CN117789394A

CN117789394A - Early fire smoke detection method based on motion history image

Info

Publication number: CN117789394A
Application number: CN202311720532.8A
Authority: CN
Inventors: 张旭; 胡英; 马新华
Original assignee: Yingkou Century Electronic Instrument Co ltd
Current assignee: Yingkou Century Electronic Instrument Co ltd
Priority date: 2023-12-14
Filing date: 2023-12-14
Publication date: 2024-03-29

Abstract

The invention discloses an early fire smoke detection method based on a motion history image, and belongs to the technical field of fire monitoring and alarming. In order to solve the problem that fire disaster early warning has defects in the aspects of accuracy, reliability and instantaneity, the accuracy of smoke to fire disaster prediction can be improved through smoke feature extraction and smoke motion analysis, the fire disaster smoke can be accurately identified through analysis of motion modes and smoke features, the instantaneity of early warning is improved through acquisition of image sequences, the image sequences are continuously acquired and are analyzed in real time, the fire disaster smoke can be timely found, the reliability is high, the problems of false alarm and missing report of a sensor can be avoided through analysis of smoke motion historical images, meanwhile, the application range is wide, different influencing factors can be combined according to different monitoring scenes through smoke motion analysis and fire disaster smoke detection, and the method is applicable to various scenes, such as inside buildings or in some outdoor scenes.

Description

Early fire smoke detection method based on motion history image

Technical Field

The invention relates to the technical field of fire monitoring and alarming, in particular to an early fire smoke detection method based on a motion history image.

Background

Fire is the most frequent disaster, and seriously threatens the public safety. In order to effectively prevent fire, it is very important to make early fire early warning. The visible smoke stage is generally generated in the process of occurrence and development of the fire, and then the smoke stage and the severe combustion stage are started, so that rapid reaction can be carried out in the early stage of occurrence of the fire by a method for detecting the smoke, early warning and control of the fire are facilitated, and the hazard degree of the fire is reduced. With the large-scale application of image type fire detection products in recent years, an image-based smoke detection method becomes a main development direction of early fire smoke detection.

Currently, the commonly used image smoke detection technology is divided into two major types, namely an image processing technology based on a traditional algorithm and an image processing technology based on deep learning, which is developed in recent years.

Conventional image processing techniques require an algorithm designer to analyze global information of an image to be processed and feature information possessed by a target. And designing a corresponding processing algorithm to complete the smoke recognition task. A detection algorithm designed in the background has larger fluctuation in performance when the detection algorithm is subjected to detection in an entirely new background. In addition, for the dynamic characteristics of smoke, the calculation amount of a method adopting an optical flow is large, and the expression capability of the method on the continuous characteristics is not strong.

The image processing technology based on deep learning utilizes a deep convolution network to extract the color and texture characteristics of the smoke image, does not need background modeling, and can be suitable for dynamic scenes. The dynamic characteristics of the smoke can assist in time modeling of the smoke by means of a 3D convolution method or a long-short-time memory network and the like, and learning of the smoke characteristics by the neural network is promoted. However, the deep learning method requires a high number of samples to train and has poor adaptability in untrained scenes. In addition, the calculation amount is large in reasoning, and the requirement on hardware is high in edge deployment.

Disclosure of Invention

The invention aims to provide an early fire smoke detection method based on a motion history image, so as to solve the problems in the background technology.

In order to achieve the above purpose, the present invention provides the following technical solutions: an early fire smoke detection method based on motion history images comprises the following steps:

the monitoring equipment is arranged in a monitoring area, and the monitoring equipment is a monitoring camera;

collecting an image sequence, realizing omnibearing monitoring in a monitoring area by using a plurality of monitoring cameras, and continuously collecting the image sequence;

Image area analysis, which is to analyze the image area of the image sequence and identify the possible abnormal smoke movement;

extracting smoke characteristics aiming at abnormal smoke movement possibly existing, wherein the smoke characteristics comprise the color, shape and movement mode of smoke;

the smoke movement analysis is carried out, and the movement trend of the smoke is analyzed based on the smoke characteristics;

fire smoke detection, namely performing multi-stage classification on the image sequence according to the extracted smoke characteristics by using a plurality of classifiers to perform fire smoke detection and judge a fire place area;

when the fire smoke is detected, an alarm signal is sent out, and various modes such as sound, light, short message and the like are used for alarming and notifying related personnel to take prevention and control measures.

