Smoke detection device, method for detecting smoke, and computer program

Abstract

Smoke detection device (1) for detecting smoke (10) from a fire (9) in a monitored area (U), with a camera interface (5) for receiving a sequence of images (3) with at least two temporally successive individual images (4), wherein the individual images (4) show the monitored area (U), with a difference module (6) for eliminating a scene background (8) of the monitoring area (U) in the individual frames (4) and for generating background-cleaned intermediate frames (13), with an evaluation module (14), wherein the evaluation module (14) is designed to perform a motion estimation to determine a movement of smoke sections of the smoke (10) in the monitoring area (U) based on the background-cleaned intermediate images (13), characterized by a background module (11), wherein the background module (11) is configured to create a background model (12) of the monitoring area (U), wherein the difference module (6) is configured to apply the background model (12) to the individual images (4) to generate the background-cleaned intermediate images (13), wherein the application of the background model (12) to the individual images (4) is configured as a division of the individual image (4) by the background model (12).

Description

State of the art

[0001] The invention relates to a smoke detection device for detecting smoke from a fire in a monitored area, comprising a camera interface for acquiring a sequence of images with at least two temporally successive individual images, wherein the individual images show the monitored area, and a differential module for eliminating a background scene of the monitored area in the individual images and for generating background-cleaned intermediate images. The invention further relates to a method for detecting smoke and a computer program.

[0002] In addition to automated fire detectors, which use temperature measurement, scattered light measurement and other measurement methods to detect characteristics of a fire and issue an alarm based on this, a large number of video-based fire detectors have become known, which can detect and report a fire using image processing based on meaningful fire characteristics such as optical emissions or smoke emissions.

[0003] In video-based fire detectors, one evaluation method focuses on the detection of smoke emissions from a fire. For example, US document 7,804,522 B2 noted that detecting thin or transparent smoke is particularly difficult because the smoke is not opaque in the corresponding video image and therefore does not completely obscure the background. Thus, it is difficult for a video-based fire detector to distinguish whether a video image shows haze, measurement inaccuracies, or transparent smoke emissions. To improve the detection capability of video-based fire detectors, the aforementioned document proposes subtracting the background from the current video image, with the background being determined, for example, over several frames. The smoke emissions are then more clearly visible in the resulting image.

[0004] WO 2008 / 037 293 A1 discloses a method and a device for detecting smoke by analyzing at least one video image recorded by a video camera monitoring an area. US 2014 / 0 099 020 A1 discloses a method for detecting smoke from a forest fire using spatiotemporal features of the smoke. Disclosure of the invention

[0005] The invention proposes a smoke detection device with the features of claim 1, a method for detecting smoke with the features of claim 6, and a computer program with the features of claim 7. Preferred or advantageous embodiments of the invention are described in the dependent claims, the following description, and the accompanying figures.

[0006] The invention relates to a smoke detection device for detecting smoke from a fire in a monitored area. The smoke is formed by smoke emissions, in particular combustion products of the fire. The monitored area can be an enclosed area, such as a hall, or alternatively, it can be an open area. The smoke detection device is designed to detect smoke, in particular smoke compartments, and thereby indirectly detect the fire. The smoke detection device can, for example, be a digital data processing device, such as a computer, a microcontroller, etc. Optionally, the smoke detection device is designed to issue a fire alarm based on the smoke detection. In particular, the smoke detection device is designed as an automatic fire alarm.

[0007] The smoke detection device has a camera interface designed to receive a sequence of at least two consecutive images. The camera interface can be wireless or wired. It is also possible for the camera interface to be configured as a network interface. The image sequence is provided by a surveillance camera directed at the monitored area. Optionally, the surveillance camera forms part of the smoke detection device. The consecutive images preferably show the same section of the monitored area. Thus, the field of view of the surveillance camera within the monitored area is preferably constant. The time interval between the at least two consecutive images is, for example, between 20 milliseconds and 300 milliseconds.For example, a standard camera with a frame rate of 30 Hertz or 60 Hertz can be used as a surveillance camera.

[0008] The smoke detection device includes a differential module designed to eliminate a scene background, particularly a static scene background, from the monitored area in the individual frames and to generate background-cleaned intermediate frames. Stationary and, optionally, quasi-stationary objects are arranged within the monitored area. The stationary objects remain in the monitored area for more than one day. The quasi-stationary objects remain in the monitored area for, e.g., more than 10 minutes, but preferably less than one day. The stationary objects can be, for example, trees, houses, paths, the ground, a mountain, etc. The quasi-stationary objects can be cars, furniture, parked objects, etc. The entirety of these stationary and optionally quasi-stationary objects forms the scene background of the monitored area.In particular, the background of the scene remains constant across the successive frames.

