Learning Model Generation for Retail Shelf Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting product shortage or display disturbance in stores face challenges due to variations in shelf configurations and product orientations, leading to false recognition and decreased detection accuracy, as they require large amounts of high-quality training data that are difficult to obtain.
Innovation Solution
A learning model generation system that includes a POS terminal and a camera to acquire inventory information and shelf images, generating a model to estimate product quantity based on these data, thereby efficiently capturing high-quality learning images and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a learning model is trained using images captured at one store location, then the learning process is simple and efficient, but false recognition occurs and detection accuracy deteriorates when applied to other stores
Solution Approach 1:
The patent segments the learning data into two distinct components: store-specific images (capturing actual shelf configurations, orientations, and display methods) and universal product images (capturing product characteristics). This segmentation allows the learning model to separately learn store-specific visual patterns and product identification features, preventing false recognition while maintaining learning efficiency.
Solution Approach 2:
The patent creates a universal learning model that can be applied across multiple stores by combining store-specific image data with product image data. The model serves multiple functions: it adapts to each store's unique shelf configuration while simultaneously recognizing products universally, eliminating the need to retrain models for each individual store.
2Measurement precision
If large amounts of high-quality training data are collected for each store, then detection accuracy improves, but the time and resources required to capture and process images increase significantly
Solution Approach 1:
The patent performs preliminary actions by capturing store-specific shelf images once during the learning phase, before actual product detection begins. These pre-captured images establish the store's visual characteristics, allowing the system to quickly adapt to new stores without requiring extensive real-time data collection. Product images are also pre-collected from databases, eliminating the need to capture them in-store.
Solution Approach 2:
The patent uses copied product images from databases or catalogs instead of capturing every product image in every store. These copied images serve as training data, significantly reducing the time and effort required for data collection while maintaining sufficient quality for accurate product recognition and quantity estimation.
3Quantity of substance
If traditional image synthesis methods are used to generate training data, then data availability improves, but the quality and realism of the training data deteriorate
Solution Approach 1:
The patent merges two types of real data: actual store-captured images showing shelf configurations and product placements, and product images from databases. This combination creates comprehensive training datasets that maintain the realism and quality of actual store environments while providing sufficient quantity for robust model training. The merged data preserves authentic visual characteristics without the artifacts of synthesized images.
Data Source
AI summary
A learning model generation device is provided with: an inventory information acquisition unit for acquiring, from a POS terminal in a store, inventory information including an inventory quantity of products for which a payment is made; an image acquisition unit for acquiring an image of a shelf of the products in the store; and a model generation unit for generating a model to estimate the number of products from the image on the basis of the image and the inventory quantity of the products.