CN-121980310-A - Video monitoring-based fire operation overall process accurate control method and system

CN121980310ACN 121980310 ACN121980310 ACN 121980310ACN-121980310-A

Abstract

The invention relates to a video monitoring-based precise control method and a video monitoring-based precise control system for the whole process of a fire operation, and belongs to the technical field of fire operation safety management. The method comprises the steps of starting a parameter acquisition device, synchronously acquiring images and environment data of an operation site, analyzing the images, identifying illegal operation behaviors of operators, calculating the risk level of a current operation area by combining thermal imaging data and environment parameters, marking specific illegal behaviors and dangerous states by combining behavior identification and risk assessment results, outputting alarm information according to the specific illegal behaviors and dangerous states, carrying out cross-correlation analysis on the identified behaviors and abnormal marking results, mining causality relations among the recognized behaviors and abnormal marking results, predicting the change trend of risk indexes by using a machine learning model based on historical and real-time monitoring data, and carrying out hierarchical sequencing on abnormal alarms according to the results of event correlation analysis and risk trend prediction, so as to realize intelligent research, judgment and closed-loop management and control on the fire operation risks.

Inventors

LIN WEI
Yang Sanpeng
YE WEI
CHEN XIANGYAN
YANG MINZHAO
Fan Chenlong
LIU LIFENG

Assignees

华能(福建)能源开发有限公司福州分公司

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. The precise control method for the whole process of the fire operation based on video monitoring is characterized by comprising the following steps of: The method comprises the steps of collecting RGB images, infrared thermal imaging images and environmental parameters in a fire operation monitoring area, carrying out pixel level alignment on the RGB images and the infrared thermal imaging images by utilizing a homogeneous coordinate transformation principle and a preset homography matrix to obtain aligned infrared thermal imaging temperature images, constructing a double-branch convolution neural network, respectively extracting appearance semantic features of the RGB images and temperature distribution features of the infrared thermal imaging temperature images, and carrying out fusion by utilizing a channel attention mechanism to obtain a fusion feature map; Performing target detection on the fusion feature map, identifying the current operation action, and matching the current operation action with a preset fire operation safety specification knowledge base to generate illegal action alarm information; comparing the temperature distribution characteristics and the environmental parameters with preset thresholds to generate an abnormal marking result, calculating comprehensive risk scores based on the offence alarm information and the abnormal marking result, inputting the comprehensive risk scores into a time sequence prediction model to obtain a risk change trend prediction value, inputting the current operation action and the abnormal marking result into a causality discovery algorithm to obtain a causality association analysis result, and grading and sequencing the offence alarm information and the abnormal marking result according to the causality association analysis result and the risk change trend prediction value to obtain a processing priority list; And executing a corresponding treatment strategy according to the treatment priority list, and outputting corresponding alarm or prompt information.
2. The video monitoring-based precise control method for the whole process of the fire operation, which is characterized in that the method utilizes a homogeneous coordinate transformation principle and a preset homography matrix to carry out pixel level alignment on an RGB image and an infrared thermal imaging image, and is specifically as follows: Expanding pixel coordinates of the RGB image and pixel coordinates of the infrared thermal imaging image into a three-dimensional column vector form by utilizing a homogeneous coordinate transformation principle; and establishing a pixel mapping relation between the RGB image and the infrared thermal imaging image based on a preset homography matrix, so that an absolute temperature value corresponding to each three-dimensional column vector form pixel coordinate in the infrared thermal imaging image is aligned with the corresponding three-dimensional column vector form pixel coordinate in the RGB image, and an aligned infrared thermal imaging temperature image is obtained.
3. The video monitoring-based precise control method for the whole process of the fire operation, which is characterized in that the method utilizes a channel attention mechanism to fuse the appearance semantic features of RGB images and the temperature distribution features of infrared thermal imaging temperature images, and is specifically as follows: Channel splicing is carried out on the appearance semantic features and the temperature distribution features to obtain a spliced feature map; Carrying out global average pooling on the spliced feature map, and calculating through at least one full-connection layer and an activation function to obtain a channel attention weight vector; and carrying out element-by-element Hadamard product operation on the spliced feature images by using the channel attention weight vector to obtain a final fusion feature image.
