Reverse Optical Flow Correction for Accurate Disparity Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical flow estimation techniques face challenges in accurately estimating pixel motion between sequential images due to errors introduced by local inconsistencies, occlusions, and dynamic changes in the environment, leading to error accumulation and poor performance.
Innovation Solution
The system employs reverse optical flow error correction by warping a current image to generate an estimated previous image, comparing it with the actual previous image to identify defective optical flows, and generating a confidence map to clean the optical flow information, using techniques such as dynamic thresholds and machine learning to adjust validity thresholds based on features and occlusions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical flow estimation is performed using conventional techniques, then processing speed is maintained at acceptable levels, but measurement precision deteriorates due to error accumulation from local inconsistencies, occlusions, and dynamic changes
Solution Approach 1:
The patent applies reverse optical flow warping to correct errors in the original optical flow estimation. By inverting the warping process and comparing the warped current image with the actual previous image, the system identifies and corrects defective optical flow vectors, thereby improving measurement precision without significantly increasing processing time
Solution Approach 2:
The system generates a confidence map through feedback from the reverse warping process, comparing estimated and actual images to identify regions with high error rates. This feedback mechanism allows selective correction of optical flow vectors based on their reliability, improving overall precision while maintaining processing efficiency
2Measurement precision
If reverse optical flow warping is applied to correct errors, then measurement precision improves, but device complexity increases due to additional processing steps
Solution Approach 1:
The patent creates a copy of the current image and warps it using the estimated optical flow to generate an estimated previous image. This copying approach allows error correction without requiring multiple physical cameras or sensors, maintaining relatively simple hardware while improving software-based measurement precision
Solution Approach 2:
The system applies different correction strategies to different regions of the image based on the confidence map. Regions with high confidence in optical flow estimates are processed differently from low confidence regions, allowing localized error correction that reduces overall processing complexity compared to uniform correction approaches
3Reliability
If confidence maps are generated and applied to correct optical flow, then reliability improves, but processing time increases
Solution Approach 1:
The patent applies partial correction by generating confidence maps that identify only the most problematic regions requiring correction. Rather than uniformly processing the entire image, the system applies correction only to regions with low confidence optical flow estimates, reducing processing time while maintaining high reliability in critical areas
Solution Approach 2:
The system performs preliminary warping and comparison to generate confidence maps before final optical flow correction. This preliminary action allows the system to pre-identify errors and plan corrections efficiently, reducing overall processing time by avoiding unnecessary computations in high-confidence regions
Data Source
AI summary
Disclosed are systems and techniques for capturing images (e.g., using an image capture) and performing reverse optical flow error correction. According to some aspects, a computing system or device can obtain first disparity information associated with a current image. The first disparity information estimates a first movement of a first feature to a first destination location in the current image. The computing system or device can warp the current image based on the first disparity information to obtain an estimated previous image, determine a confidence map associated with a confidence of the first disparity information based on a difference associated with the estimated previous image; and apply the confidence map to the first disparity information to generate updated first disparity information.