The highD Dataset: A Drone Dataset of Naturalistic Vehicle Trajectories on German Highways for Validation of Highly Automated Driving Systems
DOI: 10.1109/itsc.2018.8569552
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 highD dataset, a large-scale collection of naturalistic vehicle trajectories recorded from an aerial perspective to support the safety validation of highly automated driving (HAD) systems. The authors argue that existing data sources, such as naturalistic driving studies, field operational tests, and infrastructure sensors, suffer from limitations regarding sensor occlusion, lack of naturalistic behavior due to driver awareness, or insufficient coverage of highway scenarios. To address these gaps, the study proposes using camera-equipped drones to capture traffic from a "bird’s eye view," ensuring high longitudinal and lateral accuracy while minimizing influence on driver behavior. The dataset was collected using a DJI Phantom 4 Pro Plus drone hovering next to German highways near Cologne. Recordings were conducted at six different locations during 2017 and 2018, totaling 16.5 hours of footage across 60 sessions. The drone captured 4K video at 25 frames per second, covering road segments approximately 420 meters long. The data processing pipeline involved stabilizing videos, using an adapted U-Net neural network for semantic segmentation to detect vehicles, and applying Rauch-Tung-Striebel smoothing to refine trajectories. Static infrastructure elements were annotated manually, while dynamic maneuvers such as lane changes and vehicle following were extracted using predefined rules. The resulting dataset contains trajectories for 110,000 vehicles, including 90,000 cars and 20,000 trucks, representing a total driven distance of 45,000 km and 5,600 recorded lane changes. The authors evaluate highD against the widely used NGSIM dataset, highlighting significant advantages in quantity, variety, and quality. highD contains nearly twelve times more vehicles than NGSIM and includes a much higher proportion of trucks (23% vs. 3%), providing a more realistic representation of highway traffic. The dataset exhibits greater variety in mean speeds and truck ratios over time, whereas NGSIM lacks high-speed tracks above 75 km/h despite higher speed limits. Furthermore, the authors note that NGSIM trajectories contain physical errors and false collisions that require extensive manual correction, whereas highD’s aerial perspective and post-processing methods yield smoother, more accurate trajectories suitable for immediate use in safety validation and traffic simulation. The significance of this work lies in providing a robust, publicly available resource for validating HAD systems and developing traffic models. By demonstrating the feasibility of drone-based data collection, the paper offers a cost-effective alternative to onboard sensor campaigns that preserves naturalistic driving behavior and ensures comprehensive coverage of all road users without occlusion. The highD dataset is made available online with accompanying software tools, fostering research in automated driving, traffic analysis, and driver behavior modeling.
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-19 |
| archive | success | semantic_scholar | — | — | 6 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-19 |
| chunk | success | chunk | — | — | 1 | 2026-06-19 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-19 |
| promote | success | — | — | — | 1 | 2026-06-19 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-19 |
| verify | partial | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified_with_issues.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
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, validation psychometrics