Accurate and Accessible Motion-Capture Glove Calibration for Sign Language Data Collection
Matt Huenerfauth, Pengfei Lu · 2010 · ACM Transactions on Accessible Computing · doi:10.1145/1838562.1838564
Summary
This paper addresses a critical bottleneck in sign language technology research: the calibration of motion-capture gloves used to record hand movements for ASL animation and recognition systems. Motion-capture data enables the development of sign language animations that can make digital content accessible to deaf users with lower English literacy levels—a significant accessibility application since many deaf adults read at a fourth-grade level due to limited auditory exposure to English during language acquisition. The researchers developed a new calibration protocol for the 22-sensor CyberGlove, addressing deficiencies in existing approaches. The manufacturer's automatic calibration uses only two handshapes and produces inaccurate results, while the manual "advanced" calibration requires 75-90 minutes and lacks structured guidance. Critically, neither approach considers the needs of deaf participants: calibration is a hands-busy process that makes sign language communication impossible, yet existing protocols provide no visual or ASL-accessible instructions. The new protocol takes a median of 32 minutes and includes: step-by-step researcher instructions, a website with photos and ASL video demonstrations for each step, English text alternatives, and careful sequencing that isolates joint calibrations to avoid compounding errors. The protocol addresses all 22 sensors systematically while making the process accessible to deaf research participants—an essential consideration when the data being collected is from native ASL signers.
Key findings
Two evaluation studies demonstrated significant improvements over the standard calibration. In the handshape study, nine deaf ASL signers were calibrated using both methods and performed 20 ASL handshapes. Five native ASL judges rated animations generated from the motion-capture data. Mean judge scores were 4.52 for automatic calibration versus 5.98 for the new protocol (p<0.05). Only 41.26% of handshapes from automatic calibration were rated understandable (scores 6-10), compared to 61.43% with the new protocol. The contextual study measured real-world impact by having 12 deaf ASL signers watch animations of complete ASL stories and answer comprehension questions. Animations from the new calibration protocol received significantly higher ratings for grammatical correctness, understandability, and naturalness of movement. Comprehension question scores showed statistically significant improvement (p<0.05), demonstrating that better glove calibration translates directly to more comprehensible sign language content. Participant feedback on the protocol's accessibility was overwhelmingly positive, with Likert ratings averaging 9.78-10 across all usability dimensions including understanding directions, comfort, and organization. The protocol successfully reduced calibration time while improving accuracy—a rare combination in research methodology improvements.
Relevance
This research exemplifies accessible research methodology: designing studies that can include the very population being served by the technology. The authors recognized that collecting quality sign language data requires deaf participants, yet standard equipment setup procedures excluded or disadvantaged them. The protocol materials (ASL videos, photos, structured steps) are freely available online, enabling other researchers to conduct accessible sign language data collection. For accessibility practitioners, the work demonstrates how upstream technical infrastructure affects downstream accessibility outcomes. Poor motion-capture calibration produces sign language animations that are harder for deaf users to understand—a hidden barrier that would manifest as "the technology doesn't work well" without understanding the root cause. The evaluation methodology is also valuable: using native ASL signers as judges and measuring both isolated handshape quality and contextual comprehension provides a model for assessing sign language technology. The broader implication is that accessibility must be embedded throughout the research process, not just in the final product. Equipment calibration, participant instructions, and experimental protocols all present opportunities for exclusion or inclusion. This paper shows how thoughtful methodology design can simultaneously improve research quality and participant accessibility.
Tags: sign language · American Sign Language · deaf accessibility · motion capture · animation · sign language recognition · assistive technology · research methodology · accessible research