Local Normalization for Depth from Stereo via Midpoint Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for depth from stereo (DFS) fail to address luminance changes between left and right images, leading to signal-to-noise issues and impaired DFS quality, especially in low light and low feature strength regions.
Innovation Solution
A method that determines a target granularity for image buffers, constructs a partial image pyramid, samples it, calculates a midpoint normalization value, and normalizes the image buffers to mitigate luminance shifts and enhance DFS quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current DFS techniques are used without normalization, then processing speed is maintained, but luminance changes between left and right images cause signal-to-noise issues and impaired DFS quality
Solution Approach 1:
The patent applies preliminary normalization to the left and right images before performing DFS processing. By calculating midpoint normalization values from the images and applying them in advance, the system prepares the data to prevent luminance changes from causing signal-to-noise issues during the actual DFS computation, thereby improving DFS quality without adding complexity to the main processing pipeline
Solution Approach 2:
The patent changes the luminance parameters of the input images by applying normalization factors derived from midpoint calculations. This parameter transformation equalizes the luminance levels between left and right images, eliminating the signal-to-noise problems that would otherwise degrade DFS quality while maintaining the same processing architecture
2Reliability
If full image normalization is performed, then luminance changes are fully addressed, but processing time and computational load increase
Solution Approach 1:
The patent performs partial normalization by calculating midpoint values only from the necessary image regions and applying normalization selectively. This partial action approach addresses the luminance consistency issue in the critical areas needed for DFS while avoiding the full computational overhead of normalizing entire high-resolution images, thus reducing processing time
Solution Approach 2:
The patent segments the normalization process into discrete steps: calculating midpoint values from input images, determining normalization factors, and applying them to the image data. This segmentation allows the system to perform normalization efficiently by breaking down the complex operation into manageable computational stages that can be optimized and executed with reduced time overhead
3Reliability
If normalization is applied to handle luminance shifts, then DFS quality improves, but device complexity increases
Solution Approach 1:
The patent implements a universal normalization module that can process both left and right images through the same midpoint calculation and normalization factor application. This multi-functional approach handles luminance shifts in a unified manner, improving DFS accuracy across different image conditions without requiring separate complex processing paths for each image
Solution Approach 2:
The patent introduces midpoint normalization values as an intermediary element between the raw input images and the DFS processing stage. These intermediary values serve as normalization factors that mediate the luminance differences, allowing the DFS algorithm to operate on normalized data without directly handling the complex luminance variation problem, thus improving accuracy while keeping the processing pipeline relatively simple
Data Source
AI summary
This disclosure provides systems, devices, apparatus, and methods, including computer programs encoded on storage media, for efficient local normalization for DFS. A processor may determine a target granularity based on at least one of a first image buffer or a second image buffer. The processor may determine a partial image pyramid based on the target granularity and at least one of the first image buffer or the second image buffer. The processor may sample the partial image pyramid based on the target granularity. The processor may determine a midpoint normalization value based on the sampled partial image pyramid. The processor may normalize at least one of the first image buffer or the second image buffer based on the midpoint normalization value.