search

tomas789 / kitti2bag

Convert KITTI dataset to ROS bag file the easy way!
MIT License
702 stars 256 forks source link

tf data doesn't seem to exactly match up with gnd truth #11

Open vkee opened 6 years ago

vkee commented 6 years ago

Hi,

I generated a bag file of sequence 6 of the KITTI odometry dataset from the raw data recordings using this tool. I then wrote a simple node that uses the rosbag tf to lookup the left camera coordinate pose at each frame and then stores this trajectory. I used the KITTI evaluation tool to plot and evaluate the trajectory.

06

As you can see, it is pretty close to the ground truth which is expected. However, I would expect it to be exactly the same or very near to it as the paper (Are we ready for autonomous driving? the KITTI vision benchmark suite) says: "The ground truth for visual odometry/SLAM is directly given by the output of the GPS/IMU localization unit projected into the coordinate system of the left camera after rectification."

I did a pose by pose comparison and plotted the translation RMSE (distance between the poses) and for the most part it is close, but there is some significant error at some frames. I haven't been able to determine why there is that error as the ground truth is from the GPS/IMU data directly and the tfs are based off of that.

rmse_6_rmse

Do you have any idea what may be the cause?

Thanks, Vincent

kerimyener commented 6 years ago

@vkee hello would you mind to tell me how to convert sequence 6 of kitti odometry data to rosbag file i am a bit confused. Your help will be appreciated. Thank you

vkee commented 6 years ago

@kerimyener I used raw datasets.

brunoeducsantos commented 5 years ago

@vkee did you manage to solve the tf issue?

wxt-github commented 5 years ago

Can you tell me where the sixth sequence is in the raw data, too much raw data, and no idea where it is. Also, I would like to ask if the sixth sequence is a complete closed loop? @vkee

valgur commented 5 years ago

Related: https://github.com/ethz-asl/kitti_to_rosbag/issues/7

valgur commented 5 years ago

@wxt-github The mapping from odometry to raw datasets is provided for the first 11 sequences in the readme of the odometry devkit.

Nr. Sequence name Start End
00 2011_10_03_drive_0027 000000 004540
01 2011_10_03_drive_0042 000000 001100
02 2011_10_03_drive_0034 000000 004660
03 2011_09_26_drive_0067 000000 000800
04 2011_09_30_drive_0016 000000 000270
05 2011_09_30_drive_0018 000000 002760
06 2011_09_30_drive_0020 000000 001100
07 2011_09_30_drive_0027 000000 001100
08 2011_09_30_drive_0028 001100 005170
09 2011_09_30_drive_0033 000000 001590
10 2011_09_30_drive_0034 000000 001200
hemangchawla commented 4 years ago

Is this related to the camera rectification after the oxts to xyz?

J-Rojas commented 4 years ago

I noticed some errors with the TF when I mapped the lidar data against the vehicle's motion over time. I believe the TFs are not calculated correctly because they are missing the transformation from IMU/GPS to the velodyne sensor. The calib_imu_to_velo.txt is not used in this code and that would most likely throw off any calculations that are mapping lidar/camera data to GPS coordinates.

chubukeji commented 2 years ago

@valgur hello，the 03: 2011_09_26_drive_0067 can't be found in official website. Do you have this dataset? Or，do you know why it can't be found? my Email 2661076951@qq.com

christian-hiebl commented 1 year ago

@chubukeji This discussion suggests it was removed because parts overlapped with the public benchmark test. https://github.com/alexkreimer/odometry/issues/1#issuecomment-534451019