1. Introduction
This guide provides a detailed and comprehensive look at using and Price Reading Functionality. It presents best practices, assumptions and processes related to price reading, as well as the association of these prices with products on shelves. This document aims to facilitate understanding, to ensure the best results when using the functionality.
2. Price OCR Fundamentals
2.1 Accuracy Rate
- Expectation: Around 80% accuracy in correctly reading prices on labels.
- Conditions: This fee is based on the conditions detailed in the following sections.
2.2 Continuous Improvements
- Training: Price Reading offline was trained in a limited sample of stores and countries.
- Adjustments: We continue to make adjustments and new training to cover different situations.
- IA Online Reading: We include price reading online by AI to improve the accuracy and completeness of price recognition across different label formats and market conditions.
2.3 Methods for Reading Prices and Their Applications
2.3.1 Automatic in a Single Photo
Scenarios/Tag Types:
Recommended for captures containing labels within the traditional retail standard, with legible and well-defined characters.
Reading Flow:
Price classification occurs during image processing using Offline Reading.
2.3.2 Semi-Automatic with Individual Reading
Scenarios/Tag Types:
Suitable for promotional labels, multi-value labels and handwritten labels.
Reading Flow:
Price classification occurs automatically during image processing using Online Reading with AI.
Observation:
- Requires a good internet connection as processing is carried out online.
- The timeout for reconnection is 5 seconds.
2.3.3 Semi-Automatic for High Complexity
Scenarios/Tag Types:
Suitable for list labels and with small characters.
Reading Flow:
After initial image processing, price collection is carried out later on the price screen, for Online Reading with AI.
2.3.4 Manual Reading to Handle Exceptions and Failures
Usage Scenarios:
Applied when automatic or semi-automatic readings do not correctly identify prices or when capture errors occur.
Reading Flow:
The user captures and enters prices manually.
Intended for labels that are damaged, have poor legibility or are non-standard.
3. Best Conditions for Use
3.1 Standardized Label Models - Applicable to Price Reading offline
Description: Preference for labels within the traditional standard used in retail markets.
Ex:
4. Inappropriate Conditions for Use
4.1 Characters - Applicable to Price Reading offline
Size Difference: Large differences in character size negatively impact reading.
4.2 Decimal Separator
Visibility: Labels without a decimal separator or with the separator not clearly visible are difficult to read.
4.3 Image Quality - Applicable to all Price Reading models
Blurred Images: The precision of Price Reading drops significantly if the image is blurred.
4.4 Types of Labels
Promotional Posters: Posters or labels that are greater in height than their length.
Handwritten Prices: Handwritten price tags are not recommended.
5. Assumptions for Price Reading offline under specific conditions
5.1 Context
Text Size: The price must be significantly larger than other texts on the label.
Label Format: Labels that are larger than their width or handwritten prices are not read correctly.
6. Multi-Value Label Processing
6.1 Horizontal Labels offline
Price Reading: Only the lowest value found will be considered.
6.2 Vertical Labels offline
Treatment: Prices displayed vertically on the same label were not treated.
6.3 Horizontal Labels online
Price Reading: Defined according to the request made to AI
6.4 Vertical Labels online
Price Reading: Defined according to the request made to AI
7. Price Association for Shelf Products
7.1 Association and Validation Process
- Criteria: Association based on the shortest distance between product and label.
- Order: Association occurs shelf by shelf.
7.2 Process Steps
- Products: P1 and P2
- Available Tags: T1
- Association:
- P1 → T1 (Nearest label)
- P2 → T1 (Same label used for P1)
7.2.2 Second Shelfa
- Products: P2, P2 and P1
- Available Labels: T2 and T3
- Association:
- P2 → T2 (Nearest Tag)
- P2 (second) → T2 (shares the same label)
- P1 → T3 (Closest label)
7.2.3 Third Shelf
- Products: P3, P2 and P1
- Available Labels: T4 and T5
- Association:
- P3 → T4 (Closest label)
- P2 → T5 (Closest label)
- P1 → T5 (Shares the same label)
7.3 Last Validation
- Criteria:
- When there are equally close tags, the last one checked will be associated with the product.
- Associations always occur over shorter distances.
- In cases of equal proximity, the last validated one is chosen.
8. Premises
To ensure effective use of Price reading, it is imperative;
- Maintain a high product recognition rate, with a recognition recall of over 85%. This percentage ensures that the reading and association of prices are accurate and reliable.
- Strictly follow the best practices for using the IRE product every day, as described in the manuals. Any deviation from these practices may compromise the expected Price Reading performance. Additionally, keeping the system up to date with the latest improvements and technical recommendations provided by the product is crucial to continued performance.
- New packaging must be correctly registered in the system, respecting the guidelines of the Best practice guide, to ensure accurate price readings. Carrying out periodic reviews to ensure that all packages are correctly registered and updated in the system is a recommended practice.
- Price tags must always be positioned close to the respective products to ensure correct association. It is equally important to ensure that labels are visible and legible, avoiding positions that make reading difficult.
- Price tags must be in good condition, with no physical damage or dirt that could affect reading. Additionally, labels must have legible and clearly defined characters to ensure accurate reading.
- Maintaining the image capture environment with adequate lighting is essential to avoid shadows and reflections that could compromise label reading. Keeping the capture location clean and organized is also an essential practice to avoid visual interference when reading labels.
- Conduct regular training with prosecutors to ensure everyone is familiar with best practices and correct procedures.