JP2018055151A - Flame detector and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a flame detector capable of accurately detecting a flame by discriminating a body in motion and the flame fluctuating at the same place.SOLUTION: A flame detector, which detects a flame on the basis of a photographed image, acquires the photographed image, and extracts a vector indicating the movement of an object included in the photographed image. Then, on the basis of the vector, when the object is expanded or contracted, the object is determined as being the flame.SELECTED DRAWING: Figure 5

Description

  The present invention relates to a technique for detecting a flame portion included in a captured image.

  A technique for detecting a fire based on an image captured by a camera is known. For example, Patent Document 1 uses a color information and fluctuation information from a video image taken by a camera to detect a flame-colored moving object and measures whether or not there is a movement specific to the flame from the candidate. Identifies moving objects other than flames.

JP 2002-32872 A

  The technique of Patent Document 1 performs flame discrimination using a precondition that a moving object does not stay in the same place for a long time. For this reason, when a high-intensity object other than a flame is shaking in the same place, for example, when a red flag is shaken by the wind, this may be erroneously determined as a flame.

  The main object of the present invention is to provide a flame detection device that enables more accurate flame detection by identifying a moving object and a flame that are shaking at the same place.

  In one aspect of the present invention, the flame detection device includes: an image acquisition unit that acquires a captured image; a vector extraction unit that extracts a vector indicating a motion of an object included in the captured image; and the vector, Determination means for determining that the object is a flame when the object is expanded or contracted.

  The flame detection device detects a flame based on a captured image, acquires the captured image, and extracts a vector indicating the movement of the object included in the captured image. This vector extraction is performed by, for example, an optical flow method. And based on the extracted vector, when the target object is expanding or contracting, the target object is determined to be a flame. Since the flame expands and contracts as a specific motion, it is possible to accurately detect the flame by making a determination based on whether or not the object is expanded / contracted.

  In one aspect of the above-described flame detection apparatus, the determination unit performs determination based on an angle formed by the center of the object and the vector. In this aspect, the flame determination is performed by detecting the expansion and contraction motion based on the angle between the center of the object and the vector. In addition, the center part of a target object is the concept containing the gravity center and center of a target object. In one preferable example, the determination unit determines that the object is a flame when the distribution of the angles is biased near 0 degrees and 180 degrees.

  In another aspect of the above-described flame detection device, the vector extraction unit extracts a movement vector of a plurality of pixels constituting the object based on a plurality of captured images, and the determination unit includes the plurality of the plurality of pixels. The center portion of the object is determined for each captured image. In this aspect, the determination is performed by determining the center of the object for each of the plurality of captured images.

  In another aspect of the above-described flame detection device, the determination unit determines that the object is a flame when the direction of the vector is evenly distributed at 360 degrees. In this aspect, flame determination is performed based on the direction of the extracted vector.

  In a preferred embodiment, it is possible to configure a camera including an imaging unit that generates the captured image and the flame detection device.

  In another aspect of the present invention, a program executed by a flame detection apparatus including a computer includes: an image acquisition unit that acquires a captured image; a vector extraction unit that extracts a vector indicating a motion of an object included in the captured image; Based on the vector, when the object is inflated or contracted, the computer is caused to function as determination means for determining the object as a flame. By executing this program on a computer, the above-described flame detection apparatus can be realized.

It is a block diagram which shows the structure of the flame detection apparatus which concerns on embodiment. It is a flowchart of a flame detection process. A fluctuation pixel extracting method is schematically shown. The flame area accumulation processing is shown. It is a flowchart of a flame determination process. An example of a movement vector of a flame and a non-flame object is shown. It is a figure explaining a vector direction angle. It is a figure explaining the specific example of flame determination. The example of the movement vector of the flame which expands and contracts is shown.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[Constitution]
FIG. 1 is a block diagram showing a configuration of a flame detection apparatus 10 according to an embodiment of the present invention. The flame detection device 10 is a device that detects a flame from a captured image, and receives an image captured by the external camera 5. The flame detection device 10 is typically configured by a computer device such as a PC, and includes an interface (I / F) 11, a storage unit 12, a determination unit 13, a background model database (DB) 14, and a flame color model. DB15, the display part 16, the input part 16, and the speaker 18 are provided. Note that these components are connected via a bus 19.

  A photographed image is input from the camera 5 through the interface 11. The storage unit 12 is configured by a memory or the like, and temporarily stores a captured image input from the camera 5. The storage unit 12 also stores a program for determination processing described later, and also functions as a work memory during execution of the flame detection processing.

