技术领域technical field
本发明属于计算机视觉领域，涉及一种降低森林火灾预警误报率的方法。The invention belongs to the field of computer vision and relates to a method for reducing the false alarm rate of forest fire early warning.
背景技术Background technique
目前在使用带有电动云台的摄像头进行森林火灾监控的场景中，通常都是由前端摄像头进行可见光视频和热红外视频信息的采集，火灾预警信息的产生流程往往都是首先设定一个以红外成像参数标定的阈值，然后当当前帧的红外视频信息中有像素的数值超过预先设定的阈值时，则系统发出报警。这种方法的优点是所有超过设定阈值的异常情况基本上都可以检测出来，由于火焰的温度一般都在设定的阈值以上，所以可以保障在热红外监控下所有的火灾场景都不会被遗漏。其缺点是，可能会产生大量虚假的报警，例如当摄像头区域包含部分天空而此时太阳又恰好在该区域，则系统会不断的产生报警。另外，工厂中的机器、马路上行驶的汽车、村庄中人们使用的各种热源等都会产生这些虚假报警。At present, in the scene where a camera with an electric pan/tilt is used for forest fire monitoring, the front-end camera usually collects visible light video and thermal infrared video information. The threshold value calibrated by the imaging parameters, and then when the value of a pixel in the infrared video information of the current frame exceeds the preset threshold value, the system will send out an alarm. The advantage of this method is that all abnormal situations exceeding the set threshold can basically be detected. Since the temperature of the flame is generally above the set threshold, it can be guaranteed that all fire scenes will not be detected under thermal infrared monitoring. omission. Its disadvantage is that a large number of false alarms may be generated. For example, when the camera area includes part of the sky and the sun happens to be in this area at this time, the system will continue to generate alarms. In addition, machines in factories, cars driving on the road, various heat sources used by people in villages, etc. will all generate these false alarms.
对图像进行语义标注是近年来计算机视觉领域的研究热点之一。对图像进行语义标注有人工、半人工以及计算机自动完成等方式。目前的研究主要集中在对图像进行自动标注上。图像上下文环境理解也是计算机视觉领域的研究热点之一，对计算机视觉的应用具有重要影响，比如在一幅图像中汽车出现在马路这个上下文环境中是正常的，但是当汽车出现在天空这个上下文环境中则是异常事件；一般文本环境下，如图像或视频编码中，上下文信息的获取是比较简单的，只要得到相邻像素的像素值即可，但是在图像中就必须通过分割、识别等获取图像上下文信息。考虑到本发明应用的场景，一般一个电动云台摄像头覆盖的检测范围基本上都是固定的，因此不必通过计算机自动标注，使用人工在全景拼接完成后标注一次即可。这些标注的区域及其对应的规则就构成了检验报警是否虚假的图像上下文环境。相关研究在森林火灾视觉监控系统中尚未见报道。Semantic annotation of images is one of the research hotspots in the field of computer vision in recent years. There are manual, semi-manual, and computer-automatic methods for semantically annotating images. Current research mainly focuses on automatic annotation of images. Image context understanding is also one of the research hotspots in the field of computer vision, which has an important impact on the application of computer vision. For example, it is normal for a car to appear in the context of the road in an image, but when the car appears in the context of the sky In the middle, it is an abnormal event; in the general text environment, such as image or video coding, the acquisition of context information is relatively simple, as long as the pixel value of the adjacent pixel is obtained, but in the image, it must be obtained through segmentation, recognition, etc. Image context information. Considering the application scenarios of the present invention, generally the detection range covered by an electric pan-tilt camera is basically fixed, so it is not necessary to automatically mark by a computer, and it is only necessary to manually mark once after the panorama stitching is completed. These marked areas and their corresponding rules constitute the image context for checking whether the alarm is false. Related research has not been reported in the forest fire visual monitoring system.
发明内容Contents of the invention
针对现有技术中存在的上述技术问题，本发明提出了一种一种基于图像语义上下文环境降低森林火灾预警误报率的方法，设计合理，克服了现有技术的不足，具有良好的效果。Aiming at the above-mentioned technical problems in the prior art, the present invention proposes a method for reducing the false alarm rate of forest fire warning based on the image semantic context environment, which has a reasonable design, overcomes the deficiencies of the prior art, and has good effects.
