Unsupervised Optical Flow Estimation Robust to Intensity Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional unsupervised optical flow estimation methods are not robust to changes in light intensity between successive images, leading to errors in motion detection and object tracking.
Innovation Solution
A joint learning method that combines two models: one for estimating optical flow and another for learning intensity changes between pixels, which corrects the loss function to improve robustness to intensity variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional unsupervised optical flow estimation methods are used, then the method does not require ground truth annotation, but the estimation accuracy deteriorates under light intensity changes
Solution Approach 1:
The patent segments the optical flow estimation problem into two separate estimation tasks: one for optical flow and another for light intensity changes. This segmentation allows each model to specialize in its respective task, improving overall accuracy without requiring ground truth annotation for either task.
Solution Approach 2:
The patent introduces a mediator model that estimates light intensity changes between successive images. This intermediary model corrects the photometric error by compensating for intensity variations, thereby improving optical flow estimation accuracy without requiring ground truth data.
2Device complexity
If a single model estimates both optical flow and intensity changes, then the device complexity is reduced, but the estimation precision for both tasks deteriorates
Solution Approach 1:
The patent divides the estimation problem into separate models: one dedicated to optical flow estimation and another to light intensity change estimation. This segmentation enables each model to achieve higher precision by focusing on its specific task, avoiding the performance degradation that would result from a single multi-purpose model.
Solution Approach 2:
The patent changes the parameters being estimated by separate models - one model estimates displacement vectors (optical flow) while another estimates intensity variation parameters. This parameter separation allows each model to optimize its own loss function and architecture for its specific estimation task.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method for unsupervised learning of an optical flux model from a sequence of images, the method comprising the joint learning of a first machine learning model configured to estimate the optical flux F1,F2 between two successive images and a second machine learning model configured to estimate a difference in light intensity C1,C2 between the corresponding pixels of the two successive images, the joint learning being carried out by optimizing at least a first cost function Lph1 characteristic of an error between the first image I1 of a pair of successive images I1,I2 and a reconstructed version I1c^ of the first image I1 obtained by: - Applying the difference in light intensity C2 estimated between the second image I2 and the first image I1, to the second image I2, so as to obtain a second image I2c compensated for the difference in light intensity C2,- Applying the estimated optical flux F1 between the first image I1 and the second image I2 to the compensated second image I2c, so as to obtain the reconstructed version I1c^ of the first image I1.