ATM Tampering Detection via Adaptive Scene Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to detect specific tampering activities at ATMs, such as the installation of fake card readers or small wireless cameras, as they do not utilize domain-meaningful markers to select a reference scene model for change detection, leading to significant financial losses due to fraudulent transactions.
Innovation Solution
A video-based system that continuously learns the appearance of an ATM, using cameras to create a profile view and detect changes in real-time, with action markers for transaction start and end to adapt the scene model and suspend tampering detection during transactions, allowing for reliable identification of unauthorized alterations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general change detection systems are used at ATMs, then scene changes can be detected, but specific tampering activities cannot be identified due to lack of domain-meaningful markers
Solution Approach 1:
The system changes the parameter of scene model selection by using domain-meaningful markers (transaction start/end markers) to select appropriate reference scene models. This allows the system to adaptively choose the most relevant baseline for comparison, improving detection accuracy for specific tampering activities without requiring overly complex algorithms.
Solution Approach 2:
The patent introduces domain-meaningful markers as an intermediary element that bridges general scene changes and specific tampering detection. These markers (annotating transaction phases) serve as a mediator to select appropriate reference models, enabling the system to focus on relevant changes during specific transaction phases without increasing overall system complexity.
2Reliability
If continuous scene monitoring is performed without transaction markers, then all changes are captured, but false alarms increase due to non-tampering changes
Solution Approach 1:
The system segments the monitoring process into distinct phases using transaction markers (start and end markers). By dividing continuous monitoring into transaction-specific segments, the system can apply detection logic selectively, improving reliability by focusing on relevant phases while maintaining processing efficiency through targeted analysis.
Solution Approach 2:
The patent implements periodic action by suspending tampering detection during transaction periods (marked by domain-meaningful markers) and activating it during non-transaction periods. This periodic activation pattern reduces false alarms from transaction-related changes while maintaining detection capability during critical non-transaction phases, balancing reliability and efficiency.
3Measurement precision
If domain-meaningful markers are used to select reference scene models, then specific tampering can be detected, but the system requires complex marker-based model selection
Solution Approach 1:
The system uses domain-meaningful markers to change the parameter of reference model selection. By translating marker information into model selection criteria, the system achieves precise tampering detection while keeping the complexity manageable through a systematic parameter-based selection approach rather than ad-hoc complex algorithms.
Data Source
AI summary
A system and method of detecting tampering at an automatic teller machine includes detecting start and end indicators of a transaction. A representation of a scene at the teller machine, prior to the start of the transaction can be compared to a representation of the scene after the end of the transaction. Variations therebetween can indicate tampering at the machine.