Optical Flow Pixel Matching via Discrete Scale Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical flow processing in advanced driver assistance systems (ADAS) faces challenges in handling perspective magnification due to camera motion, leading to inaccurate pixel matching and motion estimation, especially with fisheye lenses, where existing solutions like SIFT are computationally complex and not suitable for real-time processing.
Innovation Solution
A method using scaled binary pixel descriptors and census transforms to account for perspective changes by defining binary pixel descriptors with different neighborhood pixels in each image, and modifying Hamming distance computation with multiple discrete scale ratios to determine accurate matches, implemented in an optical flow accelerator for efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SIFT is used to address perspective changes, then matching accuracy is improved, but computational complexity increases making it unsuitable for real-time processing
Solution Approach 1:
The patent transforms the continuous perspective transformation problem into a discrete set of predefined scale ratios. Instead of using complex continuous transformations like SIFT, the system defines a finite set of scale factors (e.g., 0.5, 0.75, 1.0, 1.25, 1.5) and tests these discrete scales to find the best match. This parameter discretization maintains matching accuracy while dramatically reducing computational complexity for real-time processing.
Solution Approach 2:
The patent segments the perspective transformation compensation into multiple discrete scale levels. Rather than applying a single complex transformation, the system divides the problem into testing several predefined scale ratios independently. This segmentation allows the computationally intensive matching process to be broken down into manageable discrete steps that can be efficiently processed in real-time.
2Productivity
If perspective changes are not accounted for in pixel matching, then computational complexity is reduced, but matching accuracy deteriorates leading to propagation of inaccuracy to algorithms using optical flow information
Solution Approach 1:
The patent performs preliminary scaling of pixel descriptors at multiple predefined scale ratios before the actual matching process. By pre-computing scaled versions of pixel descriptors and organizing them in a scale-space data structure, the system prepares all necessary transformation variations in advance. This preliminary action ensures that when matching occurs, the computationally expensive scaling operations have already been completed, maintaining both accuracy and real-time processing capability.
Data Source
AI summary
A method of optical flow estimation is provided that includes identifying a candidate matching pixel in a reference image for a pixel in a query image, determining a scaled binary pixel descriptor for the pixel based on binary census transforms of neighborhood pixels corresponding to scaling ratios in a set of scaling ratios, determining a scaled binary pixel descriptor for the candidate matching pixel based on binary census transforms of neighborhood pixels corresponding to scaling ratios in the set of scaling ratios, and determining a matching cost of the candidate matching pixel based on the scaled binary pixel descriptors.