CN113066077B - Flame detection method and device - Google Patents
Flame detection method and device Download PDFInfo
- Publication number
- CN113066077B CN113066077B CN202110393455.4A CN202110393455A CN113066077B CN 113066077 B CN113066077 B CN 113066077B CN 202110393455 A CN202110393455 A CN 202110393455A CN 113066077 B CN113066077 B CN 113066077B
- Authority
- CN
- China
- Prior art keywords
- flame
- images
- frame
- detection
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a flame detection method, which uses a flame detection method based on a single frame image to carry out frame-by-frame initial detection on a video, when flame is detected initially, flame area gray level images are respectively extracted from a plurality of continuous frame images starting from the current frame image in which the flame is detected, and a frame difference image of the flame area gray level images between adjacent frame images is obtained; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection. The invention also discloses a flame detection device. Compared with the prior art, the method has high detection accuracy, good real-time performance and low requirements on software and hardware resources.
Description
Technical Field
The invention relates to a digital image processing technology, in particular to a flame detection method.
Background
With the popularization of video monitoring equipment and the development of digital image processing technology, flame detection technology based on visual images is also rapidly developed. Due to the reasons of weak texture, variable form, diversity and the like of the flame image, the traditional flame detection algorithm mainly identifies the flame by color and brightness and adds logic to identify the flame by a method of multiple times of confirmation. In recent years, due to the fact that the convolutional neural network is different in military projection, flame detection can be carried out by using a detection network, and detection precision and efficiency can be greatly improved. However, the existing detection network generally detects a single image, the shape of flames existing in the single image may be very similar to that of various high-brightness red lamps, and false detection is very easy to occur in practical use. Flame and light are almost the same when the flame and the light are seen in an infrared scene, the flame can irregularly flash but the light can not, so that a single image does not distinguish the flame from the light, the confidence of false detection is extremely high, the false detection and recall rate of flame detection cannot be balanced, and the flame detection device cannot be used in actual production on a large scale.
The existing flame detection technology based on video realizes the detection of flame in video by extracting the static and dynamic characteristics of video, and can effectively detect flame by considering the characteristics of space motion, time continuity and the like of flame; however, the static and dynamic feature extraction algorithms and classification models are often very complex, for example, complex processing methods such as multilayer wavelet decomposition, Hu invariant moment, kalman filtering, markov models, etc. or neural networks with very complex structures need to be used, so that the timeliness of flame detection is poor, the requirements on software and hardware resources are very high, and the false detection rate of actual detection is also poor.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art and provide a flame detection method which has high detection accuracy, good real-time performance and low requirements on software and hardware resources.
The invention specifically adopts the following technical scheme to solve the technical problems:
a flame detection method uses a flame detection method based on a single frame image to carry out frame-by-frame initial detection on a video, when flame is detected initially, flame area gray level images are respectively extracted from a plurality of continuous frame images starting from the current frame image in which the flame is detected, and a frame difference image of the flame area gray level images between adjacent frame images is obtained; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
Preferably, the video is a visible light video or an infrared light video.
Further preferably, the frame difference image is subjected to expansion, gaussian blurring and normalization in sequence before use.
Preferably, the preliminary detection is implemented by a pre-trained neural network model.
Based on the same inventive concept, the following technical scheme can be obtained:
a flame detection device comprising:
the preliminary detection module is used for carrying out preliminary detection on the video frame by using a flame detection method based on a single-frame image;
the rechecking module is used for rechecking the detection result of the preliminary detection module according to the following method: when the flame is preliminarily detected, respectively extracting the flame area gray level images from a plurality of continuous frame images starting from the current flame detection image frame, and acquiring a frame difference image of the flame area gray level images between adjacent frame images; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
Preferably, the video is a visible light video or an infrared light video.
Further preferably, the frame difference image is subjected to expansion, gaussian blurring and normalization in sequence before use.
Preferably, the preliminary detection module is a pre-trained neural network model.
