Multi-Modal LLM Object Detection From Environment Images

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Images of environments are often stored without being utilized to determine information about objects within them, requiring manual assessment of object conditions and positions, which is inefficient and prone to errors.

Innovation Solution

Utilizing a trained multi-modal large language model to identify and locate objects within images, providing a record of their positions and conditions, and querying this record to automate monitoring and alerting for maintenance or repairs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual assessment is used to determine object conditions and positions, then flexibility in handling diverse objects is maintained, but efficiency and accuracy deteriorate due to manual labor requirements

Engineering Contradiction:
Improveefficiency of object monitoringVSAvoidcomplexity of automated monitoring system
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

A multi-modal large language model serves as an intermediary between image data and monitoring decisions. The model processes images, identifies objects, determines positions and conditions, and generates structured outputs that can be queried to provide monitoring information, eliminating the need for complex custom processing systems for each object type

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The multi-modal large language model provides universal functionality for monitoring diverse objects across different environments. A single model handles various object types (telecommunication equipment, street furniture, vegetation) and environments (indoor, outdoor, computing centers) through its general capabilities rather than requiring specialized systems for each object category

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Loss of information

If images are stored without automated processing, then storage simplicity is maintained, but loss of information occurs as object information is not extracted

Engineering Contradiction:
Improveobject information extractionVSAvoidcomplexity of image processing system
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The multi-modal large language model automatically extracts object information from images and generates structured records without requiring manual intervention. The system self-processes images, identifies objects, determines positions and conditions, and creates queryable records, transforming passive stored images into active information sources

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Manual information extraction is replaced by automated multi-modal processing. The large language model uses its trained capabilities to analyze images and generate structured outputs, substituting the mechanical process of manual assessment with automated AI-based processing

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

3Loss of time

If manual monitoring is performed, then adaptability to unique object characteristics is maintained, but time consumption increases significantly

Engineering Contradiction:
Improvetime for object monitoringVSAvoidreliability of object condition assessment
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The system enables continuous monitoring by automatically processing images and maintaining updated records of object positions and conditions. The queryable records allow continuous retrieval of monitoring information without requiring repeated manual assessment, providing ongoing reliable information about object states

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4618041A1A method for processing images to identify objects within an environment
Publication Date: 2025.09.17 BRITISH TELECOM PLC
  • EP4618041A1 patent drawingFigure 1
  • EP4618041A1 patent drawingFigure 2
  • EP4618041A1 patent drawingFigure 3

AI summary

A computer implemented method is defined. The method comprises receiving an image of an environment; providing the image to a trained multi-modal large language model trained to identify one or more objects within the image and a position of each of the one or more objects within the image and output text identifying the one or more objects and the positions of the one or more objects; receiving from the multi-modal large language model the output text; forming a record of the environment using the output text of the multi-modal large language model, wherein the record of the environment comprises an identification of the one or more objects in the environment and the position of the one or more objects in the environment; and querying the record to determine an identity and/or a position of an object of the one or more objects.