We introduce I see you a vehicle-pedestrian trajectory dataset for pedestrian trajectory prediction in near-miss situations, autonomous driving safety analysis, and pedestrian safety hazard evaluation. Please cite if it is useful in your research.
@inproceedings{quispe_i_2022,
title = {I see you: {A} {Vehicle}-{Pedestrian} {Interaction} {Dataset} from {Traffic} {Surveillance} {Cameras}},
url = {https://research.latinxinai.org/papers/neurips/2022/pdf/19_CameraReady.pdf},
doi = {10.52591/lxai2022112811},
language = {en},
booktitle = {{LatinX} in {AI} at {Neural} {Information} {Processing} {Systems} {Conference} 2022},
publisher = {Journal of LatinX in AI Research},
author = {Quispe, Hanan and Sumire, Jorshinno and Condori, Patricia and Alvarez, Edwin and Vera, Harley},
year = {2022},
}We collected vehicle and pedestrian trajectories in near-accident scenarios from surveillance cameras at signalized intersections.
For each interaction we provide:
- Processed vehicle and pedestrian trajectories in GPS coordinates.
- Vehicle-pedestrian ids
- Frame
The trajectories are provided in the following format
| clip | id | frame | latitude | longitude |
Pedestrian trajectories can be found here and vehicle trajectories here. Time is expressed as frame(FPS=30).
You can visualize each interaction trajectories using Google Maps. The individual GPS trajectory files are available here. Files with _ped in their names correspond to pedestrian trajectories, while those with _veh correspond to vehicle trajectories.
Bellow are the speed distributions for our dataset as well as number of interactions by type of interaction(refer to our paper for more information).
| Scenario | Number of Occurrences |
| Dangerous | 91 |
| Non-Dangerous | 79 |