Multi-Region Subject-Distance Inference for Obstructed Autofocus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image capturing technologies struggle to accurately focus on a subject when obstructed by blocking objects or in low-illuminance environments, leading to significant variations in focus detection results and errors, especially when using statistical values for distance information.
Innovation Solution
An inference apparatus utilizing a machine learning model trained to suppress the contribution of distance information not based on the subject, by performing inference with multiple detected distance information pieces from focus detection regions within a subject region, generating accurate distance information corresponding to the subject.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical values for distance information are used in multiple autofocus regions, then focus detection can be performed across the subject region, but the influence of blocking objects cannot be suppressed and focus accuracy deteriorates
Solution Approach 1:
The subject region is divided into multiple focus detection regions (AF regions), and distance information is independently obtained from each region. The machine learning model then processes these segmented distance information pieces to identify and suppress contributions from blocking objects, thereby resolving the contradiction between comprehensive focus detection and robustness against blocking objects.
Solution Approach 2:
A machine learning model is introduced as an intermediary between the raw distance information from multiple AF regions and the final focus detection result. This intermediary processes the distance information pieces, suppressing contributions from blocking objects while maintaining accurate focus detection on the main subject.
2Area of stationary object
If distance information from multiple focus detection regions is used, then coverage of the subject region is improved, but erroneous focus detection results with large errors cannot be suppressed
Solution Approach 1:
The subject region is divided into multiple focus detection regions, allowing comprehensive coverage while maintaining the ability to individually evaluate and suppress erroneous distance information from each region through the machine learning model.
Solution Approach 2:
The machine learning model processes distance information from multiple AF regions and generates a refined focus detection result by suppressing erroneous contributions. This feedback mechanism ensures that erroneous focus detection results with large errors are filtered out while maintaining accurate focus detection.
3Productivity
If conventional statistical value methods are used, then processing is simple and fast, but the ability to suppress blocking object influence and detection errors is insufficient
Solution Approach 1:
A machine learning model is introduced as an intermediary processing layer that enhances the conventional statistical value method. This intermediary efficiently suppresses blocking object influence and detection errors while maintaining real-time processing capability, thus resolving the contradiction between processing speed and focus detection accuracy.
Data Source
AI summary
There is provided an inference apparatus. An inference unit performs inference with use of a machine learning model based on a subject region including a subject within an image obtained through shooting, and on a plurality of distance information pieces detected from a plurality of focus detection regions inside the subject region, thereby generating an inference result indicating a distance information piece corresponding to the subject or a distance information range corresponding to the subject. The machine learning model is a model that has been trained to suppress a contribution made to the inference result by one or more distance information pieces that are not based on the subject among the plurality of distance information pieces.