Shopping Cart Terminal Item Registration via Object and Character Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing item registration systems for stores struggle to accurately identify items in images, especially when items have similar appearances.
Innovation Solution
A system comprising a terminal attached to a shopping cart with a camera that captures images of items. The terminal includes an interface circuit and a processor that executes object recognition processing to determine candidate items, and when multiple candidates are identified, character recognition processing is used to narrow down the selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If only object recognition processing is used to identify items, then the identification process is simple and fast, but identification accuracy deteriorates when items have similar appearances
Solution Approach 1:
The item identification process is segmented into two distinct stages: first object recognition processing to identify candidate items based on visual features, then character recognition processing to identify specific items based on text information. This segmentation allows each processing stage to focus on its strength, maintaining speed while improving accuracy.
Solution Approach 2:
Character recognition acts as an intermediary step between object recognition and final item identification. When multiple candidate items are identified with similar appearances, the character recognition processing serves as a mediator to differentiate between them by reading text information on packaging, thereby resolving ambiguity without requiring complete redesign of the identification system.
2Measurement precision
If character recognition processing is always executed to improve identification accuracy, then item identification accuracy improves, but processing time and system complexity increase
Solution Approach 1:
Character recognition processing is applied partially - only when object recognition identifies two or more candidate items with similar appearances. When only one candidate is identified, the system stops at object recognition, avoiding unnecessary character recognition processing and maintaining system efficiency while improving accuracy only when needed.
Solution Approach 2:
The system dynamically adjusts its processing depth based on the number of candidate items identified. The processing flow is not fixed but adapts to the specific situation: simple object recognition for clear cases, and combined object+character recognition for ambiguous cases. This dynamic approach optimizes the balance between accuracy and complexity.
Data Source
AI summary
A system for registering an item for purchase at a store includes a terminal attachable to a shopping cart and a camera attached to the cart and positioned to capture an image of an item stored therein. The terminal includes: an interface circuit connectable to the camera, and a processor configured to: control the camera to capture an image, execute object recognition processing on the image and determine one or more candidate items using a result of the recognition processing, when only one candidate item is determined by the recognition processing, register the candidate item as an item for purchase, and when two or more candidate items are determined by the recognition processing, execute character recognition processing on the image and determine a character string in the image, and register one of the candidate items as an item for purchase using a result of the character recognition processing.