[0009] The difference module is designed to eliminate the scene background in the individual frames, primarily through the implementation of digital image processing techniques. For example, the difference module sets pixels in the individual frames that correspond to the scene background to a constant value, such as 0. Instead of a constant value, a predetermined range of values ​​or a very small value can also be used. The difference module ensures that moving objects, such as smoke in the monitored area, are clearly defined. In particular, the difference module can significantly highlight transparent smoke—that is, smoke through which the scene background is still visible in the individual frames.

[0010] Within the scope of the invention, it is proposed that the smoke detection device comprises an evaluation module, wherein the evaluation module performs a motion estimation to determine the movement of at least one smoke segment within the monitored area based on background-cleaned intermediate images and / or is designed accordingly in terms of programming and / or circuitry. The evaluation module is preferably implemented as a software module. Specifically, the evaluation module determines the movement of the smoke segment within the monitored area between the two background-cleaned intermediate images. Optionally, the motion estimation also estimates the movement of other moving objects, with these additional results subsequently being discarded as non-smoke.

[0011] The smoke section can be a single pixel segment of the smoke in the individual or intermediate images, or a spatially extended sub-area of ​​the smoke. In an optional continuous operation of the smoke detection device, it is preferably provided that the determination of the movement of the smoke section within the monitoring area is continued via further individual or intermediate images.

[0012] One aspect of the invention is that combining the use of the differential module with the evaluation module creates a particularly informative analysis for transparent smoke. While the differential module isolates the transparent smoke, the evaluation module examines the movement of the smoke or smoke segments between the background-cleaned intermediate images and, if applicable, further intermediate images. Since smoke typically has a higher temperature than its surroundings, it rises vertically and / or perpendicularly. It has been observed that smoke typically rises at a rate of approximately 0.5 meters per second.This movement of the smoke, especially its speed or rate of ascent and its direction, is characteristic of smoke and constitutes a significant feature for smoke detection. The evaluation module thus enables the extraction of these significant features from the intermediate images, thereby creating a reliable basis for smoke detection.

[0013] In one possible embodiment of the invention, the motion estimation is implemented as a block-matching method. In a particularly preferred embodiment of the invention, the motion estimation is implemented as the determination of an optical flow field. The optical flow field is understood to be a vector field of the velocity vectors of visible points, in particular smoke sections, of the monitored area projected into the image plane of the individual or intermediate images in the reference frame of the imaging optics of the surveillance camera.Referring to a pixel segment of the smoke as a smoke segment in the individual or intermediate frames, or to a spatially extended sub-region of the smoke as a smoke segment, the optical flow field comprises a vector that indicates the speed and direction of the pixel segment or spatially extended sub-region of the smoke that this area has traveled between the two individual or intermediate frames. The calculation of the optical flow in individual or intermediate frames is known to those skilled in the art and is widely used in digital image processing, so that the smoke detection device can be implemented very reliably by means of calculating the optical flow.

[0014] According to the invention, the smoke detection device comprises a background module for creating a background model of the monitored area. The background model corresponds to the scene background and is, for example, designed in the same format as the individual images. The difference module uses the background model and applies it to the individual images to generate the background-cleaned intermediate images.

[0015] It may be provided that the background module is designed to create the background model from a large number of individual images in a gradual and / or accumulating manner.

[0016] For example, the background model can be averaged from earlier images, with images with older timestamps having a lower weighting in the averaging than images with newer timestamps. Alternatively, the background model can be manually assembled from individual images.

[0017] In possible embodiments of the invention, the application of the background model to the individual images is implemented as a subtraction of the background model from the respective individual image. Thus, the scene background is subtracted from the depicted monitoring area. According to the invention, the current individual images are divided by the background model, particularly pixel by pixel, to create the background-cleaned individual images.

[0018] In a possible alternative embodiment of the invention, the difference module is configured to generate background-cleaned intermediate images by creating difference images between two temporally successive individual images. Specifically, one of the individual images is subtracted from the other to generate the background-cleaned intermediate image. From a temporal perspective, the intermediate images can thus be inserted into the image sequence between the individual images, with each intermediate image containing the difference between the preceding, particularly immediately preceding, individual image and the following, particularly immediately following, individual image. Since the temporally successive individual images each show the same scene background, this background is eliminated during the generation of the difference image, leaving as image content the differences in the image areas that depict the smoke.

