Monocular Distance Estimation Using Asymmetric Blur Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the distance to an object using a monocular camera are prone to errors when blur information deviates from the modeled data, especially in cases where the object is on the near or far side of the focusing distance.
Innovation Solution
An image processing apparatus that captures images through a single optical system with a filter having point-asymmetric color filter areas, analyzing the blur information using a statistical model to estimate the distance, which includes learning methods and machine-learning algorithms like neural networks to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If blur information is used to calculate distance from a single camera image, then multiple shooting is not required and the method can be applied to moving objects, but the calculation error increases when blur information deviates from the model
Solution Approach 1:
The patent applies asymmetry by using point-asymmetric color filter areas in the filter. This creates different blur shapes for different wavelengths of light, which provides additional information for distance calculation. The asymmetric filter design allows the system to distinguish between blur caused by defocus and blur caused by other factors, thereby improving measurement accuracy while maintaining the ability to capture single images.
Solution Approach 2:
The patent changes the parameter of the filter by using multiple color filter areas with different spectral characteristics. This allows the system to capture blur information at different wavelengths simultaneously, providing multiple parameters for distance calculation. By analyzing how blur varies across different wavelengths, the system can more accurately determine distance even when the blur deviates from simple models.
2Measurement precision
If a statistical model with machine learning is used to analyze blur information, then distance estimation accuracy is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the statistical model with machine learning algorithms before actual distance measurement. The model is trained offline using labeled image-distance pairs, and the learned parameters are stored for later use. During actual operation, the pre-trained model quickly processes blur information to estimate distance, avoiding the need for complex real-time calculations and reducing processing system complexity.
Solution Approach 2:
The patent uses copying by creating a statistical model that learns the relationship between blur patterns and distance from training data. Instead of implementing complex physical models or multiple cameras, the system creates a simplified computational model that copies the essential distance-information relationship from training examples. This model can then be applied to new images with minimal processing complexity.
Data Source
AI summary
According to one embodiment, an image processing apparatus includes a memory and one or more hardware processors electrically coupled to the memory. The one or more hardware processors acquire a first image of an object including a first shaped blur and a second image of the object including a second shaped blur. The first image and the second image are acquired by capturing at a time through a single image-forming optical system. The one or more hardware processors acquire distance information to the object based on the first image and the second image, with a statistical model that has learnt previously.