Machine Learning Model for Distance Estimation with Optical State Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating distance information from defocus blur in images captured with optical systems face challenges in maintaining accuracy while managing high learning loads and stored data amounts, especially in systems with various aberrations such as high-magnification zoom lenses.
Innovation Solution
An image processing method that incorporates information about the optical system's state, including focal length, F-number, and focused object distance, into a machine learning model to accurately estimate distance information by learning weights specific to each system state, thereby reducing the learning load and stored data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the training data includes a plurality of defocus blurs with similar shapes, then the estimating accuracy of the distance information for each defocus blur lowers, but including all defocus blurs in the training data increases the stored data amount and learning load
Solution Approach 1:
The patent segments the training data by grouping defocus blurs according to their shapes. Instead of treating all defocus blurs uniformly, the method clusters similar defocus blur patterns together and selects representative samples from each cluster. This segmentation approach maintains estimating accuracy by ensuring diverse shape representations while reducing the stored data amount by eliminating redundant similar samples.
Solution Approach 2:
The patent changes the parameter of training data selection from including all defocus blurs to including only representative defocus blurs from each shape group. By transforming the selection criterion based on shape similarity analysis, the method reduces the quantity of training data while preserving the diversity needed for accurate distance estimation across different defocus conditions.
2Adaptability or versatility
If various aberrations occur in the optical system such as a high-magnification zoom lens, then the number of defocus blur groups becomes enormous, increasing the learning load and stored data amount
Solution Approach 1:
The patent merges the processing of multiple defocus blur groups by selecting representative samples from each group rather than processing all groups separately. This merging approach reduces the learning load by consolidating the training process while maintaining adaptability to various optical aberrations through the diversity of representative samples selected across different aberration types.
Solution Approach 2:
The patent changes the approach from handling each defocus blur group independently to grouping them by shape similarity and selecting representatives. This parameter change in the organization and processing method reduces the effective number of groups that need to be learned, thereby reducing learning load while preserving the system's ability to adapt to various optical aberrations.
Data Source
AI summary
An image processing method includes the steps of acquiring input data including a captured image and information about a state of an optical system that was used to capture the captured image, and estimating distance information about the captured image by inputting the input data into a machine learning model. The information about the state of the optical system includes at least one of a focal length, an F-number, and a focused object distance.