为了实现上述目的，本发明采用如下技术方案：In order to achieve the above object, the present invention adopts the following technical solutions:
一种基于图像语义上下文环境降低森林火灾预警误报率的方法，该方法步骤如下：A method for reducing the false alarm rate of forest fire warning based on image semantic context environment, the method steps are as follows:
步骤1：启动用于森林防火监控的摄像头，获得的所监控区域的全景场景图，并进行预处理，获得监控区域中不同场景的标注，形成火灾误警消除的图像语义上下文环境。Step 1: Start the camera used for forest fire monitoring, obtain the panoramic scene map of the monitored area, and perform preprocessing to obtain the annotations of different scenes in the monitored area, and form an image semantic context environment for fire false alarm elimination.
所述步骤1具体包括如下步骤：The step 1 specifically includes the following steps:
步骤1.1：按特定方法获得当前监控区域的全景场景图像，记录全景图像每一个像素点对应的采集参数，包括焦距、水平方位角和垂直方位角，其中焦距来自于摄像头，水平方位角和垂直方位角取自于云台，以云台的正北方向和水平方向对应的中心像素点作为起始零度点，确定全景图像上每一个像素点的水平方位角和垂直方位角,并将这些信息写入全景像素信息数据库Dp；Step 1.1: Obtain the panoramic scene image of the current monitoring area according to a specific method, and record the acquisition parameters corresponding to each pixel of the panoramic image, including focal length, horizontal azimuth and vertical azimuth, where the focal length comes from the camera, horizontal azimuth and vertical azimuth The angle is taken from the gimbal, and the center pixel point corresponding to the true north direction and the horizontal direction of the gimbal is used as the starting zero point to determine the horizontal azimuth and vertical azimuth of each pixel on the panoramic image, and write this information Enter the panoramic pixel information database Dp;
步骤1.2：在全景场景图像中由人工划定不同区域，包括：天空、湖泊、河流、马路、铁路、村庄、工厂等，并将不同区域的名称及其范围保存至图像场景语义数据库Ds。Step 1.2: Manually define different regions in the panoramic scene image, including: sky, lake, river, road, railway, village, factory, etc., and save the names and ranges of different regions to the image scene semantic database Ds.
步骤1.3：为图像场景语义数据库Ds各种不同区域设置相应的行为规则如村庄区域在早中晚饭时间会有明显的烟雾，在天空区域太阳在不同时间段可能出现的区域位置等。Step 1.3: Set corresponding behavior rules for different areas of the image scene semantic database Ds, such as the village area will have obvious smoke in the morning, lunch and dinner, and the area where the sun may appear in different time periods in the sky area, etc.
步骤2：对所需监控区域进行监控，获取当前监控帧，如果当前有预警信息则进入步骤3，否则继续进行监控。Step 2: Monitor the required monitoring area and obtain the current monitoring frame. If there is current warning information, go to step 3, otherwise continue monitoring.
步骤3：如果有预警信息，则将相关信息与图像语义上下文环境信息进行对比，消除虚假报警。所述步骤(3)具体包括如下步骤：Step 3: If there is warning information, compare the relevant information with the image semantic context information to eliminate false alarms. Described step (3) specifically comprises the following steps:
步骤3.1：根据当前输入的视频帧对应的焦距和全景图像的焦距，对输入视频帧进行缩小或放大，使输入视频帧与全景图像中目标大小相互匹配，然后根据输入视频帧对应的云台参数和数据库Dp中的信息进行图像位置定位；Step 3.1: According to the focal length corresponding to the currently input video frame and the focal length of the panoramic image, shrink or enlarge the input video frame so that the input video frame and the target size in the panoramic image match each other, and then according to the PTZ parameters corresponding to the input video frame Carry out image position positioning with the information in the database Dp;
步骤3.2：根据视频帧的定位信息，进行预警位置的定位；Step 3.2: According to the positioning information of the video frame, the positioning of the warning position is carried out;
步骤3.3：将重新标定过的预警位置与所述图像场景语义数据库Ds中的信息进行对比，确定预警出现的位置，并与所述相应的行为规则进行验证，如果符合规则，则消除报警要求，否则通过报警要求；如果预警位置出现在森林区域，则通过报警要求。Step 3.3: Compare the re-calibrated warning position with the information in the image scene semantic database Ds, determine the position where the warning occurs, and verify it with the corresponding behavior rules. If the rules are met, eliminate the warning requirement, Otherwise, the alarm requirement is passed; if the warning position appears in the forest area, the alarm requirement is passed.
