RU2650347C1 - Method of the natural fires recognition in sub-horizon areas - Google Patents

Abstract

FIELD: surveillance systems.

SUBSTANCE: invention relates to the field of video surveillance, mainly of open spaces, with fire hazard control, and specifically to the methods of natural fires recognition in sub-horizon areas. In the claimed method of natural fires recognition in sub-horizon areas, installing a rotating around a vertical axis video camera on a high-rise structure, making a territory filming by the camera sequential rotation around the vertical axis, performing analysis of the obtained images for the presence of objects in the sub-horizon areas of the frame, similar by the color to smoke, if any, fixing the possibility of a natural fire. According to the elevation heights maps database determining the calculated horizon line in each frame, making a correction for the visible in the frame real horizon line, analyzing only that area of the frame, which is located near the horizon line, building isolines in the analyzed area, enveloping the combined into segments by color, brightness and contrast image pixels. In the presence of the isoline section deviation from the equal distance from the horizon line, fixing the possibility of a natural fire in the place corresponding to the indicated one, and in the presence of the isoline repeating the shape of the horizon, recording the absence of a natural fire in the sub-horizon area.

EFFECT: possibility of a natural fire recognition in the sub-horizon area without the video camera zooming.

4 cl, 4 dwg

Description

FIELD OF THE INVENTION

The invention relates to the field of video surveillance, mainly open spaces, with fire hazard control, and specifically to methods for recognizing natural fires in the horizontal areas, in which a video camera rotating around a vertical axis is installed on a high-rise structure, the territory is captured by sequentially turning the video camera around a vertical axis, and analysis of the obtained images for the presence in the near-horizontal areas of the frame of objects similar in color to smoke, if any and those fixed possibility wildfire, and can be used to detect wildfires - steppe fires, forest fires (which also include peat and fires), hereinafter (prigorizontyh) areas when viewed from high-rise structures.

The following terms are used in this description:

high-rise structure - a building, a tower, any structure much higher than the height of surrounding objects, allowing observation of the surrounding terrain;

vertical - a direction parallel to gravity;

horizontal - direction perpendicular to gravity;

tilt angle - the angle between the axis of rotation of the camera and the vertical;

azimuthal angle of the video camera - the angle of orientation of the video camera relative to the north direction;

relief slice model — a graphic or digital representation of a topographic surface in the form of a raster or a regular network of cells of a given size;

elevation elevation map database — a database containing elevation data for points on the earth’s surface, for example, the SRTM (Shuttle radar topographic mission) elevation matrix — an international mission to obtain digital elevation model data (DEM) of the Earth’s territory;

calculated horizon line - the horizon line, which is calculated using a database of elevation maps of the topography based on the known height of the camera’s location,

near-horizon - located near the horizon (since we are talking about the definition of fires, then fires can be detected in the near zone, and in contrast to it in the far zone, near the horizon).

State of the art

To detect natural (steppe and forest) fires when observing from high-rise buildings using a video camera rotating around a vertical axis, methods for automatic smoke recognition are known, which work on the basis of recognition by movement, color / brightness / contrast, and model behavior of pixel groups in the analyzed frame.

Such a traditional method has a limitation due to the fact that it needs to “see” the movement - to make a decision on checking the color and / or contrast and / or brightness of a moving spot and to evaluate for compliance with the model of behavior or propagation. The limitation is the threshold number of pixels in the moving portion of the image, sufficient to determine the behavior.

Therefore, in order to recognize distant near-horizon smoke, in the traditional way, you need to zoom in on the remote scene. The specifics of a monitoring system for natural fires using PTZ cameras is as follows. The camera constantly rotates to new positions - “patrols” the territory. If the viewing angle (directly related to the zoom ratio of the lens - zoom) is very small (strongly approximate the scene being analyzed), then there will be many breakpoints in the patrol, and the time for analyzing the scene at each breakpoint does not change. This means that with a decrease in the viewing angle for a single analyzed scene, the time increases during which the camera makes a full revolution around its axis (ends the patrol cycle) and inspects the entire territory under control. In addition, multiple approximation narrows not only the horizontal viewing angle of the scene, but also the vertical viewing angle, that is, near and far objects ("fumes") do not simultaneously fall into the field of view of the camera. That is, to inspect the entire territory under control, it is necessary to go through it several times - more stopping points due to the narrow horizontal viewing angle and more “layers” or “levels” of patrolling (closer and further) due to the narrow vertical angle.

