Self-Checkout Fraud Detection Using Statistical Image Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection models for self-checkout registers exhibit unpredictable behaviors such as over-detection and under-detection when deployed in environments different from their training environment, and manually identifying target images for retraining is impractical due to high labor and cost.
Innovation Solution
A fraud detection program that automatically identifies target images for retraining by analyzing detection regions in images using statistical information and spatial consistency, enabling efficient adaptation of the object detection model to new environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of images is performed to identify target images for retraining, then identification accuracy can be achieved, but labor time and personnel expense increase significantly
Solution Approach 1:
The patent replaces the mechanical manual review process with an automated computer-based system that uses object detection models and statistical analysis to identify target images. The system automatically calculates statistical information about detection regions and identifies images with low appearance probabilities, eliminating the need for manual labor while maintaining identification accuracy.
Solution Approach 2:
The system enables self-service by allowing the object detection model to automatically identify its own training needs. Through statistical analysis of detection regions and appearance probabilities, the system autonomously determines which images should be added to training data without requiring external manual intervention or expert judgment.
2Reliability
If the object detection model is retrained with environment-specific images, then detection accuracy in new environments improves, but the complexity of the training process increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating statistical information about detection regions from a first image group before deploying the model to new environments. This statistical baseline is established in advance, enabling automatic identification of target images in new environments without requiring complex retraining procedures. The system prepares the statistical framework beforehand to simplify the adaptation process.
Solution Approach 2:
The system changes parameters by comparing appearance probabilities of detection regions against statistically derived thresholds. Instead of complex retraining, the system identifies target images by detecting parameter deviations (low appearance probabilities) relative to the statistical baseline, simplifying the adaptation process while improving detection accuracy in new environments.
3Quantity of substance
If a large number of images are manually reviewed to collect target images, then comprehensive training data can be obtained, but processing costs and time consumption increase
Solution Approach 1:
The patent replaces manual image review with an automated computational system that processes large numbers of images efficiently. The system uses object detection models and statistical analysis to automatically identify target images, enabling high-volume processing without the labor constraints of manual review. This substitution dramatically increases processing efficiency while maintaining comprehensive training data collection.
Solution Approach 2:
The system creates a statistical copy or representation of detection region patterns from the first image group. By working with this statistical model rather than manually reviewing each image, the system can efficiently process large volumes of images to identify target images, maintaining comprehensive training data collection while dramatically improving processing efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A fraud detection program causing a computer to execute a process including obtaining an object detection result by inputting a target image group including a self-checkout apparatus in an imaging range, into a model trained using a target image and an annotation, and performing fraud detection at the self-checkout apparatus based on information about an item registered thereto and the object detection result. The target image is identified by calculating statistical information of a position of a detection region of an object in each image in a first group based on positions by inputting the first group into the model, obtaining a position in each image in a second group using the model, and identifying the target image in which a region having an appearance probability equal to or less than a threshold is present, from the second group, based on the statistical information.