Self-Checkout Fraud Detection Through Multi-Event Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-checkout systems using AI image analysis for fraud detection face high false detection rates, leading to customer complaints due to non-fraudulent acts being mistakenly identified as fraudulent.
Innovation Solution
A fraudulent act detection device for cash register systems that includes a storage unit, communication interface, and processor, which processes camera images and event signals to detect reportable events, sets flag indicators, and outputs a report notification if the number of detected events exceeds a threshold, reducing false detections by displaying a warning to customers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AI image analysis is used to detect fraudulent acts at self-checkout systems, then fraud detection capability is improved, but false detection rate increases
Solution Approach 1:
The detection system segments fraud detection into multiple independent analysis dimensions: camera image analysis, weight sensor data, barcode scanning records, and transaction logs. Each dimension independently evaluates specific aspects of customer behavior, and only when multiple dimensions collectively indicate fraud is a report generated. This segmentation reduces false positives by requiring corroboration across multiple data sources rather than relying on a single AI image analysis.
Solution Approach 2:
The system introduces an intermediary analysis layer that processes and correlates data from multiple sources before generating fraud reports. This intermediary layer includes a weight verification module that compares scanned item weights with actual bagged item weights, and a transaction correlation module that cross-references camera events with point-of-sale system records. This intermediary processing acts as a filter to eliminate false detections before they reach the customer.
2Reliability
If fraud detection sensitivity is increased to catch more fraudulent acts, then fraud prevention effectiveness is improved, but customer satisfaction deteriorates due to false accusations
Solution Approach 1:
The system dynamically adjusts detection sensitivity and reporting thresholds based on contextual factors such as customer history, time of day, and transaction patterns. Rather than applying a fixed high-sensitivity threshold that triggers frequent false alarms, the system adapts its detection criteria in real-time, increasing sensitivity for high-risk scenarios while maintaining lower thresholds for low-risk customers, thereby balancing fraud prevention with customer satisfaction.
Solution Approach 2:
The system changes multiple parameters simultaneously to optimize the balance between fraud detection and customer experience: it adjusts the number of required corroborating events before reporting, modifies weight tolerance thresholds, and varies the strictness of image analysis criteria based on the specific transaction context. These parameter changes allow the system to maintain high fraud detection effectiveness while minimizing false accusations against legitimate customers.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to one embodiment, a fraudulent act detection device for cash register systems has a processor that receives images from a camera to acquire images of a customer at a point-of-sale terminal and event signals corresponding to operations of the customer at the terminal. The processor processes a camera image to detect a camera event corresponding to operations at the point-of-sale terminal, stores a terminal event indicated in the received event signals in an event log of the storage unit, then detect reportable events of different types related to possible fraudulent acts of the customer. The processor sets a flag indicator for each of different types of act to indicate the detection of reportable events and output a report to the point-of- sale terminal if the number of reportable events exceeds a reporting threshold value. The report causes the terminal to display a warning to the customer.