<![CDATA[Sign-language learning media based on 3D avatars store sign motions as bone-rotation keyframes and replay them through interpolation. Common web-based systems use linear interpolation of Euler angles (LERP-Euler), which does not follow the shortest rotational path and yields unnatural angular velocity. Because movement is a meaning-bearing phonological parameter in sign language, interpolation quality affects sign legibility. This study objectively compares four methods LERP-Euler, quaternion SLERP, SLERP with smoothstep easing, and Catmull-Rom without human raters, grounding quality on the minimum-jerk and shortest-path principles. Eighty dictionary sign words were tested with an automated harness recording each bone's orientation and the fingertip position at 240 Hz. Eight metrics (Normalized Jerk, SPARC, path-length ratio, geodesic deviation, angular-speed CV, C1 jump, fingertip path length, and computation cost) were analysed using the Friedman test with post-hoc Wilcoxon (Bonferroni). No method dominates all dimensions: Catmull-Rom and eased SLERP are the smoothest (p < 0.001), while SLERP and eased SLERP best follow the shortest path (ratio ? 1.000) at the lowest cost; Catmull-Rom overshoots (ratio ? 1.13) and is costliest. Overall, eased SLERP is recommended as the best compromise for real-time web deployment.]]>
Copyrights © 2026
