MAD database
The MAD database was recorded using a Microsoft Kinect sensor in an indoor environment. The database contains:
The multimodal activity of 20 subjects recoded with the Microsoft Kinect camera. The modalities include: RGB video (240*320), 3D depth (240*320), and skeleton (3D coordinates of 20 joints per frame).
The activity of 20 subjects performing 35 sequential actions (see actionlist). Each subject repeats the set of 35 actions twice. Each video is about (4000-7000 frames) (See actionlist).
Ground truth labels: start and end of the action has been labeled, suitable for detection (also for classification).
Dong Huang, Yi Wang*, Shitong Yao* and F. De la Torre. Sequential Max-Margin Event Detectors, ECCV 2014