3D Image Synthesis via Depth-Visible Light Region Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image capturing systems face challenges in accurately synthesizing three-dimensional images due to non-coincidence regions between the field of views of visible light and IR cameras, which lack depth or visible light information, affecting the quality and accuracy of the synthesized 3D images.
Innovation Solution
An image processing method and apparatus that acquire a coincidence region between depth and visible light images, remove non-coincidence regions from both images, and generate a three-dimensional image only from the coincidence region, ensuring accurate synthesis by matching feature points and determining the coincidence degree based on camera field views and relative positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If non-coincidence regions are included in image synthesis, then the field of view coverage is improved, but the synthesis accuracy deteriorates due to missing depth or visible light information
Solution Approach 1:
The patent divides the image synthesis process into two distinct parts: coincidence region synthesis and non-coincidence region handling. The coincidence region (where both depth and visible light information exist) is processed with full synthesis accuracy, while non-coincidence regions are handled separately by filling missing data, thus segmenting the problem to maintain high accuracy in the critical overlapping area while still providing complete coverage.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. In the coincidence region, accurate depth-aware synthesis is performed using both depth and visible light information. In non-coincidence regions, a simplified filling strategy is used where missing depth information is replaced with default values or missing visible light information is filled from surrounding areas, thus adapting the synthesis quality to the local availability of data.
2Loss of information
If complete images including non-coincidence regions are synthesized, then the information completeness is improved, but the processing complexity increases due to handling missing information
Solution Approach 1:
The patent segments the synthesis process into distinct stages: first identifying coincidence and non-coincidence regions, then processing coincidence regions with full depth-aware synthesis, and finally handling non-coincidence regions through simpler filling operations. This segmentation reduces overall processing complexity by avoiding complex operations in regions where they are not needed.
Solution Approach 2:
The patent applies a simplified filling strategy for non-coincidence regions rather than attempting full depth-aware synthesis everywhere. Missing information is filled using default values or simple interpolation, which is less computationally intensive than full synthesis, thus reducing processing complexity while still providing complete information coverage.
3Manufacturing precision
If depth information is used in synthesis, then the three-dimensional effect is improved, but the handling of non-coincidence regions becomes more difficult due to missing depth data
Solution Approach 1:
The patent segments the image region into coincidence and non-coincidence areas based on depth information availability. In coincidence regions where depth data exists, full depth-aware synthesis is performed to maximize three-dimensional quality. In non-coincidence regions where depth data is missing, the system switches to a simplified mode using visible light information and default depth values, thus managing the complexity of handling missing depth data while maintaining high quality where possible.
Solution Approach 2:
The patent applies different synthesis strategies to different spatial regions based on local depth information availability. Where depth information is present, high-quality depth-aware synthesis is used. Where depth information is absent, a fallback strategy using visible light intensity and simplified depth estimation is applied, thus adapting the three-dimensional quality to local conditions while managing overall system complexity.
Data Source
AI summary
An image processing method, an image capturing apparatus (100), a computer device (1000) and a non-volatile computer-readable storage medium (200) are provided. The image processing method includes that: a first depth image of a present scene is acquired; a first visible light image of the present scene is acquired; a coincidence region between the first depth image and the first visible light image, a non-coincidence region of the first depth image and a non-coincidence region of the first visible light image are acquired; the non-coincidence region of the first depth image is removed to obtain a second depth image corresponding to the coincidence region; a non-coincidence region of the first visible light image is removed to obtain a second visible light image corresponding to the coincidence region; and a three-dimensional image is synthesized according to the second depth image and the second visible light image.