Image Recognition Device Using Separate Exposure Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition technologies face accuracy issues when recognizing subjects from high dynamic range (HDR) images due to artifacts generated by combining short-exposure and long-exposure images, which can lead to inaccurate subject recognition.
Innovation Solution
An image recognition device that captures multiple images with different sensitivities in one frame period, using a recognition unit with deep neural networks (DNNs) to process each image separately before HDR combination, improving accuracy by recognizing subjects from image data that does not include artifacts and adjusting exposure times based on recognition scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If HDR image is generated by combining multiple images with different sensitivities, then dynamic range is improved, but subject recognition accuracy deteriorates due to artifacts
Solution Approach 1:
The recognition process is segmented into two distinct paths: one for recognizing subjects from individual exposure images (short-exposure and long-exposure images) and another for generating HDR images. The subject recognition is performed on the individual exposure images before HDR combination, avoiding the artifact problem entirely. This segmentation allows the system to maintain high dynamic range imaging capability while achieving accurate subject recognition by processing recognition and HDR generation as separate functional streams.
2Loss of information
If subject is recognized from HDR image, then comprehensive image data is used, but recognition accuracy deteriorates due to artifacts from combining exposures
Solution Approach 1:
Subject recognition is performed as a preliminary action before HDR image generation. The system first performs subject recognition on the individual short-exposure and long-exposure images, then generates the HDR image separately. This preliminary recognition approach ensures that subject detection is based on clean, artifact-free exposure images while still utilizing the comprehensive data from multiple exposures for the final HDR output, thus maintaining both information completeness and recognition accuracy.
Data Source
AI summary
Provided are an image recognition device, a solid-state imaging device, and an image recognition method capable of improving accuracy in recognizing a subject. The image recognition device according to the present disclosure includes an imaging unit and a recognition unit. The imaging unit captures a plurality of images having different sensitivities in one frame period to generate image data of the plurality of images. The recognition unit recognizes the subject from the image data of each of the images, and recognizes the subject captured in an image of one frame based on a result of recognizing the subject.