Cloud-Mobile Image Processing for Supermarket Product Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition technologies for retail product identification in supermarkets suffer from accuracy issues, particularly in distinguishing visually similar products and high false-positive errors, and are hindered by battery life and data connectivity limitations in mobile device implementations.
Innovation Solution
Enhanced neural networks with training image augmentation techniques, parallel output sections for dynamic inventory management, and distributed recognition systems that leverage both mobile devices and cloud resources to optimize accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If image recognition is performed on mobile devices, then response speed is improved, but battery life is reduced
Solution Approach 1:
The image recognition system is segmented into two parts: a cloud-based deep neural network that performs complex calculations and a mobile device that captures images and displays results. This segmentation allows the computationally intensive tasks to be performed in the cloud, preserving mobile device battery life while maintaining fast response speeds through optimized cloud-compute architecture.
Solution Approach 2:
A cloud-based processing system serves as an intermediary between the mobile device camera and the final recognition results. The cloud system receives images from the mobile device, processes them through deep neural networks, and returns results, thereby offloading computational burden from the mobile device while maintaining responsiveness.
2Use of energy by moving object
If image recognition is performed in the cloud, then battery life is improved, but response speed and data connectivity requirements worsen
Solution Approach 1:
The system performs preliminary image capture and basic processing on the mobile device before transmitting to the cloud. The mobile device prepares the image data in an optimized format, reducing the amount of data that needs to be transmitted and processed in the cloud, thereby improving response speed while maintaining battery efficiency.
3Measurement precision
If conventional barcodes are used, then identification accuracy is improved, but checkout process complexity and time consumption worsen
Solution Approach 1:
The mechanical process of manually locating, orienting, and scanning barcodes is replaced with an automated image recognition system. The mobile device camera captures images of products, and the cloud-based neural network automatically identifies products without requiring manual barcode manipulation, thereby maintaining accuracy while significantly improving checkout speed.
Solution Approach 2:
The system changes the fundamental parameter of product identification from barcode-based (requiring specific geometric orientation and physical contact) to image-based recognition (tolerant of various orientations and distances). This parameter change eliminates the need for manual product manipulation while maintaining or improving identification accuracy.
4Device complexity
If cloud-based image recognition is used, then device complexity is reduced, but data connectivity and usage costs worsen
Solution Approach 1:
The complex deep neural network algorithms and computational resources are extracted from the mobile device and placed in the cloud. The mobile device only needs basic camera functionality and image processing capabilities, while the cloud system handles the complex recognition tasks, thereby reducing device complexity while managing data usage through efficient image compression and selective transmission.
Data Source
AI summary
Aspects of the detailed technologies concern training and use of neural networks for fine-grained classification of large numbers of items, e.g., as may be encountered in a supermarket. Mitigating false positive errors is an exemplary area of emphasis. Novel network topologies are also detailed—some employing recognition technologies in addition to neural networks. A great number of other features and arrangements are also detailed.