  The determination unit 13 detects a flame included in the captured image by analyzing the captured image input from the camera 5. The captured image is a moving image composed of a plurality of frame images. The flame detection process executed by the determination unit 13 will be described later.

  The background model DB 14 stores a background model used for extracting a moving object from a captured image. Specifically, the background model DB 14 stores models of background images that do not move between a plurality of frame images.

  The flame color model DB 15 stores data indicating the color corresponding to the flame. Specifically, a look-up table in which the color corresponding to the flame is represented by coordinates in the LAB color space is stored by learning the color data of various flames by LAB conversion.

  The display unit 16 is configured by a liquid crystal display, for example, and displays a captured image during monitoring, and displays a warning or the like when the determination unit 13 detects a flame. The input unit 17 is a keyboard, a mouse, or the like, and is operated by a person who performs monitoring. The speaker 18 outputs a warning sound or the like when the determination unit 13 detects a flame.

  In the above configuration, the interface 11 is an example of an image acquisition unit of the present invention, the determination unit 13 is an example of a vector extraction unit and a determination unit of the present invention, and the camera 5 is an example of a photographing unit in the present invention.

[Flame detection processing]
Next, the flame detection process performed by the flame detection apparatus 10 will be described. FIG. 2 is a flowchart of the flame detection process. This process is realized mainly by the determination unit 13 of the flame detection apparatus 10 executing a program prepared in advance.

  First, the determination unit 13 performs background difference processing using the background difference method (step S11). Specifically, the determination unit 13 refers to the background model stored in the background model DB 14 to identify the background region in the captured image input from the camera 5 and performs a difference process on a portion other than the background region. To extract as a moving object region. At that time, the determination unit 13 extracts the moving object region while appropriately updating the background model stored in the background model DB 14 based on the input captured image. Thereby, the moving body area | region with the possibility of a flame is extracted among the target objects contained in a picked-up image.

  Next, the determination unit 13 performs color-based classification processing (step S12). The color-based classification process is a process for extracting a flame region based on color information. Specifically, the determination unit 13 uses a machine learning method such as SVM (Support Vector Machine), for example, to extract a flame color area included in the photographed image with reference to the flame color model stored in the flame color model DB 15. To do. At this time, the determination unit 13 extracts a flame color region in the moving object region obtained by the background difference process. Thereby, the area | region with high possibility of a flame is extracted among the moving body area | regions obtained by the background difference process.

  Next, the determination unit 13 performs a fluctuation measurement process (step S13). Specifically, the determination unit 13 extracts, as a fluctuation pixel, a portion where the color has changed between frame images in the flame color region extracted by the color-based classification process. FIG. 3 schematically shows a fluctuation pixel extraction method. As shown in FIG. 3, by taking the difference between the (i-1) th frame and the ith frame, the flame that was not flame in the (i-1) th frame (non-flame color) was flamed in the ith frame. Pixels that have become colored and pixels that are flame colors in the (i-1) th frame but are no longer flame colors (non-flame color) in the i-th frame are extracted. Such a pixel whose frequency of change between flame color and non-flame color exceeds a predetermined threshold is extracted as a fluctuation pixel. Then, the determination unit 13 accumulates fluctuation pixels for a plurality of frames to generate a fluctuation accumulated image. This fluctuation accumulated image is an image showing the position of the fluctuation pixel, and is an image showing the shape of the peripheral part (outline part) of the flame. Specifically, the determination unit 13 binarizes the luminance of each flame color pixel in each frame to generate a fluctuation region binary image, accumulates the fluctuation region binary image for a plurality of frames, and generates a fluctuation accumulated grayscale image. .

  Next, the determination unit 13 performs flame color area accumulation processing (step S14). In this process, a flame color accumulation image is generated by accumulating a plurality of frames of flame color area images obtained by the color-based classification process. The flame color cumulative image is an image showing the entire shape of the flame, and in particular, the region corresponding to the center portion of the flame has a higher luminance. FIG. 4 shows an example of flame color area accumulation processing. The determination unit 13 binarizes the luminance of the flame color pixel in each frame to generate a flame color region binary image, and accumulates the flame color region binary image for a plurality of frames to generate a flame color accumulation grayscale image.

