Fraud Detection via Reliability Scoring in Self-Checkout Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing self-checkout systems struggle to accurately detect fraudulent acts, such as 'label switching' and 'banana tricks,' due to variations in imaging environments like lighting and direction, which can lead to incorrect item recognition by machine learning models.
Innovation Solution
An information processing system that uses a machine learning model, like the CLIP model, to analyze videos from monitoring cameras, calculate reliability scores for items, and generate alerts when discrepancies are detected between recognized items and registered items, thereby enhancing fraudulent act detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a machine learning model is used to recognize items from monitoring camera videos, then automated fraud detection is enabled, but recognition accuracy deteriorates due to variations in imaging environments such as lighting and direction
Solution Approach 1:
The patent introduces an information processing device as an intermediary between the machine learning model and the fraud detection system. This device calculates reliability scores for multiple possible items, compares them against registered items, and generates alerts only when discrepancies are found. This intermediary layer compensates for the machine learning model's inaccuracies by providing a verification mechanism that reduces false positives while maintaining automated detection.
2Measurement precision
If multiple possible items are considered with reliability scores, then detection accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent changes the parameter of item recognition from a single definitive identification to a probabilistic scoring system. By calculating reliability scores for multiple possible items and comparing them against registered items, the system achieves higher detection accuracy. The information processing device manages this complexity by systematically evaluating score differences and generating alerts only when thresholds are exceeded, making the complex processing可控 (controllable).
Data Source
AI summary
A non-transitory computer-readable recording medium stores therein an information processing program that causes a computer to execute a process including, acquiring a video of a person who grasps an item to be registered in an accounting machine, by analyzing the acquired video, calculating a score indicating a level of reliability of the item that is contained in the video with respect to each of a plurality of possible items that are set previously, acquiring information on the item that is registered in the accounting machine by the person by operating the accounting machine, based on the calculated score, selecting a possible item from the possible items, and based on the selected possible item and the acquired information on the item, generating an alert indicating abnormality of the item that is registered in the accounting machine.