Search

CN-122024447-A - Lightweight fire detection and emergency evacuation device, method and storage medium

CN122024447ACN 122024447 ACN122024447 ACN 122024447ACN-122024447-A

Abstract

The invention discloses a medium and small place fire detection method which comprises the following steps of performing medium and small place scene modeling and equipment deployment, completing scene digital modeling and light-weight equipment deployment, generating a scene topological graph, wherein a light-weight AI fire detection model adopts MobileNetV as a basic backbone network, the model parameter amount is 1.8M, training and deployment of the light-weight AI fire detection model, and the light-weight AI fire detection model further comprises a feature fusion module, wherein the feature fusion module extracts RGB channel difference values of flame colors, smoke texture LBP features, regional temperature average values and mutation rates for feature fusion, real-time fire detection and risk assessment, dynamic evacuation route planning and updating, multi-dimensional emergency warning and evacuation guiding and fire data analysis and detection model updating. The invention can realize the extremely early and accurate detection of fire disaster in small and medium places, the rapid transmission of alarm information and the intelligent guiding of evacuation routes, and meets the core requirements of low cost, easy deployment and high reliability in small and medium places.

Inventors

  • DENG XING
  • XU JIAN

Assignees

  • 深圳沃特华安全技术有限公司

Dates

Publication Date
20260512
Application Date
20260224

Claims (10)

  1. 1. The fire detection method for the middle and small places is characterized by comprising the following steps of: Step 101, modeling and equipment deployment of a small and medium place scene are completed, and scene digital modeling and lightweight equipment deployment are completed to generate a scene topological graph; step 102, training and deploying a lightweight AI fire detection model, constructing the lightweight AI model suitable for edge-end operation, and realizing multi-feature accurate identification of extremely early fire; step 103, real-time fire detection and risk assessment, wherein an edge computing terminal processes camera image data in real time to complete fire identification and risk level judgment; 104, dynamically planning and updating an evacuation route, and generating an optimal evacuation route by adopting an improved path planning algorithm based on a scene topological graph and real-time fire data; Step 105, multi-dimensional emergency warning and evacuation guiding, namely realizing warning and guiding through the multi-dimensions of vision, hearing and mobile terminals by means of deployed lightweight equipment to form closed loop response; And 106, analyzing fire data, updating a detection model, collecting fire data for analysis after evacuation is completed, transmitting the fire data to a lightweight AI fire detection model for training and updating, and improving the accuracy of fire detection.
  2. 2. The medium-small fire detection method according to claim 1, wherein the lightweight AI fire detection model adopts MobileNetV as a basic backbone network, and the model parameter is 1.8M.
  3. 3. The medium-small fire detection method of claim 2, wherein the lightweight AI fire detection model further comprises a feature fusion module, wherein the feature fusion module extracts RGB channel differences of flame colors, smoke texture LBP features, and region temperature mean and mutation rate for feature fusion.
  4. 4. The medium-small fire detection method according to claim 2, wherein the training accuracy of the lightweight AI fire detection model is not less than 99.2%, and the model volume is further compressed to 0.8MB by INT8 quantization compression.
  5. 5. The medium-small fire detection method according to claim 1, wherein the path planning calculates the path cost using a formula p=d× (1+Δ), wherein P is the path cost, D is the physical distance, and Δ is the risk level coefficient, wherein the level 1 risk coefficient Δ is 0.1,2, the level 3 risk coefficient Δ is 0.5, and the level 3 risk coefficient Δ is 1.0.
  6. 6. A fire evacuation method for small and medium places, which is characterized by comprising the following steps: step 201, fire identification is performed on medium and small places by adopting a light AI model to connect a camera and linkage equipment; Step 202, calculating a dynamic evacuation path, when an extremely early fire condition is identified, shooting and transmitting the extremely early fire condition to a lightweight AI model by a monitoring camera, further extracting smoke texture characteristics of pictures, simultaneously externally connecting temperature measuring points to monitor temperature change of a region, further acquiring continuous 10-frame images, detecting the spreading trend of a fire region, and generating an evacuation path; step 203, intelligent indicator lights turn to control the indicator lights to turn to according to the evacuation path output by the light AI model; Step 204, voice guidance is started, voice warning and mobile terminal APP information pushing are further started; step 205, carrying out diversion evacuation on indoor personnel according to the evacuation path; And 206, continuously monitoring the medium and small places, returning to the step 201, and continuously monitoring environmental parameters in the places by using the cameras and the lightweight AI model after the fire is effectively controlled.
  7. 7. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the medium-small fire detection method according to any one of claims 1 to 5.
  8. 8. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the medium-small fire evacuation method of claim 6.
  9. 9. A fire detection and emergency treatment device comprising: one or more processors; Memory, and One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, wherein the execution of the computer programs by the processors implements the steps of the medium-small venue fire detection method of claim 7.
  10. 10. A fire detection and emergency evacuation apparatus comprising: one or more processors; Memory, and One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, wherein the execution of the computer programs by the processors implements the steps of the medium-small venue fire evacuation method of claim 8.