  Next, the determination unit 13 performs a flame candidate area selection process (step S15). This process is a process of determining a flame candidate region based on the fluctuation accumulated grayscale image obtained in step S13 and the flame color accumulated grayscale image obtained in step S14. In one method, as illustrated in FIG. 4, the determination unit 13 includes a rectangle that includes pixels that exceed a predetermined threshold (“4” in the example of FIG. 4) among pixels included in the flame color cumulative grayscale image. The region is determined as a flame candidate region. Then, the determined outline of the flame candidate region is corrected as necessary based on the fluctuation accumulated grayscale image. In another method, the determination unit 13 generates a combined cumulative grayscale image by combining the flame color accumulated grayscale image and the fluctuation accumulated grayscale image, and includes pixels that exceed a predetermined threshold among the pixels included in the image. A rectangle is determined as a flame candidate region. Thus, the determination unit 13 determines a flame candidate region from the captured image.

  Next, the determination part 13 performs a flame determination process (step S16). The flame determination process is a process for determining whether or not the flame candidate region selected in step S15 is a flame. Here, in the present embodiment, the determination is performed by paying attention to the movement peculiar to the flame, specifically, the expansion / contraction movement (flame breath) of the flame. More specifically, the expansion and contraction motion of the flame is represented by a movement vector using optical flow, and it is determined whether or not the flame is based on the angle formed with the center of the flame.

  FIG. 5 shows a flowchart of the flame determination process. First, the determination unit 13 extracts a flame color area in the flame candidate area selected in step S15 (step S20). Next, the determination unit 13 acquires a movement vector between frames before and after using the optical flow method (step S21). FIG. 6A shows an example of the movement vector of each pixel constituting the flame candidate region at different time points. Since the flame repeats expansion / contraction, a movement vector that basically goes outward from the center of the flame in all directions and a movement vector that goes from the outside to the center of the flame are generated at the periphery of the flame. Therefore, if the movement vector of the flame candidate region includes many vectors that go from the center of the flame to the outside and vectors that gather from the outside to the center of the flame, it is estimated that this is the expansion and contraction motion of the flame. And it can be determined that the flame candidate region is a flame.

  On the other hand, for example, a moving body other than a flame, such as a flag having a high-luminance color, often translates in a specific direction, so that the vector in a specific direction increases as illustrated in FIG. 6B. Therefore, it is possible to determine whether or not a flame is present by obtaining movement vectors of a plurality of points constituting the flame color area and analyzing the tendency thereof.

  Specifically, as illustrated in FIG. 7, the determination unit 13 determines, for each of the plurality of movement vectors obtained in step S <b> 21, an angle formed by the direction of the center of gravity of the flame area and the movement vector (hereinafter, “vector direction”). (Referred to as "angle") (step S22). Thereby, the vector direction angle is obtained for each frame by the number of vectors in the flame color region.

  Next, the determination unit 13 aggregates the vector direction angle distribution for a plurality of frames (step S23). FIG. 8A shows an example of distribution of vector direction angles obtained by aggregation. As described above, since the flame expands and contracts, when the flame color region in the flame candidate region is a flame, the vector direction angle has a distribution biased to 0 degrees or around 180 degrees. On the other hand, when the flame color area in the flame candidate area is a moving object other than the flame, the vector direction angle is distributed in a certain specific direction. Therefore, the determination part 13 determines whether it is a flame based on distribution of a vector direction angle.

  That is, as shown in FIG. 8A, the determination unit 13 determines whether or not the vector direction angle distribution is biased to 0 degrees or around 180 degrees (step S24). When the vector direction angle is biased to 0 degrees or 180 degrees (step S24: Yes), the determination unit 13 determines that the flame color area is a flame (step S25). On the other hand, when the vector direction angle is not biased to 0 degrees or 180 degrees (step S24: No), the determination unit 13 determines that the flame color area is not flame (step S26).

In one specific example of the determination in step S24, the determination unit 13 performs the determination by regarding the graph indicating the vector direction angle distribution as illustrated in FIG. 8A as a quadratic curve and calculating an approximate expression. Specifically, as shown in parentheses along the horizontal axis of FIG. 8A, a quadratic curve representing the distribution of vector direction angles by expressing 0 to 180 degrees with 10-step numerical values every 20 degrees. An approximate expression (y = ax ² + bx + c) is obtained. When the flame color region is flame, the quadratic curve indicating the distribution of the vector direction angle is convex downward as shown in FIG. 8A, and the vertex is the center of the horizontal axis (0 to 180 degrees). Should be located nearby. Therefore, the determination unit 13 first determines whether the coefficient “a” of the obtained approximate expression is a> 0 and whether the X coordinate of the vertex of the quadratic curve is near the center of the X axis. For example, as illustrated in FIG. 8B, the determination unit 13 determines that the coefficient a of the approximate expression is a predetermined threshold value a> a1 (> 0), and the vertex X coordinate x _ap is a predetermined threshold value x1 < When x _ap <x2, the flame color area is determined to be flame.

