AutomatumData/automatum_full_highway_drone_dataset

Automatum Data: Overview and Getting Started

Introduction

The Automatum Data datasets contain high-precision movement data of traffic participants (vehicles, trucks, vans) extracted from drone recordings (bird's eye view). They are excellently suited for the simulation, validation, and development of algorithms for automated driving (e.g., Level 3 functions) as well as for traffic analysis purposes.

Research Use and Extended Data Pool

Please note that these publicly available datasets are intended exclusively for research purposes.

These datasets represent just a small excerpt from the comprehensive Automatum Data Pool, which contains over 1,000 hours of processed videos. For commercial use or if you are interested in accessing further datasets, please send us a request via our website:
👉 automatum-data.com/de

Note: The complete data is available as ZIP archives. We have also provided a Sample_Data directory so you can explore the structure directly here on Hugging Face before downloading the larger ZIP files.

Available Scenarios

Our open-source data covers various traffic scenarios:

Data Structure

Each dataset represents a recording of about 10 to 18 minutes in length at a specific location. The structure of a recording usually includes:

• dynamicWorld.json: Contains the dynamic behavior of all objects (trajectories, velocities, accelerations, bounding boxes, lane assignments).
• staticWorld.xodr: Contains the static road geometry in OpenDRIVE format.
• recording_name.html: An overview of the recording with basic metadata.

Installation of the Python Library

To work with the data, Automatum Data provides an open-source library. Installation is done via pip:

pip install openautomatumdronedata
# To upgrade:
pip install --upgrade openautomatumdronedata

Getting Started / Using the Data

Here is a simple example to load a dataset and access the dynamic objects:

from openautomatumdronedata.dataset import droneDataset
import os

# Specify the path to the recording folder
path_to_dataset = os.path.abspath("PATH_TO_FOLDER_E.G._Highway-A9-Allersberg")
dataset = droneDataset(path_to_dataset)

# Access the dynamic world
dynWorld = dataset.dynWorld

# Retrieve general metadata
print("UUID:", dynWorld.UUID)
print("Duration:", dynWorld.maxTime, "seconds")
print("Number of frames:", dynWorld.frame_count)

# Number of all objects in the dataset
print("Number of vehicles:", len(dynWorld))

# Retrieve all objects visible at second 1.0
objects_at_1s = dynWorld.get_list_of_dynamic_objects_for_specific_time(1.0)
if len(objects_at_1s) > 0:
    obj = objects_at_1s[0]
    print(f"Vehicle {obj.UUID} - Type: {obj.type}")
    print(f"X-Positions: {obj.x_vec[:5]}")

Important Object Properties (dynObject):

• x_vec, y_vec: X and Y positions over time.
• vx_vec, vy_vec: Velocities.
• ax_vec, ay_vec: Accelerations.
• lane_id_vec: Assignment to lanes over time.
• length, width: Dimensions of the object.

Advanced Analysis

The library allows for complex analysis, such as:

• Detecting lane changes: By evaluating changes in lane_id_vec.
• Visualizing traffic density: Generating heatmaps using the X/Y coordinates.
• Identifying critical driving maneuvers: Finding vehicles with high longitudinal or lateral acceleration (ax_vec, ay_vec).
• Object relationships (TTC, TTH): Calculating distances and critical times to leading or adjacent vehicles.