Current Directions in Biomedical Engineering · Published 2025-09-01 · DOI 10.1515/cdbme-2025-0180
Hand gesture recognition is an important task in human-machine interaction, enabling intuitive and accessible control methods across various applications, including assistive technologies, virtual reality or sign language translation. The gesture recognition task relies on sensory input from different modalities and in order to be processed, the data needs to be labeled with the corresponding hand gesture classes. This is often done manually which remains a major bottleneck in developing robust gesture recognition systems. This paper presents an Auto-Labeling Pipeline designed to automate the annotation process for multi-modal sensor data, reducing the time and effort required for labeling input data. The proposed system uses data from the Ultraleap Leap Motion Controller, and classifies predefined gesture models based on inter-joint angles and the cosine similarity between their pose vectors. The pipeline proposes a post-processing step to filter misclassifications and enhance gesture recognition reliability. The developed open-source toolbox enables researchers to collect and label gesture datasets efficiently, making hand gesture recognition more accessible and scalable.
Abstract from DOAJ. Public domain (CC0 1.0).
Read the article at the publisher →