  When the flame determination process is completed as described above, the process returns to the main routine of FIG. And the determination part 13 performs an output process based on the result of a flame determination process (step S17). Specifically, when it is determined that the flame is determined by the flame determination process, the determination unit 13 outputs a warning or the like using the display unit 16 and / or the speaker 18. On the other hand, when it is determined that it is not flame, the determination unit 13 does not output anything or outputs display / sound indicating that it is not flame. Thus, the flame detection process ends.

  As described above, in the present embodiment, the flame is detected by paying attention to the expansion and contraction motion that is a specific motion of the flame. Therefore, for example, it is possible to identify a moving object and a flame swaying at the same place, such as a red flag, with high accuracy, and the reliability of flame detection can be improved.

[Modification]
(Modification 1)
In the above embodiment, in the flame determination process, the angle formed by the direction of the center of gravity of the flame color area and the movement vector is the vector direction angle, but instead of the center of gravity, the direction of the center of the flame color area and the movement vector The formed angle may be a vector direction angle. In other words, in the present invention, the center of gravity or the center of the flame-colored region is set as the “center”, and the angle formed by the direction of the center of the flame-colored region and the movement vector may be used as the vector direction angle.

(Modification 2)
In the above embodiment and the first modification, the angle formed by the center of gravity or the center direction of the flame color area and the movement vector is the vector direction angle, and flame determination is performed based on the distribution. Flame determination may be performed based on the direction of the vector. FIG. 9A schematically shows the direction of the movement vector when the flame expands, and FIG. 9B schematically shows the direction of the movement vector when the flame contracts. Basically, when the flame expands, each movement vector becomes a vector toward the outside of the flame as shown in FIG. 9A, and when the flame contracts, each movement as shown in FIG. 9B. The vector goes to the center of the flame. That is, it is considered that the direction of the movement vector is distributed almost uniformly over the entire circumference (360 degrees) regardless of whether the flame expands or contracts. Therefore, in the flame determination process, the direction of the movement vector in the flame color area of the flame candidate area is totaled, and when the distribution is almost evenly distributed in the direction of 360 degrees, the flame color area is determined to be flame. It is good to do.

(Modification 3)
In the above embodiment, the flame detection device 10 is configured by a PC or the like, and a captured image from the external camera 5 is supplied. In another configuration example, the flame detection device 10 may be downsized and built in the camera 5 to be configured as a camera with a flame detection function. In still another configuration example, a system may be constructed in which the detection result from the flame detection device 10 is transmitted to the server device through a network, and the server device performs overall control. In still another configuration example, the flame detection device 10 may be configured as a server device. In this case, a photographed image from the camera 5 is transmitted to the server device through the network, and the server device executes a flame detection process to detect the flame. As a result, it is possible to construct a system in which a plurality of cameras are provided for each monitoring location, and a captured image is transmitted to the server device, and a large number of locations are simultaneously monitored by the server device.

DESCRIPTION OF SYMBOLS 5 Camera 10 Flame detection apparatus 12 Memory | storage part 13 Judgment part 14 Background model database 15 Flame color model database

Claims (7)

Image acquisition means for acquiring a captured image;
Vector extraction means for extracting a vector indicating the movement of the object included in the captured image;
Based on the vector, when the object is inflated or contracted, determination means for determining the object as a flame,
A flame detection apparatus comprising:   The flame detection apparatus according to claim 1, wherein the determination unit performs the determination based on an angle formed by a center portion of the object and the vector.   The flame detection device according to claim 2, wherein the determination unit determines that the object is a flame when the angle distribution is biased to around 0 degrees and 180 degrees. The vector extracting means extracts movement vectors of a plurality of pixels constituting the object based on a plurality of captured images,
The flame detection apparatus according to claim 2, wherein the determination unit determines a central portion of the object for each of the plurality of captured images.   The flame detection device according to claim 2, wherein the determination unit determines that the object is a flame when the direction of the vector is evenly distributed at 360 degrees. Photographing means for generating the photographed image;
The flame detection device according to claim 1,
A camera comprising: A program executed by a flame detection device including a computer,
Image acquisition means for acquiring a captured image;
Vector extraction means for extracting a vector indicating the movement of the object included in the captured image;
A determination means for determining the object as a flame when the object is inflated or contracted based on the vector;
A program for causing the computer to function as: