Autofocus Confidence Measure Using Weighted Depth Estimates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autofocus systems often fail to converge on a proper focus, are slow, or continue 'hunting' for the focal position, especially in noisy conditions, leading to reduced accuracy and increased lens movements.
Innovation Solution
A method that combines past and present depth estimation results using statistical models and confidence measures to determine the next lens position, adjusting focus based on a weighted mean of depth and variance, and generating a confidence measure to control the autofocus process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional autofocus mechanisms search for the peak of the autofocus curve using image gradients, then the system can achieve focus, but the system may never converge on a proper focus, is slow, or continues hunting for the proper focal position
Solution Approach 1:
The patent applies preliminary action by using depth estimation from previous image pairs to predict the next lens position before actually capturing the next image. This allows the system to proactively move the lens to a predicted optimal position rather than reactively searching, significantly reducing the time to achieve focus while maintaining accuracy.
Solution Approach 2:
The patent implements feedback by using the depth estimation results from previously captured image pairs to inform and adjust the position of the next lens movement. This closed-loop feedback mechanism allows the system to learn from past measurements and continuously refine its focus approach, preventing hunting behavior and ensuring convergence to the proper focal position.
2Speed
If the autofocus system uses only the most recent depth estimation result to determine the next lens position, then the system responds quickly, but the focus accuracy degrades in noisy conditions
Solution Approach 1:
The patent merges multiple depth estimation results from different image pairs to determine the next lens position. By combining information from multiple sources rather than relying on a single recent estimation, the system achieves more robust and accurate focus determination that is resistant to noisy conditions, while still maintaining responsive performance.
Data Source
Figure 1~2B
Figure 3
Figure 4~5
AI summary
Autofocusing is performed in response to a weighted sum of previous blur difference depth estimates at each focus adjustment iteration. Variance is also determined across both past and present estimations providing a confidence measure on the present focus position for the given picture. Focus adjustment is repeated until the variance is sufficiently low as to indicate confidence that a proper focus has been attained. The method provides more accurate and rapid focusing than achieved by the best current depth-based techniques, such as those utilizing most recent depth estimation to determine the next lens position. In contrast to this, the present apparatus and method combines all previous depth estimation results in the autofocus process to determine the next lens position based on statistical models and confidence measure.