MIVIA Gesture (FELICE Project)

The gesture command dataset facilitates human-robot gesture communication. It consists of videos or sequences of images in which people reproduce the gestures of interest. To increase the representativeness of the dataset, people are framed from different distances from the camera.

Data acquisition:

The dataset includes gesture samples recorded with the camera installed on board the robot. To reproduce the various possible conditions that the method must deal with in the real environment, people that reproduce the gestures are framed at different distances from the camera.

MIVIA Gesture:

Video samples or sequences of JPEG images framing the gestures, organized in subfolders that allow to determine the metadata. At the first level of nesting the folders are divided by subject and at the second level by gesture and distance from the camera. The specific gesture type is encoded in the folder name, while the bounding boxes of the hand(s) are stored in a separate JSON file.