阅读说明：本技术 一种基于YOLOv5模型进行精准火焰判别、灭火点定位与快速灭火的方法与装置 (Method and device for carrying out accurate flame discrimination, fire extinguishing point positioning and rapid fire extinguishing based on YOLOv5 model ) 是由 王天一 兰振东 施泽平 刘燕娜 于 2021-09-29 设计创作，主要内容包括：本发明是一种基于YOLOv5模型进行精准火焰判别、灭火点定位与快速灭火的方法与装置。该装置包括由红外阵列测温传感器阵列组成的温度采集装置,集成图像处理单元的图像采集与分析装置,拥有报警与灭火功能的执行装置三部分,并分为三步执行,第一步由温度采集装置采集室内温度,并对该阵列分布的温度数据进行分析,初步定位异常温度位置,第二步调用摄像头采集图像,通过YOLOv5模型对当前场景进行分析并对火焰进行精确定位,第三步通过执行机构执行报警和灭火行为。其中使用YOLOv5模型不仅进行对火焰位置的精确检测,且能够提前预测当前室内场景的危险行为,有效降低火灾发生的几率并降低灭火的错误率。(The invention relates to a method and a device for carrying out accurate flame discrimination, fire extinguishing point positioning and quick fire extinguishing based on a YOLOv5 model. The device comprises a temperature acquisition device consisting of an infrared array temperature measurement sensor array, an image acquisition and analysis device of an integrated image processing unit, an execution device with alarming and fire extinguishing functions, and is divided into three steps for execution, the first step is to acquire indoor temperature by the temperature acquisition device and analyze the temperature data distributed by the array, the abnormal temperature position is preliminarily positioned, the second step is to call a camera to acquire images, the current scene is analyzed and flame is accurately positioned by a YOLOv5 model, and the third step is to execute alarming and fire extinguishing behaviors through an execution mechanism. The YOLOv5 model is used for accurately detecting the flame position, the dangerous behaviors of the current indoor scene can be predicted in advance, the probability of fire occurrence is effectively reduced, and the fire extinguishing error rate is reduced.)

1. The invention relates to a method and a device for carrying out accurate flame discrimination, fire extinguishing point positioning and quick fire extinguishing based on a YOLOv5 model, which is characterized in that: the device comprises a temperature acquisition device consisting of an infrared array temperature measurement sensor array, an image acquisition and analysis device integrated with an image processing unit, and an execution device with alarming and fire extinguishing functions.

2. The method and apparatus for accurate flame discrimination, fire-extinguishing point location and rapid fire-extinguishing based on the YOLOv5 model as claimed in claim 1, wherein the method comprises the steps of:

(1) preliminarily positioning the abnormal temperature position, collecting the temperature through a temperature collecting device, wherein the temperature data are distributed in an array form, scanning and judging whether the abnormal temperature occurs by utilizing an algorithm to obtain the position of the abnormal temperature, and resolving to obtain the actual position of the abnormal temperature;

(2) the image acquisition and analysis device calls a camera to acquire an image of an abnormal temperature position, inputs the image into a YOLOv5 model to perform flame scene recognition and rapidly and accurately positions the generated flame;

(3) the executing device takes corresponding executing actions according to the analysis result in the step 2: the fire extinguishing device has the advantages that the fire extinguishing device can give an alarm to dangerous scenes possibly generating fire, and if flame occurs, the water gun is started to conduct quick and accurate fire extinguishing action on the position of the flame.

3. The method and the device for distinguishing the accurate flame, locating the fire extinguishing point and extinguishing the fire rapidly based on the YOLOv5 model as claimed in claim 2, wherein the specific method of the step 1 is as follows: the temperature acquisition device is composed of an infrared array temperature measurement sensor array, can realize temperature detection in a certain range, wherein each sensor covers W multiplied by L area to obtain a multiplied by b pixel temperature array, acquires temperature array data through the temperature acquisition device, and maps each pixel position in the temperature array data to an actual area to obtain a position resolving function:

wherein i and j are pixel positions of the detected abnormal temperature, H is the height of the current device, alpha and beta are two infrared light incidence angles of each infrared temperature sensor, then traversing the temperature value of each pixel in the temperature array, marking the pixels exceeding a set threshold value, comparing the temperature of the marked pixels with the temperature of surrounding pixels, finally determining that the temperature of the pixels is abnormal, obtaining the pixel positions (i and j) of the abnormal temperature, and obtaining the real positions (x and y) of the abnormal temperature according to a position calculation function.

