CN106845410B - Flame identification method based on deep learning model - Google Patents
Flame identification method based on deep learning model Download PDFInfo
- Publication number
- CN106845410B CN106845410B CN201710047239.8A CN201710047239A CN106845410B CN 106845410 B CN106845410 B CN 106845410B CN 201710047239 A CN201710047239 A CN 201710047239A CN 106845410 B CN106845410 B CN 106845410B
- Authority
- CN
- China
- Prior art keywords
- image
- spherical
- flame
- fisheye
- model
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000013136 deep learning model Methods 0.000 title claims abstract description 13
- 239000000779 smoke Substances 0.000 claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims abstract description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 230000003287 optical effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000002265 prevention Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G06T3/08—
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
- G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
Abstract
The invention relates to a flame identification method based on a deep learning model, which comprises the steps of (1) acquiring video information, reading each frame of image, and then performing Gaussian filtering to obtain a filtered fisheye image; (2) correcting internal parameters of the fisheye image; (3) correcting external parameters of the fisheye image; (4) constructing a spherical model, projecting the corrected fisheye image onto the spherical model, and then removing a repeated region projected on the spherical model to form a spherical image; (5) removing interference information of smoke in the spherical image to the identification of the flame part by using a smoke passing model; (6) acquiring a dynamic area on the spherical image; (7) generating a perspective image from the dynamic area part; (8) normalizing the perspective image, using the perspective image as the input of the trained seven-layer architecture convolutional neural network, identifying whether the dynamic area is flame, if the dynamic area is flame, entering the step (9), otherwise, ending the operation; (9) and displaying the identification result and generating alarm information.
Description
Technical Field
The invention belongs to the field of communication, and particularly relates to a flame identification method based on a deep learning model.
Background
With the continuous improvement of industrialization and town level in China, large public buildings with modern facilities are developed towards diversification with large space, wide depth and complex functions, which puts higher requirements on the development of smoke and fire prevention towards diversification with large space, wide depth and complex functions and the reliable, stable and high-precision design and operation of smoke and fire prevention and fire protection safety systems.
At present, the important research problem in the field of safety engineering is effectively improving the sensing capability of fire safety to the scene and the rapidity and the accuracy level of fire safety to the early warning treatment of the fire scene through the related technology in the field of information.
At present, a subway train fire detection alarm system based on the fuzzy characteristic of fire smoke images is designed by Chen civilian brightness and the like, and an image spatial domain differential method is adopted to detect the occurrence of fire. The Li Shi Rev et al adopts an infrared filter to filter and obtain a filtered image to realize the extraction of the flame target. The flame recognition is carried out by the steps of image segmentation, image enhancement, feature extraction and the like. Wangholon et al analyze the dynamic and static characteristics of flames including edge jitter, area variation and shape, color, culture, etc. However, the design system of the flame recognition system has the following common problems:
(1) the probability of false detection or missed detection is higher in the traditional method;
(2) whether flame characteristics exist in a field dynamic region or not is considered, and the algorithm is high in complexity;
(3) if the detected area is not a flame or is similar to the texture feature of the flame, the false detection is likely to occur;
(4) the observable fire field is small.
Disclosure of Invention
The purpose of the invention is as follows: the invention improves the problems in the prior art, namely, the invention discloses a flame identification method based on a deep learning model.
The technical scheme is as follows: a flame identification method based on a deep learning model comprises the following steps:
(1) acquiring video information through two back-to-back fisheye cameras, reading each frame image of the acquired video information, and performing Gaussian filtering on the acquired fisheye image to obtain a filtered fisheye image;
(2) correcting the internal parameters of the fisheye image obtained in the step (1), and then entering the step (3), wherein the internal parameters comprise a tangential error, a radial error and an optical center error;
(3) correcting external parameters of the fisheye image, and entering the step (4);
(4) constructing a spherical model, projecting the fisheye image corrected in the step (3) onto the spherical model, removing a repeated region projected on the spherical model to form a spherical image, and then entering the step (5);
(5) removing interference information of smoke on the flame part identification in the spherical image obtained in the step (4) through a smoke removing model, and then entering the step (6);
(6) acquiring a dynamic region on the spherical image by adopting an improved Codebook method, and then entering the step (7);
(7) generating a perspective image of the dynamic region part acquired in the step (6);
(8) normalizing the perspective image obtained in the step (7), using the perspective image as the input of a convolutional neural network with a trained seven-layer architecture, identifying whether the dynamic area is flame, if the dynamic area is flame, entering the step (9), otherwise, ending the operation;
(9) and transmitting the identification result to a lower computer, and displaying the identification result and generating alarm information on the lower computer.
