L3Pilot Data

OpenDD: A Large-Scale Roundabout Drone Dataset


The OpenDD dataset introduced in the following is an extensive anonymized trajectory dataset, covering 7 roundabouts in Wolfsburg, Germany, and Ingolstadt, Germany.

Supported by the L3Pilot project, the dataset is published by VW Group Innovation, part of the Volkswagen AG (→ external link).

The dataset aims at providing a relevant dataset to improve trajectory prediction algorithms, as well as to provide naturalistic data for the simulation of other traffic participants.

OpenDD covers trajectories and high precision bounding boxes of over 80,000 different road users tracked with a unique object id in over 62h of data, as well as HD map information of the 7 covered roundabouts.

A more detailed introduction of the data and a description of the suggested splits for training, test, and evaluation purposes is provided in the accompanying publication, mentioned below.


For some of the roundabouts the drivable area image was missing from the zip, as well as small corrections in the split definition have been made.

Please make sure that you have the data corresponding to version 2 when using the dataset. The version can be seen in the name of the trajectory database.


The corresponding publication can be accessed at arXiv (→ external link).

When using the dataset, please refer to the following submission at the 23rd IEEE International Conference on Intelligent Transportation Systems (ITSC) (→ external link).

@inproceedings{openDD dataset,
title={OpenDD: A Large-Scale Roundabout Drone Dataset},
author={Breuer, Antonia, Termöhlen, Jan-Aike, Homoceanu, Silviu, and Fingscheidt, Tim},
booktitle = {Proceedings of International Conference on Intelligent Transportation Systems},
month = {September},
year = {2020}}


The dataset is distributed under the CC BY-ND 4.0 license (→ external link). A human-readable summary of the license can be found here (→ external link).


Structure and content of the Dataset Downloads

The dataset is split up in several downloads, each of which has the following structure.

  • $ROOT
    • README.txt – Short information on the dataset.
    • LICENSE_by_nd_4.0.txt - The CC BY-ND 4.0 license.
    • trajectories_$ROOT.sqlite - SqLite database containing the trajectories of the given sub-dataset.
    • map_$ROOT - Map information for the roundabout covered in $ROOT.
      • map_$ROOT_UTM32N.xml - XML representation of the roundabout covered in $ROOT, in the UTM32N coordinate system.
      • map_$ROOT.sqlite – SqLite database containing most of the information from map_$ROOT.xml.
      • shapefiles_trafficlanes - Shapefiles describing the trafficlanes.
      • shapefiles_borderlines - Shapefiles describing the lane borders.
      • shapefiles_drivableareas - Shapefiles describing the areas in the roundabout.
    • geo-referenced_images_$ROOT Geo-referenced images for visualization/training purposes.
      • $ROOT.jpg - Bird's eye view image taken by a drone of $ROOT.
      • $ROOT_drivable_area.png - A black-and white image of the drivable area, computed from $ROOT_map.xml
      • $ROOT.tfw - World file locating $ROOT.jpg and $ROOT_drivable_area.png in the UTM coordinate system.
    • split_definition - A collection of files defining which .sqlite database tables of the total dataset belongs to which split described in the paper (indepedent of subdataset $ROOT).
      • r1_train.txt
      • r1_val.txt
      • r2_train.txt
      • r2_val.txt
      • r3_train.txt
      • r3_val.txt
      • rA.txt
      • rB.txt
      • rC.txt

Dataset Description

A detailed description of the dataset can be found in the aforementioned publication. Here we will only briefly describe the data included in the dataset. The complete dataset can be downloaded below, split up over the different roundabouts, as well as one small example sub-dataset. For using the splits defined in the paper, the definitions of the splits reside in $ROOT/split_definition, as shown in Contents.


Trajectory Databases (.sqlite)

The actual dataset contains one .sqlite database for each of the 7 roundabouts included in the dataset. Each such database includes a table for each recording of the given roundabout.

An exemplary database, containing only one recording (thus only one table) for rdb1 can be found in the example_data download provided below.

Each table contains information about all objects present during the corresponding recording. A row in the table represents one instance of a given object at a given time instant.









Row Id.




Timestamp of the given object instance. Counted from the beginning of the given recording.




Id of the object this instance belongs to. Unique over the dataset.




X-Part of the UTM 32N coordinate of the center of the object instance.




Y-Part of the UTM 32N coordinate of the center of the object instance.




Bounding Box angle relative to UTM 32N x-axis.




Velocity of the object.




Acceleration of the object.




Tangential acceleration of the object.




Latitudinal acceleration of the object.



Pedestrian, Bicycle, Car, Van, Truck, Motorcycle, Trailer

Class the object belongs to.




Width of the object's bounding box.




Length of the object's bounding box.


String in form of [Int]


Array of trailers that belong to the object.

UTM32N Geo-Referenced Bird's Eye View Images (.png, .jpg and .tfw)

The actual dataset contains one .png image, one .jpg image and one .tfw World File (→ external link) for each of the 7 roundabouts covered by the dataset. The $ROOT.jpg contains the image of the roundabout, where private areas are anonymized by black polygons. The $ROOT_drivable_area.png contains a visualization of the drivable area of the $ROOT roundabout, which is directly derived from the areas defined in map_$ROOT_UTM32N.xml. The corresponding .tfw World File (→ external link) contains 6 lines, as follows:

0.02466059 # x-Component of the pixel width
-0.00230412 # y-Component of the pixel width
-0.00215162 # x-Component of the pixel height
-0.02544322 # y-Component of the pixel height
619245.36794307 # UTM x-Coordinate of the center of the upper-left pixel
5809189.40151734 # UTM y-Coordinate of the center of the upper-left pixel


Exemplary Data Visualisation

Visualisation of all trajectories in a given recording

Here all trajectories of the exemplary dataset on rdb1, provided below, can be seen. The trajectories are overlaid on the geo-referenced image using the UTM coordinates and the provided world file:



Visualisation of a given time instant in a recording

Visualisation of the a given time instant in a recording of rdb6 in the original video. Here, the high accuracy of the provided tracks can be seen.

The color of the OBJID visualizes the object's class.

  • Red - Car
  • Pink - Truck
  • Yellow - Trailer
  • Green - Pedestrian
  • Blue - Bicycle

For object 59 on top of the image it can be seen how trailers are detected and tracked separately, while the correspondence to the towing object is still provided.



The whole dataset is split up in 8 downloads, as listed below. The downloads contain exemplary data or the all data on one of the 7 roundabouts, and follow the file structure listed in Contents. For using the splits defined in the paper, the definitions of the splits reside in $ROOT/split_definition, as shown in Contents.

We additionally provide a smaller download "exemplary_data" which includes the trajectories of one recording and the related images and map. This exemplary dataset should suffice to get a first impression of the data contained in OpenDD without the need for a larger download.

example_data (14MB) - a small exemplary dataset, please download this before downloading the whole dataset

rdb1 (679MB) - files corresponding to rdb1

rdb2 (186MB) - files corresponding to rdb2

rdb3 (135MB) - files corresponding to rdb3

rdb4 (136MB) - files corresponding to rdb4

rdb5 (74MB) - files corresponding to rdb5

rdb6 (356MB) - files corresponding to rdb6

rdb7 (143MB) - files corresponding to rdb7