4. The method and the device for performing accurate flame discrimination, fire extinguishing point positioning and rapid fire extinguishing based on the YOLOv5 model as claimed in claim 2, wherein the specific method of the step 2 is as follows: acquiring an image of indoor flame, marking the flame and a fire source point, constructing a data set in a YOLO format, and enabling the data set to be as follows: 2, dividing the ratio of the model into a training set and a testing set, selecting a YOLOv5s6 model, preprocessing an input layer by adopting a Mosiac data enhancement method, training the model by using weights pre-trained on COCO train2017 until the model converges, stopping training, pruning the trained model, setting the pruning rate to be 30%, re-training the obtained result by using an original data set until the model converges, repeating the process for 2 times to obtain a final recognition model, wherein in order to ensure that the model can be suitable for various indoor environments and has less errors, when an indoor scene is changed, a blank background picture collected by a camera and a part of pictures randomly extracted from the original data set form a new data set, simultaneously freezing the weights of the first 85% layers in the network, carrying out migration training to obtain a detection model more suitable for the current indoor environment, and inputting image information collected by the camera into the model for reasoning, and judging whether flame appears in the current scene or a scene with fire danger appears, and acquiring the actual position of the flame appearing.

5. The method and the device for performing accurate flame discrimination, fire extinguishing point positioning and rapid fire extinguishing based on the YOLOv5 model as claimed in claim 2, wherein the specific method of the step 3 is as follows: this actuating device includes two-dimentional cloud platform, squirt, singlechip and the wiFi module that constitutes by the steering wheel, by the motion of single chip microcomputer control cloud platform to carry out network communication through the wiFi module, according to the testing result that step 2 obtained, if not detect flame, report to the police to the current scene that has potential risk of catching fire through wiFi, if detect flame, will call the squirt and carry out point-to-point putting out a fire.

Technical Field

The invention relates to the technical field of fire fighting equipment, in particular to a method and a device for carrying out accurate flame discrimination, fire extinguishing point positioning and quick fire extinguishing based on a YOLOv5 model.

Background

With the increasing value of residences and frequent fires caused by other people's forefires, battery overcharging, blowers, heaters and the like, people desire a real-time, fast and automatic response fire extinguisher capable of minimizing loss. Particularly, in case of fire in modern high-rise buildings, due to the chimney effect of high-rise residences, the fire rapidly develops, the amount of used social resources is large, and the problem of untimely fire extinguishment exists, but most of the fire extinguishments are caused by weak fire protection consciousness of owners. Along with the development and popularization of the AI artificial intelligence technology, the fire extinguishing system has a decision-making function by carrying out operation processing on data acquired by each sensor, so that the real-time, quick and automatic response intelligent system integrating fire prevention and fire extinguishing becomes a demand.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a method and a device for carrying out accurate flame discrimination, fire extinguishing point positioning and quick fire extinguishing based on a YOLOv5 model, solves the problems in the background technology, and realizes an intelligent system integrating fire prevention and fire extinguishing in real time, quickly and automatically.

In order to achieve the purpose, the invention is realized by the following technical scheme:

the device includes the temperature acquisition device who constitutes by infrared array temperature measurement sensor array, and integrated image processing unit's image acquisition and analytical equipment possess the execution device triplex of warning and function of putting out a fire to organize and control the triplex device by the singlechip, wherein the execution part is including the squirt, wiFi module etc. and the camera among image acquisition and the analytical equipment binds with the squirt, can move along with the squirt. The method comprises the following concrete implementation steps:

(1) preliminarily positioning abnormal temperature position, and collecting temperature distribution array data by a temperature collecting device, wherein the temperature sensing device is composed of a plurality of infrared array temperature measuring sensor arrays, the actual area of each infrared array temperature measuring sensor is WxL, and the resolution is a_s×b_sWhen the abnormal temperature occurs in the area, if the pixel position in the temperature array is (i)_s，j_s) According to the incident angle alpha of the infrared light of the sensor_s，β_sThe actual position of the abnormal temperature can be calculated according to the calculation formula of the current height H of the device

Therefore, the temperature sensing device consisting of a plurality of infrared array temperature measurement sensor arrays can realize the preliminary positioning of abnormal temperature positions;

(2) according to the preliminary position (x) in step 1_p，y_p) Controlling the pan-tilt to aim the camera at this position, wherein the angle alpha of the two axes of rotation of the pan-tilt_p，β_pIs composed of

Wherein L is the central distance between the execution device and the temperature acquisition device, a camera is called to acquire image information of an abnormal temperature position, a YOLOv5 model is used for flame scene recognition, and the pixel position (i, j) of the generated flame is obtained;

(3) the executing device carries out corresponding behaviors according to the detection result in the step 2, if flame is not detected, the WiFi module is used for alarming a scene with potential fire risk, if flame is detected, the cradle head is controlled to move until the pixel position (i, j) of the flame falls in the area of 10 multiplied by 10 of the image center, at the moment, the cradle head keeps the position, and the water gun is started to carry out point-to-point fire extinguishing.