4. The method for accurately controlling the whole process of the fire operation based on video monitoring according to claim 1, wherein the method is characterized in that the fusion feature map is subjected to target detection to identify the current operation action, and specifically comprises the following steps: Inputting the fusion feature map into a preset YOLOv target detection model, and outputting a target detection result by the model, wherein the target detection result comprises a detected target category, a target boundary frame coordinate and a confidence coefficient; removing repeated target boundary frames by adopting a non-maximum suppression algorithm, and reserving a target boundary frame with highest confidence; The target feature vector sequence is subjected to standardization processing, and the target feature vector sequence after the standardization processing is input into an LSTM network; The method comprises the steps of inputting time sequence feature vectors into a full-connection layer, setting 2 layers, setting 128 output dimensions of a first layer, setting ReLU as an activation function, setting 6 output dimensions of a second layer, respectively corresponding to 6 operation types, namely welding, cutting, polishing, walking, climbing and standing, respectively calculating the probabilities of six actions under a current fire operation scene by using an activation function softmax, and selecting the action with the largest probability value as the current operation action.
5. The video monitoring-based fire operation overall process accurate control method according to claim 4, wherein the method is characterized in that the current operation action is matched with a preset fire operation safety specification knowledge base, and specifically comprises the following steps: the safety specification knowledge base is stored in a structured data format, and each rule comprises preconditions, judgment logic and alarm information; converting the current operation action and the target detection result into fact data with the same format as the precondition; And matching the fact data with preconditions of all rules in the safety specification knowledge base, and triggering and outputting alarm information corresponding to the illegal behaviors if the rule judgment logic is met.
6. The video monitoring-based fire operation overall process accurate control method according to claim 1, wherein the abnormal marking results comprise an environment parameter abnormal marking result and a temperature abnormal marking result.
7. The video monitoring-based fire operation overall process accurate control method according to claim 6, wherein the method calculates a comprehensive risk score based on the offence alarm information and the abnormal marking result, specifically: Constructing a multi-factor risk assessment model, and taking the illegal behaviors and the alarm information, the environment parameter abnormal marking result and the temperature abnormal marking result as assessment factors by using a weighted scoring method, wherein the illegal behaviors and the alarm information are taken as behavior factors, the environment parameter abnormal marking result is taken as an environment factor, the temperature abnormal marking result is taken as a temperature factor, and weights are respectively distributed to the three assessment factors; formulating a scoring rule of the evaluation factor, calculating a comprehensive risk score according to scores of the behavior factor, the environment factor and the temperature factor and combining weights of the factors, and expressing the comprehensive risk score as a formula: ; In the formula, The overall risk score is represented as such, Representing the score of the behavioural factor, Representing the environmental factor score(s), Representing the temperature factor score(s), The weights of the three evaluation factors are represented respectively.
8. The fire operation overall process accurate control system based on video monitoring is characterized by comprising the following modules: the data fusion preprocessing module is used for synchronously acquiring an RGB image, an infrared thermal imaging image and environmental parameters in a fire operation monitoring area, carrying out pixel level alignment on the RGB image and the infrared thermal imaging image by utilizing a homogeneous coordinate transformation principle and a preset homography matrix to obtain an aligned infrared thermal imaging temperature image; The data analysis module is used for carrying out target detection on the fusion feature map to identify current operation, matching the current operation with a preset fire operation safety standard knowledge base to generate offence alarm information, comparing temperature distribution features and environment parameters with a preset threshold to generate an abnormal mark result, calculating comprehensive risk scores based on the offence alarm information and the abnormal mark result, inputting the comprehensive risk scores into a time sequence prediction model to obtain a risk variation trend prediction value, inputting the current operation and the abnormal mark result into a causality discovery algorithm to obtain a causality association analysis result, and carrying out hierarchical sequencing on the offence alarm information and the abnormal mark result according to the causality association analysis result and the risk variation trend prediction value to obtain a processing priority list; and the strategy response module is used for executing a corresponding treatment strategy according to the treatment priority list and outputting corresponding alarm or prompt information.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 7 when the program is executed by the processor.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1 to 7.

