Image Correspondence Analysis Using Signature Strings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing correspondences in image data sets, particularly in monocular camera systems and stereo image processing, are computationally intensive and struggle with large optical flows, limiting their ability to process data efficiently, especially in resource-constrained environments like motor vehicles during aggressive steering or high speeds.
Innovation Solution
A process and device that transform image data using a signature operator to generate signature strings for each pixel, which are stored in tables and compared to identify corresponding points, allowing for efficient correspondence analysis and hypothesis generation, even with large optical flows, by utilizing a census transformation and hash tables to optimize memory and computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computationally intensive correlation extensions are used for correspondence analysis, then measurement precision of optical flow is improved, but processing speed deteriorates and computational resources are excessively consumed
Solution Approach 1:
The patent segments the image into multiple blocks and processes each block independently to generate correspondence hypotheses. This segmentation reduces the computational complexity of comparing entire images while maintaining measurement precision at the block level, thereby improving processing speed without sacrificing optical flow measurement accuracy.
Solution Approach 2:
The patent performs preliminary actions by generating correspondence hypotheses based on block comparisons before conducting full image processing. This preliminary hypothesis generation filters out obvious non-corresponding regions early in the process, reducing the computational burden of subsequent processing steps and improving overall processing speed while maintaining precision.
2Measurement precision
If traditional correspondence analysis algorithms are used, then measurement precision is maintained, but the system cannot handle large optical flows occurring during aggressive steering or high vehicle speeds
Solution Approach 1:
The patent implements dynamic processing by adjusting the block size and processing parameters based on the detected optical flow magnitude. When large optical flows are detected (as during aggressive steering or high speeds), the system dynamically adapts its processing strategy to maintain correspondence identification precision while handling the increased motion complexity.
Solution Approach 2:
The patent introduces a hierarchical processing dimension by first analyzing image blocks at a coarse level to detect large displacements, then refining correspondence identification at finer levels. This multi-dimensional approach enables the system to handle both small and large optical flows effectively, expanding its adaptability across different driving conditions while maintaining precision.
3Measurement precision
If comprehensive correspondence analysis is performed on entire images, then measurement precision is improved, but device complexity and computational resource requirements increase
Solution Approach 1:
The patent divides the image into multiple manageable blocks and processes each block independently for correspondence analysis. This segmentation reduces the computational resource requirements and device complexity by breaking down the complex task of entire image comparison into simpler, parallelizable block-level operations, while maintaining overall measurement precision through aggregated block results.
Solution Approach 2:
The patent performs partial correspondence analysis by focusing computational resources on comparing representative blocks rather than every pixel in the entire image. This partial action approach reduces device complexity and computational resource requirements while still achieving sufficient measurement precision for optical flow determination in practical driving scenarios.
Data Source
AI summary
Processing of image data relating to moving scenarios, especially for recognizing and tracking objects located therein, requires identifying corresponding pixels or image areas in the individual successive image data sets. Likewise, processing of stereo images requires identifying the data areas which correspond to each other in two images that are recorded substantially at the same time from different angles of vision. According to the novel method of analyzing correspondences in image data sets, the image data sets that are to be compared are transformed using a signature operator such that a signature string is calculated for each pixel and is stored in a signature table allocated to the individual image data sets along with the pixel coordinates in a first step. A correspondence hypothesis is then generated for the signature strings identified in both tables and is stored in a list of hypothesis is then generated for the signature strings identified in both tables and is stored in a list of hypotheses for further processing. The inventive method advantageously makes it possible to analyze correspondences in a very efficient manner regarding the computing time while allowing fast processing of image pairs even when individual objects are presented at very different points in the two data sets.