Automated Consignment Sorting via Image Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically sorting and distributing consignments fail when characters in address fields are partially hidden or use non-Latin scripts, such as Arabic, Cyrillic, Greek, or Asian characters, as they cannot be accurately recognized or evaluated, necessitating manual intervention.
Innovation Solution
A method that captures image data of a consignment's surface using an image sensor, determines metadata representing image features through segmentation algorithms, and uses these features to determine distribution and sorting information independently of character recognition, enabling automated sorting and distribution by comparing image features with a database or neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If character recognition methods are used for automated sorting, then sorting automation is achieved for standard cases, but sorting fails when characters are hidden or use non-Latin scripts
Solution Approach 1:
The patent segments the address field recognition task into two independent parts: extracting visual features (shape, size, position, color) of address elements and characters separately, and then combining these features to determine sorting information. This segmentation allows the system to process addresses even when character recognition fails, by relying on the visual features of the address field as a whole.
Solution Approach 2:
The patent creates a universal sorting system that handles multiple types of address fields simultaneously - fully visible Latin script addresses, partially hidden addresses, and non-Latin script addresses - using a single feature extraction and combination approach. The system evaluates both the address field structure and character content together, making it adaptable to various address formats without requiring separate processing methods.
2Reliability
If manual evaluation is used for hidden or non-Latin addresses, then sorting accuracy is maintained, but productivity decreases due to manual intervention
Solution Approach 1:
The patent enables the system to automatically evaluate and sort all address types without human intervention. By combining visual feature extraction with character recognition results, the system independently determines sorting information for all consignments, including those with hidden or non-Latin characters that previously required manual evaluation. This self-service approach eliminates manual processing while maintaining accuracy.
3Measurement precision
If complete character capture is required for accurate sorting, then sorting precision is improved, but the system becomes vulnerable to hidden or corner-afixed addresses
Solution Approach 1:
The patent performs preliminary extraction of visual features (shape, size, position, color) of address fields and characters before attempting character recognition. By preparing these visual characteristics in advance, the system can compensate for incomplete character capture and still determine sorting information based on the pre-extracted features, making it robust against hidden or partially visible addresses.
Data Source
AI summary
A method performed by at least one apparatus is disclosed in which image data is obtained that represents an image of a surface of a consignment captured by an image sensor. At least partially on the basis of the image data, metadata associated with the image data is determined. The metadata represent a plurality of image features of the image represented by the image data. Each image feature of the image features represented by the metadata is an image component of the image represented by the image data. Distribution and/or sorting information is determined for the automated distribution and/or sorting of the consignment at least partially on the basis of the image features represented by the metadata.