本发明所带来的有益技术效果如下：The beneficial technical effects brought by the present invention are as follows:
本发明使用在全景拼接图像基础上建立的全景像素信息数据库Dp和图像场景语义数据库Ds获得消除虚假报警的图像语义上下文环境信息；较好地消除了虚假报警，大幅度的提高了火灾预警的准确率。The invention uses the panoramic pixel information database Dp and the image scene semantic database Ds established on the basis of panoramic mosaic images to obtain the image semantic context environment information for eliminating false alarms; the false alarms are better eliminated, and the accuracy of fire early warning is greatly improved Rate.
附图说明Description of drawings
图1为基于图像语义上下文环境降低森林火灾预警误报率的方法的流程框图。Figure 1 is a flowchart of a method for reducing the false alarm rate of forest fire warning based on image semantic context.
图2为用于消除虚假报警的图像语义上下文环境的构建的流程框图。Fig. 2 is a flowchart of the construction of the image semantic context environment for eliminating false alarms.
图3为对预警信息进行虚假消除的流程框图。Fig. 3 is a flow chart of false elimination of early warning information.
具体实施方式Detailed ways
下面结合附图以及具体实施方式对本发明作进一步详细说明。The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.
针对使用带有电动云台的摄像头进行森林火灾监控的场景中虚假报警次数较多，如何提升报警准确的问题，本发明提出了一种基于图像语义上下文环境降低森林火灾预警误报率的方法，其流程框图如图1所示。该方法主要包括两部分，一部分是用于消除虚假报警的图像语义上下文环境的构建，包括：监控区域的全景像素信息数据库Dp和图像场景语义数据库Ds的建立，另一部分是对预警信息进行虚假消除。下面展开具体说明。Aiming at the problem of how to improve the accuracy of the alarm when the number of false alarms is high in the forest fire monitoring scene using a camera with an electric pan/tilt, the present invention proposes a method for reducing the false alarm rate of forest fire early warning based on the image semantic context environment. Its flowchart is shown in Figure 1. The method mainly includes two parts, one part is the construction of the image semantic context environment for eliminating false alarms, including: the establishment of the panoramic pixel information database Dp and the image scene semantic database Ds of the monitoring area, and the other part is the false elimination of the early warning information . The following is a detailed description.
1、用于消除虚假报警的图像语义上下文环境的构建，其流程框图如图2所示。1. The construction of the image semantic context environment for eliminating false alarms, the flowchart of which is shown in Figure 2.
首先根据特定方法获得当前监控区域的全景场景图像，同时记录全景图像每一个像素点对应的采集参数Q(f,ωh,ωv)，包括焦距f、水平方位角ωh和垂直方位角ωv，其中焦距来自于摄像头，水平方位角和垂直方位角取自于云台，以云台的正北方向和水平方向对应的中心像素点作为起始零度点，确定全景图像上每一个像素点的水平方位角ωh和垂直方位角ωv,并将这些信息全景像素信息数据库Dp中。First, obtain the panoramic scene image of the current monitoring area according to a specific method, and record the acquisition parameters Q(f,ωh ,ωv ) corresponding to each pixel of the panoramic image, including focal length f, horizontal azimuth ωh and vertical azimuth ωv , where the focal length comes from the camera, the horizontal azimuth and vertical azimuth are taken from the gimbal, and the center pixel point corresponding to the true north direction and the horizontal direction of the gimbal is used as the starting zero point to determine each pixel on the panoramic image The horizontal azimuth ωh and the vertical azimuth ωv , and put these information into the panoramic pixel information database Dp.
在得到每一个像素点的所需信息后，就需要对全景场景图像中由人工划定不同场景，主要包括天空、湖泊、河流、马路、铁路、村庄、工厂等。在这里设想的是通过基于上下文感知的图像标注算法。对于输入的未加标注的全景图和带有标签图像的混合区域集，使用支持向量机制识别分类图像区域并标注。对于未知区域建立上下文描述符，并结合图像视觉特征进行聚类标注，在此之后引入标签共生信息对每幅图的标签集进行修正，最后将不同区域的名称及其范围保存至图像场景语义数据库Ds中。After obtaining the required information for each pixel, it is necessary to manually delineate different scenes in the panoramic scene image, mainly including sky, lake, river, road, railway, village, factory, etc. What is envisioned here is through context-aware based image annotation algorithms. For input unlabeled panoramas and a mixed region set with labeled images, support vector mechanisms are used to identify and label classified image regions. Create context descriptors for unknown areas, and perform clustering and labeling in combination with image visual features, then introduce label co-occurrence information to correct the label set of each image, and finally save the names and ranges of different areas to the image scene semantic database Ds.
