Data-driven Inverse Kinematics using Laban Movement Analysis - ScienceDirect
Fig. 1. The two-layer workflow of the proposed approach: first layer: filling the gaps between sparse input, second layer: mapping end-effector positions to human motion. Fig. 2. The proposed CVAE architecture, designed to recreate dense end-effector sequences with conditions of initial and final end-effector positions with LMA-based style descriptors. Fig. 3. The proposed Synthesizer architecture, designed to synthesize full-body motion using given dense end-effector positions and LMA-based style descriptors. Fig. 4. Summary of the effect of defined style descriptors on final motion generation. (a). . (b). . Fig. 5. (a) Visual breakdown of the skeletal structure and (b) comparison of the generated low (blue) and high (red) 𝒱 style descriptor poses. Fig. 6. Comparison of the generated low (blue) and high (red) ℋ style descriptor poses. Fig. 7. Comparison of the generated low (blue) and high (red) 𝒫 style descriptor poses. Fig. 8. Comparison of the generated low (blue) and high (re