Further, the monitoring equipment layout specifically further comprises the following steps:

according to the scene and the purpose to be monitored, selecting a waterproof and dustproof outdoor monitoring camera in an outdoor monitoring area, communicating the outdoor monitoring camera with a solar power supply system, selecting an indoor camera in an indoor monitoring area, and connecting the indoor camera with a circuit;

according to the size and shape of the monitoring area, the number and positions of monitoring cameras to be arranged are determined, the adjacent monitoring cameras are arranged at intervals of 10-20 meters and cover the whole monitoring area, and the visual angle of each monitoring camera is adjusted, so that the monitoring cameras can clearly shoot all important areas in the monitoring area;

The monitoring camera is connected with the monitoring center in a wired or wireless mode.

Further, the image acquisition sequence specifically further comprises the following steps:

the monitoring camera monitors the monitoring area in all directions and acquires an image sequence, and the monitoring image sequence is transmitted in real time;

the collected monitoring image sequence is backed up through a storage device;

regular maintenance and management of backup data.

Further, the image area analysis specifically further includes the following steps:

a1: differentiating the current image and the background image to obtain a differential image containing motion information;

a2: dividing the image difference image into image blocks with the size of N multiplied by N, wherein the value of N is determined by the image resolution, removing part of the image blocks, specifically calculating the number of pixel points in the image blocks, and counting the number N of pixel points meeting the motion characteristics in each block _i The results were updated as follows:

where T1 is the proportional threshold for the moving pixel point, for B _n The pixel block with the value of 0 in (x, y) does not carry out the next calculation;

a3: calculate B _n The average pixel value P for each motion block labeled 1 in (x, y) increases when the average is greater than the threshold T2, and decreases otherwise, as shown in the formula:

Wherein tMHI _σ (x, y) is the pixel value of the motion block corresponding to the motion history image;

a4: continuously repeating steps A1 and A3 within a set time threshold sigma for tMHI _σ (x, y) to obtain motion history image MHI _σ ；

The edge contour of the smoke area has higher complexity and is characterized by contour curvature entropy;

the calculation formula of the contour curvature entropy is as follows:

wherein H is _c Is the discrete curvature entropy of the smoke edge profile, P (C _i ) Is of curvature C _i Probability of total collection, H _c The larger the edge curvature, the more complex it is;

based on A1-A4, the specific steps of the smoke detection method are as follows:

s1: initializing a system, reading each parameter threshold value, and setting an image count which accords with smoke characteristics to 0;

s2: collecting a frame of image I, and taking the first frame of image as a background image I _b Returning to the step S2, otherwise, turning to the step S3;

s3: i is combined with background image I _b Obtaining a differential image I by differencing _diff And integrates the motion history image MHI according to the algorithm _σ Step S4 is carried out;

s4: traversing motion history image MHI _σ When the corresponding pixel value tMHI in the motion block _σ When (x, y) is greater than the threshold T3, the block is determined to be a suspected smoke region, and the process proceeds to step S5;

s5: counting the number N of suspected smoke area blocks _s If N _s If the background image I is larger than the threshold value T4, the step S6 is carried out, otherwise, the background image I is updated _b Step S2 is carried out;

S6：filling and searching B by using the suspected smoke area block as a seed area and adopting an area searching algorithm _n Non-0 region of (x, y) communicating with the suspected smoke region block and calculating the region area A _s If A _s If the threshold value T5 is larger than the threshold value T5, the step S7 is carried out, otherwise, the step S2 is carried out;

s7: calculating the sharp angle number S of the edge of the connected region _c Entropy of curvature H _c If S _c Greater than threshold T6 and H _c If the threshold value T7 is larger than the threshold value T7, the step S8 is carried out, otherwise, the step S2 is carried out;

s8: adding 1 to the image count conforming to the smoke characteristics, and if the image count conforming to the smoke characteristics is greater than T8, entering step S9; otherwise, the count value of the number of images conforming to the smoke characteristics is set to 0, and the step S2 is returned;

s9: and outputting a fire alarm signal and marking the position of the smoke area in the image.

