Self-Checkout Non-Scan Detection via Camera and POS Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional self-checkout POS terminals are vulnerable to product loss due to customers intentionally or unintentionally failing to scan items, with existing measures like overhead cameras being susceptible to manipulation, leading to concealed non-scanning activities.
Innovation Solution
A system comprising a scanner, a video camera disposed perpendicularly to the scanner, proximity sensors to define an Area of Action, and an artificial neural network to detect scan and non-scan events by correlating video data with POS data, generating alerts for non-scan events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If overhead cameras are used to monitor scanning activity, then detection capability is improved, but the system becomes susceptible to manipulation and false detection
Solution Approach 1:
The patent introduces an intermediary processing system that combines multiple data sources (camera images, POS data, weight sensor data) rather than relying on a single overhead camera. This intermediary layer analyzes correlations between different data streams to detect non-scan events, making the system resistant to manipulation of any single component.
Solution Approach 2:
The system changes the parameters of detection by using multiple sensing modalities (visual, positional, weight) rather than relying solely on visual detection. This multi-parameter approach allows the system to cross-validate information and detect discrepancies that indicate manipulation or non-scan events.
2Device complexity
If traditional scanning monitoring is used, then system simplicity is maintained, but non-scan events go undetected leading to product loss
Solution Approach 1:
The patent merges multiple existing components (scanner, camera, proximity sensors, weight sensors, POS system) into an integrated monitoring system. By combining these components and their data streams, the system achieves reliable non-scan detection without requiring entirely new complex hardware, thus balancing complexity and effectiveness.
Solution Approach 2:
The system makes existing components serve multiple functions: the camera not only monitors scanning but also tracks item movement; the weight sensor not only measures product weight but also detects when items are placed in the bagging area; the proximity sensor not only detects item presence but also triggers video capture. This multi-functionality improves reliability without adding dedicated components for each function.
3Measurement precision
If video capture is continuous, then monitoring coverage is maximized, but processing load and energy consumption increase
Solution Approach 1:
The system uses periodic action by triggering video capture only when proximity sensors detect that an item has entered the scanning area. This event-driven periodic capture maximizes monitoring coverage at critical moments while avoiding continuous recording, thereby reducing energy consumption and processing load significantly.
Solution Approach 2:
The proximity sensors perform preliminary detection to identify when an item enters the scanning area before triggering the video camera. This preliminary action ensures that video capture is activated only when necessary, optimizing both monitoring coverage and energy efficiency by avoiding unnecessary continuous recording.
Data Source
AI summary
A system for detecting scan and non-scan events in a self-check out (SCO) process includes a a scanner for scanning objects and generating point of sale (POS) data, a video camera for generating a video of the scanning region, proximity sensors proximal to the video camera for defining an Area of Action (AoA), wherein the video camera starts capturing scanning region, when the objects enter the AoA, and the POS data includes non-zero values, an Artificial neural network (ANN) for receiving an image frame and generating one or more values, each indicating a probability of classification of the image frame into one or more classes respectively, and a processing unit for processing the POS data, and probabilities of one or more classes to detect a correlation between video data and POS data, and detect one of: scan and non-scan event in the image frame based on the correlation.