[0019] In a preferred embodiment of the invention, the smoke detection device comprises a detection module for detecting fire smoke based on motion estimation. If the motion estimation includes, for example, the determination of the optical flow field, the smoke can be detected as a feature by a characteristic cluster of flow vectors that are vertically or substantially vertically oriented and have a predetermined length or range of lengths, since the smoke is usually arranged in a continuous area, rises vertically, and exhibits a uniform rate of ascent or a uniform rate of ascent as its natural motion. Other moving objects that do not exhibit this natural motion of smoke are rejected as non-smoke.

[0020] In particular, the detection module is designed to determine a smoke probability based on motion estimation and, for example, detect smoke when a predefined probability is exceeded. Optionally, it can also be provided that further parameters, such as the size of the area in the intermediate images with a sufficient smoke probability and / or the duration of the detected smoke probability, are included in the calculation.

[0021] Another aspect of the invention relates to a method for detecting smoke in a monitored area, preferably using the smoke detection device described above. In this method, individual images of the monitored area are captured. Subsequently, a background scene, particularly consisting of static and optionally quasi-static objects, is removed from the individual images to generate background-cleaned intermediate images. In a subsequent step, the movement of smoke segments in the intermediate images is estimated in order to detect the smoke.

[0022] Another object of the invention is a computer program with program code means for carrying out the previously described method, preferably on a smoke detection device or on a data processing system.

[0023] Further features, advantages, and effects of the invention will become apparent from the following description of a preferred embodiment of the invention and the accompanying figures. These figures show: Fig. 1 a schematic block diagram of a smoke detection device as an embodiment of the invention; Fig. 2a,b two different representations of optical flux fields of the smoke detection device for further description of the function or the method.

[0024] The Fig. Figure 1 shows a highly schematic representation of a smoke detection device 1 as an embodiment of the invention. The smoke detection device 1 is designed as an automatic fire alarm and can, for example, be implemented as a computer or a digital data processing device with a computer program for implementing the individual modules.

[0025] Furthermore, in the Fig. Figure 1 shows a surveillance camera 2, which is directed with its field of view towards a surveillance area U, such that the surveillance area U is imaged by the surveillance camera 2. The surveillance camera 2 can, for example, be a black and white camera or a color camera. The surveillance camera 2 can optionally form an integral part of the smoke detection device 1.

[0026] The surveillance camera 2 provides a sequence of images 3 via a camera output, consisting of a multitude of individual images 4 that follow each other sequentially. For example, the individual images 4 are provided at a fixed frequency of, say, 30 Hertz. The sequence of images 3, or the individual images 4, are transmitted via a camera interface 5 of the smoke detection device 1.

[0027] The smoke detection device 1 has a differential module 6, which is designed for image processing of the individual images 4. Two individual images 4 are shown as examples in the smoke detection device 1, depicting the monitoring area U. Within the monitoring area U, a stationary object 7, in this example a house, is shown, which together with other stationary objects and the entire surroundings of the monitoring area U forms a scene background 8. Furthermore, a fire 9 with a smoke flare is shown in a highly schematic way as smoke 10.

[0028] The difference module 6 is configured to eliminate the scene background 8 in the individual images 4. For this purpose, the smoke detection device 1, for example, has a model module 11 which models the background 8 as a background model 12. The background model 12 can be created, for example, by averaging a majority of the individual images 4. To obtain a current background model 12, it can be provided, for example, that current individual images 4 are weighted more heavily than older individual images during the averaging process. In this way, it is also possible, for example, to take into account gradual changes, such as changes in brightness due to day-night cycles or changes due to weather changes.

[0029] The difference module 6 applies the background model 12 to the individual images 4 available for processing. This application can be performed, for example, by subtraction, so that the scene background 8 is subtracted pixel by pixel from the individual images 4. Alternatively, matching pixels or pixel values ​​can be set to a predefined value.

[0030] In an alternative configuration, the difference module 6 generates difference images of two temporally consecutive individual images 4 as the background-cleaned intermediate images 13. Here, the difference module 6 subtracts the two temporally consecutive individual images from each other to obtain a difference image as the background-cleaned intermediate image 13 for this pair of individual images 4. In this configuration, the model module 11 can therefore be omitted.

[0031] As a result, the difference module outputs 6 background-cleaned intermediate images 13, which show only the smoke 10 as their main image content. However, this is barely recognizable, since, for example, in the case of transparent smoke 10, only faint streaks remain in the background-cleaned intermediate images 13.