Further, when the fisheye image in step (3) is converted into a 640 × 480 pixel image, if the number of corner points detected on the fisheye image is greater than 300, step (3) includes the following steps:
(31) generating a normalized image block with a certain perspective angle and a certain size on the spherical model by taking the detected feature points as centers;
(32) calculating the correlation values of the feature point pairs within a certain range on two adjacent frame images,
(33) and finally, obtaining the optimal matching pair and optimizing the relative posture of the camera.
Further, when the fisheye image in step (3) is converted into a 640 × 480 pixel image, if the number of corner points detected on the fisheye image is less than 300, step (3) includes the following steps:
(31) vanishing point pairs are obtained on the spherical model by utilizing the edges and the geometric structures detected on the image,
(32) and obtaining a transformation relation of a structure on the image on the basis of the vanishing point pairs so as to calculate and optimize the relative posture of the camera and further optimize the parameters.
Further, when the angle of the repeated field of view of the large-field-of-view image in step (3) is greater than 15 degrees, step (3) includes the steps of:
(31) a large-angle perspective image is generated through a spherical model,
(32) and calculating the correlation degree on the repeated visual field of the perspective image so as to feed back and adjust the posture of the camera, and generating the perspective image again for optimization after adjustment.
Has the advantages that: the flame identification method based on the deep learning model disclosed by the invention has the following beneficial effects:
1. the false detection rate or the missing detection rate is low;
2. the observable fire range has a large field of view.
Drawings
FIG. 1 is a flow chart of a flame identification method based on a deep learning model according to the present disclosure;
FIG. 2 is a schematic view of a spherical model;
fig. 3 is a perspective image schematic diagram based on a spherical image.
The specific implementation mode is as follows:
the following describes in detail specific embodiments of the present invention.
As shown in fig. 1 to 3, a flame recognition method based on a deep learning model includes the following steps:
(1) acquiring video information through two back-to-back fisheye cameras, reading each frame image of the acquired video information, and performing Gaussian filtering on the acquired fisheye image to obtain a filtered fisheye image;
(2) correcting the internal parameters of the fisheye image obtained in the step (1), and then entering the step (3), wherein the internal parameters comprise a tangential error, a radial error and an optical center error;
(3) correcting external parameters of the fisheye image, and entering the step (4);
(4) constructing a spherical model, projecting the fisheye image corrected in the step (3) onto the spherical model, removing a repeated region projected on the spherical model to form a spherical image, and then entering the step (5);
(5) removing interference information of smoke on the flame part identification in the spherical image obtained in the step (4) through a smoke removing model, and then entering the step (6);
(6) acquiring a dynamic region on the spherical image by adopting an improved Codebook method, and then entering the step (7);
(7) generating a perspective image of the dynamic region part acquired in the step (6);
(8) normalizing the perspective image obtained in the step (7), using the perspective image as the input of a convolutional neural network with a trained seven-layer architecture, identifying whether the dynamic area is flame, if the dynamic area is flame, entering the step (9), otherwise, ending the operation;
(9) and transmitting the identification result to a lower computer, and displaying the identification result and generating alarm information on the lower computer.
