3D Rendering Engine for Ground Segmentation Jitter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods for augmented reality applications suffer from jitter at the junctions of ground regions and other areas, resulting in poor segmentation effects due to differences in real-world environments captured in image frames.
Innovation Solution
A method and device that utilize a 3D rendering engine to map initial segmentation results and accumulation maps to a 3D virtual scene, aligning plane models and cameras to determine a ground segmentation result, which improves the stability and coherence of segmentation by referencing previous image segmentation results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional image segmentation methods are used for augmented reality applications, then the segmentation process is simple and fast, but jitter occurs at the junctions of ground regions and other areas resulting in poor segmentation effects
Solution Approach 1:
The patent transitions from 2D image plane segmentation to 3D virtual scene segmentation by constructing a 3D virtual scene with ground planes and buildings. The segmentation is performed in this elevated third dimension and then projected back to the 2D image plane, eliminating jitter at junctions by leveraging the additional spatial dimension.
Solution Approach 2:
The patent introduces a 3D virtual scene as an intermediary between the input image and the final segmentation result. This virtual scene acts as a mediator that reconciles the conflicting requirements by providing a stable geometric framework that eliminates jitter while maintaining segmentation accuracy.
2Reliability
If a 3D rendering engine is used to map segmentation results to a virtual scene, then jitter is reduced and segmentation quality is improved, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary construction of the 3D virtual scene including ground planes and building models before executing the segmentation task. By pre-establishing this stable geometric framework, the actual segmentation process benefits from reduced computational complexity and faster processing while maintaining high stability and eliminating jitter.
Solution Approach 2:
The patent divides the complex segmentation task into multiple independent components: ground region segmentation, building region segmentation, and sky region segmentation. Each component is processed separately in the 3D virtual scene and then integrated, reducing overall computational complexity and processing time while maintaining segmentation stability.
3Productivity
If segmentation is performed without considering 3D spatial relationships, then the processing is faster and simpler, but the segmentation results show jitter at frame boundaries
Solution Approach 1:
The patent resolves the contradiction between processing speed and frame-to-frame consistency by moving the segmentation process to a 3D virtual scene. This dimensional elevation provides inherent spatial relationships that ensure consistent segmentation across frames while maintaining processing efficiency through the structured 3D environment.
Solution Approach 2:
The patent changes the spatial parameters of the segmentation process by introducing 3D coordinates and transformations. By working in three-dimensional space with defined ground planes and building models, the system achieves stable frame-to-frame consistency while maintaining reasonable processing speeds through efficient 3D rendering techniques.
Data Source
Figure 1(A)~1(C)
Figure 2
Figure 3
AI summary
The present disclosure discloses a method and device for segmenting an image, and a storage medium. The method includes: acquiring the image; obtaining an initial segmentation result by segmenting a target ground region of the image; and determining a ground segmentation result of the image based on the initial segmentation result and a current accumulation map, where the current accumulation map includes a ground segmentation result of previous image.