Further, the smoke feature extraction specifically further comprises the following steps:

extracting frames with abnormal smoke movement in a monitoring image sequence, setting the frames as abnormal key frames, and dividing image areas of the abnormal key frames;

preprocessing the image in the abnormal key frame, removing noise and abnormal values, and obtaining a processed image;

extracting characteristics from the processed image, and extracting color characteristics, shape characteristics and motion characteristics, wherein the color characteristics comprise average color and color contrast of smoke, the shape characteristics comprise edges and textures of the smoke, and the motion characteristics comprise flow modes and speed of the smoke;

The feature extraction of the processed image comprises the following steps:

reading the processed image based on a computer, and determining a red channel, a green channel and a blue channel of the processed image in a preset color space;

setting target color thresholds of a background image and a foreground image in a processed image based on distribution states of a red channel, a green channel and a blue channel of the processed image in a preset color space;

image segmentation is carried out on the processed image according to the target color threshold value, and a background image and a foreground image are obtained;

acquiring an extraction type for extracting features of the processed image, and determining a feature extraction index corresponding to the extraction type;

performing first extraction on the background image based on the extraction type and an extraction index of the extraction type to obtain first characteristic information of the background image in the processed image;

performing second extraction on the foreground image based on the extraction type and the extraction index of the extraction type to obtain second characteristic information of the foreground image in the processed image;

and integrating the first characteristic information and the second characteristic information to obtain third characteristic information, and completing characteristic extraction of the processed image based on the third characteristic information.

Further, the smoke movement analysis specifically further comprises the following steps:

calculating the direction and speed of the movement of the smoke based on the color features, the shape features and the movement features of the smoke in combination with the abnormal key frames;

when a monitoring area with abnormal smoke movement is indoors, building structure nodes of the monitoring area are obtained, and the spreading path of the smoke is analyzed by combining the building structure nodes and a monitoring image sequence;

when a monitoring area with abnormal smoke movement is outdoors, acquiring electronic map information and current day wind direction information of the monitoring area, and analyzing the diffusion flow speed and flow direction of the smoke by combining the electronic map information and the wind direction information;

based on the above flow, the movement path of the smoke is judged.

Further, the fire smoke detection specifically further comprises the following steps:

determining a fire occurrence point based on a motion path of smoke;

and analyzing the fire spreading path and the fire situation based on the fire occurrence point and the movement path of the smoke.

Further, determining a fire occurrence based on the movement path of the smoke includes:

the method comprises the steps of obtaining a movement path of the obtained smoke, and carrying out structural analysis on the movement path of the smoke to obtain tree structure characteristics of the movement path of the smoke;

Discretizing the motion path based on the tree structure characteristics to obtain discrete motion tracks of the motion path at different moments, and determining relative position offset of the smoke at different moments based on the discrete motion tracks;

constructing a rectangular coordinate system, mapping discrete motion paths at different moments to the rectangular coordinate system based on relative position offset, and determining aggregation points of the motion paths of the smoke in different directions based on mapping results and tree structure characteristics;

determining the gathering points of the motion paths of the smoke as a candidate fire occurrence point set, and respectively determining the motion trend of each candidate fire occurrence point in the candidate fire occurrence point set based on the diffusion flow speed and the flow direction of the smoke;

tracing the gathering point of the movement path of the smoke based on the movement trend and the opposite direction of the flow direction of the smoke, and obtaining the gathering point source of the movement path of the smoke based on the tracing result;

analyzing the aggregation point sources based on the extracted smoke characteristics and a preset fire disaster occurrence point evaluation index to obtain the confidence coefficient of the aggregation point sources, and judging the aggregation point sources as fire disaster occurrence points when the confidence coefficient is in a preset threshold value interval;

extracting a frame image corresponding to a fire disaster occurrence point, preprocessing the frame image, determining an edge contour of open fire at the fire disaster occurrence point, and extracting pixel characteristic values in the edge contour;

Determining color saturation thresholds of open fire and smoke based on the pixel characteristic values, and respectively matching the color saturation thresholds of the open fire and the smoke with a preset reference list to obtain a first fire evaluation parameter and a second fire evaluation parameter;

and carrying out weighted average processing on the first fire evaluation parameter and the second fire evaluation parameter based on the weight of the influence of the open fire and the smoke on the fire, and obtaining the current fire condition of the fire occurrence point based on the weighted evaluation processing result.

Further, the fire smoke alarm specifically further comprises the following steps:

marking a fire hazard zone based on the fire zone location and the path zone of abnormal smoke movement determined by the fire smoke detection;

performing sound-light fire evacuation alarm on a fire hazard area;

and carrying out fire alarm based on the position of the fire hazard area, and providing a detailed address of fire and fire analysis conditions.

Further, when detecting fire smoke, sending out an alarm signal, including:

reading the image sequence, and determining the distribution area of smoke in the image sequence;

reading the pixel points corresponding to the smoke distribution area, and determining the color distribution characteristics of the pixel points in the smoke distribution area;

Splitting a smoke distribution area according to color distribution characteristics of pixel points to obtain m sub-distribution areas, and splitting each sub-distribution area according to a preset unit to obtain n area blocks corresponding to the sub-distribution areas;

the color-concentration estimation table is called, n area blocks corresponding to the sub-distribution areas are matched in the color concentration estimation table, and smoke concentration estimation values corresponding to unit areas in the n area blocks corresponding to the sub-distribution areas are determined;

calculating the concentration value of fire smoke in the smoke concentration estimated values corresponding to unit areas in n area blocks corresponding to the sub-distribution areas;

wherein C represents the concentration value of fire smoke; j represents the sequence number value of the sub-distribution area; m represents the total number of sub-distribution areas;a region volume representing a jth sub-distribution region; i represents the sequence number value of the split region block of the sub-distribution region according to a preset unit; n represents the total number of region blocks; s represents the area of the region block; h is a _i Representing a height value corresponding to the i-th region block; c _ij A smoke concentration estimation value representing an i-th region block in the j-th sub-distribution region; ρ represents an error factor, and the value range is (0.01,0.02); pi represents a constant of 3.14;

Acquiring a preset fire smoke concentration alarm threshold interval [ a, b ], and comparing a concentration value C of fire smoke with the preset fire smoke concentration alarm threshold interval [ a, b ];

determining an alarm time length for sending an alarm signal based on the comparison result;

when C is less than or equal to a, the alarm time length of the alarm signal is 0, and a represents the lower threshold value of the preset fire smoke concentration alarm threshold value interval;

when a is less than or equal to b, the alarm time length of the alarm signal is a first preset time length, and b represents the element and the upper threshold value of the fire smoke concentration alarm threshold value interval;

when C > b, the alarm time length for sending out the alarm signal is a second preset time length, and the first preset time length is smaller than the second preset time length.

Compared with the prior art, the invention has the beneficial effects that:

1. the method can improve the accuracy of smoke to fire prediction through smoke feature extraction and smoke motion analysis, can accurately identify fire smoke through analysis of motion modes and smoke features, improves early warning instantaneity through acquisition of image sequences, continuously acquires the image sequences and carries out real-time analysis, can discover fire smoke in time, has high reliability, can avoid the problems of false alarm and missing report of a sensor through analysis of smoke motion historical images, and has wide application range.

2. The invention processes the monitoring image sequence by the difference method of two adjacent frames of images, the dynamic characteristics of the smog are embodied by the motion history image, false alarm can be effectively eliminated, especially static background image and non-diffusion cloud interference, the motion history image is calculated by adopting a single frame accumulation method, the calculated amount is small, the edge end is easy to realize, the identification method does not relate to the smog color characteristics, and the invention is effective to smog with various colors and has wide applicability.

3. According to the invention, whether a fire disaster exists can be accurately judged by extracting the smoke characteristics in the image, a large amount of smoke can be generated at the initial stage of the fire disaster, and the fire disaster signs can be timely found by extracting the smoke characteristics in the image and analyzing, so that early warning is performed at the early stage, and more escape and fire extinguishing time is provided for people.

4. The invention combines different influencing factors to the smog in different areas, thereby more accurately judging the movement path of the smog, further improving the accuracy of predicting the occurrence of the fire disaster, taking preventive measures in advance, such as closing electric appliances, prohibiting smoking and the like, so as to avoid the occurrence of the fire disaster, simultaneously determining the position of the fire disaster, being beneficial to finding the fire source in time, taking corresponding fire extinguishing measures, providing direction for fire extinguishment and improving the fire extinguishing efficiency.

5. The method has the advantages that the discretization and the structural analysis are carried out on the motion paths of the smoke, the visual display of the discrete motion paths of the smoke at different moments in a constructed rectangular coordinate system is realized, the analysis of the fire occurrence point according to the visual display result is facilitated, the focusing point of the motion paths of the smoke under different conditions is determined according to the visual display result, the accurate and effective judgment of the fire occurrence point according to the focusing point is realized, finally, the frame image of the fire occurrence point is analyzed, the effective determination of the fire condition of the fire occurrence point is realized, the corresponding alarm operation is facilitated on the fire occurrence point and the fire condition of the fire occurrence point, and the accuracy, the reliability and the real-time performance of smoke monitoring are improved.