接下来就是为图像场景语义数据库Ds各种不同区域设置相关行为规则，如村庄区域在早中晚饭时间会有明显的烟雾，在铁路上的固定时刻都有火车经过，在天空区域太阳在不同时间段可能出现的区域位置等。行为规则一般包括：对象、时间、事件特征等不同属性，一个区域的行为规则可能有很多条。The next step is to set relevant behavior rules for different areas of the image scene semantic database Ds, such as the village area will have obvious smog in the morning, lunch and dinner, trains pass by at fixed times on the railway, and the sun in the sky area at different times The location of the region where the segment may appear, etc. Behavior rules generally include: different attributes such as object, time, and event characteristics, and there may be many behavior rules in an area.
2、对预警信息进行虚假消除，其流程框图如图3所示。2. False elimination of early warning information, the flow chart is shown in Figure 3.
根据当前输入的视频帧I对应的焦距fa和全景图像P的焦距fp，对输入视频帧进行缩小或放大，其公式为：According to the focal length fa corresponding to the currently input video frame I and the focal length fp of the panoramic image P, the input video frame is reduced or enlarged, and the formula is:
I'＝Z(I,fa,fp)I'=Z(I,fa,fp)
上式中Z表示采取的具体的缩放技术，在具体实现时需要根据参数fa，fp考虑最终生成图像I'的平滑性问题。在使输入视频帧与全景图像中目标大小相互匹配后，根据输入视频帧对应的云台参数和数据库Dp中的信息进行图像位置定位，其公式为：In the above formula, Z represents the specific scaling technology adopted, and the smoothness of the final generated image I' needs to be considered according to the parameters fa and fp during specific implementation. After matching the size of the target in the input video frame and the panoramic image, the image position is positioned according to the pan/tilt parameters corresponding to the input video frame and the information in the database Dp, the formula is:
I”＝Loc(P,I',Q)I"=Loc(P,I',Q)
上式中I”是定位后I'在全景图像P中的信息。根据I”的信息对预警位置进行重新定位，获得其在全景拼接图像中的位置。In the above formula, I" is the information of I' in the panoramic image P after positioning. Relocate the warning position according to the information of I", and obtain its position in the panoramic stitching image.
将重新标定过的预警位置与图像场景语义数据库Ds中的信息进行对比，确定预警位置出现的区域，并对该区域的各种行为规则进行验证，如果通过则消除报警要求，否则通过报警要求。Compare the re-calibrated warning position with the information in the image scene semantic database Ds, determine the area where the warning position appears, and verify the various behavior rules of the area. If it passes, the alarm requirement is eliminated, otherwise the alarm requirement is passed.
当然，上述说明并非是对本发明的限制，本发明也并不仅限于上述举例，本技术领域的技术人员在本发明的实质范围内所做出的变化、改型、添加或替换，也应属于本发明的保护范围。Of course, the above descriptions are not intended to limit the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or replacements made by those skilled in the art within the scope of the present invention shall also belong to the present invention. protection scope of the invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710079911.1ACN106815960B (en) | 2017-02-15 | 2017-02-15 | A method of reducing Forest Fire Alarm rate of false alarm |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710079911.1ACN106815960B (en) | 2017-02-15 | 2017-02-15 | A method of reducing Forest Fire Alarm rate of false alarm |
| Publication Number | Publication Date |
|---|---|
| CN106815960A CN106815960A (en) | 2017-06-09 |
| CN106815960Btrue CN106815960B (en) | 2018-10-02 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710079911.1AActiveCN106815960B (en) | 2017-02-15 | 2017-02-15 | A method of reducing Forest Fire Alarm rate of false alarm |
| Country | Link |
|---|---|
| CN (1) | CN106815960B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563471B (en)* | 2020-05-14 | 2024-01-05 | 杭州海康威视系统技术有限公司 | Factory risk assessment method and device |
| CN111667656A (en)* | 2020-06-05 | 2020-09-15 | 国网电力科学研究院有限公司 | Screening system and method for fire point of mountain fire in transmission line |
| CN115294531B (en)* | 2022-08-23 | 2025-02-07 | 浙江大华技术股份有限公司 | A smoke identification method, a smoke identification device and a storage medium |
| CN116863675A (en)* | 2023-07-28 | 2023-10-10 | 郑州畅威物联网科技有限公司 | Methods, systems, equipment and media for distinguishing true and false alarms of gas alarms |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101320487A (en)* | 2008-07-07 | 2008-12-10 | 中国科学院计算技术研究所 | A Scene Preprocessing Method for Fire Simulation |
| CN102164270A (en)* | 2011-01-24 | 2011-08-24 | 浙江工业大学 | Intelligent video monitoring method and system capable of exploring abnormal events |
| CN103152558A (en)* | 2013-03-29 | 2013-06-12 | 西南交通大学 | Intrusion detection method based on scene recognition |
