Deaf and Hard of Hearing

The study aims to automatically apply anonymization to the German Sign Language dataset, DGS Korpus. Multi-camera video recordings of conversations in German sign language were collected, annotated, and translated.

Authors capture eye tracking and motion capture data to explore how the dominant hand, the head, the chest and the eyes are used to sign short Finnish Sign Language stories. MoCap and eye tracking data is collected for 7 stories signed in Finnish sign language. ELAN is used to align the data.

A 3D video dataset is collected to explore sign language recognition using linguistic subunits. Each participant performed 20 signs 5 times, this was recorded using 3D Kinect sensors and the skeleton tracking data was collected.

A Russian sign language corpus is collected that focussed on a format, which used handshape, location, hand trajectory and hand orientation as the key features to be captured. Each sign was repeated to build a machine learning friendly dataset. Kinect 2.0 camera recorded the gestures was used to collect the corpus that was aligned and annotated afterwards.

A MoCap dataset of facial expressions and hand configurations for French sign language is collected and will be used for animating a signing avatar. Different words and verbs are collected with Motion Capture hardware worn by signers, 1 signer each for 3 datasets that focussed on different sets of words.

A 3D video dataset is collected to explore sign language recognition using linguistic subunits. Each participant performed 40 signs 5 times, this was recorded using 3D Kinect sensors and the skeleton tracking data was collected.

A dataset of 3D depth video and skeleton data is collected for Chinese sign language. 4 signers performed 26 alphabets and 10 number signs twice each and 4 signers performed them once, this was recorded using a 3D Kinect sensor.

This dataset is introduced for continuous sign language recognition with sentence-level annotations. A Microsoft Kinect camera is used for all recording, providing RGB, depth and body joints modalities in all videos.

The work aims to compile a representative corpus of SSL and then transcribe and annotate the video data to get the information needed for describing the lexicon and grammar of the language. Data from 72 informants have been collected from different Slovenian regions and thorugh the support of deaf clubs. Recording sessions include daily life conversations and video elicited conversations.

A video dataset is proposed for automatic recognition of American sign language. High resolution cameras and different lighting conditions were used to record videos of motion primitives and shapes, and of subjects signed two or more sentences including prosody patterns.