The implementation method of the YOLOv5 model used in step 2 is as follows:

firstly, acquiring an image of indoor flame, marking the flame and a fire source, constructing a data set in a YOLO format, and enabling the data set to be as follows: 2, dividing the ratio into a training set and a testing set;

preprocessing an input layer by selecting a YOLOv5s6 model and adopting a Mosiac data enhancement method, and training the model by using weights pre-trained on COCO train2017 until the model converges;

thirdly, pruning the trained model, setting the pruning rate to be 30%, training the obtained result again by using the original data set until convergence, and repeating the process for 2 times to obtain a final recognition model;

and fourthly, in order to ensure that the model can be used in various indoor environments and has less errors, when an indoor scene is changed, acquiring a blank background picture and a part of pictures randomly extracted from the original data set by a camera to form a new data set, and simultaneously freezing the weights of the front 85 percent layers in the network to perform transfer learning to obtain a detection model more suitable for the current indoor environment.

The invention provides a method and a device for carrying out accurate flame discrimination, fire extinguishing point positioning and quick fire extinguishing based on a YOLOv5 model, which have the following beneficial effects:

(1) acquiring temperature distribution array data according to a temperature acquisition device consisting of an infrared array temperature measurement sensor array, and resolving the position of the abnormal temperature by applying geometric mathematical knowledge;

(2) further predicting indoor scenes by using a YOLOv5 model, reducing misjudgment rate, realizing accurate and rapid positioning of flames and timely discovering potential fire risks;

(3) the executing device carries out corresponding action according to the detection result, can realize accurate positioning, in time put out a fire to the flame that has produced, reports to the police through wiFi to the scene that has the potential risk of catching a fire, has realized controlling before catching a fire, puts out a fire after catching a fire fast.

Drawings

FIG. 1 is a perspective view of a main body part of a device for performing accurate flame discrimination, fire point positioning and rapid fire extinguishing based on a YOLOv5 model;

FIG. 2 is a top view of the main body of the device for accurate flame discrimination, fire-extinguishing point location and fast fire-extinguishing based on the YOLOv5 model.

Description of the labeling: 1. a single chip microcomputer; 2. an x-degree-of-freedom steering engine; 3. a y degree of freedom steering engine; 4. an x-degree-of-freedom steering engine fixing piece; 5. a y-degree-of-freedom steering engine fixing piece; 6. a steering engine connecting piece; 7. a camera; 8. a high pressure water gun; 9. a high-pressure water gun fixing piece; 10. an infrared array temperature measurement sensor; 11. an image processing module; 12. and a WiFi module.

Detailed Description

The device includes the temperature acquisition device who constitutes by infrared array temperature measurement sensor 10 array, the image acquisition and the analytical equipment of integrated image processing module 11, possess this triplex of the final controlling element who reports to the police and put out a fire the function, carry out program control to triplex device by singlechip 1, wherein, the final controlling element is still including high-pressure squirt 8, wiFi module 12 etc. and camera 7 among image acquisition and the analytical equipment binds mutually with high-pressure squirt 8, can move along with high-pressure squirt 8.

The device comprises the following concrete implementation steps: the temperature acquisition device acquires temperature distribution array data, the singlechip 1 traverses the temperature distribution array data and marks the position (i, j) of abnormal temperature according to the position by a formula

Calculating to obtain the actual position (x) of the abnormal temperature_p，y_p) According to the actual position, can be formulated by

Calculating the angle alpha of two rotation axes of the head_p，β_pThe single chip microcomputer 1 controls the x freedom steering engine 2 and the y freedom steering engine 3 of the pan-tilt to respectively rotate an angle alpha_p，β_pThe high-pressure water gun 8 is aligned to the abnormal temperature position, the image processing module 11 is awakened to call the camera 7 to acquire image information of the abnormal temperature position, the YOLOv5 model is operated to perform flame scene recognition, if the flame is not detected, the scene with potential fire risk is alarmed through the WiFi module 12, if the flame is detected, the pixel position (i, j) of the flame is fed back to the single chip microcomputer 1, the single chip microcomputer 1 controls the movement of the x-degree-of-freedom steering engine 2 and the y-degree-of-freedom steering engine 3 of the cradle head according to the position until the pixel position of the flame falls into the area of 10 x 10 of the image center, the cradle head keeps the position at the moment, and the high-pressure water gun 8 is started to perform point-to-point fire extinguishing.