Dynamic morphable object reconstruction method and system for embodied intelligence three-dimensional perception
By introducing flow matching and joint optimization of objectives, the problems of low computational efficiency and deformation discontinuity in the reconstruction of non-rigid deformable targets by dynamic neural radiation fields are solved, and efficient and stable 3D reconstruction of dynamically deformable targets in embodied intelligent systems is realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INFORMATION SCI & TECH UNIV
- Filing Date
- 2026-06-08
- Publication Date
- 2026-07-17
AI Technical Summary
Existing dynamic neural radiation field methods suffer from low computational efficiency, lack of continuous physical constraints in deformation modeling, spatiotemporal discontinuity, reconstruction distortion, and poor rendering effects when dealing with non-rigid deformation targets, making it difficult to meet the needs of embodied intelligent systems for efficient perception and real-time optimization.
By introducing the concept of flow matching, the continuous migration of sampling points in the observation space to a unified reference space is described by parameterized velocity field. This establishes a continuous correspondence between the states of the dynamic target at different times and constructs a joint optimization objective, including image rendering consistency constraints, continuous motion field constraints, and spatiotemporal structure consistency constraints, to optimize the dynamic implicit radiation representation network.
It improves the continuous deformation modeling capability of complex non-rigid dynamic targets, reduces high-frequency detail loss and temporal flicker, and realizes high-quality, spatiotemporally continuous 3D reconstruction of dynamic scenes, adapting to the needs of embodied intelligent environment perception.
Smart Images
Figure CN122415897A_ABST