Binocular Depth Estimation Mapping for Feature-Poor Image Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image depth estimation methods, such as binocular ranging, can only determine depth for image portions with prominent feature points, failing to estimate depth for areas lacking distinct features.
Innovation Solution
An image depth estimation method that calculates relative depth values for multiple pixels in a binocular image, identifies a target region with consistent bidirectional disparity, determines absolute depth values within this region, constructs a mapping relationship between relative and absolute depth values, and applies this mapping to estimate depth across the entire image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If binocular ranging method is used to estimate depth, then depth can be determined for image portions with prominent feature points, but depth cannot be determined for portions with non-prominent feature points
Solution Approach 1:
The patent segments the image into different regions based on feature point prominence. Prominent feature point regions are processed using traditional binocular ranging methods, while non-prominent feature point regions are processed using a different approach that leverages spatial relationships and depth maps from prominent regions to infer depths in these areas, thereby achieving comprehensive depth estimation across the entire image.
Solution Approach 2:
The patent introduces an intermediary mechanism that uses the depth information and spatial relationships derived from prominent feature point regions as mediators to infer and estimate depths in non-prominent feature point regions. This intermediary approach allows the system to bridge the gap between regions with and without prominent features, enabling unified depth estimation across diverse image types.
2Productivity
If depth estimation is performed only on image portions with prominent feature points, then processing complexity is reduced, but complete depth estimation for the entire image is not achieved
Solution Approach 1:
The patent performs preliminary processing to identify and segment prominent feature point regions first, establishing depth information and spatial relationships in these key areas. This preliminary action creates a foundation that simplifies the subsequent inference process for non-prominent regions, allowing the system to achieve complete image coverage without proportionally increasing overall processing complexity.
Solution Approach 2:
The patent applies different processing strategies to different local regions of the image based on their feature point characteristics. Prominent feature regions receive detailed binocular ranging processing, while non-prominent regions receive simplified inference-based processing. This local quality differentiation allows the system to maximize depth estimation coverage while managing processing complexity through region-specific approaches.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Embodiments of the present disclosure disclose an image depth estimation method and apparatus, an electronic device, and a storage medium. The method includes: obtaining relative depth values of a plurality of pixels in a binocular image; obtaining a bidirectional disparity of a same scene point in the binocular image, and determining a target region in which the bidirectional disparity is consistent; determining an absolute depth value of a pixel in the target region based on a bidirectional disparity of the pixel in the target region; constructing a mapping relationship between a relative depth value and an absolute depth value based on the absolute depth value of the pixel in the target region and a relative depth value of the pixel in the binocular image; and determining, based on the mapping relationship and a relative depth value of a pixel to be determined in the binocular image, an absolute depth value of the pixel to be determined in the binocular image.