Deaf and Hard of Hearing

This work presents the sign language corpus sourced from the Air Travel Information System (ATIS) dataset, which aims to support machine translation and automatic sign language recognition. 595 sentences were chosen from the ATIS datasets as a base and translated into Irish sign language (ISL), German sign language (DGS1) and South African sign language (SASL).

A dataset of ASL phrases obtained from five deaf children playing an educational game called CopyCat was collected. Each child wearing colored gloves with 3-axis accelerometers attached and seated facing a camera was asked to sign the appropriate phrase given scenarios presented on the game.

The KETI sign language dataset is constructed to develop a neural network model for Korean sign language translation. The dataset consists of 14,672 full high definition videos that captured 524 different signs, potentially used in various emergency situations, and were recorded at 30 frames per second and from two camera angles: front and side.

A dataset of 3D video streams and hand tracking of ASL signers is collected and annotated. A predefined set of phrases are communicated in ASL, while a kinect sensor is used to collect 3D data along with accelerometer and marker based hand tracking system.

A video dataset is collected to improve Polish sign language recognition. Two signers performed 101 words and 35 sentences signs 20 times, this was recorded using a video camera.

The Czech sign language corpus is intended for training and testing of sign language recognition systems. A dataset of video streams and hand tracking of sign language numbers, days, month names, alphabets, town names and selected signs is collected.

A dataset of facial expressions, videos and lip reading, for capturing Arabic sign language and non-manual signs is collected. 10 signers under controlled lighting conditions contributed to the image and video dataset that contains hand shapes, Arabic finger spelling alphabets, numbers, indivdual signs and signs in continuous sentences, lip movement, and facial expressions.

A MoCap dataset of body movements for French sign language is collected and will be used for animating a signing avatar. Dataset is evaluated by people, including native signers. 2 signers made guestures for 4 subsets of words and actions which were captured by Motion capture system and annotated using ELAN software.

A dataset of video streams of American sign language sentences is collected. Multiple RGB camera captured signs as speakers made them.