Product Detection Device Using Adaptive Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing product detection systems face challenges in accurately detecting stockout and display disturbances across different stores due to variations in shelf types, product orientations, and display modes, leading to false recognition and degraded detection accuracy when using a single learned model.
Innovation Solution
A product detection device that acquires images of shelves, determines product display information, selects a suitable model based on this information, and detects display states using a combination of image acquisition, determination, selection, and detection units, allowing for improved accuracy by adapting to specific store conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single learned model is used for product detection across different stores, then device complexity is reduced, but detection accuracy deteriorates due to variations in shelf types, product orientations, and display modes
Solution Approach 1:
The patent segments the detection system by creating multiple learned models, each specialized for specific shelf types, product orientations, and display modes. Instead of using one general model, the system divides the detection task into specialized sub-tasks handled by different models, thereby improving detection accuracy for each specific scenario while managing complexity through organized model selection based on store conditions.
2Measurement precision
If multiple specialized models are used for different store conditions, then detection accuracy is improved, but device complexity increases due to model selection and management requirements
Solution Approach 1:
The system performs preliminary action by pre-establishing multiple learned models for different store conditions before actual detection begins. Each model is pre-trained for specific scenarios (different shelf types, orientations, display modes), and the system prepares a model selection mechanism that matches current store conditions to the appropriate pre-trained model, thereby improving detection accuracy without requiring complex real-time adjustments.
3Ease of operation
If a model learned at one location is applied to different stores, then ease of operation is improved, but reliability deteriorates due to false recognition from varying display conditions
Solution Approach 1:
The patent applies local quality by tailoring the detection model to match local store conditions. Instead of using a universal model everywhere, the system selects or adapts models based on specific local characteristics such as shelf type, product orientation, and display mode at each store location. This ensures high reliability by using models trained on data from similar conditions, reducing false recognition while maintaining ease of operation through automated model selection.
Data Source
AI summary
A product detection device is provided with an image acquisition unit, a determination unit, a selection unit, and a detection unit. The image acquisition unit acquires an image of a shelf on which products are displayed. The determination unit determines, from the image, product display information including at least one of a shape of shelf, shapes of the products, and a display condition. The selection unit selects, on the basis of the determined product display information, a model to be used to detect the image. The detection unit uses the selected model to detect the state of the display of the products displayed on the shelf from the image.