Motion-Based Analysis of Personalization and Kinematic Features in Japanese Sign Language Video Data

Recent advancements in sign language technology have sparked interest in automatically distinguishing individual signers based on kinematic features.While it is widely recognized that individuality exists in sign languages as it does in spoken languages, few studies have investigated this inter-signer variability.This study examined the extent to which Japanese Sign Language motion data captures personal traits that can inform signer identification.We extracted 2D keypoint-based kinematic features from video data and evaluated them across multiple classification models, using feature selection to isolate the most influential cues.Results indicate that even without explicit depth information, 2D kinematic features provide sufficient individuality to distinguish signers, highlighting the feasibility of personalized signer recognition.Furthermore, these findings underscore the necessity for developing anonymization technologies that can mitigate privacy risks by preventing the unintended leakage of personally identifiable information inherent in sign language motion data.