Image Sensor Recognition Switching for High Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recent digital cameras with dual recognition sections in the image sensor and signal processor face challenges in optimizing recognition accuracy and efficiency due to the separate circuits and differing recognition targets.
Innovation Solution
An image capturing apparatus with an image sensor section and a signal processor that can switch between recognition and learning modes, where the image sensor section performs initial recognition and the signal processor performs advanced recognition using a learning model, and also updates the learning model based on recognition results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If two recognition sections are provided (one in the image sensor section and one in the signal processor), then recognition coverage and functionality are improved, but the two recognition sections are not necessarily adapted to the same recognition target due to being separate circuits
Solution Approach 1:
The patent implements feedback by having the first recognition unit in the image sensor section provide recognition results to the second recognition unit in the signal processor. The second recognition unit uses this feedback to correct and refine its recognition results, ensuring both recognition sections are adapted to the same recognition target while maintaining high recognition accuracy.
2Measurement precision
If high-level recognition processing is performed by a signal processor using a learning model, then recognition accuracy is improved, but the learning model requires continuous updates and training which consume time and resources
Solution Approach 1:
The patent applies preliminary action by pre-storing multiple learning models with different characteristics in the storage section before recognition processing begins. The control section selects the most appropriate pre-trained learning model based on the specific recognition task, avoiding the need for time-consuming training during actual operation.
Solution Approach 2:
The patent changes parameters by selecting different pre-trained learning models with varying characteristics (e.g., different architectures, training datasets, or optimization parameters) depending on the recognition requirements. This allows the system to adapt to different recognition scenarios without retraining, saving time while maintaining high accuracy.
3Adaptability or versatility
If an image sensor section and signal processor are used as separate circuits, then functional flexibility is improved, but coordination and adaptation between the two recognition sections becomes difficult
Solution Approach 1:
The patent merges the functions of the two separate recognition sections by establishing a feedback loop where the first recognition unit in the image sensor section and the second recognition unit in the signal processor work together. The control section coordinates them to process the same image data, ensuring both sections are adapted to the same recognition target despite being separate circuits.
Data Source
AI summary
An image capturing apparatus includes an image sensor section and a signal processor for processing an image input from the image sensor section via a signal line. A control arithmetic unit switches the signal processor to a recognition mode or a learning mode. The image sensor section includes an image capturing section for generating the image, and a sensor recognition section for performing recognition processing based on the image. The signal processor includes a recognition section for performing, in the recognition mode, recognition processing by inputting the image input from the image sensor section to a learning model, and a learning section for performing, in the learning mode, machine learning of the learning model based on a recognition result obtained by the sensor recognition section and the image input from the image sensor section.