Emergency Image Hazard Detection Using CNN and Relational Reasoning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Images sent during emergency incidents to PSAPs may be of poor quality or contain indistinguishable objects, leading to missed critical information about hidden dangers.

Innovation Solution

A computer-implemented method using Convolutional Neural Networks (CNN) and Relational Networks (RN) with Long-Short-Term-Memory (LSTM) architectures for object recognition and classification, combined with deep learning algorithms, to identify and alert about secondary hazardous situations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If images are sent from emergency location to PSAP, then visual information is provided to assist emergency dispatchers, but the images may be of poor quality or contain indistinguishable objects leading to missed critical information

Engineering Contradiction:
Improvecritical information about hidden dangersVSAvoidimage quality
Core Design Contradiction:
Loss of informationVSMeasurement precision

Solution Approach 1:

A deep learning processing system acts as an intermediary between the emergency scene and the PSAP dispatcher. The system receives images from the emergency location, applies CNN for object detection and RN-LSTM for relational reasoning to identify hidden dangers, then presents processed information to the dispatcher. This intermediary processing compensates for poor image quality by extracting meaningful information that would otherwise be indistinguishable to human eyes.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If deep learning algorithms are applied to process emergency images, then hidden dangers are identified, but processing time and computational complexity increase

Engineering Contradiction:
Improveidentification of hidden dangersVSAvoidemergency response time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The system performs preliminary processing by detecting objects and pre-processing images before full analysis. The CNN identifies candidate objects first, then the RN-LSTM applies relational reasoning only to these detected objects rather than the entire image. This staged approach reduces computational complexity and processing time while still identifying hidden dangers effectively.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4036791B1Computer-implemented method of handling an emergency incident, communication network, and emergency processing unit
Publication Date: 2026.03.04 ATOS PUBLIC SAFETY LLC
  • EP4036791B1 patent drawingFigure 1A~1B
  • EP4036791B1 patent drawingFigure 2
  • EP4036791B1 patent drawingFigure 3

AI summary

The present invention relates to a computer-implemented method of handling an emergency incident reported to a Public Safety Answering Point PSAP (4) the method comprising the steps of receiving, at the PSAP (4), information on an emergency incident, the information comprising at least one image (1) of the emergency incident, the emergency incident being a primary hazardous situation; applying a Convolutional Neural Network, CNN, object recognition and classification process for identifying and marking objects (2) on the at least one image that are related to the emergency incident and that may cause at least one secondary hazardous situation; processing the data relating to the identified and marked objects (2) by applying a deep learning algorithm to the data in a Relational Network, RN architecture, in particular, a Long-Short-Term-Memory, SLTM, architecture, wherein the image (1),on the basis of the identified and marked objects (2), is correlated to a set of recognized objects in a database that are essential for classifying the emergency. Further, the present invention relates to a communication network (4) according to claim 8, wherein the emergency processing unit (8) adapted for handling emergency incidents, the communication network comprising an ESInet via which information on emergency incidents are transmitted to a PSAP (7), wherein the communication network (4) further comprises an emergency processing unit (8) adapted to carry out the computer-implemented method of handling an emergency incident. Further, the present invention relates to an emergency processing unit (8) comprising a deep learning unit (3) adapted to carry out the method of handling an emergency incident.