the b is above a, so imu_T_Lidar is right, rather than Lidar_T_imu. Based on this supporse, each camera's extr means camera_T_Lidar, and each y-axis translation should be negative value since camera's y-axis defines the vertical direction, right? But the Cam4 and Cam5 is not.

My discription may seem complicated, but i think i'm talking about some simple issues. So can u do me a favour?

" According to pose convention a_T_b, the z-axis translation 7.62e-02 means the b is above a," it should means a is above b, thus 7.62e-02 should be negative, right?