Description

Video monitoring-based fire operation overall process accurate control method and system Technical Field The invention relates to a video monitoring-based precise control method and a video monitoring-based precise control system for the whole process of a fire operation, and belongs to the technical field of fire operation safety management. Background The fire operation is widely applied to industrial production, power production and building construction, the traditional fire operation management mainly relies on manual supervision and on-site recording, and manual inspection of operation personnel operation normalization, monitoring of the environment safety state of an operation area and manual recording of operation flow and abnormal conditions are required in the operation process. With the development of video monitoring technology, part of the existing schemes introduce single video monitoring equipment, video data of an operation area is collected through a camera to assist in manually observing the operation state, and few schemes attempt to fuse the video data with simple environment sensor data, but only can realize basic abnormality alarming function. For example, in the fire-moving operation such as the overhaul of a boiler, an effective and timely monitoring and alarming mechanism is not available under the condition of overhigh peripheral environment temperature. In the prior art, most video monitoring equipment is designed by a single camera or a fixed visual angle, the coverage range is limited, image correction and enhancement processing are lacked, the data quality is poor in a complex operation environment, the data analysis link depends on manual judgment, only a few schemes introduce simple algorithm to identify obvious illegal behaviors, the refined classification of operation actions and the quantitative evaluation of risks cannot be realized, the control execution link is single alarm output, the functions of equipment linkage control, resource dynamic scheduling and the like are omitted, and the requirements of system energy consumption optimization and fault rapid diagnosis are not considered. In addition, the prior art generally lacks coverage of the whole flow of the fire operation, closed loop management from risk prejudging to post-tracing is difficult to realize, and particularly, the risk caused by abnormal peripheral environment temperature in the overhaul process is lack of targeted monitoring and early warning. The traditional management mode relies on manual monitoring and judgment, has pain points such as insufficient real-time performance, difficult data tracing and the like, and is difficult to meet the requirement of modern safety production on digital management of fire operation. Especially in the equipment maintenance process, such as operation of boiler welding and the like, if the peripheral environment temperature is abnormally increased, the traditional mode cannot realize rapid and accurate temperature monitoring and alarming. Therefore, a method and a system for accurately controlling the whole process of the fire operation based on video monitoring are needed. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a video monitoring-based fire operation overall process accurate control method and a video monitoring-based fire operation overall process accurate control system. The technical scheme of the invention is as follows: in one aspect, the invention provides a video monitoring-based precise control method for the whole process of a fire operation, which comprises the following steps: The method comprises the steps of synchronously collecting RGB images, infrared thermal imaging images and environmental parameters in a fire operation monitoring area, carrying out pixel level alignment on the RGB images and the infrared thermal imaging images by utilizing a homogeneous coordinate transformation principle and a preset homography matrix to obtain aligned infrared thermal imaging temperature images, constructing a double-branch convolution neural network, respectively extracting appearance semantic features of the RGB images and temperature distribution features of the infrared thermal imaging temperature images, and carrying out fusion by utilizing a channel attention mechanism to obtain a fusion feature map; The method comprises the steps of carrying out target detection on the fusion feature map to identify current operation, matching the current operation with a preset fire operation safety specification knowledge base to generate offence alarm information, comparing temperature distribution features and environment parameters with a preset threshold to generate an abnormal marking result, calculating comprehensive risk score based on the offence alarm information and the abnormal marking result, inputting the comprehensive risk score into a time sequence prediction model to obtain a risk variation trend pred