Object-Boundary Depth Estimation for Stable Multi-View Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-view depth estimation methods produce inconsistent and artifact-prone depth information, which affects the quality of synthesized views in immersive video applications.
Innovation Solution
A method for depth estimation that involves segmenting 2D images to identify objects, determining depth values sequentially from object boundaries inward, using iterative processes to refine depth values based on similarity metrics and variance, and adjusting parameters for convergence, while utilizing known camera parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing multi-view depth estimation methods are used, then depth estimation can be performed, but the depth information is inconsistent and contains artifacts
Solution Approach 1:
The patent segments the image into multiple objects based on segmentation results, then processes depth estimation separately for each object. This segmentation approach allows the system to handle depth discontinuities between different objects, reducing artifacts and improving consistency in the final depth map while maintaining measurement precision for each segmented object region.
Solution Approach 2:
The patent applies different depth estimation strategies to different regions of the image based on object segmentation. By treating each object separately and applying local processing rules, the system achieves higher quality depth estimation for each region while maintaining overall consistency, resolving the contradiction between reliability and measurement precision.
2Device complexity
If depth values are determined for all pixels simultaneously, then processing is simpler, but depth discontinuities cause artifacts
Solution Approach 1:
The patent segments the image into distinct objects and processes depth estimation for each object separately. This segmentation enables the system to handle depth discontinuities at object boundaries without propagating errors across the entire image, thereby improving depth continuity and reducing artifacts while maintaining manageable processing complexity through modular object-based processing.
Solution Approach 2:
The patent performs preliminary object segmentation before depth estimation, preparing the image by identifying object boundaries and regions. This preliminary action allows the subsequent depth estimation process to focus on each object independently, handling discontinuities more effectively and producing more reliable depth maps with fewer artifacts.
Data Source
Figure 1~2
Figure 3~4
AI summary
A system and method for depth estimation. Segmentation is performed on a set of 2D images of a scene, each 2D image having a different view of the scene. For at least one object identified in the segmentation, a depth value is determined for each of a plurality of pixels of the object in at least one 2D image in which the object appears by: defining an initial estimation region for the object, the initial estimation region comprising a boundary of the object as identified in the segmentation; and iteratively repeating the steps of determine a depth value for each of a plurality of pixels in a current estimation region and defining an updated estimation region for the object, until a depth value has been determined for a predetermined proportion of the object.