6. Through the distribution area of smog in the image sequence is read, the distribution area of smog is divided and split, the effective division of the distribution area can be effectively realized, so that the smog concentration value of each sub-distribution area can be effectively evaluated through a color-concentration estimation table, the effectiveness and convenience of calculating the smog concentration value of a fire disaster are further improved, the measurement of the concentration of the smog of the fire disaster by personnel can be effectively improved by setting different alarm time lengths, different smog early warning measures are made, and the effectiveness of detecting the smog of the fire disaster is improved.

Drawings

FIG. 1 is a schematic flow chart of the method of the present invention;

FIG. 2 is a flow chart of an early fire smoke detection method;

FIG. 3 is a schematic view of smoke from a smoke movement history image;

FIG. 4 is a schematic diagram of a motion block corresponding to a smoke motion history image;

FIG. 5 is a schematic view of a motion history of a smoke motion history image;

fig. 6 is a schematic diagram of smoke detection results.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In order to solve the problems that the conventional image processing technology requires algorithm designers to analyze global information of an image to be processed and feature information of a target, the performance of a detection algorithm often has larger fluctuation, and in addition, for dynamic features of smoke, a method adopting an optical flow has large calculation amount, has weak expression capability on continuous features, and meanwhile, the method adopting deep learning has high sample number required during training, has poor adaptability in untrained scenes, has large calculation amount during reasoning and has high requirement on hardware during edge end deployment, the invention provides the following technical scheme with reference to figures 1-6:

An early fire smoke detection method based on motion history images comprises the following steps:

Specifically, can improve the accuracy of smog to the conflagration prediction through smog feature extraction and smog motion analysis, through analysis motion pattern and smog feature, can accurately discern conflagration smog, the real-time of early warning has been improved through gathering the image sequence, continuous collection image sequence and carry out real-time analysis, can in time discover conflagration smog, the reliability is high, through analyzing smog motion history image, can avoid the problem that the sensor was mistaken to report and was missed to report, and wide application scope simultaneously, can combine different influence factors according to different monitored scenes through smog motion analysis and conflagration smog detection for be applicable to various scenes, like in the building inside or some outdoor scenes.

The monitoring equipment layout specifically further comprises the following steps:

Specifically, can constantly monitor the smog in the monitored area through setting up the surveillance camera head for 24 hours, in case the abnormal situation is found, can in time send out the alarm, remind relevant personnel to handle to improve early warning ability, simultaneously through the image sequence that surveillance camera head gathered, can discern, analyze, early warning the smog in the monitored area, accomplish in time discovery and the alarm to the conflagration, avoid taking place bigger calamity and loss.

The image acquisition sequence specifically further comprises the following steps:

the collected monitoring image sequence is backed up through a storage device;

regular maintenance and management of backup data.

Specifically, after abnormal conditions such as fire occur, the specific position and reason of the fire can be found out through replaying the monitoring video, the follow-up fire investigation and treatment are facilitated, meanwhile, the rule and the characteristic of the fire occurrence can be found through analyzing the monitoring video, so that more effective fire prevention and control measures are formulated, moreover, the monitoring video can be stored for a long time, when unexpected conditions occur, the monitoring video can be restored to the previous state through backup data, and the integrity and the reliability of the data are ensured.

The image area analysis specifically further comprises the following steps:

a2: dividing the image difference image into image blocks with the size of N multiplied by N, wherein the value of N is determined by the image resolution, eliminating partial image blocks in order to improve algorithm instantaneity and reduce system calculation amount, specifically calculating the number of pixel points in the image blocks, and counting the number N of pixel points meeting the motion characteristic in each block _i The results were updated as follows:

In a video sequence, the motion area corresponding to the current frame is overlapped in the motion history image, the old motion area in the current motion history image is affected less and less, the brightness is gradually darkened, the diffused smoke leaves a track from light to dark in the motion history image in the overlapping process of the motion history image, the increased pixel value of the track is far greater than the reduced pixel value, and other moving targets such as pedestrians and vehicles are opposite, so that the phenomenon does not exist in the motion history image;

in addition to dynamic features, smoke has some obvious static features, and the edge profile of the smoke area has higher complexity and can be characterized by profile curvature entropy;

