README.md

---
license: other
pretty_name: GeoTraj-LC Video Dataset
task_categories:
- object-detection
- video-classification
- other
language:
- en
tags:
- lane-change-detection
- tracking
- time-series-classification
- trajectory-prediction
- highway-surveillance
- vehicle-trajectory
- homography
- intelligent-transportation
gated: true
extra_gated_heading: "Request access to the GeoTraj-LC video dataset"
extra_gated_description: "Access is limited to non-commercial academic research. Requests are manually reviewed by the dataset maintainers."
extra_gated_prompt: "Please provide accurate affiliation and research-use information. Approved users must not redistribute the videos, annotations, download links, or derived identifiable content."
extra_gated_button_content: "Submit access request"
extra_gated_fields:
  Full name: text
  Institution: text
  Institutional email: text
  Position:
    type: select
    options:
    - Student
    - Researcher
    - Faculty
    - Engineer
    - Other
  Country or region: country
  Research purpose:
    type: select
    options:
    - Academic research
    - Reproducibility study
    - Benchmark comparison
    - Education
    - Other
  Project or paper title: text
  Expected use of the dataset: text
  I agree to use the dataset only for non-commercial academic research: checkbox
  I agree not to redistribute the dataset or share download links: checkbox
  I agree not to attempt to identify individuals, vehicles, or sensitive locations from the videos: checkbox
  I agree to cite the associated paper and dataset page in publications: checkbox
---

# GeoTraj-LC Video Dataset

GeoTraj-LC is a fixed-camera highway surveillance dataset for geometry-rectified
vehicle trajectory analysis and online lane-change event detection. The dataset
contains video clips, MOT-style vehicle tracks, lane IDs, manually refined
frame-level lane-change labels, lane maps, homography matrices, and metric scale
factors.

Access to the video data is gated. The dataset is provided only for
non-commercial academic research after manual review by the maintainers.

## Dataset Contents

The release is organized around eight fixed roadside highway scenes:

```text
G0512-K111
G4202-K32
G4215-K12
S2-K152
S2-K188_0
S2-K188_1
S2-K201
S3-K316
```

Each scene contains multiple clips named with the following convention:

```text
{sequence_name}_clip_{clip_number}.txt
```

Example:

```text
G0512-K111_clip_0000.txt
S2-K188_0_clip_0001.txt
```

The expected dataset layout is:

```text
RLC_Dataset/
|-- annotations/
|   |-- events/
|   `-- frame_labels/
|-- maps/
|   |-- centerlines/
|   |-- homography/
|   |-- lane_boundaries/
|   `-- scale_factors.json
|-- tracks/
|   |-- mot_raw/
|   `-- refined/
`-- videos/
```

## Track Format

Refined track files are stored under:

```text
RLC_Dataset/tracks/refined/*.txt
```

Each row follows a MOT-style format with an additional lane ID column:

```text
frame_id, track_id, x1, y1, w, h, score, cls, lane_id
```

The lane-change detector uses the bottom-center point of each vehicle bounding
box as the trajectory point. The point is projected from image coordinates to
the road plane using the scene homography and then converted to metric
coordinates using the sequence-specific scale factor.

## Annotation Format

Frame-level labels are stored under:

```text
RLC_Dataset/annotations/frame_labels/*.txt
```

Each row extends the track format with manually refined lane state and
evaluation labels:

```text
frame_id, track_id, x1, y1, w, h, score, cls, lane_id, gt_state, eval_label
```

The lane-change state is:

```text
0: stable lane keeping
1: left lane change
2: right lane change
```

A complete lane-change event starts when the vehicle first crosses the lane
marking and ends when the vehicle fully enters the target lane. Some clips
contain incomplete lane-change events because vehicles may enter the camera view
while already changing lanes or leave the camera view before the maneuver is
finished.

## Scene-Based Split

The default split is scene-based so that training, validation, and test clips do
not share the same camera scene:

| Split | Scene prefixes |
| --- | --- |
| train | G4202-K32, S2-K188_0, S2-K188_1, S3-K316 |
| val | G4215-K12, S2-K201 |
| test | G0512-K111, S2-K152 |

Dataset statistics:

| Split | Clips | Tracks | Left LC | Right LC | Total LC |
| --- | ---: | ---: | ---: | ---: | ---: |
| train | 38 | 9102 | 482 | 308 | 790 |
| val | 17 | 3126 | 98 | 137 | 235 |
| test | 21 | 10509 | 389 | 351 | 740 |
| total | 76 | 22737 | 969 | 796 | 1765 |

## Scale Factors

The road-plane scale factors convert projected road-plane coordinates to metric
coordinates:

| Scene | Scale factor |
| --- | ---: |
| G0512-K111 | 20.0 |
| G4202-K32 | 9.5 |
| G4215-K12 | 9.17 |
| S2-K152 | 5.17 |
| S2-K188_0 | 3.5 |
| S2-K188_1 | 3.5 |
| S2-K201 | 6.0 |
| S3-K316 | 6.83 |

## Access Conditions

By requesting access, users agree to:

- use the dataset only for non-commercial academic research;
- not redistribute the dataset, videos, annotations, download links, or modified
  copies;
- not attempt to identify individuals, vehicles, institutions, or sensitive
  locations from the videos;
- store the dataset securely and restrict access to approved project members;
- cite the associated paper and dataset page in publications or public reports.

See `DATASET_TERMS.md` and `LICENSE.txt` for the full terms.

## Citation

If you use this dataset, please cite the associated paper and this dataset page.

```bibtex
@misc{geotrajlc_dataset,
  title        = {GeoTraj-LC Video Dataset},
  author       = {REPLACE_WITH_AUTHOR_LIST},
  year         = {2026},
  howpublished = {\url{REPLACE_WITH_DATASET_URL}},
  note         = {Accessed under a non-commercial academic research license}
}
```

## Contact

For dataset access questions, contact:

```text
REPLACE_WITH_CONTACT_NAME
REPLACE_WITH_CONTACT_EMAIL
REPLACE_WITH_INSTITUTION
```