Radial Basis Function Neural Network Pose Interpolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for animating virtual characters in VR, AR, and MR environments face challenges in accurately interpolating between captured poses due to resource constraints and data storage limitations, leading to inefficient computing and potential errors like jitter and popping.
Innovation Solution
The use of radial basis function neural networks (RBFNNs) and other neural networks to separate and manipulate linear and angular parameters of a digital character's pose, enabling more accurate interpolation and reducing computational resources required.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional interpolation methods are used to animate digital characters between captured poses, then the system can handle resource constraints and data storage limitations, but the animation accuracy deteriorates causing errors like jitter and popping
Solution Approach 1:
The patent segments the pose data into linear parameters (position, scale) and angular parameters (rotations). This segmentation allows each parameter type to be interpolated using methods optimized for its mathematical properties, with linear parameters using Euclidean distance and angular parameters using spherical linear interpolation (slerp), thereby resolving the contradiction between computational efficiency and interpolation accuracy
Solution Approach 2:
The patent changes the parameter representation by separating linear and angular components and applying different interpolation strategies to each. Linear parameters are interpolated directly in Euclidean space while angular parameters are interpolated on the rotation manifold, which maintains mathematical consistency and prevents interpolation errors that cause jitter and popping
2Manufacturing precision
If more training poses are captured to improve animation accuracy, then the interpolation quality improves, but resource constraints and data storage limitations are exacerbated
Solution Approach 1:
The patent creates a mathematical model (interpolation function) that copies the essential characteristics of the training poses and uses this model to generate intermediate poses. Instead of storing all possible poses, the system stores a compact set of training poses and uses the interpolation algorithm to synthesize the full range of motion, significantly reducing data storage requirements while maintaining animation accuracy
3Productivity
If traditional interpolation methods are used, then the computational process is simpler, but the computing efficiency deteriorates due to improper handling of angular parameters
Solution Approach 1:
The patent changes the mathematical treatment of parameters by applying appropriate distance metrics and interpolation methods to each parameter type. Linear parameters use standard Euclidean operations while angular parameters use rotation manifold operations, which improves computational efficiency by avoiding the mathematical inconsistencies that arise from treating all parameters uniformly
Data Source
AI summary
Systems and methods are provided for interpolation of disparate inputs. A radial basis function neural network (RBFNN) may be used to interpolate the pose of a digital character. Input parameters to the RBFNN may be separated by data type (e.g. angular vs. linear) and manipulated within the RBFNN by distance functions specific to the data type (e.g. use an angular distance function for the angular input data). A weight may be applied to each distance to compensate for input data representing different variables (e.g. clavicle vs. shoulder). The output parameters of the RBFNN may be a set of independent values, which may be combined into combination values (e.g. representing x, y, z, w angular value in SO(3) space).