You can reduce the time required for a full survey of the controlled area by using several patrol routes for near and far smoke, combining a wide viewing angle to search for nearby smoke and a narrow angle for distant.

But this method creates an increased load on the camcorder zoom, and it quickly fails.

A method is known from the prior art for recognizing natural fires in horizontal areas, in which a video camera rotating around a vertical axis is mounted on a high-rise structure, a territory is taken by sequentially turning the video camera around a vertical axis, and the obtained images are analyzed for objects in the horizontal areas of the frame by color similar to smoke, if any, fix the possibility of a natural fire, see the description of the patent for invention No. 2534827, published 12/10/2014.

This method is the closest in technical essence and the achieved technical result and is selected for the prototype of the invention.

The disadvantage of this prototype is the low practical applicability for determining a natural, for example, forest fire, due to the fact that:

a) the distance at which smoke is detected is limited - it is impossible to apply the specified method for determining distant smokes without approaching remote scenes due to the absence of a key determination factor - the small pixel size of the object in the frame and the lack of detectable pixel movement;

b) the time for a full inspection of the territory is significant. This is bad, first of all, because the speed of detection of a natural fire significantly affects the economic component: more forest burns and the cost of extinguishing increases, because the area of the fire at the time of detection is larger, which means it needs more effort and means to extinguish it;

c) the most unreliable element of modern cameras is precisely the zoom system (zoom, zoom). Due to the constant change in the viewing angle, the load on this system is growing, which means that the likelihood of a camera breakdown is also growing;

d) when approaching a remote scene (reducing the viewing angle), the “recognition” of the territory decreases due to the small number of landmarks that simultaneously fall into the frame — the automated system makes a “mark” on the detected object and “applies” the frame in which it found this “object” "

Thus, the problem to which the present invention is directed is the possibility of determining a natural fire in the horizontal region without zooming the camera.

Disclosure of invention

Based on this original observation, the present invention mainly aims to propose a method for recognizing natural fires in the near-horizon areas, which can at least alleviate the above drawback, namely, to allow the possibility of recognizing a natural fire in the near-horizon area without zooming the camera, which is set technical task.

To achieve this goal, the method further comprises steps in which:

using the database of elevation maps of the relief, the calculated horizon line in each frame is determined,

make a correction for the real horizon line visible in the frame,

analyze only that area of the frame that is near the horizon line,

build in the analyzed area contours that envelope the pixels in segments combined by color, brightness and contrast,

if there is a deviation of the contour section from an equal distance from the horizon, the possibility of a natural fire is fixed in the place corresponding to the specified one, and if there is a contour that repeats the shape of the horizon, the absence of a natural fire in the near-horizon region is recorded.

Thanks to these non-obvious characteristics, it becomes possible to compensate for the errors in the real angle of inclination of the vertical axis of rotation of the camera, automatically detect the horizon line in the frame, and it becomes possible to analyze only the area that is next to the horizon line. In the analyzed area, it becomes possible to automatically identify places that correspond to fires. This can be done without zooming the camcorder, which in turn increases its service life, as this is the most common cause of a video camera breakdown.

In addition, this method does not accept haze or clouds for smoke in a natural fire, which increases the reliability of recognition of a natural fire.

Wherein:

but. the speed of the camcorder's patrol route is high, which means that the speed of fire detection is high,

b. the zoom resource of the PTZ camera is saved,

at. the recognition of the territory by the frame in which the object is detected due to the greater number of landmarks in the frame is preserved.

There is a variant of the proposed method, in which only the region of the frame that is located near the horizon is analyzed, and the analyzed region is defined as a set of points located at a distance from the highest and lowest points of the horizon not more than 30 pixels.

