Motion Vector Decoding for Panoramic Video via Polyhedron Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion vector decoding technologies for panoramic videos face challenges due to geometrical distortion when converting three-dimensional panorama images to two-dimensional planar images, leading to inaccurate motion vector prediction and reduced encoding efficiency.
Innovation Solution
A motion vector decoding method that adjusts motion vectors based on the layout information of polyhedrons, including face quantity, arrangement manner, and rotation, to accurately determine the motion vectors of adjacent blocks, thereby improving decoding efficiency by accounting for geometrical deformations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the panorama image is converted to a geographic map for ease of encoding, then the encoding process becomes simpler, but severe distortion and data redundancy occur in areas adjacent to the north and south poles
Solution Approach 1:
The patent divides the spherical panorama image into multiple planar face images by mapping to a polyhedron surface. Each face image represents a portion of the panorama without the severe distortion that occurs in geographic map projections, particularly near the poles. This segmentation allows each face to be encoded independently with minimal distortion.
Solution Approach 2:
The patent transitions from a two-dimensional geographic map projection to a three-dimensional polyhedron-based representation. By mapping the spherical panorama to the surface of a polyhedron and then unfolding it into multiple planar faces, the system preserves geometric accuracy while maintaining ease of encoding for each individual face.
2Ease of operation
If motion vector prediction is performed using adjacent blocks in the distorted geographic map, then the decoding process is straightforward, but the motion vector prediction accuracy deteriorates due to geometrical distortion
Solution Approach 1:
The patent segments the distorted panorama into multiple undistorted planar face images. Motion vector prediction is then performed within each face image where geometric relationships are preserved, ensuring accurate motion estimation. The segmentation allows standard motion compensation techniques to work effectively without the interference of projection distortion.
3Manufacturing precision
If the panorama image is mapped to a polyhedron surface to avoid distortion, then image quality is improved, but the encoding and decoding complexity increases due to multiple geometric transformations
Solution Approach 1:
The patent performs the polyhedron mapping and face image generation as a preliminary step during encoding. The layout information describing the polyhedron structure and face arrangements is pre-calculated and stored. During decoding, this pre-prepared layout information guides the reconstruction process, eliminating the need for complex real-time geometric transformations and reducing decoding complexity.
4Adaptability or versatility
If multiple geometric transformations are applied to obtain the two-dimensional planar image from the polyhedron, then the panorama image can be represented, but the correlation between motion vectors of adjacent blocks is reduced
Solution Approach 1:
The patent segments the polyhedron surface into multiple planar face images, each representing a local region of the panorama. Adjacent blocks within the same face maintain strong geometric and photometric correlations, ensuring reliable motion vector prediction. The segmentation prevents the correlation degradation that would occur if entire distorted regions were treated as single units.
Data Source
Figure 1a~2
Figure 3m~4
Figure 5~7
AI summary
Embodiments of the present invention disclose a motion vector decoding method and a decoder. The method includes: determining a prediction motion vector of a to-be-decoded unit based on a motion vector of a prediction unit of the to-be-decoded unit; when a face image in which the to-be-decoded unit is located and at least one of a face image in which a first reference unit is located and a face image in which the prediction unit is located are not face images in a same orientation, performing a first update on the prediction motion vector, where the first update is used to determine a mapping vector that is of the prediction motion vector and that is in a plane of the face image in which the to-be-decoded unit is located; and obtaining a motion vector of the to-be-decoded unit based on the prediction motion vector obtained after the first update. Based on the embodiments of the present invention, encoding efficiency can be improved.