Checkout Image Verification for Fast Ticket Switching Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision-based approaches for detecting ticket switching during self-checkout and cashier-assisted transactions are inefficient, taking too long to recognize discrepancies between scanned items and their actual prices, leading to intolerable transaction times and high false positive rates.
Innovation Solution
A hybrid machine-learning model using a modified Convolutional Neural Network (CNN) for general feature extraction, combined with item-specific classifiers, to rapidly and accurately identify potential ticket switching by analyzing item images and barcodes, with a feedback loop for rescanning when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing computer vision-based approaches are used to detect ticket switching, then detection capability is provided, but transaction time increases to intolerable levels and false positive rates become too high
Solution Approach 1:
The detection system is segmented into multiple specialized components: a CNN model for feature extraction from item images, and multiple item-specific classifiers (e.g., for bananas, apples, oranges) that each specialize in detecting specific item types. This segmentation allows each component to be optimized for its specific function, improving overall detection accuracy while maintaining fast processing speeds.
Solution Approach 2:
The system performs preliminary action by pre-training the CNN model and all item-specific classifiers before actual transaction processing. During transactions, the pre-trained models immediately process item images and barcodes without requiring real-time training, enabling fast detection while maintaining high accuracy. The feedback loop also performs preliminary correction by identifying false positives before they disrupt the transaction flow.
2Reliability
If existing computer vision-based approaches are used to detect ticket switching, then detection capability is provided, but false positive rates become too high causing transaction interruptions
Solution Approach 1:
Each item-specific classifier is trained with high-quality, item-specific data and features, giving it local expertise in detecting that particular item type. For example, the banana classifier is specialized with banana-specific features, making it highly accurate for bananas but not wasting computational resources on other items. This local quality optimization reduces false positives while maintaining detection capability.
Solution Approach 2:
The system implements a feedback loop where detection results are continuously monitored and used to adjust the system's operation. When false positives are detected, the feedback mechanism triggers rescanning or model retraining, correcting errors in real-time. This feedback ensures high detection capability is maintained while minimizing false positives that would interrupt transactions and reduce productivity.
3Measurement precision
If a hybrid machine-learning model with feedback loop is implemented, then detection precision and efficiency are improved, but system complexity increases
Solution Approach 1:
The complex hybrid model is segmented into distinct, manageable modules: a CNN backbone for feature extraction, multiple item-specific classifier modules, and a feedback control module. Each module has a specific, well-defined function, making the overall complex system easier to train, deploy, and maintain. The segmentation allows independent optimization of each component without requiring complete retraining of the entire system.
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
A machine-learning algorithm is trained on images with a set of diverse items to produce as output feature vectors in a feature-vector space derived for the set. New item images for new items are passed to the algorithm and new feature vectors are projected into the vector space. A classifier for each new item is trained on the new feature vectors to determine whether the new item is new item or is not that new item. During a transaction, an item code scanned for an item and an item image are obtained. The item image is passed to the algorithm, a feature vector is obtained, a corresponding classifier for the item code is retrieved, the feature vector is passed to the classifier, and a determination is provided as to whether the item image and item code matches a specific item that should be associated with the item code.