Thanks to these non-obvious characteristics, it becomes possible to specifically indicate the size of the analyzed area, since this is quite enough for a qualitative analysis.

There is also a variant of the proposed method, in which the possibility of a natural fire is detected when an isoline is detected that is longer than the horizon line, and deviations in this isoline from an equal distance from the horizon line are substantially shorter in horizontal measurement than the horizon line.

Thanks to these non-obvious characteristics, it becomes possible to cut off haze and mists from a real natural fire, since they give the deviation of isolines along the entire length of the horizon.

There is another variant of the proposed method, in which the analysis of pixels in the frame area, limited by the contour area with a deviation from equal distance from the horizon for the similarity in color with smoke.

Thanks to these non-obvious characteristics, it is possible to search for smoke by color, since smoke has a unique color.

The set of essential features of the present invention is unknown from the prior art for devices of similar purpose, which allows us to conclude that the criterion of "novelty" for the invention as a device is met. Also, the set of essential features of the invention does not follow explicitly from the prior art for devices of a similar purpose, which allows us to conclude that the criterion of "inventive step" for the invention is met.

Brief Description of the Drawings

Other distinguishing features and advantages of this invention clearly follow from the description below for illustration and not being restrictive, with reference to the accompanying drawings, in which:

- figure 1 depicts a schematic side view of the location of the camera aimed at the horizon, according to the prior art, shows the direction of gravity,

- figure 2 depicts a view on the screen of the image obtained by the camera in the recognition of natural fires in the near and horizontal areas, according to the prior art,

- figure 3 depicts a schematic view on the screen of the image obtained by the video camera when recognizing natural fires in the horizontal areas, according to the invention,

- figure 4 schematically depicts the steps of a method for recognizing natural fires in the horizontal areas according to the invention.

In the figures indicated:

1 - video camera

2 - high-rise structure, such as a mast,

3 - slice of the terrain,

4 - vertical

5 - axis of rotation of the video camera,

6 - the direction of the eyepiece of the camera to the horizon,

7 - visible horizon,

8 - recognizable near zone,

9 - recognizable near-horizon area

10 - isoline

11 - area deviation of the contour

α is the angle of the camera

β - correction to the angle of inclination, the angle between the vertical and the axis of rotation of the camera,

g is the vector of gravity.

As a camcorder, you can use any camcorder such as a Pan-tilt-zoom camera (PTZ camera) - that is, a camcorder that supports remote control of direction and zoom. It should have power and a signal transmission unit to a remote server, where information is analyzed, and, if necessary, output to the screen.

The implementation of the invention

A method for recognizing natural fires in near-horizontal areas is as follows. Here is the most comprehensive example of the invention, bearing in mind that this example does not limit the application of the invention.

According to figure 4:

Stage A1. Installation A video camera rotating around a vertical axis 5 is installed on a high-rise structure 2.

Stage A2. Snapping. Its exact coordinates and suspension height are determined. The coordinates of the video camera are specified according to several sources of orthophotomaps - Google maps, Yandex maps, etc. Thus, it becomes possible to bind the coordinates of the installation of the camera - up to a meter.

Stage A3. Correction of the angle of rotation. Calculate the correction of the azimuth of the camera’s rotation relative to the north direction by comparing the direction to a visible landmark.

Stage A4. Tilt Correction.

Stage A41. Create a relief map with a radius of 45 km around the coordinates of the installation of the camera (if you don’t have your own, you can take SRTM - a public map of the elevation of the relief for the whole world).

Stage A42. They build a model of sections of the relief around the video camera with a step of 0.1 degrees (which corresponds to the usual error of the rotation mechanism of the video camera) over the entire model depth of 45 km, i.e. these are 3600 vectors containing 500 elevations, each. The step between the points along the cut is 90 meters.

Stage A43. Knowing the relief around the camera, the height of the camera’s suspension and assuming that it is standing strictly upright, we get the calculated horizon line (in each frame, while the camera rotates). Anchor points on the horizon line are calculated as the farthest visible elevation points.

Stage A44. A line composed of points of the calculated horizon, the location of which is converted from angular to pixel coordinates, is superimposed on the image from the video camera.