The calculation formula of the contour curvature entropy is as follows:

based on the theory, the specific steps of the smoke detection method are as follows:

s6: filling and searching B by using the suspected smoke area block as a seed area and adopting an area searching algorithm _n Non-0 region of (x, y) communicating with the suspected smoke region block and calculating the region area A _s If A _s If the threshold value T5 is larger than the threshold value T5, the step S7 is carried out, otherwise, the step S2 is carried out;

Specifically, by carrying out differential processing on two adjacent frames of images of the monitoring image sequence, the dynamic characteristics of smoke are embodied through the motion history image, false alarm can be effectively eliminated, especially static background images and non-diffusion cloud interference, the motion history image is calculated by adopting a single frame accumulation method, the calculated amount is small, the edge end is easy to realize, the recognition method does not relate to the smoke color characteristics, and the method is effective to smoke with various colors and has wide applicability.

Referring to fig. 3-6, the surveillance area of fig. 3 is an outdoor scene with a video resolution of 1920 x 1080. Firstly, setting the image count which accords with the smoke characteristics to be 0, collecting images under normal conditions as background images, and then starting circulation processing; accumulating motion history images according to the method step S4 described above, wherein n=16, t1=0.1, t2=20; traversing the motion-history image MHI by taking the threshold t3=60 according to the previous method step S5 _σ Marking all suspected smoke areas as shown in fig. 4-5; counting the number N of suspected smoke area blocks _s Taking the threshold t4=128, if N _s >T4 meets the dynamic characteristics of the smoke and further judges the static characteristics of the smoke; searching and calculating the area A of the connected area according to the step S7 of the method _s The method comprises the steps of carrying out a first treatment on the surface of the Taking the threshold t5=2000, if a _s >T5 then calculates the number of sharp corners S of the connected region according to the method step S8 _c And curvature entropy H _c The method comprises the steps of carrying out a first treatment on the surface of the Taking the threshold t6=20, t7=0.8, if S _c Greater than threshold T6 and H _c If the image count is larger than the threshold value T7, the image count conforming to the smoke characteristics is increased by 1; if the eligible pattern count value is greater than T8 (t8=15 If the fire is judged to occur, an alarm signal is given and the alarm signal is displayed in a frame in the image smoke area as shown in fig. 6.

The smoke feature extraction specifically further comprises the following steps:

and extracting features from the processed image, namely extracting color features, shape features and motion features, wherein the color features comprise average colors and color contrast of smoke, the shape features comprise edges and textures of the smoke, and the motion features comprise flow modes and speed of the smoke.

Specifically, smoke is one of important characteristics of fire, whether the fire exists can be accurately judged by extracting smoke characteristics in the image, a large amount of smoke can be generated at the initial stage of the fire, and the fire signs can be timely found by extracting and analyzing the smoke characteristics in the image, so that early warning is performed in an early stage, and more escape and fire extinguishing time is provided for people.

The smoke movement analysis specifically further comprises the following steps:

based on the above flow, the movement path of the smoke is judged.

Specifically, by combining different influencing factors with the smoke in different areas, the movement path of the smoke can be accurately judged, the accuracy of predicting the occurrence of the fire is improved, preventive measures such as closing an electric appliance and prohibiting smoking are adopted in advance, the occurrence of the fire is avoided, meanwhile, the position of the occurrence of the fire can be roughly determined, the fire source can be found in time, corresponding fire extinguishing measures are adopted, the direction is provided for fire extinguishment, and the fire extinguishing efficiency is improved.

The fire smoke detection specifically further comprises the following steps:

determining a fire occurrence point based on a motion path of smoke;

The fire smoke alarm specifically further comprises the following steps:

performing sound-light fire evacuation alarm on a fire hazard area;

based on the position of the fire hazard area, the fire alarm is carried out, and the detailed address of the fire and the fire analysis condition are provided

Specifically, the fire spreading way and the fire analysis situation of the staff are informed in advance through prediction, so that the firefighter is helped to treat the fire more safely, reasonably and efficiently, the firefighter is helped to rapidly position the fire source, the fire extinguishing efficiency is improved, the fire loss is reduced, meanwhile, the firefighter is prevented from entering a dangerous area, and the personal safety of the firefighter is guaranteed.

The embodiment also provides an early fire smoke detection method based on the motion history image, which is used for extracting features from the processed image and comprises the following steps:

In this embodiment, the target color threshold may be determined based on the distribution states of the red channel, the green channel, and the blue channel of the processed image in the preset color space, and used as a measure for dividing the background image from the foreground image in the processed image.

