Image Georectification with Nonstationary Object Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image georectification methods face difficulties in accurately aligning images due to nonstationary objects like vehicles, people, and animals, which cause inconsistencies and make the process more challenging, especially when these objects are in different positions across multiple images.
Innovation Solution
A system and method for improved image georectification that involves receiving and comparing input images to identify nonstationary portions, either through error matrix analysis or machine learning, and removing these portions to enhance alignment accuracy by generating an error matrix or using machine learning to recognize and exclude pixels from mobile objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image georectification methods are used to align images, then the process can be completed using standard algorithms, but the presence of nonstationary objects like vehicles, people, and animals causes inconsistencies and reduces alignment accuracy
Solution Approach 1:
The patent extracts and removes nonstationary portions (mobile objects like vehicles, people, animals) from the image data before performing georectification. By separating these problematic elements from the stationary background, the registration process can focus only on stable features, thereby improving both alignment accuracy and consistency without being disrupted by moving objects.
2Measurement precision
If nonstationary portions are removed from images before georectification, then alignment accuracy is improved, but the complexity of the processing system increases
Solution Approach 1:
The patent segments the image processing task into distinct stages: first identifying and removing nonstationary portions, then performing georectification on the cleaned data. This segmentation allows each subsystem to be optimized independently—the removal module handles mobile objects while the georectification module focuses on stationary features—reducing overall system complexity compared to a monolithic approach that must handle all image features simultaneously.
Solution Approach 2:
The patent performs preliminary removal of nonstationary portions before the main georectification process. By preparing the image data in advance by eliminating problematic mobile objects, the subsequent georectification algorithm can operate more efficiently and accurately without needing to account for moving features, thereby reducing the computational complexity of the main processing pipeline.
3Measurement precision
If error matrix analysis is used to identify nonstationary portions, then the method can detect mobile objects, but the processing time and computational resources increase
Solution Approach 1:
The patent applies error matrix analysis selectively to identify nonstationary portions rather than analyzing every pixel in the entire image. By focusing computational resources only on regions where mobile objects are likely to appear or where errors indicate movement, the system achieves adequate detection accuracy while significantly reducing overall processing time and computational resource requirements.
Data Source
AI summary
Method comprises receiving, by at least one memory, from at least one imaging system, at least two input images. The method includes comparing, by a processor, the at least two input images to each other such that nonstationary portions of the at least two input images are determined by either separating, by the at least one processor, each of the two input images into multiple pixel regions, and generating an error matrix for each of said multiple pixel regions. If an error value in the error matrix falls within a predetermined range, the pixel region is a nonstationary portion of the input images; or identifying, by a machine learning system, nonstationary portions of the input images. The pixels from nonstationary portions are removed from the at least two input images.