Kinect Skeleton Dataset for Fall Detection

Data was recorded with a Kinect v1 sensor in a furnished bedroom and kitchen in an apartment in Braganca, Portugal, in March 2025, and with an Azure Kinect sensor in a laboratory room at the Université Marie et Louis Pasteur, in Besancon, France, between June and July 2026, following approval from the IPB and UMLP Ethics Committees, respectively.

Each participant performed 20 scripted activities, covering 15 fall types (backward, forward, and lateral falls, and syncope) and five activities of daily living that share kinematic features with falls, such as sitting down or lying in bed. Each CSV file corresponds to one participant performing one activity and contains, for every frame, the timestamp, the floor plane equation, and the 3D coordinates and tracking status