Compared with the prior art, the technical scheme and the optimal scheme thereof have the following beneficial effects:
on the basis of a flame detection result based on a single frame image, according to a flame abnormal motion physical principle, an image frame time sequence is simply processed, a high-frequency jumping flame detection result is reserved, low-frequency abnormal motion such as personnel movement, background movement and the like is eliminated, and compared with the existing flame detection based on the single frame image, the flame detection accuracy is greatly improved, and the false detection rate is reduced; compared with the existing video flame detection technology, the method has the remarkable advantages of good real-time performance, low requirements on software and hardware resources and easiness in implementation.
Drawings
FIG. 1 illustrates preliminary flame detection results in an exemplary embodiment;
FIG. 2 is an image of a flame region in a video frame at time t in an exemplary embodiment;
FIG. 3 is a frame difference image of a flame region gray scale image between a video frame and a previous frame image at time t in an embodiment;
FIG. 4 is a frame difference image after sequential dilation, Gaussian blur, and normalization in an embodiment;
FIG. 5 is a graph of probability of flame detection obtained in an exemplary embodiment.
Detailed Description
Aiming at the defects in the prior art, the invention solves the problem that the false detection rate is greatly reduced by simply processing the image frame time sequence and reserving the high-frequency jumping flame detection result and eliminating low-frequency abnormal motion such as personnel movement, background movement and the like on the basis of the flame detection result based on a single frame image according to the flame abnormal motion physical principle, thereby keeping the advantages of good real-time performance, low requirements on software and hardware resources and easy realization of the single frame image flame detection technology.
Specifically, the flame detection method provided by the invention specifically comprises the following steps:
performing frame-by-frame initial detection on a video by using a flame detection method based on a single frame image, when flame is initially detected, respectively extracting flame area gray level images from a plurality of continuous frame images starting from an image frame of the currently detected flame, and acquiring a frame difference image of the flame area gray level images between adjacent frame images; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
The flame detection device provided by the invention comprises:
the preliminary detection module is used for carrying out preliminary detection on the video frame by using a flame detection method based on a single-frame image;
the rechecking module is used for rechecking the detection result of the preliminary detection module according to the following method: when the flame is preliminarily detected, respectively extracting the flame area gray level images from a plurality of continuous frame images starting from the current flame detection image frame, and acquiring a frame difference image of the flame area gray level images between adjacent frame images; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
In the above technical solution, the preliminary detection may adopt various existing flame detection methods based on a single frame image, such as SSD, YOLO, RCNN, and the like, and preferably adopts an implementation scheme based on a neural network, that is, a pre-trained neural network model is used to implement the preliminary flame detection of a single frame image.
In order to further improve the detection effect, the frame difference image is preferably subjected to expansion, gaussian blurring and normalization in sequence before use.
The technical scheme of the invention comprehensively completes accurate detection of dynamic flame by utilizing single image detection, time sequence and physical rules, is suitable for visible light videos and infrared light videos, can effectively avoid false detection caused by lamps similar to flame, red background and the like, and can accurately identify even flame printed images.
To facilitate understanding of the public, the technical solution of the present invention is further described in detail by a specific embodiment in combination with the attached drawings:
the preliminary detection module of this embodiment adopts the deep learning neural network model, and its flame detection frame that adopts is the rectangle detection frame, and its process of realizing flame detection specifically is as follows:
1. training a deep learning neural network for detecting flame in advance by using the existing single-frame image detection method to obtain a flame detection network M;
2. in the actual detection process, a gunlock, a camera and other modes are used for acquiring a video stream, the video stream is analyzed into a frame-by-frame image, and an image I is obtainedt(t represents the current time) input into the flame detection netObtaining a preliminary flame detection result D ═ M (I) shown in FIG. 1 by complexing with Mt) (D represents a set of multiple possible flame zones [ [ x1, y1, x2, y2 ]],....,[x1,y1,x2,y2]]And x1, y1, x2, y2 indicate the upper left and lower right coordinates of the detected flame region).
3. From image I on the basis of detection result DtAnd then respectively cutting off images of the flame region part in a continuous series of frame images (the specific number of the frame images can be flexibly selected according to actual conditions, for example, image frames with the time length of 1 second-2.5 seconds can be taken), so as to generate a flame region image sequence, wherein the flame region image at the time t is represented as It(D) Converting all images in the flame area image sequence from RGB images into gray level images or extracting R channel gray level images to obtain a flame area gray level image sequence, wherein the flame area gray level image at the time t is represented as S(h,w,1) t。
4. Calculating a flame region gray level image S(h,w,1) tAnd the gray level image S of the flame region of the previous frame(h,w,1) t-1Frame difference image diff of(h,w,1) t＝||S(h,w,1) t-S(h,w,1) t-1And | | l, the obtained frame difference image is as shown in fig. 3, thereby obtaining a frame difference image sequence.