Stage A45. The operator on the image from the video camera indicates the points of the real, not calculated, visible horizon (10-12 reference points at 360 degrees in different azimuthal directions evenly). The more reference points, the more accurate the calculation.

Stage A46. In the database, degrees β are recorded - deviations of the real horizon from the calculated one at the reference points, and an interpolation calculation is made for all other points between the reference points. The error is recorded in degrees, and not in pixels, so that the calculation does not depend on the approximation - removal (zoom) of the camera, when one degree has a different number of pixels. The obtained coefficients are used as the correction angle β to the angle of inclination α of the camera to calculate the coordinates of the intersection of the vector corresponding to the line of sight of the camera 8 (formed by drawing a conditional line between the centers of the matrix and the camera lens) with a relief in real work.

Stage A5. Shooting. Survey the territory by sequentially turning the camcorder around a vertical axis. To do this, determine the optimal viewing time for the patrol route of the territory and the sufficiency of time to analyze the scene, the viewing angle, for example: 6 degrees horizontally and 10-12 minutes for a full revolution of the camera around its axis. Moreover, the analysis of the scene at each stopping point takes from 10 to 15 seconds.

Stage A6. Framing. Using the database of elevation heights, the calculated horizon line is determined in each frame, taking into account corrections for the real horizon line, only the area of the frame that is located near the horizon line is analyzed, and the analyzed area is set as a set of points located at a distance from the highest and highest low horizon points of no more than 30 pixels.

Knowing where the horizon should be (calculated and adjusted for real), and all its features (dips and elevations), cut off the area with a small margin up and down from the horizon. In this area, initially there are not enough pixels to determine movement and behavior.

Stage A7. Analysis.

Divide the observed scene into sections:

- above the horizon (there are clouds in this zone and the method does not respond to them)

- horizon zone 9 (distances from the camera installation point from 8 to 12 km)

- Near zone 8 (from 1 to 8 km).

In the near zone 8, a classic algorithm works - enough information about the movement in the frame. See FIG. 2.

In the analyzed near-horizon region (see Fig. 3):

Stage A71. Isolines 10 are constructed in the analyzed near-horizon region, enveloping the pixels of the image in segments combined by color, brightness and contrast. The lines that outline such segments are contours (lines between areas with similar color, brightness, and contrast).

Stage A72. Of the many contours, a contour 10 with the longest length is selected containing a portion 11 with the greatest deviation from the horizon line, which is substantially shorter in horizontal dimension than the horizon line.

Stage A73. Pixels are analyzed in the frame region bounded by section 11 of the selected contour 10 with the greatest deviation from an equal distance from the horizon for similarity in color with smoke.

Stage A74. If there is a deviation of 11 sections of the contour from an equal distance from the horizon line 7, the possibility of a natural fire is fixed in the place corresponding to the indicated one, and if there is a contour that repeats the shape of the horizon, the absence of a natural fire in the near-horizon region is recorded.

In the zone close to the horizon, there will be a sharp transition from bright sky to dark earth, forest, etc. The sky is always brighter than the earth.

If in the near-horizon zone there are no features unaccounted for in the calculation model of the horizon, then there will be at least one contour that repeats the line of the calculated horizon as a whole. In this case, any feature (anomaly) of this contour can be considered potentially dangerous.

The method allows you to consider:

- smoke located closer to the horizon with respect to the camera will give a “failure” to the contour;

- Smoke located in relation to the camera beyond the horizon will give rise to the contour above the horizon.

The smoke in this method is considered to be a group of pixels of substantially shorter horizontal dimension than the horizon line, that is, the smoke should be grouped and not “smeared” along the entire length of the horizon line. "Smudged Smoke" is considered morning / evening / weather haze or fog.

Check the color / brightness / contrast of pixels from a real frame inside the area bounded by the horizon and contour.

False alarms may be due to inaccuracies in the elevation model constructed from outdated data that do not take into account changes that visually change the horizon (deforestation on hills, opencast mining, etc.). Such places can be covered by “masks” and processed in the subsequent stages of the analysis of smoke detection algorithms.

