Purchase Auditing With ML Item Matching on Standard Cameras
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional customer purchase auditing methods in stores are inefficient and costly, often leading to customer frustration and are not easily implementable in smaller retailers due to the need for high-resolution cameras.
Innovation Solution
Implementing a system using computer vision with machine learning models on existing lower resolution cameras to identify items and match them to transaction receipts, allowing for discrepancy audits at self-service kiosks or mobile devices, with optional higher resolution verification by audit stations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high resolution cameras are used to monitor customers and verify purchases, then measurement precision of item identification is improved, but device complexity and cost increase significantly
Solution Approach 1:
The patent uses image copying and processing techniques where multiple images are captured from different angles and combined to create a composite view. This allows standard cameras to achieve high-resolution identification capability without requiring expensive specialized hardware. The system captures images at regular resolutions and uses computational methods to enhance identification accuracy.
Solution Approach 2:
The patent makes the auditing system universally applicable to different store types and camera configurations. The system can operate with standard cameras already present in most stores, eliminating the need for specialized high-resolution camera infrastructure. The machine learning model adapts to various camera qualities and resolutions, making the solution accessible to both large and small retailers.
2Measurement precision
If manual auditing by employees is performed, then measurement precision of purchase verification is improved, but productivity decreases due to slow processing
Solution Approach 1:
The patent replaces the mechanical manual auditing process with an automated computer vision system. Machine learning models automatically analyze images, identify items, compare them against transaction data, and detect discrepancies without human intervention. This substitution maintains verification accuracy while dramatically increasing processing speed and eliminating customer wait times.
Solution Approach 2:
The system enables self-service auditing where the computer vision system autonomously performs purchase verification without requiring employee involvement. The automated system handles image capture, analysis, comparison, and discrepancy detection independently, freeing employees from manual auditing tasks while maintaining continuous operation at high speed.
3Device complexity
If existing camera systems are used instead of high resolution cameras, then device complexity is reduced, but measurement precision of item identification deteriorates
Solution Approach 1:
The patent changes the parameters of image processing by capturing multiple images at different angles, resolutions, and time points. Instead of relying on high camera resolution, the system uses computational parameter adjustments including image enhancement, noise reduction, and feature extraction algorithms to achieve accurate item identification from standard camera inputs.
Solution Approach 2:
The patent creates a composite auditing system that combines standard cameras with advanced machine learning processing. Rather than relying on a single high-resolution camera, the system integrates multiple standard cameras and combines their outputs through computational methods, achieving superior identification accuracy equivalent to or better than expensive specialized hardware.
Data Source
AI summary
Techniques for auditing purchases are provided. An item for purchase is scanned to determine a first identity of the item. The scanned item is added to a transaction. A first image of the scanned item is captured using a first camera. A second identity of the scanned item from the first image is determined using a trained machine learning (ML) model. Responsive to determining the first and second identities of the item do not match, an audit of the transaction is performed.