Self-Service Checkout Fraud Detection via Video Analysis

The patent replaces the mechanical weight sensor system with an optical imaging system using cameras and image processing algorithms. Instead of measuring physical weight, the system captures images of commodity products and uses computer vision to identify and verify products, thereby detecting fraud without the high cost and complexity of weight sensors.

The patent creates a visual copy (image) of the commodity product and compares it with the registered product information. By capturing an image of the product being scanned and analyzing its visual characteristics, the system verifies whether the scanned product matches the actual product, enabling fraud detection through information copying rather than physical measurement.

The patent develops a universal image recognition model that can identify multiple types of commodity products across different categories and stores. Instead of training separate AI models for each product type or store, the system uses a single multi-functional model that adapts to various products, thereby reducing training complexity and data requirements while maintaining high detection accuracy.

The patent changes the approach from extensive supervised training to a more efficient training methodology that uses fewer data samples. By modifying training parameters and using techniques such as transfer learning or few-shot learning, the system achieves high accuracy with reduced training data requirements, making it suitable for environments with limited product variety or short product lifecycles.

The patent implements a self-service fraud detection system where the imaging device automatically captures images and the processing unit automatically analyzes them without requiring user intervention. The system autonomously compares the captured product image with registered information and generates alerts for potential fraud, maintaining ease of operation while significantly improving detection reliability compared to manual verification methods.

The patent introduces a feedback mechanism where the system continuously monitors the scanning process, compares scanned products with actual products using image analysis, and provides real-time feedback through alerts when discrepancies are detected. This closed-loop feedback system maintains simple operation for users while reliably detecting fraudulent activities by automatically correcting or flagging errors.