Dense Motion Tracking via Deformable Model and Signed Distance Field
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for processing volumetric video content face challenges in tracking motion trajectories due to topology changes and geometric deformations in human performance captures, particularly with adaptive mesh re-sampling being slow and weakening feature localization.
Innovation Solution
A motion tracking mechanism that converts point clouds to a deformable model representation, merging meshes and level sets, allowing for dense motion tracking by deforming and re-sampling the model to add or remove surface sections, and propagating this representation across frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If adaptive mesh re-sampling is used to address geometric changes, then geometric adaptability is improved, but processing speed deteriorates and feature localization precision weakens
Solution Approach 1:
The patent uses a signed distance field (SDF) as a continuous implicit representation that copies geometric information from the point cloud without requiring mesh re-sampling. The SDF captures surface geometry through distance values, allowing the system to adapt to geometric changes while maintaining processing efficiency and avoiding the computational overhead of adaptive mesh re-sampling operations
Solution Approach 2:
The patent transforms the geometric representation from explicit mesh structures to implicit SDF parameters. By representing geometry through distance field parameters rather than mesh vertices and faces, the system achieves geometric adaptability through parameter updates without the computational cost of mesh re-sampling, thus improving processing speed while maintaining adaptability
2Device complexity
If mesh down-sampling is performed to handle geometric changes, then processing complexity is reduced, but feature localization precision deteriorates
Solution Approach 1:
The signed distance field continuously represents geometric information at any resolution without losing detail. Instead of down-sampling meshes which discards feature information, the SDF copies and preserves complete geometric data in implicit form, allowing precise feature localization to be maintained while reducing processing complexity through efficient distance queries
Solution Approach 2:
The SDF representation maintains high geometric fidelity locally by storing precise distance information for each point in space. This allows the system to query local geometric features with high precision without requiring fine mesh resolution, thus maintaining feature localization precision while operating with lower overall processing complexity
3Stability of the object's composition
If mesh smoothing is applied to handle topology changes, then geometric continuity is improved, but correspondence mapping accuracy deteriorates
Solution Approach 1:
The signed distance field continuously captures the exact geometric composition at each time step without smoothing or approximation. By copying the precise distance information from the point cloud to the SDF, the system maintains both geometric continuity for stable representation and high correspondence mapping accuracy through direct distance comparisons between frames
Solution Approach 2:
The SDF is dynamically updated at each time step to reflect the current geometric configuration without requiring smoothing operations. This dynamic updating preserves the true geometric continuity of the object's composition while maintaining accurate correspondence mapping through consistent distance field representations across time steps
Data Source
AI summary
An apparatus to facilitate motion tracking is disclosed. The apparatus includes one or more processors to receive video data including a plurality of point clouds in a time series, convert each of the plurality of point clouds into a deformable model representation and generate an inter-frame mapping between the deformable model representations to track motion between the plurality of point clouds.