Self-Service Checkout Fraud Detection via Visual Article Counting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Self-service checkout systems face challenges in detecting unfair actions, such as inadvertent mistakes or intentional cheating, without the practicality of installing weight sensors due to high implementation costs, especially in large or widespread stores.

Innovation Solution

An information processing program and apparatus that captures image data using a camera, processes it with a machine learning model trained for article identification and storage recognition, enabling the detection of unfair actions by counting articles and comparing scanned items with those interacted with by the user.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If weight sensors are installed to detect unfair actions, then detection accuracy is improved, but implementation cost increases significantly

Engineering Contradiction:
Improveunfair action detection accuracyVSAvoidimplementation cost
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent replaces the mechanical weight sensing system with an optical imaging system combined with machine learning algorithms. The camera captures images of articles on the conveyor belt, and the machine learning model identifies and counts articles based on visual features, eliminating the need for expensive weight sensors while maintaining detection capability.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates a visual copy (image) of the physical article placement on the conveyor belt. By capturing and analyzing images of the articles, the system creates a digital representation that can be processed to detect unfair actions, replacing the need for direct physical measurement through weight sensors.

Inventive Principle:
Principle #26Copying

2Reliability

If machine learning models are used for article identification, then unfair action detection capability is improved, but processing time increases

Engineering Contradiction:
Improveunfair action detection capabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-training the machine learning model with extensive article images before deployment. The model learns to recognize various article types, patterns, and characteristics in advance, enabling rapid identification during actual checkout operations without requiring complex real-time processing.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent optimizes processing speed by adjusting parameters such as image resolution, processing batch size, and model complexity. By finding the optimal balance between model accuracy and processing speed, the system achieves reliable unfair action detection while maintaining acceptable processing times for self-service checkout operations.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4160559A1Information processing program, information processing method, and information processing apparatus
Publication Date: 2023.04.05 FUJITSU LTD
  • EP4160559A1 patent drawingFigure 1
  • EP4160559A1 patent drawingFigure 2
  • EP4160559A1 patent drawingFigure 3

AI summary

An information processing apparatus (100) obtains image data in which a predetermined area in front of an accounting machine, which is used by a user to register an article and pay the bill, is captured. The information processing apparatus (100) inputs the image data in a machine learning model (104) that is trained to identify an article and a storage for the article, and obtains the output result. The information processing apparatus (100) refers to the article and the storage specified in the output result, and identifies the action taken by the user with respect to the article.