Further, when the fisheye image in step (3) is converted into a 640 × 480 pixel image, if the number of corner points detected on the fisheye image is greater than 300, step (3) includes the following steps:
(31) generating a normalized image block with a certain perspective angle and a certain size on the spherical model by taking the detected feature points as centers;
(32) calculating the correlation values of the feature point pairs within a certain range on two adjacent frame images,
(33) and finally, obtaining the optimal matching pair and optimizing the relative posture of the camera.
Further, when the fisheye image in step (3) is converted into a 640 × 480 pixel image, if the number of corner points detected on the fisheye image is less than 300, step (3) includes the following steps:
(31) vanishing point pairs are obtained on the spherical model by utilizing the edges and the geometric structures detected on the image,
(32) and obtaining a transformation relation of a structure on the image on the basis of the vanishing point pairs so as to calculate and optimize the relative posture of the camera and further optimize the parameters.
Further, when the angle of the repeated field of view of the large-field-of-view image in step (3) is greater than 15 degrees, step (3) includes the steps of:
(31) a large-angle perspective image is generated through a spherical model,
(32) and calculating the correlation degree on the repeated visual field of the perspective image so as to feed back and adjust the posture of the camera, and generating the perspective image again for optimization after adjustment.
Fig. 2 is a schematic view of a spherical model. In the calculation process, each pixel point on the fisheye image is projected to the azimuth angle on the spherical model
And an elevation angle theta. As shown in fig. 2, the spatial point P is projected to the point P on the spherical model. The azimuth angle can be calculated according to the position of the projection point
And an elevation angle theta. Any point on the spherical surface
Calculating coordinates (x) in three-dimensional spacep,yp,zp). And respectively calculating three-dimensional points on the two fisheye images in the processing process. Direct projection to space coordinates (x) calculation for pre-imagep,yp,zp) The transformation of the spatial coordinates for projection of the back image onto a spherical surface, as shown in equation (1), is as shown in equation (2), i.e. the coordinates are rotated 90 degrees counterclockwise.
From this, it can be seen that a repeated theta is obtained on both images,
in the range of (a), if the repetition of theta is obtained,
the range of (2) is that only the pixel points on one image are taken for projection. The method can effectively solve the problem that the repeated area exists in the projection.
As shown in fig. 3, a perspective image schematic diagram based on a spherical image. O is the center of a circle, willAnd generating a perspective image block with fixed length and width by taking any point p on the spherical surface as a center. P is the center point on the perspective image block, OpP is collinear. The radius of the sphere is rsGeneral case rs1, namely a unit sphere. lsThe arc length from the projection point of the perspective image boundary to the intersection point of the sphere center and the spherical surface to the point p is phi, which is the included angle between the projection line of the perspective image boundary and the OpP line. If the size of the perspective image is fixed, the larger phi is, the larger the visual field of the perspective image is, and the closer the perspective image is to the spherical surface, otherwise, the perspective image is far away from the spherical surface. The zoom effect can be achieved by adjusting the phi angle. The establishment of the spherical model can acquire perspective image blocks with variable focal length in any viewpoint direction without dead angles in 360 degrees. And a foundation is laid for subsequent flame identification.
The embodiments of the present invention have been described in detail. However, the present invention is not limited to the above-described embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present invention.
Claims (4)
1. A flame identification method based on a deep learning model is characterized by comprising the following steps:
(1) acquiring video information through two back-to-back fisheye cameras, reading each frame image of the acquired video information, and performing Gaussian filtering on the acquired fisheye image to obtain a filtered fisheye image;
(2) correcting the internal parameters of the fisheye image obtained in the step (1), and then entering the step (3), wherein the internal parameters comprise a tangential error, a radial error and an optical center error;
(3) correcting external parameters of the fisheye image, and entering the step (4);
(4) constructing a spherical model, projecting the fisheye image corrected in the step (3) onto the spherical model, removing a repeated region projected on the spherical model to form a spherical image, and then entering the step (5);
(5) removing interference information of smoke on the flame part identification in the spherical image obtained in the step (4) through a smoke removing model, and then entering the step (6);
(6) acquiring a dynamic region on the spherical image by adopting an improved Codebook method, and then entering the step (7);
(7) generating a perspective image of the dynamic region part acquired in the step (6);
(8) normalizing the perspective image obtained in the step (7), using the perspective image as the input of a convolutional neural network with a trained seven-layer architecture, identifying whether the dynamic area is flame, if the dynamic area is flame, entering the step (9), otherwise, ending the operation;
(9) and transmitting the identification result to a lower computer, and displaying the identification result and generating alarm information on the lower computer.