Description

Lightweight fire detection and emergency evacuation device, method and storage medium Technical Field The invention relates to the technical field of fire prevention and control and emergency evacuation, in particular to a lightweight fire detection and emergency evacuation device, a lightweight fire detection and emergency evacuation method and a storage medium. Background On the one hand, traditional fire detection equipment such as a point smoke detector is easily interfered by oil smoke, dust, steam and the like, the false alarm rate is as high as 10% -15%, so that the situation personnel generate numbness to alarms, on the other hand, the traditional AI image fire detection system is mainly designed for large venues, has complex models, has more than or equal to 100M of parameter quantity, high hardware requirements, needs to deploy a special GPU server, has high deployment cost, has more than or equal to 5 ten thousand yuan investment, and is difficult to adapt to the requirements of low cost and easy maintenance of the small and medium places. The emergency evacuation in middle and small places is dependent on fixed indicator lamps and paper plans, has obvious limitations, the fixed indicator lamps cannot dynamically adjust evacuation routes according to fire positions and smoke spreading directions, people can be guided to enter dangerous areas easily, the paper plans lack of real-time performance, can not be quickly transmitted to present people when a fire occurs, and most of middle and small places are not provided with professional emergency broadcasting systems, alarm information is not timely and accurately transmitted, so that evacuation confusion and even safety accidents occur. The Chinese authorized patent number 202411514696X discloses a fire identification alarm method and a fire detector based on an AI image, the technical scheme utilizes a model library platform for centralized management, the platform side is responsible for collecting, establishing and maintaining a fire identification model library containing various scene adaptations, and the models are subjected to strict test and optimization, so that the fire identification model library has higher identification precision and generalization capability. When a user puts forward fire monitoring demands, the platform side can rapidly analyze, match and recommend the most suitable fire recognition model according to the demands of the user. The technical scheme cannot be flexibly suitable for various middle and small places, and the model library needs a long time to update, so that the flexibility is required to be improved. The existing light AI image recognition technology is still insufficient in application in the field of fire detection, is mostly recognized by adopting single characteristics (such as flame color and smoke outline) for pursuing model simplification, ignores the composite characteristics of weak smoke and abnormal temperature in early stage of fire, causes the extremely early detection accuracy to be less than or equal to 85 percent, lacks scene self-adaptation capability, has serious performance attenuation in different light (such as weak light at night and direct light at strong light) and shielding (such as shielding smoke of a goods shelf) scenes, and is not fused with the depth of an evacuation guiding link to form a detection-guiding split situation. In summary, the prior art cannot meet the requirements of small and medium places, and has the main defects of high deployment cost and poor compatibility of ① AI detection models, single characteristic dependence on ② fire identification, low extremely early detection precision, weak anti-interference capability, solidification in ③ evacuation guiding mode, incapability of dynamically adapting to fire scenes, ④ detection-warning-guiding process cutting, and low emergency response efficiency. Therefore, it is necessary to provide an integrated method that is lightweight, high-precision, low-cost, and compatible with detection and guidance. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a lightweight fire detection and emergency evacuation device, a lightweight fire detection and emergency evacuation method and a lightweight storage medium. In a first aspect, the invention provides a method for detecting fire in a medium or small place, comprising the following steps: Step 101, modeling and equipment deployment of a small and medium place scene are completed, and scene digital modeling and lightweight equipment deployment are completed to generate a scene topological graph; step 102, training and deploying a lightweight AI fire detection model, constructing the lightweight AI model suitable for edge-end operation, and realizing multi-feature accurate identification of extremely early fire; step 103, real-time fire detection and risk assessment, wherein an edge computing terminal processes camera image data in