Unidentified Baggage Identification Using Multi-View Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Airline baggage identification systems face challenges in identifying lost luggage without tags, due to the high volume of similar-looking bags and the difficulty in quickly capturing distinct characteristics, leading to costly and inefficient retrieval processes.

Innovation Solution

A method using convolutional neural networks to classify unidentified baggage by obtaining multiple images from different viewpoints and data types (visible, infrared, and 3D) to determine the probability of belonging to a specific class, followed by a search procedure in a database to match with potential candidates and identify the owner.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional manual identification methods are used for lost luggage, then operators can examine each bag individually, but the process becomes extremely time-consuming and costly given the high volume of similar-looking bags

Engineering Contradiction:
Improveidentification accuracyVSAvoidretrieval time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces manual mechanical examination of luggage with an automated image processing system using convolutional neural networks. Multiple images of luggage are captured and processed through CNN algorithms that automatically extract features and compare them against a database, eliminating the need for operators to manually examine each bag while maintaining high identification accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system creates digital copies of luggage through multiple images captured from different viewpoints and data types (visible, infrared, 3D). These digital copies are stored in a database and used for automated comparison and identification, allowing rapid matching without physical handling of the actual luggage.

Inventive Principle:
Principle #26Copying

2Measurement precision

If multiple images from different viewpoints and data types are obtained for each bag, then the system can capture comprehensive characteristics for accurate identification, but the data processing complexity and storage requirements increase

Engineering Contradiction:
Improveluggage classification accuracyVSAvoidimage processing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex identification task into distinct processing stages: image acquisition from multiple viewpoints and data types, feature extraction through convolutional neural networks, probability calculation for each baggage class, and final matching against the database. This segmentation allows the system to handle complex multi-dimensional data systematically while maintaining accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system transforms physical luggage characteristics into multiple data representations by capturing images in different spectral domains (visible, infrared) and dimensional formats (2D, 3D). The convolutional neural networks then transform these images into probability distributions across baggage classes, changing the parameter space to make identification more robust and accurate.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3712807B1Method for identifying luggage
Publication Date: 2021.12.01 IDEMIA IDENTITY & SECURITY FRANCE SAS
  • EP3712807B1 patent drawingFigure 1~2
  • EP3712807B1 patent drawingFigure 3~5
  • EP3712807B1 patent drawingFigure 6A~6B

AI summary

Method for identifying unidentified baggage based on: acquiring at least two images of the unidentified baggage of different types, the type of image of baggage depending on a viewpoint on said baggage in said image and/or on a nature of the data representing said baggage in said image taken from a plurality of different data types including data representing the visible domain and/or infrared data and/or data representing three-dimensional information; and, for each different type of image, using a neural network adapted to said type to classify the baggage represented by said image into a class of a plurality of baggage classes defined for said type.Once obtained, the classes allow for searching for luggage matching each identified class in a luggage database, with each piece of luggage in that database associated with a recipient. Each matching piece of luggage is then compared to an unidentified piece of luggage to identify the recipient.