Accessible Motion-Capture Glove Calibration Protocol for Recording Sign Language Data from Deaf Subjects
Pengfei Lu, Matt Huenerfauth · 2009 · Proceedings of the 11th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '09) · doi:10.1145/1639642.1639658
Summary
This paper addresses a critical bottleneck in sign language technology research: the calibration of motion-capture gloves used to record handshape data from deaf signers. Motion-capture gloves like the CyberGlove contain sensors that measure finger joint angles, but they require calibration each time they are worn — mapping raw sensor readings to actual joint positions. The manufacturer-provided automatic calibration software is fast but inaccurate, while existing manual calibration approaches are time-consuming, inaccessible to deaf participants (instructions typically assume English literacy), and imprecise. The authors designed a new manual calibration protocol specifically optimized for recording American Sign Language (ASL) handshapes from deaf participants. The protocol was developed iteratively over several months with input from deaf users and includes ASL video instructions, a participant-facing website with images and videos, and structured steps organized to minimize confusion and maximize accuracy. Key design decisions include focusing on one joint type at a time to reduce errors, using ASL handshapes with extreme angles to make gain values easier to judge, employing foam wedges to standardize finger abduction angles, and sequencing calibration steps to avoid compounding errors across related joints.
Key findings
The evaluation involved 5 deaf ASL signers who were calibrated using both the new protocol and the manufacturer's automatic calibration, then asked to perform 20 ASL handshapes each. A native ASL signer judged the correctness of the resulting handshape data on a 1-10 scale. The new protocol received a mean score of 5.92 compared to 4.94 for the automatic calibration — a statistically significant improvement (p=0.026). Using the new protocol, 6% to 35% of handshapes scored above the threshold for being considered non-misidentifiable as another ASL handshape, compared to 6% to 65% with the automatic calibration. The protocol took approximately 49 minutes per participant, including about 9 minutes of ASL demonstration videos and 12 minutes for the automatic calibration baseline. While the automatic calibration alone took only 1-2 minutes, its quality was substantially lower. The judge noted particular difficulty with curved finger handshapes in the automatic calibration results. Participants rated the protocol positively on questionnaires, finding the directions understandable (9.4/10), the process comfortable (8.8/10), and the overall organization satisfactory (10/10).
Relevance
This work has important implications for the broader field of sign language technology, which underpins accessibility applications like automatic sign language recognition and generation of sign language animations. High-quality motion-capture data is foundational — poor calibration means poor training data, which cascades into less accurate recognition systems and less natural animations. The paper's emphasis on making the calibration process accessible to deaf participants is itself a valuable accessibility contribution: too often, research tools designed to benefit deaf users are themselves inaccessible during data collection. The freely available protocol (with ASL video instructions) sets a model for inclusive research methodology. For practitioners working on sign language technology, the specific calibration insights — such as which joints the automatic software handles poorly, how foam wedges improve abduction calibration, and why step sequencing matters — are directly actionable. The work also highlights the gap between fast automated approaches and careful manual calibration, suggesting opportunities for improved semi-automated solutions.
Tags: sign language · motion capture · American Sign Language · deaf users · accessibility technology · data collection · sign language animation · CyberGlove · calibration