PEViD-UHD: Privacy Evaluation Ultra High Definition Video Dataset

Introduction

To understand the implications of using UHD video in video surveillance, a dataset suitable for privacy evaluation is necessary. Existing public UHD datasets are mostly designed for evaluation of video compression and do not carry privacy sensitive information. And datasets dedicated to the evaluation of privacy issues, such as PEViD-HD (mmspg.epfl.ch/pevid-hd), contain at most HD video sequences. 

Dataset in details

The dataset consists of 26 4K UHD video sequences, each 13 seconds long, with a frame resolution of 3840 x 2160 pixels and captured at 30 fps using a Samsung Galaxy Note 3 smartphone. Resized sequences in HD (1920 x 1080 pixels) and SD (720 x 404 pixels) are also provided together with the original 4K UHD sequenes.

Video sequences in the dataset depict several typical surveillance scenarios: walking, exchanging bags, fighting, and stealing, which were shot in outdoor and indoor environments. Participants appearing in the video have various gender and race, they are dressed differently and carry various personal items and accessories. Their silhouettes were manually annotated and the annotations are provided in XML format. All participants have read and signed a consent form, allowing free usage of these video sequences for research purposes.

Here are some screnshot examples of the scenes available in this dataset: 

 

 

 

Subjective experiment and data

To better understand the degree to which an UHD video impacts privacy intrusiveness when compared to lower resolutions, we conducted subjective experiments comparing privacy intrusiveness of UHD to HD and SD video. The sequences were evaluated using the methodology proposed in [1] and extended in [2].

To avoid the memory effect, subjects first viewed all video sequences in SD (the least number of visible details), followed by HD, and finally by UHD. Hence, the test was divided into three separate sessions corresponding to their resolutions. 

A total of 20 naïve subjects took part in the experiments, with 25% being female and overall age ranging from 18 to 27 years old. Test consisted of three sessions of 15 minutes each (SD, HD, and UHD content) with a few minutes breaks in-between. During each session, subjects assessed 20 selected video sequences. Test subjects were then given 30 seconds to answer specific questions about privacy and intelligibility for each video.

All video sequences were played on a 56-inch professional high performance Sony Trimaster SRM-L560 4K/QFHD LCD reference monitor. Viewing conditions for the participating subjects were set according to recommendation ITU-R BT.2022 [3]. 

The resulted subjetive data is also available to download with the dataset.

[1] P. Korshunov, C. Araimo, F. De Simone, C. Velardo, J. Dugelay, and T. Ebrahimi, “Evaluation of visual pri- vacy filters impact on video surveillance intelligibility,” in International Workshop on Quality of Multimedia Ex- perience (QoMEX), Yarra Valley, Australia, July 2012, pp. 150–151.

[2] P. Korshunov, C. Araimo, F. De Simone, C. Velardo, J. Dugelay, and T. Ebrahimi, “Subjective study of pri- vacy filters in video surveillance,” in IEEE International Workshop on Multimedia Signal Processing (MMSP), Banff, AB, Canada, Sept. 2012, pp. 378–382.

[3] ITU-R BT.2022, “General viewing conditions for sub- jective assessment of quality of SDTV and HDTV tele- vision pictures on flat panel displays,” International Telecommunication Union, August 2012.

Download

You can use PEViD-UHD dataset in your research without any conditions, as long as you clearly mention and acknowledge the following paper :

P. Korshunov, and T. Ebrahimi. UHD Video Dataset for Evaluation of Privacy. Sixth International Workshop on Quality of Multimedia Experience (QoMEX 2014), Singapore, 2014.

Bibtex file:
@InProceedings{KorUHD2014,
    author    = {Pavel Korshunov and Touradj Ebrahimi},
    title     = {UHD Video Dataset for Evaluation of Privacy},
    booktitle = {Sixth International Workshop on Quality of Multimedia Experience (QoMEX 2014)},
    address  = {Singapore},
    year      = {2014},
    month    = {18–20 September},
}

You can download all video files, associated XML-based annotations, and the raw data of the subjective experiments from the following FTP (please use dedicated FTP clients, such as FileZilla or FireFTP):

FTP address: tremplin.epfl.ch
User name: PEViD@grebvm2.epfl.ch
Password: m@l>A_@$1(G?Z+;C
 
The dataset is inside folder ‘PEViD-UHD’. Please see the readme.txt file inside the download folder for details about the dataset and the annotation format. You may also check the above paper for some other helpful information.
 
In case of any problems or questions, please send an email to pavel.korshunov (at) epfl.ch

 

Permission is hereby granted, without written agreement and without license or royalty fees, to use, copy, modify, and distribute the data provided and its documentation for research purpose only. The data provided may not be commercially distributed. In no event shall the Ecole Polytechnique Fédérale de Lausanne (EPFL) be liable to any party for direct, indirect, special, incidental, or consequential damages arising out of the use of the data and its documentation. The Ecole Polytechnique Fédérale de Lausanne (EPFL) specifically disclaims any warranties. The data provided hereunder is on an “as is” basis and the Ecole Polytechnique Fédérale de Lausanne (EPFL) has no obligation to provide maintenance, support, updates, enhancements, or modifications.