ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Pointing Gesture Recognition via Self-supervised Regularization for ASD Screening
Cited 3 time in scopus Download 90 time Share share facebook twitter linkedin kakaostory
Authors
Cheol-Hwan Yoo, Jang-Hee Yoo, Ho-Won Kim, ByungOk Han
Issue Date
2023-10
Citation
International Conference on Computer Vision Workshops (ICCVW) 2023, pp.3036-3043
Language
English
Type
Conference Paper
DOI
https://dx.doi.org/10.1109/ICCVW60793.2023.00326
Abstract
The ability to point to objects for sharing social purpose or attention is known as one of the key indicators in distinguishing children with typical development (TD) from those with autism spectrum disorder (ASD). However, there is a lack of datasets specifically tailored for children’s pointing gestures. This lack of training data from the target domain becomes a major factor in the performance degradation of conventional supervised CNNs due to domain shift. Toward an effective and practical solution, we propose an end-to-end learning scheme for domain generalized pointing gesture recognition adopting self-supervised regularization (SSR). To prove the effectiveness of our method in real-world situations, we designed a Social Interaction-Inducing Content (SIIC)-based ASD diagnostic system and collected an ASD-Pointing dataset consisting of 40 TD and ASD children. Through extensive experiments on our collected datasets, we achieved an ASD screening accuracy of 72.5%, showing that pointing ability can play a vital role as an indicator in distinguishing between ASD and TD.
KSP Keywords
ASD children, End to End(E2E), End-to-end learning, Key indicators, Pointing gestures, Real-world, Target domain, autism spectrum disorder, diagnostic system, gesture recognition, performance degradation