2. The flame identification method based on the deep learning model as claimed in claim 1, wherein when the fisheye image in step (3) is converted into a 640 × 480 pixel map, if the number of corner points detected on the fisheye image is greater than 300, step (3) comprises the following steps:
(31) generating a normalized image block with a certain perspective angle and a certain size on the spherical model by taking the detected feature points as centers;
(32) calculating the correlation values of the feature point pairs within a certain range on two adjacent frame images,
(33) and finally, obtaining the optimal matching pair and optimizing the relative posture of the camera.
3. The flame identification method based on the deep learning model as claimed in claim 1, wherein when the fisheye image in step (3) is converted into a 640 x 480 pixel map, if the number of corner points detected on the fisheye image is less than 300, step (3) comprises the following steps:
(31) vanishing point pairs are obtained on the spherical model by utilizing the edges and the geometric structures detected on the image,
(32) and obtaining a transformation relation of a structure on the image on the basis of the vanishing point pairs so as to calculate and optimize the relative pose of the camera and further optimize the external parameters.
4. The flame recognition method based on the deep learning model as claimed in claim 1, wherein when the angle of the repeated field of view of the large-field-of-view image in the corrected fisheye image in step (3) is greater than 15 degrees, step (3) comprises the following steps:
(31) a large-angle perspective image is generated through a spherical model,
(32) and calculating the correlation degree on the repeated visual field of the perspective image so as to feed back and adjust the posture of the camera, and generating the perspective image again for optimization after adjustment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710047239.8A CN106845410B (en) | 2017-01-22 | 2017-01-22 | Flame identification method based on deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710047239.8A CN106845410B (en) | 2017-01-22 | 2017-01-22 | Flame identification method based on deep learning model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106845410A CN106845410A (en) | 2017-06-13 |
| CN106845410B true CN106845410B (en) | 2020-08-25 |
Family
ID=59120999
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710047239.8A Expired - Fee Related CN106845410B (en) | 2017-01-22 | 2017-01-22 | Flame identification method based on deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106845410B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830143A (en) * | 2018-05-03 | 2018-11-16 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of video analytic system based on deep learning |
| CN108985374A (en) * | 2018-07-12 | 2018-12-11 | 天津艾思科尔科技有限公司 | A kind of flame detecting method based on dynamic information model |
| CN109165575B (en) * | 2018-08-06 | 2024-02-20 | 天津艾思科尔科技有限公司 | Pyrotechnic recognition algorithm based on SSD frame |
| US11927944B2 (en) * | 2019-06-07 | 2024-03-12 | Honeywell International, Inc. | Method and system for connected advanced flare analytics |
| CN111310662B (en) * | 2020-02-17 | 2021-08-31 | 淮阴工学院 | Flame detection and identification method and system based on integrated deep network |
| CN113450530A (en) * | 2021-06-03 | 2021-09-28 | 河北华电石家庄热电有限公司 | Safety early warning system based on intelligent video analysis algorithm |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101441712A (en) * | 2008-12-25 | 2009-05-27 | 北京中星微电子有限公司 | Flame video recognition method and fire hazard monitoring method and system |
| CN101625789A (en) * | 2008-07-07 | 2010-01-13 | 北京东方泰坦科技有限公司 | Method for monitoring forest fire in real time based on intelligent identification of smoke and fire |
| CN102163358A (en) * | 2011-04-11 | 2011-08-24 | 杭州电子科技大学 | Smoke/flame detection method based on video image analysis |
| CN104200468A (en) * | 2014-08-28 | 2014-12-10 | 中国矿业大学 | Method for obtaining correction parameter of spherical perspective projection model |
