Neural-Network Identification Codes for Long-Range Warehouse Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Barcode readers in industrial trucks are limited in range, making it difficult to detect and identify objects at distances greater than 4 to 5 meters, which is necessary for effective control of forklift trucks in warehouses.

Innovation Solution

A data processing device equipped with a neural network that generates adversarial patches, enhancing the detectability of identification codes like barcodes and QR codes, allowing recognition at greater distances by training the network to process images in reverse and adding these patches to the codes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a barcode reader is used to detect identification codes, then object identification is enabled, but the detection range is limited to short distances (4-5 meters)

Engineering Contradiction:
Improveidentification code detection accuracyVSAvoiddetection distance
Core Design Contradiction:
Measurement precisionVSLength of stationary object

Solution Approach 1:

The patent modifies the identification code structure by adding adversarial patches that change the visual parameters of the code. These patches alter the optical characteristics of the barcode or QR code, enabling the neural network to recognize the code from greater distances while maintaining identification accuracy. The adversarial patches transform the appearance of the identification code without compromising its encodable information.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If conventional identification codes are used, then objects can be identified, but the codes are not easily detectable by neural networks at great distances

Engineering Contradiction:
Improveobject identification reliabilityVSAvoidneural network detection difficulty
Core Design Contradiction:
ReliabilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent inverts the traditional approach by designing identification codes specifically optimized for neural network detection rather than human reading. Adversarial patches are generated by processing identities through the neural network in reverse, creating codes that are inherently suited for machine recognition. This inversion makes the codes more detectable by neural networks while maintaining their identification function.

Inventive Principle:
Principle #13The other way round (Inversion)

3Measurement precision

If the truck approaches closer to read the barcode, then detection accuracy improves, but the operational efficiency and control capability are reduced

Engineering Contradiction:
Improvebarcode reading accuracyVSAvoidtruck operational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-modifying the identification codes with adversarial patches before the truck approaches. The codes are designed in advance to be detectable from a distance, eliminating the need for the truck to slow down or position itself precisely for reading. This allows the truck to maintain its approach trajectory and speed, improving operational efficiency while ensuring accurate detection.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4672164A1Object identification apparatus, method and system
Publication Date: 2025.12.31 STILL GMBH
  • EP4672164A1 patent drawingFigure 1
  • EP4672164A1 patent drawingFigure 2
  • EP4672164A1 patent drawingFigure 3a~3c

AI summary

The invention relates to a data processing device (110) for identifying objects (141; 143), particularly in a warehouse. The data processing device (110) comprises an interface (113) configured to receive a plurality of images of a plurality of optically detectable identification codes for identifying a respective object (141; 143). Furthermore, the data processing device (110) comprises a processor (111) configured to implement a neural network, wherein the neural network is trained to identify, in a forward processing direction, a respective optically detectable training identification code based on an image of the optically detectable training identification code.The processor (111) is further configured to process the neural network with a multitude of identities as input in a reverse processing direction in order to generate as output the multitude of optically detectable identification codes for identifying a respective object. Furthermore, a corresponding procedure for identifying a respective object (141; 143) is described.