/3d-groundtruth-traffic-sign-positions

Ground Truth 3D positions of traffic signs in the KITTI sequences 00 to 10 (except 03)

3d Ground truth Traffic Sign Positions in KITTI

This respository contains the annotated Ground Truth (GT) traffic sign 3D positions annotated using the LiDAR and images from the KITTI raw dataset.

Usage

If you use this dataset, please cite as:

H. Chawla, M. Jukola, T. Brouns, E. Arani, and B. Zonooz, "Crowdsourced 3D Mapping: A Combined Multi-View Geometry and Self-Supervised Learning Approach," 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),Las Vegas, USA, IEEE (in press), 2020.

Distribution of Signs

A total of 73 signs are annotated across sequences 00 -> 10 (except 03 which is missing from KITTI raw)

Seq Date_Drive # Signs Annotated
00 2011_10_03_drive_0027 15
01 2011_10_03_drive_0042 16
02 2011_10_03_drive_0034 12
04 2011_09_30_drive_0016 1
05 2011_09_30_drive_0018 4
06 2011_09_30_drive_0020 5
07 2011_09_30_drive_0027 2
08 2011_09_30_drive_0028 8
09 2011_09_30_drive_0033 7
10 2011_09_30_drive_0034 3

Annotation Process

Inputs

  • LiDAR scans
  • Rectified and Synced Images
  • GT path (metric)
  • OXTS GPS
  • Calibration (Camera, Camera-to-GPS/IMU, Camera-to-Velodyne)

Method

Details Coming soon!

Outputs

  • Relative Positions of signs in each sequence in format
  • Absolute Metric Positions of signs (wrt origin frame)
  • Absolute GPS Positions of signs

All annotation txt files (per Sequence) are of the format:

Image ID x y z GT ID
... ... ... ... ...

Annotation Variance

Since each raw sign position is annotated from multiple coresponding pairs of image and lidar scans, it's annotated position may vary slightly. Then the raw sign position annotations are averaged as explained above. There mean difference in the final positions of each sign and raw annotations is given below:

Var x y z Overall
Mean 0.035 0.021 0.034 0.061

Usage

  • Each folder contains txt files describing the raw (available), relative (available) and absolute (coming soon!) annotated traffic sign positions for the corresponding KITTI sequence.
  • The annotations can be loaded in python using pandas. 
    import pandas as pd
pd.read_csv(ground_truth_position_file,sep=';')

Additional

Detection Annotations

  • GT annotations for the bounding box detections have also been provided.
  • Per-image annotation files contain the (x,y) coordinates of the 4 corners of the bounding boxes (8 values), with each row representing a different traffic sign.