Hand Gestures Database

This is a set of hand gestures images captured by Kinect device. It includes a set of images for training and testing machine learning techniques. It also includes a database of gestures transitions.

Download »

Disclaimer

The LNCC HGR Database is available for research purposes only. Permission to use but not distribute or reproduce this database is granted to all researchers.

Description

This database is divided in three parts (train, test and transitions) and it is composed by segmented hand images. Both hands were used to the database construction and combines depth data and infrared laser speckle pattern (ILSP) images, captured from a Kinect device.

The training set is composed by 6841 ILSP segmented images like the ones pictured in the figure below which define the classes for training. Each image pictured on this figure represents one class or a static gesture, named: “C”, “Closed”, “V”, “Opened” e “Pointer”. However, these classes have the pictured gesture made with both of the hands. The test set is composed by 21336 ILSP images.

To generate the database images, 15 users were placed in the CAVE system of the LNCC, one at a time, and asked to perform the hand gestures of figure below. The video stream is captured at 30 fps and the depth and infrared images are processed to segment the ILSP image. A 64 × 64 image containing each one of the hands is recorded from each video image. Each image is centered in the corresponding hand position. The participants are asked to change the hand posture and position during the acquisition process to allow recording in different orientations and scales. In this way, we get a more representative sampling of the space of ILSP hand images.

Classes

In real-time applications it is important to use hand gestures transitions to navigate and manipulate virtual objects in a virtual worlds. Hand gesture transition means that a user is using a sequence of two gestures to make an action. For example, to pick up a virtual object, the user must first perform the gesture "Opened" and in a second time he needs to close his hand to represent the "Closed" gesture. Our transitions database consists of 439 samples generated by 15 participants. Users are asked to change the hand posture inside the CAVE, from one static gesture to another one.

1. “Opened” → “Closed” (74 transitions)

2. “Closed” → “Opened” (73 transitions)

3. “V” → “Closed” (76 transitions)

4. “V” → “Pointer to right” or “‘Pointer to left” (68 transitions at all)

5. “Pointer to right” or “‘Pointer to left” → “Opened” (74 transitions at all)

6. “C” → “Pointer to right” or “‘Pointer to left” (74 transitions at all)

Contact

Prof. Gilson A. Giraldi

Email: gilson@lncc.br