5. Sequentially performing expansion operation and Gaussian blur on the frame difference images in the frame difference image sequence, and finally normalizing and outputting the processed frame difference image sequence and the processed frame difference image diff(h,w,1) tIs converted into diff by the above treatment(h,w,1) t', as shown in FIG. 4;
6. all the images in the processed frame difference image sequence are superposed to be used as the outline image of the object detected in an accumulation way
The processed frame difference image sequence diff(h,w,1) t' all the images are inverted and then superimposed, and the images are used as the non-detected area transaction accumulated images
The two are weighted and summed to obtain a flame detection probability map P shown in FIG. 5(h,w,1)＝αP+ (h,w,1)+βP- (h,w,1)Indicating a frequently occurring abnormal movement region; wherein, the weights alpha and beta are respectively used for controlling the flame sensitivity and the background anti-interference capability.
7. According to the flame abnormal movement physical principle, the flame does not move greatly in the initial stage, and the airflow disturbance can present irregular high-frequency disturbance during the combustion of the flame; with the continuous accumulation of the frame difference diff before and after, the moving area is continuously accumulated, the non-moving area is continuously subtracted, after a period of accumulation, the probability density of the frequent high-frequency abnormal movement area is gradually increased, the low-frequency abnormal movement (such as personnel movement, background movement and the like) hardly changes at the same position in high frequency, and after the period of probability accumulation, the high-frequency jumping flame can be well detected. Based on the principle, the preliminary detection result can be rechecked according to the pixel distribution condition in the flame detection probability image, specifically, if the number of pixels of which the pixel values exceed the preset threshold value in the flame detection probability image exceeds the preset number, the preliminary detection result is judged to be correct, otherwise, the preliminary detection result is false detection. The preset threshold and the preset number can be determined by calculation or actual experiments according to actual conditions.
By adopting the technical scheme, all possibly false-reported flame detection results of the existing flame detection technology based on single-frame images can be almost eliminated, and flame-like objects can not be detected by mistake any more. Compared with the existing video flame detection technology, the method has the remarkable advantages of good real-time performance, low requirements on software and hardware resources and easiness in implementation.
Claims (8)
1. A flame detection method is characterized in that a flame detection method based on a single frame image is used for carrying out frame-by-frame preliminary detection on a video, when flame is detected preliminarily, a flame area gray level image is respectively extracted from a plurality of continuous frame images starting from an image frame of the currently detected flame, and a frame difference image of the flame area gray level image between adjacent frame images is obtained; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
2. The flame detection method of claim 1, wherein the video is a visible light video or an infrared light video.
3. The flame detection method of claim 1, wherein the frame difference image is subjected to dilation, gaussian blurring, and normalization in sequence before use.
4. The flame detection method of claim 1, wherein the preliminary detection is performed by a pre-trained neural network model.
5. A flame detection device, comprising:
the preliminary detection module is used for carrying out preliminary detection on the video frame by using a flame detection method based on a single-frame image;
the rechecking module is used for rechecking the detection result of the preliminary detection module according to the following method: when the flame is preliminarily detected, respectively extracting the flame area gray level images from a plurality of continuous frame images starting from the current flame detection image frame, and acquiring a frame difference image of the flame area gray level images between adjacent frame images; and carrying out weighted summation on the superposed images of all the frame difference images and the superposed images of the reverse images of all the frame difference images to obtain a flame detection probability image, if the number of pixels of which the pixel values exceed a preset threshold value in the flame detection probability image exceeds a preset number, judging that the preliminary detection result is correct, and otherwise, carrying out false detection.
6. The flame detection device of claim 5, wherein the video is a visible light video or an infrared light video.
7. The flame detection device of claim 5, wherein the frame difference image is subjected to dilation, Gaussian blur, and normalization in sequence prior to use.
