Camera-Based Product Recognition for Storage Compartment Stock Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for managing product stock in electronic apparatuses, such as refrigerators, face challenges with barcode dependency, sensor complexity, and neural network training requirements, leading to increased processing power, size, and recognition failures due to obstructions or varying product poses.
Innovation Solution
An electronic apparatus equipped with a camera, memory, and processor that uses multiple neural network models to capture and analyze product features, updating a database with identified features for input and output products, utilizing training data that accounts for different poses and product similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If barcode recognition method is used for product management, then product identification can be achieved, but barcodes must be attached to all products and the method cannot identify products without barcodes
Solution Approach 1:
The patent replaces the mechanical barcode reading system with an image-based recognition system using a camera and neural network. The camera captures images of products, and the neural network processes these images to identify products without requiring physical barcodes, thus substituting a mechanical system with an optical and computational one.
Solution Approach 2:
The patent creates a digital copy of product appearance features through camera imaging and stores these features in a database. Instead of relying on printed barcodes, the system captures and stores visual representations of products, enabling identification through image matching rather than barcode scanning.
2Measurement precision
If detecting sensors (camera, RFID) are mounted on the apparatus to manage stock, then product detection capability is improved, but processing power and device size increase
Solution Approach 1:
The patent extracts only the essential visual features needed for product identification from the captured images, rather than processing entire high-resolution images. This feature extraction approach reduces the computational burden and data storage requirements while maintaining product detection capability.
Solution Approach 2:
The system uses a single camera positioned to capture one side of the storage compartment, performing partial observation rather than complete 360-degree monitoring. This partial action approach achieves sufficient product identification without requiring multiple sensors or complex multi-angle imaging systems.
3Extent of automation
If RFID sensors are used for product recognition, then automatic identification is achieved, but the device requires additional sensors every time an item is moved, increasing processing power and size
Solution Approach 1:
The camera serves multiple functions: it captures images for product identification, tracks product movement by comparing sequential images, and updates the database. This single multi-functional component replaces the need for separate RFID sensors and processing systems, reducing overall device complexity.
Solution Approach 2:
The system uses the existing camera infrastructure to perform automatic identification and tracking without requiring additional dedicated sensors. The camera continuously captures images that are processed to identify products and their movement status, making the system self-sufficient with minimal additional hardware.
4Measurement precision
If neural network model is used for product recognition, then recognition accuracy can be improved, but large amount of training data and repeated retraining are necessary, increasing storage and processing cost
Solution Approach 1:
The patent segments the product recognition task into distinct phases: initial neural network training with comprehensive data, followed by operation phase where the trained model is applied to new images. This segmentation allows the heavy training process to be performed once offline, while the operational system uses the pre-trained model with minimal additional processing.
Solution Approach 2:
The neural network is trained in advance with comprehensive product data before deployment. This preliminary training action prepares the model to handle various product poses and appearances, so that during operation, the system can quickly identify products without requiring repeated retraining, thus reducing ongoing storage and processing costs.
5Adaptability or versatility
If image training for all poses is performed in advance, then product recognition from various angles is achieved, but large amount of storage and processing are required, resulting in high cost
Solution Approach 1:
The patent applies parameter changes to the training data by transforming images into different poses, angles, and lighting conditions. Instead of collecting and storing separate images for each possible pose, the system uses computational transformations to generate varied training samples from a smaller set of original images, reducing storage requirements while maintaining pose-invariant recognition capability.
Data Source
AI summary
An electronic apparatus is disclosed. The electronic apparatus includes: electronic apparatus including: a storage compartment; a camera; a memory configured to store a database corresponding to products provided in the storage compartment; and a processor communicably coupled to the camera and the memory to control the electronic apparatus, wherein the processor is configured to: control the camera to capture images of one side of the storage compartment; based on identifying that a first product is input to the storage compartment, obtain a first feature of the first product; update the database by storing the first feature in the memory; based on identifying that a second product is output from the storage compartment, obtain a second feature of the second product; identify a product corresponding to the second feature from the database; and update the database based on the product corresponding to the second feature.