The inD Dataset: A Drone Dataset of Naturalistic Road User Trajectories at German Intersections

Bock, Julian; Krajewski, Robert; Moers, Tobias; Runde, Steffen; Vater, Lennart; Eckstein, Lutz · 2020 · OpenAlex-citations

DOI: 10.1109/iv47402.2020.9304839

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper introduces the inD (intersection Drone) dataset, a large-scale collection of naturalistic road user trajectories recorded at German urban intersections. The research addresses a critical gap in automated driving development: the lack of high-quality, public datasets containing diverse urban traffic scenarios. While existing datasets often focus on highways, university campuses, or limited pedestrian interactions, they fail to provide the comprehensive, naturalistic data required for training prediction models and validating safety systems for complex intersection environments. The authors argue that ground-level sensors suffer from occlusion and observer effects, whereas drones offer a superior bird’s-eye perspective that captures natural behavior with minimal interference. To create the dataset, the authors employed camera-equipped drones to record traffic at four distinct unsignalized intersections in Aachen, Germany. The recordings, totaling approximately 10 hours, were captured using a DJI Phantom 4 Pro in 4K resolution at 25 frames per second. The processing pipeline utilized deep learning algorithms for automated detection and tracking. Specifically, the authors used semantic segmentation based on U-Net architectures, training separate networks for small and large objects to handle the significant size disparity between pedestrians and vehicles. This approach allowed for pixel-accurate extraction of trajectories, surpassing the precision of bounding-box methods. Post-processing involved Bayesian smoothing to refine positions, speeds, and accelerations, converting image coordinates into metric coordinates for each site. The resulting inD dataset contains trajectories for more than 11,500 road users, including cars, trucks, buses, pedestrians, and bicyclists. It features a diverse mix of infrastructure layouts, traffic rules, and interaction types across the four recording sites. The dataset achieves a positioning error of less than 0.1 meters, significantly higher accuracy than comparable datasets like the Stanford Drone Dataset, which relies on lower-resolution imagery and bounding boxes. Unlike other public datasets that are limited to specific road user types or controlled experiments, inD provides a representative distribution of all road user types interacting naturally on public roads. The significance of this work lies in providing the first large-scale, high-precision dataset of naturalistic urban intersection trajectories. By offering detailed data on the interactions between vehicles and vulnerable road users, the inD dataset enables robust training and validation of data-driven methods for automated driving, such as behavior prediction and scenario-based safety testing. The authors conclude that the drone-based approach and the resulting dataset surpass existing resources in size, accuracy, and ecological validity, thereby fostering advancements in automated driving research.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• naturalistic crash near crash

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: dataset resource, tool software