WorkoutSU-10 Exercise Dataset

Team: Farhood Negin, Fırat Özdemir, Ceyhun Burak Akgül, Kamer Ali Yüksel, Aytül Erçil
Organizations: Sabancı University VPA Lab (SU-VPA), Vistek ISRA Vision (VIV)
Web: http://vpa.sabanciuniv.edu/downloads/WorkoutSU-10/

Download:

• The minimal dataset may be freely downloaded from here (84 MB).
• The full database (142 GB) may be obtained by application from the contact person above.

The WorkoutSU-10 exercise dataset comprises a collection of sequences of human body movements represented by 3D positions of skeletal joints. The dataset was collected at Sabancı University in Istanbul as part of the research project ViPSafe supported partially by TÜBITAK (agreement 109E134).

If you use this dataset, the following paper should be cited in any resulting publication:

The dataset comprises of 1500 sequences in total, collected from 15 people performing 10 different exercises. The motion files contain tracks of 20 joints estimated using the MS Kinect SDK. The body pose is captured at a sample rate of 30Hz.

The participants were recruited at the Computer Vision Laboratory at Sabancı University (VPALAB). All participants filled a consent form. Each recording session has taken 50 to 90 minutes. Some of the participants were already familiar with the domain of machine learning and computer vision. Relevant statistics of the participants are listed below:

·         73% male,

·         94% right-handed,

·         153-191cm tall with an average height of 179 cm,

·         20 to 30 years old with an average of 24 years of age.

• [2013] - Farhood Negin, Firat Ozdemir, Ceyhun Burak Akgul, Kamer Ali Yuksel, Aytul Ercil, "A Decision Forest Based Feature Selection Framework for Action Recognition from RGB-Depth Cameras," ICIAR 2013.