GPJATK DATASET – Calibrated and synchronized multi-view video and motion capture dataset for evaluation of gait recognition

=======================
Summary
=======================
GPJATK DATASET – MULTI-VIEW VIDEO AND MOTION CAPTURE DATASET
The GPJATK dataset has been designed for research on vision-based 3D gait recognition. It can also be used for evaluation of the multi-view (where gallery gaits from multiple views are combined to recognize probe gait on a single view) and the cross-view (where probe gait and gallery gait are recorded from two different views) gait recognition algorithms. In addition to problems related to gait recognition, the dataset can also be used for research on algorithms for human motion tracking and articulated pose estimation. The GPJATK dataset is available only for scientific use. 
All documents and papers that use the dataset must acknowledge the use of the dataset by including a citation of the following paper: 
B. Kwolek, A. Michalczuk, T. Krzeszowski, A. Switonski, H. Josinski, and K. Wojciechowski, „Calibrated and synchronized multi-view video and motion capture dataset for evaluation of gait recognition,” Multimedia Tools and Applications, vol. 78, iss. 22, p. 32437–32465, 2019, doi:10.1007/s11042-019-07945-y

=======================
Data description
=======================
The GPJATK dataset contains data captured by 10 mocap cameras and four calibrated and synchronized video cameras. The 3D gait dataset consists of 166 data sequences, that present the gait of 32 people (10 women and 22 men). In 128 data sequences, each of the individuals was dressed in his/her own clothes, in 24 data sequences, 6 of the performers (person #26-#31) changed clothes, and in 14 data sequences, 7 of the performers attending in the recordings had a backpack on his/her back. Each sequence consists of four videos with RGB images with a resolution of 960×540, which were recorded by synchronized and calibrated cameras with 25 frames per second, together with the corresponding MoCap data. The mocap data were registered at 100 Hz by a Vicon system consisting of 10 MX-T40 cameras.
During the recording session, the actor has been requested to walk on the scene of size 6.5 m × 4.2 m along a line joining the cameras C2 and C4 as well as along the diagonal of the scene. In a single recording session, every performer walked from right to left, then from left to right, and afterward on the diagonal from upper-right to bottom-left and from bottom-left to upper-right corner of the scene. Some performers were also asked to attend additional recording sessions, i.e. after changing into another garment, and putting on a backpack.

=======================
Dataset structure
=======================
* Gait_Data - data for gait recognition containing 32 subjects. The data was obtained using both marker-less and marker-based motion capture systems.
    * Markerless motion tracking algorithm - dataset obtained using a markerless motion tracking algorithm
    * MoCap - dataset obtained using the Vicon motion capture system
    Each dataset contains:
        * Arff - motion data after smoothing, normalization, and MPCA in Weka ARFF format
        * AsfAmc - motion data saved in Acclaim ASF/AMC format
        * Csv - motion data saved in CSV format. Each row contains data for one frame and each column represents a different attribute. Unit for angles attributes are degrees and unit for distances are millimeters.
    * Mat - Matlab .mat files
* Sequences - 166 video sequences with 32 subjects. Each sequence consists of 4 video streams and MoCap data. Video is recorded with a frequency of 25 Hz, and MoCap data is recorded at 100 Hz. Both systems are synchronized. 
    Each sequence contains: 
        * Background - sequences with a background in AVI format
        * Calibration - camera calibration data (Tsai model)
        * Edges - images with detected edges
        * Videos - sequences in AVI format
        * MoCap - data from motion capture system in formats: c3d and Acclaim ASF/AMC
        * Silhouettes - images with person silhouettes
* Matlab_scripts - Matlab scripts for generating .arff files 
    It requires scripts:
    * Tensor Toolbox
    * Matlab Toolbox for Multilinear Principal Component Analysis (MPCA) by Haiping LU (https://www.mathworks.com/matlabcentral/fileexchange/26168-multilinear-principal-component-analysis--mpca-?s_tid=prof_contriblnk)
* ListOfSequences.txt - file with information about video sequences (start frames, frames numbers, offsets)
* ActorsData.txt - file with information about recorded persons (age, gender, height, width) 
* GPJATK_Release_Agreement.pdf - GPJATK dataset release agreement which must be accepted to use the database

=======================
Project participants
=======================
Konrad Wojciechowski (Polish-Japanese Academy of Information Technology)
Bogdan Kwolek <bkw@agh.edu.pl> (AGH University of Science and Technology)
Adam Świtoński <adam.switonski@polsl.pl> (Silesian University of Technology)
Tomasz Krzeszowski <tkrzeszo@prz.edu.pl> (Rzeszow University of Technology)
Henryk Josiński <henryk.josinski@polsl.pl> (Silesian University of Technology)
Agnieszka Michalczuk <agnieszka.michalczuk@polsl.pl> (Silesian University of Technology)

=======================
Acknowledgements
=======================
The recordings were made in the years 2012-2014 in the Human Motion Lab (Research and Development Center of the Polish-Japanese Academy of Information Technology) in Bytom as part of the projects: 1) „System with a library of modules for advanced analysis and an interactive synthesis of human motion” co-financed by the European Regional Development Fund under the Innovative Economy Operational Programme – Priority Axis 1; 2) OR00002111 financed by the National Centre for Research and Development (NCBiR).

=======================
Privacy statement
=======================
Data of human subjects is provided in coded form (without personal identifying information and with blurred faces to prevent identification).

=======================
Further information
=======================
For any questions, comments or other issues please contact Tomasz Krzeszowski <tkrzeszo@prz.edu.pl>.