Joint Attention in Autonomous Driving (JAAD) Dataset

Download Paper

JAAD is a dataset for studying joint attention in the context of autonomous driving. The focus is on pedestrian and driver behaviors at the point of crossing and factors that influence them. To this end, JAAD dataset provides a richly annotated collection of 346 short video clips (5-10 sec long) extracted from over 240 hours of driving footage. These videos filmed in several locations in North America and Eastern Europe represent scenes typical for everyday urban driving in various weather conditions.
Bounding boxes with occlusion tags are provided for all pedestrians making this dataset suitable for pedestrian detection.
Behavior annotations specify behaviors for pedestrians that interact with or require attention of the driver. For each video there are several tags (weather, locations, etc.) and timestamped behavior labels from a fixed list (e.g. stopped, walking, looking, etc.). In addition, a list of demographic attributes is provided for each pedestrian (e.g. age, gender, direction of motion, etc.) as well as a list of visible traffic scene elements (e.g. stop sign, traffic signal, etc.) for each frame.

JAAD Features


Video Footage

346 HD video clips 5-10 seconds long are recorded with on-board camera at 30 FPS.

Spatial Annotations

Bounding boxes are provided for all 2793 pedestrians present in the videos.

Behavioral Annotations

Behavior tags are provided for each pedestrian per frame, including actions like walking, standing, crossing, looking (at the traffic), etc.

Pedestrian Attributes

A list of attributes is provided for each pedestrian. Attributes include age, gender, clothing and accessories, direction of motion, crossing location, number of people in the group.

Context Annotations

Each video is annotated with weather and time of day attribute. Text annotations for visible elements of infrastructure such as crosswalks, traffic lights, stop lights, etc., are provided per frame.

Variety of scenarios

Videos were recorded in several locations in North America and Europe under different weather conditions.

JAAD Stats


Total number of frames 82,032
Total number of annotated frames 82,032
Number of pedestrians with behavior annotations 686
Total number of pedestrians 2,786
Number of pedestrian bounding boxes 378,643
Average length of pedestrian track 121 frames

Pedestrian counts
Number of pedestrians who cross the street 495
Number of pedestrians who do not cross the street 191

Download JAAD


Videos

Get MP4 files (3.1GB)


Annotations

Annotations are hosted on our github page


Cite us

If you found our dataset useful in your research please consider citing our papers:

@inproceedings{rasouli2017ICCVW,
title={Are they going to cross? A benchmark dataset and baseline for pedestrian crosswalk behavior},
author={Rasouli, Amir and Kotseruba, Iuliia and Tsotsos, John K},
booktitle={Proceedings of the IEEE International Conference on Computer Vision Workshops},
pages={206--213},
year={2017}
}

@inproceedings{Rasouli2017IV,
title={Agreeing to cross: How drivers and pedestrians communicate},
author={Rasouli, Amir and Kotseruba, Iuliia and Tsotsos, John K},
booktitle={IEEE Intelligent Vehicles Symposium (IV)},
pages={264--269},
year={2017}
}

With questions regarding the dataset please contact Amir Rasouli (aras@eecs.yorku.ca) and Iuliia Kotseruba (yulia_k@eecs.yorku.ca).

The videos and annotations are released under the MIT License.