12 May 2026

Photo: UHH/IDGS

We present SignCollect, an automated workflow for capturing, processing, annotating, and publishing sign language datasets.

Recording happens in two stages at separate locations. First, glosses, sentences, or texts are captured in a five-camera studio with QR code based frame-level synchronization, and the raw footage is then automatically converted into post-processed video suitable for the general public. The video is annotated through our web-based tool with SignBank integration.

In a separate stage of the workflow, selected items are re-recorded in 3D in a 21-camera Vicon motion capture studio.

Datasets are released under CC BY-NC 4.0 through Figshare for the research community. Motion capture is published either as raw marker data or in retargeted form on the SignLab avatar in male or female variants, and video is published either as raw or post-processed material. The data is made accessible to the Deaf and Hard-of-Hearing community through the Signio application.

We illustrate the workflow with three projects that have used it or using it now: Zin in NGT (Josje Ritmeester - 4500 sentences), BAK (3000 signs), and SignBeach (Scheifner et al. - 5600 videos based on 1400 signs).

The pipeline is meant to be picked up by other sign language research groups, and it addresses recurring practical problems around synchronization, throughput, and getting data out of the lab and into use.