Learning positioning from scratch --- kaist data set experience

1 Introduction to Kaist dataset

I chose the data set for several days, and finally chose the kaist data set. This data set includes wheel speedometer, imu, gps, and 16-line radar. These sensors meet the needs of multi-sensor fusion positioning in my opinion.

A brief introduction to the Kaist dataset

The dataset is oriented towards urban environments (e.g. metropolitan areas, complex buildings and residential areas). [1] Provides the above sensor data

Dataset URL: https://sites.google.com/view/complex-urban-dataset/download-lidar#h.sa42osfdnwst
Dataset Paper Name: Complex Urban LiDAR Data Set

1.1 Sensor data

provided

• 2 veldyne 16-line radars with 16 lines
• 2 sick single line radar
• Odometer
• 1 consumer grade imu
• A 3-axis fiber optic gyroscope (can be used as the true value of attitude)
• a gps
• a vrs gps(rtk)
• stereo camera

The specific sensor configuration can refer to the screenshots in the paper below.

1.2 Sensor location

Two 16-line radars are installed at the rear of the car, obliquely downward, so that you can see more of the ground.
The timestamp in the data is the timestamp of the last packet in the radar circle.

The positions of all sensors are relative to the reference coordinate system, which is Vehicle Center in the figure.

2 Dataset download

I put the connection of the data set in the Tencent document. The address of the Tencent document is as follows:
https://docs.qq.com/sheet/DVElRQVNlY0tHU01I?tab=BB08J2

Now downloading needs to go over the wall, and if necessary in the future, the data can also be uploaded to Alibaba cloud disk or the like.

3 Use of the dataset

This dataset provides a tool to use. As long as the folder of the dataset is set in a fixed format and loaded through the tool, the topic data in ros format can be directly obtained, which is very convenient.

This tool is file_player, and its github address is
https://github.com/irapkaist/file_player

3.1 Project repository

It needs to be downloaded and compiled in your own workspace. I have downloaded the dependencies and source code and put them in an open source project. The github address is
https://github.com/xiangli0608/Learning_localization_from_scratch_ws

This warehouse is already a working space, just download it and directly cd into it, and then compile it with catkin_make.

There are 3 folders in the kaist_tool folder in the src folder, which are

• file_player is a tool that reads datasets and publishes datasets through messages in ros format
• irp_sen_msgs some custom message types
• kaist_description The urdf I wrote based on the calibration file provided by the dataset

Since I only downloaded the Urban08 dataset for the time being, there is only one urban08.urdf. I don’t know if the calibration parameters of the other datasets are the same. If they are the same, this urdf can use all the data.

The current urdf does not have a camera link, and it will be added when it is used.

3.2 Folder settings for datasets

Since file_player is a tool for reading folders, there are requirements for the placement of folders and directories. If you do not set it according to the specified location, you will not be able to read the data pairs.

The settings for the folder are as follows:

Urban08

• calibration
  • EncoderParameter.txt
  • Vehicle2BackSick.txt
  • Vehicle2FOG.txt
  • Vehicle2GPS.txt
  • Vehicle2IMU.txt
  • Vehicle2LeftVLP.txt
  • Vehicle2MiddleSick.txt
  • Vehicle2RightVLP.txt
  • Vehicle2VRS.txt
• sensor_data
  • VLP_left folder
  • VLP_right folder
  • SICK_back folder
  • SICK_middle folder
  • csv files for other sensor data

When you use the official link to download, you need to merge the Urban08 folder, and finally merge it into the folder shown above.

3.3 start file_player

After compiling the code, first source, then rospack prodfile, and then execute

roslaunch file_player file_player.launch

The following interface will pop up

Click the Load button, select the Urban08 folder you just set up, and then print it in the terminal

Stamp data are loaded
Altimeter data are loaded
Encoder calibrated parameter
Encoder resolution: 4096
Encoder left wheel diameter: 0.623803
Encoder right wheel diameter: 0.623095
Encoder wheel base: 1.52683
Encoder data are loaded
Fog data are loaded
Gps data are loaded
Vrs gps data are loaded
IMU data are loaded
scandir: No such file or directory
[ INFO] [1658414406.186192494]: camera calibration URL: file:///media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/left.yaml
[ INFO] [1658414406.187175921]: Unable to open camera calibration file [/media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/left.yaml]
[ WARN] [1658414406.187216202]: Camera calibration file /media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/left.yaml not found.
[ INFO] [1658414406.187247816]: camera calibration URL: file:///media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/right.yaml
[ INFO] [1658414406.187349750]: Unable to open camera calibration file [/media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/right.yaml]
[ WARN] [1658414406.187373500]: Camera calibration file /media/trunk/Trunk/0-LX/Kaist/Urban08/calibration/right.yaml not found.

It means that the data is loaded successfully, if there is no prompt, it means that the loading is not successful. Then press the Play button to publish the data through ROS.

3.3 Sensor Data Topics

There will be many topics released, some are in the standard format of ros, and some are in the format of irp_sen_msgs

The data in the ros standard format is as follows:

• /imu/data_raw 110hz
• /odom 100hz
• /gps/fix 5hz
• /ns1/velodyne_points 10hz
• /ns2/velodyne_points 10hz

in

• /vrs_gps_data is the GPS of rtk, which can be used as the true value of the location, in the format of irp_sen_msgs/vrs
• /xsens_imu_data is a fiber optic gyroscope, which can be used as the true value of the attitude, in the format of irp_sen_msgs/imu

There are a lot of visual topics, let’s talk about it when we use vision

3.4 Point cloud visualization

roslaunch kaist_description kaist_description.launchVisualization of point clouds is possible by executing

The visualization between the coordinate systems is as follows

The visualization of the point cloud is as follows

4 Summary

Now that the data set is selected, the data can also be released through ROS. The next step is to build a map first, and then start multi-sensor fusion positioning.

REFERENCES

[1] KAIST URBAN DATA SET/Complex Urban LiDAR Data Set data set parameters
https://blog.csdn.net/Iqun_LAN/article/details/106445884
[2] Use of Kaist data set
https://blog.csdn.net /weixin_43846627/article/details/113510014