In this embodiment, the preset color space may be an RGB color space.

In this embodiment, the extraction type may be one including: color features, shape features, and motion features.

In this embodiment, the feature extraction index corresponding to the extraction type may be extraction content corresponding to each extraction type, where when the extraction type is a color feature, the corresponding extraction index is: when the average color, color contrast and extraction type of the smoke are shape features, the corresponding extraction indexes are as follows: edges, textures of smoke; when the extraction type is shape characteristics, the corresponding extraction indexes are as follows: flow pattern, speed of smoke.

In this embodiment, the first characteristic information is characteristic information of smoke in a background image of the processed image.

In this embodiment, the second characteristic information is characteristic information of smoke in a foreground image of the processed image.

In this embodiment, the third feature information is composed of the first feature information and the second feature information, and is used to characterize the feature extraction result of the processed image.

The working principle and the beneficial effects of the technical scheme are as follows: the background image and the foreground image of the processed image are extracted, so that the extraction error rate during feature extraction can be avoided, and the feature extraction of the processed image is finer and the accuracy of feature extraction of the processed image is improved through the extraction of the first feature information of the foreground image and the second feature information of the background feature respectively.

The embodiment also provides an early fire smoke detection method based on a motion history image, which determines a fire occurrence point based on a motion path of smoke, and comprises the following steps:

In this embodiment, the structural analysis is to determine the relative positional relationship between the movement paths of the smoke at different positions, for example, when there is an external interference factor, the movement paths of the smoke may be abrupt or the smoke may be dispersed.

In this embodiment, the tree structure feature may be a positional relationship between motion paths characterizing smoke in different directions at different times and under different states, i.e. may be a state three-way condition characterizing smoke at different times.

In this embodiment, the discrete movement track may be a movement path that is discrete into a plurality of points, so that the movement path of the smoke is conveniently displayed in a rectangular coordinate system.

In this embodiment, the relative position offset may be the offset direction and the offset distance of the smoke position at the adjacent time compared to the offset direction and the offset distance occurring at the previous time.

In this embodiment, the focal point may be the point of coincidence of the paths of movement of the smoke in different directions, and when the paths of movement of the smoke coincide, there is a possibility that a fire may occur.

In this embodiment, the candidate set of fire points may be, and not only, a location where the aggregate points of the motion paths of all the smoke are determined to be possible fire points.

In this embodiment, the movement trend may be the occurrence direction of the candidate occurrence point at different time, for example, may be a change along with the change of the flow direction of the smoke or less along with the flow direction of the smoke, for example, when the movement trend of the candidate fire occurrence point is consistent with the flow direction of the smoke, it is determined that the focal point in the current movement direction is not the fire occurrence point, and the smoke paths caused by external interference factors are collected.

In this embodiment, tracing the focus of the smoke's path of motion based on the motion trend and the opposite direction to the smoke's flow direction may be to find the source focus of the smoke path, since the more the focus is opposite to the smoke's flow direction, the closer to the fire point.

In this embodiment, the point of focus source refers to the point of coincidence of the paths of motion of the smoke that produce the smoke, i.e., all the different directions, i.e., the point of initial spread of the smoke from the point of fire.

In this embodiment, the preset fire occurrence point evaluation index is set in advance, and may be, for example, the concentration of smoke, the overlap ratio of the smoke movement paths, or the like.

In this embodiment, the confidence is a confidence value that characterizes the point of aggregation source as being able to characterize the point of fire, the greater the value the more likely it is that the point of fire will be.

In this embodiment, the preset threshold interval is set in advance, and is used to measure that the minimum criterion for determining the fire occurrence point is met, and can be adjusted.

In this embodiment, the preprocessing may be denoising processing or the like of the frame image.

In this embodiment, the edge profile may be the shape that the fire point appears in the frame image.

In this embodiment, the pixel characteristic value may be the sharpness of the pixel point, the color threshold of the pixel point, and the like.

In this embodiment, the color saturation threshold is an image parameter used to characterize the intensity of an open fire at the point of fire occurrence as well as the smoke concentration.

In this embodiment, the preset reference list is set in advance, and is used to characterize the fire conditions corresponding to the color saturation thresholds of different open fires and smoke.

