3D View Interpolation via Sphere Grid and Optical Flow

The patent divides the 3D model into multiple grid regions on a sphere grid, where each grid region corresponds to a specific viewing angle range. Instead of storing complete views from all angles, only key grid regions are pre-rendered and stored. When a user requests a view, the system segments the sphere grid to identify the corresponding grid region and generates the view by interpolating between stored key views, thereby reducing storage requirements while maintaining view accuracy.

The patent performs preliminary rendering and storage of views from key grid regions on the sphere grid before user interaction. These pre-computed views are stored in advance, eliminating the need to render all possible views in real-time. When a user interacts with the 3D model, the system quickly retrieves and interpolates between the pre-stored key views to generate the requested angle, balancing preparation work with storage efficiency.

The patent applies partial action by only rendering and processing the specific grid region that corresponds to the user's current viewing angle, rather than rendering the entire 3D model from all angles. The system identifies the target grid region on the sphere grid and performs interpolation only for that region using optical flow information, significantly reducing the computational load while maintaining real-time responsiveness to user interactions.

The patent replaces the traditional mechanical rendering system with an optical flow-based interpolation system. Instead of performing full 3D rendering computations in real-time, the system uses optical flow information to estimate pixel movements between key views and interpolates the target view mathematically. This substitution of computational mechanics dramatically reduces device performance requirements while maintaining response speed.

The patent introduces optical flow information as an intermediary element between the stored key views and the user's requested view. The optical flow data captures pixel movement patterns between adjacent views, enabling accurate interpolation even when only a limited number of key views are stored. This intermediary information allows the system to generate smooth transitions and accurate views at intermediate angles, maintaining user experience while improving storage efficiency.

The patent changes the parameter representation from storing complete image data at all angles to storing key views with associated optical flow parameters. By transforming the storage format to include optical flow information (which describes pixel displacement vectors), the system can reconstruct intermediate views with high accuracy using fewer stored images, thereby improving storage efficiency without sacrificing user experience.