8. The flame detection device of claim 5, wherein the preliminary detection module is a pre-trained neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110393455.4A CN113066077B (en) | 2021-04-13 | 2021-04-13 | Flame detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110393455.4A CN113066077B (en) | 2021-04-13 | 2021-04-13 | Flame detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113066077A CN113066077A (en) | 2021-07-02 |
| CN113066077B true CN113066077B (en) | 2021-11-23 |
Family
ID=76566540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110393455.4A Active CN113066077B (en) | 2021-04-13 | 2021-04-13 | Flame detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113066077B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113743378B (en) * | 2021-11-03 | 2022-02-08 | 航天宏图信息技术股份有限公司 | Fire monitoring method and device based on video |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997461A (en) * | 2017-03-28 | 2017-08-01 | 浙江大华技术股份有限公司 | A kind of firework detecting method and device |
| CN112001375A (en) * | 2020-10-29 | 2020-11-27 | 成都睿沿科技有限公司 | Flame detection method and device, electronic equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7505604B2 (en) * | 2002-05-20 | 2009-03-17 | Simmonds Precision Prodcuts, Inc. | Method for detection and recognition of fog presence within an aircraft compartment using video images |
-
2021
- 2021-04-13 CN CN202110393455.4A patent/CN113066077B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997461A (en) * | 2017-03-28 | 2017-08-01 | 浙江大华技术股份有限公司 | A kind of firework detecting method and device |
| CN112001375A (en) * | 2020-10-29 | 2020-11-27 | 成都睿沿科技有限公司 | Flame detection method and device, electronic equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于火焰动态模型融合的火焰检测算法研究;张瑞;《万方数据知识服务平台》;20160129;正文第2-4章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113066077A (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022099598A1 (en) | Video dynamic target detection method based on relative statistical features of image pixels | |
| CN107085714B (en) | Forest fire detection method based on video | |
| CN104796582B (en) | Video image denoising and Enhancement Method and device based on random injection retinex | |
| CN111723644A (en) | Method and system for detecting occlusion of surveillance video | |
| CN108898132B (en) | Terahertz image dangerous article identification method based on shape context description | |
| CN105513053B (en) | One kind is used for background modeling method in video analysis | |
| CN112184759A (en) | Moving target detection and tracking method and system based on video | |
| CN108198206A (en) | The multi-object tracking method combined based on multiple features combining and Camshift algorithms | |
| CN108280409B (en) | Large-space video smoke detection method based on multi-feature fusion | |
| CN111611907B (en) | Image-enhanced infrared target detection method | |
| CN107392095A (en) | A kind of small IR targets detection algorithm based on mask image | |
| CN115937237A (en) | Local feature extraction method based on edge transform domain | |
| CN111047624A (en) | Image dim target detection method, device, equipment and storage medium | |
| CN113066077B (en) | Flame detection method and device | |
| CN111460964A (en) | Moving target detection method under low-illumination condition of radio and television transmission machine room | |
| JP7096175B2 (en) | Object extraction method and device | |
| CN113902694A (en) | Target detection method based on dynamic and static combination | |
| CN107729811B (en) | Night flame detection method based on scene modeling | |
| Wu et al. | Video surveillance object recognition based on shape and color features | |
| CN113205494A (en) | Infrared small target detection method and system based on adaptive scale image block weighting difference measurement | |
| CN110502968B (en) | Method for detecting infrared small and weak moving target based on track point space-time consistency | |
| Miura et al. | The examination of the image correction of the moving-object detection for low illumination video image | |
| Qin et al. | A shadow removal algorithm for ViBe in HSV color space | |
| CN111797761B (en) | Three-stage smoke detection system, method and readable medium | |
| Chuang et al. | Moving object segmentation and tracking using active contour and color classification models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 210000 Longmian Avenue 568, High-tech Park, Jiangning District, Nanjing City, Jiangsu Province Patentee after: Xiaoshi Technology (Jiangsu) Co.,Ltd. Address before: 210000 Longmian Avenue 568, High-tech Park, Jiangning District, Nanjing City, Jiangsu Province Patentee before: NANJING ZHENSHI INTELLIGENT TECHNOLOGY Co.,Ltd. |