In this embodiment, the first fire evaluation parameter may be a fire size determined by considering only the color saturation threshold of an open fire.

In this embodiment, the second fire assessment parameter may be a fire level determined by considering only the color saturation threshold of the smoke.

In this embodiment, the impact weight may be a measure characterizing the impact of open fire and smoke, respectively, on the fire size.

The working principle and the beneficial effects of the technical scheme are as follows: the method has the advantages that the discretization and the structural analysis are carried out on the motion paths of the smoke, the visual display of the discrete motion paths of the smoke at different moments in a constructed rectangular coordinate system is realized, the analysis of the fire occurrence point according to the visual display result is facilitated, the focusing point of the motion paths of the smoke under different conditions is determined according to the visual display result, the accurate and effective judgment of the fire occurrence point according to the focusing point is realized, finally, the frame image of the fire occurrence point is analyzed, the effective determination of the fire condition of the fire occurrence point is realized, the corresponding alarm operation is facilitated on the fire occurrence point and the fire condition of the fire occurrence point, and the accuracy, the reliability and the real-time performance of smoke monitoring are improved.

The embodiment also provides an early fire smoke detection method based on the motion history image, when detecting fire smoke, sending out an alarm signal, comprising:

In this embodiment, the area of each region block in the sub-distribution region is equal.

In this embodiment, the color distribution characteristics of the pixels can distinguish the distribution areas of the smoke according to different colors, so that the obtained pixels of the sub-distribution areas have consistent colors, and when the colors of the pixels are consistent, the smoke concentrations in the sub-distribution areas are consistent.

In this embodiment, the preset unit may be set in advance, for example, may be 1.

In this embodiment, the area of each region block in the sub-distribution region is the same.

In this embodiment, the color-concentration estimation table may be set in advance, and the estimated value of the corresponding smoke concentration in each color may be effectively estimated, which is determined based on the historical experimental data.

In this embodiment, the preset fire concentration alarm threshold interval may be a criterion that is set in advance and used to measure the alarm duration.

In this embodiment, the first preset duration and the second preset duration are set in advance, where the first preset duration is smaller than the second preset duration.

The working principle and the beneficial effects of the technical scheme are as follows: the method has the advantages that the distribution areas of the smoke in the image sequence are read, the distribution areas of the smoke are divided and split, the effective division of the distribution areas can be effectively realized, the smoke concentration value of each sub-distribution area can be effectively evaluated through the color-concentration estimation table, the effectiveness and convenience of computing the fire smoke concentration value are further improved, the measurement of the fire smoke concentration by personnel can be effectively improved through setting different alarm time periods (first preset time periods and second preset time periods), different smoke early warning measures are made, and the effectiveness of fire smoke detection is improved.

The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should be covered by the protection scope of the present invention by making equivalents and modifications to the technical solution and the inventive concept thereof.

Claims

1. An early fire smoke detection method based on motion history images is characterized by comprising the following steps:

When the fire smoke is detected, an alarm signal is sent out, and various modes such as sound, light and short message are used for alarming and notifying related personnel to take prevention and control measures.

2. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the monitoring equipment layout specifically further comprises the following steps:

3. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the image acquisition sequence specifically further comprises the following steps:

the collected monitoring image sequence is backed up through a storage device;

regular maintenance and management of backup data.

4. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the image area analysis specifically further comprises the following steps:

the calculation formula of the contour curvature entropy is as follows:

s4: traversing motion historyImage MHI _σ When the corresponding pixel value tMHI in the motion block _σ When (x, y) is greater than the threshold T3, the block is determined to be a suspected smoke region, and the process proceeds to step S5;

5. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the smoke feature extraction specifically further comprises the following steps:

the feature extraction of the processed image comprises the following steps:

6. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the smoke movement analysis specifically further comprises the following steps:

based on the above flow, the movement path of the smoke is judged.

7. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the fire smoke detection specifically further comprises the following steps:

determining a fire occurrence point based on a motion path of smoke;

8. The method for detecting early fire smoke based on motion history image according to claim 7, wherein: determining a fire occurrence based on a motion path of smoke, comprising:

9. The method for detecting early fire smoke based on motion history image as recited in claim 1, wherein: the fire smoke alarm specifically further comprises the following steps:

performing sound-light fire evacuation alarm on a fire hazard area;

10. A method of detecting early fire smoke based on motion history images as defined in claim 1, wherein upon detection of fire smoke, emitting an alarm signal comprises: