Single-Camera Depth Estimation with Aberration-Driven Bokeh Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for acquiring distance from a single-camera image suffer from accuracy issues due to optical aberrations, particularly chromatic aberration, which affect the bokeh in captured images, leading to non-linear variations in point spread function shapes that complicate distance estimation.
Innovation Solution
An image processing device utilizes a statistical model trained on bokeh patterns caused by optical aberrations, incorporating aberration maps and lens information to select appropriate models for accurate distance estimation, using machine learning algorithms like convolutional neural networks to analyze bokeh variations and position dependencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If distance is acquired from a single-camera image, then device complexity is reduced, but measurement precision deteriorates due to optical aberrations affecting bokeh patterns
Solution Approach 1:
The patent converts the harmful effect of optical aberrations (chromatic aberration causing non-linear bokeh variations) into a beneficial feature by training statistical models to recognize these aberration patterns. The system learns to identify specific bokeh characteristics caused by lens aberrations and uses them as distinctive features for accurate distance estimation, thereby transforming what was previously a source of error into a reliable distance cue
Solution Approach 2:
The patent changes the approach from attempting to eliminate or correct optical aberrations to utilizing them by adjusting the parameters of statistical models. Different statistical models are trained with specific parameters optimized for different types of lens aberrations, allowing the system to adapt to various optical characteristics and maintain high measurement precision across different camera lenses
2Measurement precision
If statistical models trained on bokeh patterns are used, then measurement precision improves, but device complexity increases due to multiple models and aberration maps
Solution Approach 1:
The patent applies preliminary action by pre-training multiple statistical models offline with abundant bokeh image data and lens aberration information before actual deployment. The complex task of learning intricate bokeh patterns and aberration characteristics is performed in advance, creating ready-to-use models that can be efficiently deployed in real-time applications without requiring complex processing during actual distance estimation
Solution Approach 2:
The patent segments the complex problem of distance estimation by creating multiple specialized statistical models, each optimized for specific types of lens aberrations or focal length ranges. Instead of using one complex universal model, the system divides the problem into smaller segments and uses the appropriate segmented model based on the specific imaging conditions, thereby managing complexity through modular organization
3Measurement precision
If aberration maps and lens information are incorporated, then measurement precision improves, but ease of operation deteriorates due to model selection requirements
Solution Approach 1:
The patent implements self-service by enabling the system to automatically identify the appropriate statistical model and aberration map based on the captured image characteristics. The system autonomously determines lens type, focal length, and aberration patterns from the image data itself, then selects and applies the corresponding pre-trained model without requiring manual configuration or user intervention, thereby maintaining ease of operation while achieving high precision
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The solution enhances the accuracy of distance estimation by leveraging bokeh patterns and aberration maps, providing precise distance information despite variations in lens types and focal settings, improving the reliability of single-camera depth estimation.
Implementation Method 1
the bokeh that occurs in the image captured by the optical system and varies non-linearly in accordance with the distance to the subject in the image
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
According to one arrangement, an image processing device (3) includes means (31) for storing a statistical model generated by learning of bokeh that occurs in a first image affected by aberration of a first optical system and varies non-linearly in accordance with a distance to a subject in the first image, means (37) for acquiring a second image affected by aberration of a second optical system, and means (38) for inputting the acquired second image into the statistical model corresponding to a lens used in the second optical system and acquiring distance information indicating a distance to a subject in the second image.