Pedestrian Scenario Dataset

This file represents a sample of the Pedestrian Scenario Dataset.

The Pedestrian Scenario Dataset was derived from the video footage from the Caltech Pedestrian Dataset and Joint Attention for Autonomous Driving Dataset. The videos from these datasets were all viewed, and then labelled manually as per the labels in the labels readme file. The raw images were translated into human poses which have been provided in this sample.

In total there are 171 individual pedestrian sequences provided in the form of extracted human pose. This represents 20% of the total commercial dataset.
There are also 171 .json label files which correspond to the sequences.
There is another .txt file which explains the classes in each label for each sequence.

The pose estimation used to extract the pose in this dataset is Alpha Pose.
The pose included in this dataset will be updated from time to time.

This dataset comes with several advantages, mainly being the size of the simplified pedestrian crossing data, along with the rich labels which offer far more insight into the actions and behaviour of each pedestrian.
By using simplified data in the form of human pose, it makes it significantly easier to apply machine learning and extract knowledge from the pedestrian movements.

For more information on the dataset and the uses of this dataset, please see the associated paper at

For citiation of this dataset, please use:
AUTHOR = {Spooner, James and Palade, Vasile and Cheah, Madeline and Kanarachos, Stratis and Daneshkhah, Alireza},
TITLE = {Generation of Pedestrian Crossing Scenarios Using Ped-Cross Generative Adversarial Network},
JOURNAL = {Applied Sciences},
VOLUME = {11},
YEAR = {2021},
NUMBER = {2},
URL = {},
ISSN = {2076-3417},
DOI = {10.3390/app11020471}

For access to the full dataset, please contact [email protected]