TST Fall detection dataset v2

The dataset is composed by ADLs (Activity of Daily Living) and falls simulated by 11 young actors. The following actions are part of ADL category:

• sit, the actor sits on a chair;
• grasp, the actor walks and grasps an object from the floor;
• walk, the actor walks back and forth;
• lay, the actor lies down on the floor;

The following actions are part of the fall category:

• front, the actor falls from the floor and ends up lying;
• back, the actor falls backward and ends up lying;
• side, the actor falls to the side and ends up lying;
• EUpSit, the actor falls backward and ends up sitting.

Each actor repeated each action 3 times, generating a total number of 264 sequences. For each sequence, the following data are available:

• depth frames, resolution of 512x424, captured by Kinect v2;
• two raw acceleration streams, provided by IMUs constrained to the waist and right wrist of the actor;
• skeleton joints in depth and skeleton space, captured by Microsoft SDK 2.0 (see JointType enumeration for the order: https://msdn.microsoft.com/en-us/library/microsoft.kinect.kinect.jointtype.aspx)
• timing information, timestamps of Kinect frames and acceleration samples, useful for synchronization

This code can be used to read data with MATLAB:

http://www.tlc.dii.univpm.it/dbkinect/FallDetection/MatlabCode.zip

If you use the dataset, please cite the following paper:

S. Gasparrini, E. Cippitelli, E. Gambi, S. Spinsante, J. Wahslen, I. Orhan and T. Lindh, “Proposal and Experimental Evaluation of Fall Detection Solution Based on Wearable and Depth Data Fusion”, ICT Innovations 2015, Springer International Publishing, 2016. 99-108, doi: 10.1007/978-3-319-25733-4_11.