3DPES - PEople Surveillance Dataset

The growing interest on People Re-identification for surveillance forensics multimedia search action analysis calls for public and challenging datasets to train and test proposed solutions. Re-identification can exploit appearance features extracted by images but also motion and spatio-temporal features when videos and possibly camera calibration are available. Currently available datasets are not sufficient to cover all the aspects and the requirements of state of the art techniques.

3D-PES would like to give annotation data for people segmentation, people tracking, people 3D reconstruction under calibrated scenarios, people search and re-identification by providing more videos with the same individuals captured from different points of view in real urban scenario.

The new benchmark extends the previously published Visor-SARC3D dataset and now contains:

• More than 600 videos of 200 people from 8 static cameras
• Segmented key-frames for each person
• Reference background images
• Camera calibration data
• Structured annotation on DB: key frames people detections
• 3D scene reconstruction

Related publication:
D. Baltieri R. Vezzani R. Cucchiara, "3DPes: 3D People Dataset for Surveillance and Forensics", in Proceedings of the 1st International ACM Workshop on Multimedia access to 3D Human Objects Scottsdale Arizona USA Nov 28 - Dec 1 2011

This dataset is part of the ViSor project.