Hand motion dataset

Question

I am looking for a dataset of hand motions, e.g. take (with hand), throw (with hand), wave (with hand), etc. I imagine that a hand has a number of sensors attached, say, at an elbow, at a wrist, at the center of a palm, on each finger in four places.

A dataset would have the following format then:

• motion tag, set of frames:

  • frame index, wall-clock time, 3-d coordinate of each sensor (either absolute or relative to the body);

All above is equivalent to the fact that dataset could be visualized using simple 3d application and motions are recognizable in the resulting video.

P.S. It would also be nice to know, if such a dataset could be easily collected.

Answer 1

Take a look at the Ninapro project here: http://ninapro.hevs.ch/node/7

You'll need to create a username and password to access the databases but don't worry, you never get any annoying emails, this is totally a research-driven project to share data and ideas.

They have several databases full of time series, labelled datasets created during hand movement experiments. The data consists of EMG, accelerometer, gyroscope, and other IMU device data. The accelerometer data consists of X, Y, Z values for each sensor. Databases 2, 3, and 7 definitely contain accelerometer data if you think that might be useful.

Each database has details on the experiment performed, a description of the dataset contents, and a link to the research paper the data relates to. I use MATLAB to manipulate the datasets and then feed them into whatever classifier I'm working with.

Answer 2

It sounds like you are describing motion-capture technology (e.g. LOTR Gollum).

In which case, CMU Graphics Lab Motion Capture Database may contain some of the data you want. It’s a collection of various motion capture recordings of well over 100 subjects.

UIC has as Motion Capture Hand Postures Data Set. Smaller set of motion capture data, but perhaps more towards your specific interest.

Or you can grab some hardware to make a dataset with something like the Cyberglove or Wrist + Hand + Finger Straps.