Character Animation Retargeting With Automatic Reference-Point Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing character animation retargeting techniques face challenges in transferring motion sequences between models with different bodily proportions, leading to misalignments and requiring manual corrections, which are unsuitable for real-time adjustments and result in unnatural motion sequences.
Innovation Solution
A technique that automatically adjusts the location of misaligned reference points in an avatar model by calculating distance vectors between corresponding points in an actor model, using precise and transition zones to maintain relative positional relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If motion sequences are transferred directly between models with different bodily proportions, then the retargeting process is simple and fast, but misalignments occur between body parts (e.g., hands intruding into head or failing to contact cheeks)
Solution Approach 1:
The retargeting process is divided into two distinct phases: a fast retargeting phase that transfers motion sequences between models with different proportions, and a correction phase that adjusts specific misaligned body parts. This segmentation allows the system to achieve both speed and precision by handling alignment issues only where necessary rather than requiring complete manual correction of all body parts.
Solution Approach 2:
Instead of uniformly correcting all body part alignments, the system identifies and corrects only the specific local misalignments that occur during retargeting (such as hands relative to face or body parts). This localized correction approach maintains the overall speed benefit while fixing critical alignment issues, applying precision only where the motion transfer creates problems.
2Manufacturing precision
If manual correction of misalignments is performed, then body part alignment precision is improved, but the process requires user intervention and is not suitable for real-time adjustments
Solution Approach 1:
The system automatically identifies misaligned body parts and applies corrections without requiring user intervention. The correction process is fully automated, with the system analyzing the motion sequence, detecting misalignments, and adjusting body parts automatically. This enables real-time adjustments to avatar models while maintaining precise body part alignment, eliminating the need for manual user correction.
3Device complexity
If motion sequences are restricted to minimize visual effects of misalignments, then the retargeting process is simpler, but the motion sequences appear artificial or unnatural
Solution Approach 1:
The system performs preliminary retargeting of motion sequences between models with different proportions, then automatically identifies and corrects misalignments before the motion is displayed. This preliminary correction action allows natural human-like motions (including hand-to-face contacts) to be transferred without restriction, while the system proactively fixes alignment issues that would otherwise create artificial appearances.
Solution Approach 2:
The system converts the harmful effect of proportion differences (which cause misalignments) into a beneficial automatic correction mechanism. Instead of restricting motion sequences to avoid misalignment problems, the system uses the detected misalignments as signals to automatically adjust body parts, transforming what was previously a limitation into an opportunity for intelligent, real-time correction that preserves natural motion behavior.
Data Source
AI summary
One embodiment of the present invention sets forth a technique for retargeting character animation from a given actor model to a given avatar model. The technique includes receiving a set of correspondence relationships between one or more reference points in the actor model and one or more reference points in the avatar model, determining a misaligned reference point in the avatar model relative to the actor model, and identifying, based on the set of correspondence relationships, a reference point in the actor model corresponding to the misaligned reference point in the avatar model. The technique further includes calculating a distance vector between the identified reference point in the actor model and another reference point in the actor model and adjusting the location of the misaligned reference point in the avatar model based on the distance vector.