The sequence of steps is approximate and allows you to rearrange, decrease, add or perform some operations at the same time without losing the ability to determine the coordinates of objects. For example, step A45 may be performed automatically by automatically recognizing a real horizon line on a display.

Industrial applicability

The proposed method for recognizing natural fires in horizontal areas can be carried out by a specialist in practice and upon implementation ensures the implementation of the declared purpose, which allows us to conclude that the criterion of "industrial applicability" for the invention is met.

In accordance with the proposed invention, a prototype of a plant was tested that implements a method for recognizing natural fires in horizontal areas, consisting of a video camera mounted on the tower, a data transmission unit to a server connected to a computer device with a monitor.

Tests of the system showed that it provides the ability to:

- determine the calculated horizon line in each frame from the database of elevation maps of the relief,

- make an adjustment for the real horizon line visible in the frame,

- analyze only that area of the frame that is near the horizon line,

- build in the analyzed area contours that envelope the pixels of the image combined in color, brightness and contrast, into segments,

- if there is a deviation of the contour section from an equal distance from the horizon, to fix the possibility of a natural fire in the place corresponding to the specified one, and if there is a contour that repeats the shape of the horizon, to fix the absence of a natural fire in the near-horizon area.

Thus, due to the fact that the calculated horizon line in each frame is determined from the database of elevation height maps, the correction is made for the real horizon line visible in the frame, only the area of the frame that is located near the horizon line is analyzed, built in the analyzed area contours enveloping the pixels of the image combined in color, brightness and contrast into segments, in the presence of a deviation of the contour section from an equal distance from the horizon, fix the possibility of a natural fire in the place corresponding to azannomu, and in the presence of contours, repeating the shape of the horizon, the absence of fixed natural fire in prigorizontnoy region and the claimed technical result is achieved, namely the ability to recognize the natural fire in the art without prigorizontnoy zoom camcorder.

An additional positive effect of the analysis of horizontal contours is the ability to automatically calculate the daily deviations of high-rise structures (masts), on which PTZ cameras are installed.

These deviations are associated with uneven heating of the metal structure of the mast from its different sides (solar and shadow).

Visually, contours constructed in the near-horizon region should generally repeat the slope of the calculated horizon line, taking into account corrections, since they accurately repeat the visible relief. If the mast deviated due to one-sided heating, the axis of rotation of the camera will deviate along with it, since the camera is rigidly fixed to the mast structures. The contour that runs along the visible horizon line will intersect the calculated line (one end will be higher than the calculated line, and the other lower). Fixing the discrepancy between the angles of inclination of the calculated horizon and isolines at different points of the patrol stop is used to clarify the correction angle of the axis of rotation of the camera.

Claims (9)

1. A method for recognizing natural fires in near-horizontal areas, in which a video camera rotating around a vertical axis is installed on a tall building, the territory is taken by sequentially turning the video camera around a vertical axis, and the obtained images are analyzed for objects in the horizontal areas of the frame that are similar in color to smoke, if any, fix the possibility of a natural fire, characterized in that using the database of elevation maps of the relief, the calculated horizon line in each frame is determined, make a correction for the real horizon line visible in the frame, analyze only that area of the frame that is near the horizon line, build in the analyzed area contours that envelope the pixels in segments combined by color, brightness and contrast, if there is a deviation of the contour section from an equal distance from the horizon, the possibility of a natural fire is fixed in the place corresponding to the specified one, and if there is a contour that repeats the shape of the horizon, the absence of a natural fire in the near-horizon region is recorded. 2. The method according to claim 1, characterized in that they analyze only that region of the frame that is near the horizon, and the analyzed region is set as a set of points located at a distance from the highest and lowest points of the horizon not more than 30 pixels . 3. The method according to claim 1, characterized in that they fix the possibility of a natural fire when an isoline is detected longer than the horizon line and deviations in this isoline from an equal distance from the horizon line, a substantially shorter horizontal dimension than the horizon line. 4. The method according to claim 1, characterized in that they analyze the pixels in the frame area limited by the contour section with a deviation from equal distance from the horizon for similarity in color with smoke.

Images