Stored Product Identification Across Appliance Camera Configurations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep-learning based identification algorithms for household storage appliances require complex re-training and adaptation to handle changes in image data caused by different appliance models or camera models, leading to impaired identification accuracy and increased development costs.
Innovation Solution
A method using a standard identification algorithm adapted by an upstream adaptation algorithm trained on the household storage apparatus configuration, reducing the need for extensive re-training by limiting parameter adjustments to a few tens to hundreds, thus facilitating use across different configurations with improved identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep-learning based identification algorithms are re-trained to adapt to changes in image data caused by different appliance models or camera models, then identification accuracy is improved, but development outlay and complexity increase significantly
Solution Approach 1:
The identification algorithm is divided into two independent parts: a standard identification algorithm that remains unchanged and an adaptation algorithm that handles configuration-specific adjustments. This segmentation allows the standard algorithm to be trained once on standard configuration images, while adaptation algorithms are trained separately for different appliance or camera models, reducing overall development complexity while maintaining identification accuracy.
Solution Approach 2:
An adaptation algorithm is introduced as an intermediary component between the image data capture and the standard identification algorithm. This adaptation algorithm processes images from different configurations and transforms them into a format compatible with the standard identification algorithm, eliminating the need to re-train the main identification algorithm for each new configuration.
2Device complexity
If the standard identification algorithm is used across different appliance configurations without adaptation, then development costs are reduced, but identification accuracy deteriorates due to changes in image brightness or distortions
Solution Approach 1:
The adaptation algorithm adjusts parameters such as image brightness, contrast, and geometric distortions to normalize images from different appliance or camera configurations. By modifying these parameters, the adaptation algorithm ensures that images from various sources meet the requirements of the standard identification algorithm, maintaining high identification accuracy without requiring changes to the core algorithm.
3Adaptability or versatility
If appliance-specific features of the identification algorithm are operated in parallel to ensure function across all appliance models, then adaptability is improved, but device complexity increases
Solution Approach 1:
The standard identification algorithm is designed to be universal and can process images from different appliance configurations without modification. The adaptation algorithm serves as a multi-functional component that handles various configuration-specific adjustments, allowing the system to maintain a single standard identification algorithm while achieving broad compatibility across different appliance models.
Data Source
AI summary
A method for identifying stored products in household storage devices. A set of images of a stored product of the storage device is captured, and the stored product is identified by evaluating image data of the set of images by a standard identification algorithm. The standard identification algorithm has been trained by way of a standard set of images captured with a standard configuration. Before the standard identification algorithm is used, a data characteristic of the image data is adapted to a standard data characteristic of the standard set of images by way of an adaptation algorithm that is at least partially trained based on a configuration of the household storage device.