[0032] The background-cleaned intermediate images 13 are passed to an evaluation module 14, which performs a motion estimation to determine the movement of the smoke 10, in particular smoke sections, i.e., subsections of the smoke 10, in the monitoring area U based on the background-cleaned intermediate images 13. For example, the motion estimation is implemented by calculating an optical flow field 18 between the background-cleaned intermediate images 13. When calculating the optical flow or the optical flow field, vectors are generated that include the change in position and the speed and / or length of the change in position between the two temporally successive background-cleaned intermediate images 13. In the Fig. Figure 1 shows the optical flow field 18 in coded form, with vectors of the optical flow field 18, which have a direction and length corresponding to the natural movement of smoke 10, being colored dark.

[0033] In a detection module 15, the optical flow field 18, or the optical flow, is evaluated for the detection of smoke 10 from the fire 9. This evaluation is based on the assumption that the smoke 10 moves naturally in a perpendicular and / or vertical direction at a rate of approximately 0.5 meters per second. Therefore, a dark area, such as the one shown in the coded representation of the optical flow field 18, is a characteristic of the smoke 10.

[0034] The smoke detection device 1 has an output interface 17 which transmits an alarm from the detection module 15, for example, to an alarm device 19, an operations center, or the like.

[0035] In the Fig. Figures 2a and 2b show an exemplary evaluation of the optical flow field 18, where pixels with vectors corresponding to the natural movement of the smoke 10 are highlighted in light colors in this example. Fig. 2 a is an optical flow field 18, which was created on the basis of non-background-cleaned individual images 4. In the Fig. In Figure 2b, an optical flow field 18 was calculated for the same individual images 4 based on background-cleaned intermediate images 13. Background cleaning was performed using the OpenCV Background Subtraction tool. A comparison of the two representations shows that the optical flow field 18 calculated based on the background-cleaned images 13 is significantly more pronounced than the optical flow field 18 calculated solely from the uncleaned individual images 4.

[0036] Thus, the interplay of background correction with the calculation of the optical flow field is an advantageous evaluation of individual images 4 of an image sequence 3, which can particularly exploit their advantages when the smoke 10 is designed as a transparent and / or see-through smoke.

Claims

[1] Smoke detection device (1) for detecting smoke (10) from a fire (9) in a monitored area (U), with a camera interface (5) for receiving a sequence of images (3) with at least two temporally successive individual images (4), wherein the individual images (4) show the monitored area (U), with a difference module (6) for eliminating a scene background (8) of the monitoring area (U) in the individual frames (4) and for generating background-cleaned intermediate frames (13), with an evaluation module (14), wherein the evaluation module (14) is designed to perform a motion estimation to determine a movement of smoke sections of the smoke (10) in the monitoring area (U) based on the background-cleaned intermediate images (13), characterized by a background module (11), wherein the background module (11) is configured to create a background model (12) of the monitoring area (U), wherein the difference module (6) is configured to apply the background model (12) to the individual images (4) to generate the background-cleaned intermediate images (13), wherein the application of the background model (12) to the individual images (4) is configured as a division of the individual image (4) by the background model (12). [2] Smoke detection device (1) according to claim 1, characterized by , that the evaluation module (14) for determining an optical flow field (18) is designed as a motion estimator for determining the motion. [3] Smoke detection device (1) according to claim 1, characterized by , that the background module (11) is designed for the gliding and / or accumulating creation of the background model (12) from a large number of individual images (4). [4] Smoke detection device (1) according to any one of the preceding claims, characterized by a detection module (15) for detecting smoke (10) of the fire (9) based on motion estimation. [5] Smoke detection device (1) according to claim 4, characterized by , that the detection module (15) is designed to determine a probability of smoking based on the motion estimation. [6] Method for detecting smoke (10), preferably with a smoke detection device (1) according to one of the preceding claims, wherein individual images (4) of a monitoring area (U) are recorded, wherein a scene background (8) is subsequently removed from the individual images (4) to generate background-cleaned intermediate images (13), and wherein a movement of smoke sections of the smoke (10) is estimated in the background-cleaned intermediate images (13), wherein a background model (12) of the monitoring area (U) is created, wherein the background model (12) is applied to the individual images (4) to generate the background-cleaned intermediate images (13), wherein the application of the background model (12) to the individual images (4) is designed as a division of the individual image (4) by the background model (12). [7] Computer program with program code means to perform all steps of a method according to claim 6 when the program is executed on a computer and / or on a smoke detection device (1) of any one of claims 1 to 5.

Images