Depth-Fill Algorithm for Low-Complexity Stereo Vision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-complexity stereo vision algorithms face challenges in estimating scene depth due to 'missing pixels' in texture-less regions, which existing methods attempt to address by increasing computational requirements, making them unsuitable for low-power embedded processors.
Innovation Solution
A depth-fill algorithm that utilizes temporal information from stereo camera images, exploiting events like shadows and moving objects to fill in missing depth pixels by updating a background depth model, allowing for robust and efficient depth recovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If global constraints are imposed on pixel matches to reflect smoothness of 3D surfaces, then depth estimation accuracy is improved, but computational requirements increase tremendously
Solution Approach 1:
The patent segments the depth estimation process into two stages: first, a low-complexity stereo algorithm generates initial depth estimates (even with holes in texture-less regions); second, a temporal filling algorithm processes sequences of depth images to fill missing pixels using temporal consistency. This segmentation allows the system to achieve complete depth maps without requiring computationally expensive global constraints in the matching process itself.
Solution Approach 2:
The patent performs preliminary depth estimation using a simple stereo algorithm before applying temporal filling. By first obtaining a rough depth map with identified hole locations, the system can then focus computational effort only on filling missing pixels using temporal information from previous and future frames, rather than applying complex global constraints to all pixels simultaneously.
2Measurement precision
If more constraints are imposed on pixel matches, then depth estimate accuracy is improved, but deployment on low-power embedded processors becomes infeasible
Solution Approach 1:
The patent separates depth estimation into a low-power initial stereo matching stage and a temporal filling stage that operates on sparse hole data rather than full image sequences. This segmentation enables deployment on low-power processors by avoiding the need to implement computationally intensive global constraint algorithms in real-time.
Solution Approach 2:
The patent processes continuous sequences of video frames to fill depth holes temporally. By accumulating temporal information across multiple frames and using motion compensation to track hole positions, the system achieves accurate depth estimation through continuous processing of available data rather than requiring complex per-frame analysis.
Data Source
AI summary
A method and apparatus for depth-fill algorithm for low-complexity stereo vision. The method includes utilizing right and left images of a stereo camera to estimate depth of the scene, wherein the estimated depth relates to each pixel of the image, and updating a depth model with the current depth utilizing the estimated depth of the scene.