Self-Service Terminal Manipulation Detection Using Dual Classifier Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting manipulation attempts at self-service terminals, such as ATMs, require high hardware and software complexity to achieve reliable results, often leading to false alarms from personal items left on the control panel.
Innovation Solution
A method using a parallel structure of two classifiers to evaluate image data from cameras, where the first classifier detects changes and the second classifier checks the plausibility of these changes, ensuring that an alarm is only triggered if both classifiers confirm a manipulation attempt, thereby reducing false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single classifier is used to detect manipulation attempts, then the detection speed is fast, but the reliability is low and false alarms occur frequently
Solution Approach 1:
The detection system is segmented into two independent classifiers with distinct functions: a first classifier for detecting manipulation attempts and a second classifier for verifying plausibility. This segmentation allows each classifier to specialize in its specific task, improving overall reliability while keeping individual classifier complexity manageable.
Solution Approach 2:
The second classifier acts as an intermediary verification layer between the first classifier's detection and the final alarm trigger. It evaluates whether detected changes are plausible manipulation attempts or false alarms, thereby improving reliability without requiring the first classifier to be overly complex.
2Reliability
If multiple sensors and complex classifiers are used to improve detection reliability, then the reliability improves, but the hardware complexity and cost increase
Solution Approach 1:
The system uses a single camera performing multiple functions: capturing images for both manipulation detection and plausibility verification. This multi-functionality approach achieves high reliability without requiring multiple different sensors, thereby reducing hardware complexity and cost.
Solution Approach 2:
The software is segmented into two classifiers with different evaluation criteria. The first classifier focuses on detecting changes in operating elements, while the second classifier evaluates contextual plausibility. This software segmentation achieves high detection reliability without proportionally increasing hardware complexity.
3Reliability
If traditional surveillance systems with multiple cameras are used, then comprehensive monitoring is achieved, but the system complexity and cost increase
Solution Approach 1:
A single camera is configured to capture images that serve dual purposes: monitoring for manipulation attempts and providing contextual information for plausibility assessment. This eliminates the need for multiple cameras while maintaining comprehensive surveillance reliability.
Solution Approach 2:
Instead of adding more cameras (spatial dimension), the system adds a temporal and analytical dimension by using two different evaluation approaches (classifiers) on the same image data. This transforms the problem from hardware multiplication to software intelligence, reducing physical complexity while maintaining reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method (100) for detecting and verifying attempts to manipulate a self-service terminal which has at least one operating element (for example a keyboard) for users, onto which at least one camera is directed. The generated image data (IN) are supplied to a first classifier (CF) that checks, on the basis of the image data, whether a manipulation attempt on the operating element, for example keyboard, can be detected. Moreover, simultaneously with being supplied to the first classifier (CF), the image data (IN) are also supplied to a second classifier (SC) that checks, on the basis of the image data, whether the detection of a manipulation attempt is plausible. Only when both classifiers give a positive indication, that is confirm a manipulation attempt and the plausibility thereof, is an alarm triggered. It is thus possible to prevent false alarms, which occur with conventional self-service terminals when users or customers leave personal items (for example wallets) on the control panel.