| CN104581076A (en) * | 2015-01-14 | 2015-04-29 | 国网四川省电力公司电力科学研究院 | Mountain fire monitoring and recognizing method and device based on 360-degree panoramic infrared fisheye camera |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7202794B2 (en) * | 2004-07-20 | 2007-04-10 | General Monitors, Inc. | Flame detection system |
-
2017
- 2017-01-22 CN CN201710047239.8A patent/CN106845410B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101625789A (en) * | 2008-07-07 | 2010-01-13 | 北京东方泰坦科技有限公司 | Method for monitoring forest fire in real time based on intelligent identification of smoke and fire |
| CN101441712A (en) * | 2008-12-25 | 2009-05-27 | 北京中星微电子有限公司 | Flame video recognition method and fire hazard monitoring method and system |
| CN102163358A (en) * | 2011-04-11 | 2011-08-24 | 杭州电子科技大学 | Smoke/flame detection method based on video image analysis |
| CN104200468A (en) * | 2014-08-28 | 2014-12-10 | 中国矿业大学 | Method for obtaining correction parameter of spherical perspective projection model |
| CN104581076A (en) * | 2015-01-14 | 2015-04-29 | 国网四川省电力公司电力科学研究院 | Mountain fire monitoring and recognizing method and device based on 360-degree panoramic infrared fisheye camera |
Non-Patent Citations (2)
| Title |
|---|
| Computer vision based method for real-time fire and flame detection;B. Ugˇur To¨reyin 等;《Pattern Recognition Letters 27》;20050826;第49-58页 * |
| 基于视觉传感器的火焰识别与高压细水雾喷淋实验研究;丁飞;《中国优秀硕士学位论文全文数据库 工程科技II辑》;20140815(第08期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106845410A (en) | 2017-06-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106845410B (en) | Flame identification method based on deep learning model | |
| CN105225230B (en) | A kind of method and device of identification foreground target object | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN109118523A (en) | A kind of tracking image target method based on YOLO | |
| CN110956661B (en) | Method for calculating dynamic pose of visible light and infrared camera based on bidirectional homography matrix | |
| CN111680588A (en) | Human face gate living body detection method based on visible light and infrared light | |
| CN109308718B (en) | Space personnel positioning device and method based on multiple depth cameras | |
| CN108596193B (en) | Method and system for building deep learning network structure aiming at human ear recognition | |
| JP2017534046A (en) | Building height calculation method, apparatus and storage medium | |
| CN110189375B (en) | Image target identification method based on monocular vision measurement | |
| CN111723801B (en) | Method and system for detecting and correcting target in fisheye camera picture | |
| CN106033614B (en) | A kind of mobile camera motion object detection method under strong parallax | |
| CN106875419A (en) | Small and weak tracking of maneuvering target based on NCC matching frame differences loses weight detecting method | |
| CN109086724A (en) | A kind of method for detecting human face and storage medium of acceleration | |
| CN107038714B (en) | Multi-type visual sensing cooperative target tracking method | |
| CN107560592A (en) | A kind of precision ranging method for optronic tracker linkage target | |
| JP2009288885A (en) | Lane detection device, lane detection method and lane detection program | |
| CN106295657A (en) | A kind of method extracting human height's feature during video data structure | |
| Xu et al. | Dynamic obstacle detection based on panoramic vision in the moving state of agricultural machineries | |
| CN106915303A (en) | Automobile A-column blind area perspective method based on depth data and fish eye images | |
| CN109919832A (en) | One kind being used for unpiloted traffic image joining method | |
| CN114998737A (en) | Remote smoke detection method, system, electronic equipment and medium | |
| CN112652020A (en) | Visual SLAM method based on AdaLAM algorithm | |
| CN111047636A (en) | Obstacle avoidance system and method based on active infrared binocular vision | |
| Chowdhury et al. | Robust human detection and localization in security applications |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200825 Termination date: 20210122 |