| CN103530995A (en)* | 2013-10-12 | 2014-01-22 | 重庆邮电大学 | Video monitoring intelligent early-warning system and method on basis of target space relation constraint |
| KR20160126248A (en)* | 2015-04-23 | 2016-11-02 | 전상원 | Forest fire early warning system |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050103506A1 (en)* | 2003-11-18 | 2005-05-19 | Warrack Malcolm J. | Fire protection method |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101320487A (en)* | 2008-07-07 | 2008-12-10 | 中国科学院计算技术研究所 | A Scene Preprocessing Method for Fire Simulation |
| CN102164270A (en)* | 2011-01-24 | 2011-08-24 | 浙江工业大学 | Intelligent video monitoring method and system capable of exploring abnormal events |
| CN103152558A (en)* | 2013-03-29 | 2013-06-12 | 西南交通大学 | Intrusion detection method based on scene recognition |
| CN103530995A (en)* | 2013-10-12 | 2014-01-22 | 重庆邮电大学 | Video monitoring intelligent early-warning system and method on basis of target space relation constraint |
| KR20160126248A (en)* | 2015-04-23 | 2016-11-02 | 전상원 | Forest fire early warning system |
| Publication number | Publication date |
|---|---|
| CN106815960A (en) | 2017-06-09 |
| Publication | Publication Date | Title |
|---|---|---|
| CN106815960B (en) | A method of reducing Forest Fire Alarm rate of false alarm | |
| CN104966372B (en) | The forest fire intelligent identifying system and method for multi-data fusion | |
| JP3466169B2 (en) | Management system for roads and surrounding facilities | |
| CN113743256B (en) | Intelligent early warning method and device for site safety | |
| CN111462451A (en) | Straw burning detection and alarm system based on video information | |
| CN110348348A (en) | One kind personnel of taking part in building march into the arena identity method for quickly identifying and early warning system | |
| JP2008059319A (en) | Object recognition device and video object positioning device | |
| CN106331639A (en) | A method and device for automatically determining the position of a camera | |
| CN103152558B (en) | Based on the intrusion detection method of scene Recognition | |
| CN105608209B (en) | Video annotation method and video annotation device | |
| CN113867406B (en) | Unmanned aerial vehicle-based line inspection method, unmanned aerial vehicle-based line inspection system, intelligent equipment and storage medium | |
| CN113569956A (en) | Mountain fire disaster investigation and identification method based on AI algorithm | |
| CN117809441B (en) | Mobile sentinel danger early warning system for forest fire prevention | |
| CN114998737A (en) | A long-distance smoke detection method, system, electronic device and medium | |
| CN114566056A (en) | Highway tunnel driving safety risk identification, prevention and control method and system | |
| CN116468974B (en) | Smoke detection method, device and storage medium based on image generation | |
| CN103888731A (en) | Structured description device and system for mixed video monitoring by means of gun-type camera and dome camera | |
| CN117523426A (en) | Low-altitude target detection algorithm integrating feature pyramids | |
| KR101937582B1 (en) | Safety System for Walkway Monitor | |
| CN106960027B (en) | The UAV Video big data multidate association analysis method of spatial information auxiliary | |
| US11703820B2 (en) | Monitoring management and control system based on panoramic big data | |
| CN112668397A (en) | Fire real-time detection and analysis method and system, storage medium and electronic equipment | |
| CN114506221A (en) | Tunnel fire environment detection system and method based on high temperature superconducting maglev | |
| CN111428695B (en) | Straw combustion detection method based on deep learning and regional characteristics | |
| CN117237448B (en) | Train fusion positioning method and device based on machine vision |
| 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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right | Effective date of registration:20190709 Address after:266590 No. 579, Qian Wan Gang Road, Qingdao economic and Technological Development Zone, Shandong Co-patentee after:Qingdao Mount Ke Zhihui Information technology Co., Ltd Patentee after:Shandong Univ. of Science & Technology Address before:266590 No. 579, Qian Wan Gang Road, Qingdao economic and Technological Development Zone, Shandong Patentee before:Shandong Univ. of Science & Technology |