JPH04191999A - Fire detector - Google Patents

Abstract

PURPOSE:To heighten detecting accuracy economically by extracting a change area from a visible video obtained in time series, unifying the change area at every same event, and deciding whether or not it is fire based on the feature quantity of a unified change area. CONSTITUTION:A start-up means 3 starts up either a smoke detecting means 4 or a flame detecting means 5 by supplying different parameters corresponding to the detection of either smoke or flame based on the judging result of a judging means 2. The occurrence of the fire can be detected by fetching a monitor image from an image pickup means 1 based on the parameters supplied from the start-up means 3 in time series by the smoke detecting means 4 and the flame detecting means 5, and detecting a smoke or flame area by performing image processing. Since decision for the occurrence of the fire is detected by judging whether it is due to the smoke or the flame in such way, and also, since the change area is unified at every same event and it is decided whether or not it is the fire by the feature quantity of the change area, it is possible to detect the fire with high accuracy economically.

Description

[Detailed Description of the Invention] CObject of the Invention] (Industrial Application Field) This invention automatically detects the occurrence of fires in urban areas by analyzing surveillance images from cameras fixed at high places. This invention relates to a fire detection device for detecting fire.

(Prior Art) In recent years, the occurrence of large earthquakes has been predicted, and awareness of disaster prevention has increased, and several urban disaster prevention information systems are being developed at the national and local government levels. In particular, there is a strong desire for a fire prevention system that can prevent simultaneous, large-area fires that occur in conjunction with large earthquakes.

In fire detection systems that cover wide areas such as urban areas,
- Since it is possible to monitor a wide area at once, an effective method is to detect the occurrence of a fire through image processing of surveillance video captured by a sensor such as a monitor camera. This method can be roughly divided into a method that uses infrared images captured by a similar sensor such as an infrared imaging device, and a method that uses visible images captured by a normal ITV. However, sensors such as infrared imaging devices have problems in terms of cost and maintenance. On the other hand, when visible images are used, it is difficult to detect smoke and flames using conventional processing on a single image because they can be identified using feature quantities such as shading information and shape. Furthermore, simple differential processing between images could not provide sufficient detection performance due to noise such as swaying trees or flashing lights such as neon lights.

(Problems to be Solved by the Invention) As described above, when an infrared imaging device or a similar device is used, there are problems in terms of qualification and maintenance. Furthermore, when a normal ITV was used, sufficient performance could not be obtained.

This invention was made in view of these conventional problems, and an object thereof is to provide a fire detection device that is economical and can detect fires with high accuracy using a normal ITV. .

[Structure of the Invention] (Means for Solving the Problems) This invention firstly captures surveillance images from a camera in chronological order and performs predetermined image processing to detect the occurrence of a fire. In the apparatus, means for determining and switching between a first mode in which fire occurrence is detected by smoke or a second mode in which fire is detected by flame; and when the first mode is selected by the means, detecting a smoke area from the image, and detecting a flame area from the image when the second mode is selected, and detecting a fire based on generated smoke or flames. It is a device.

Second, there is a means for capturing monitoring images from the camera in time series, a means for detecting a changed area in the image captured by this means, and an integrated storage of the changed areas detected by this means for each same event. This is a fire detection device characterized by comprising means and means for determining whether or not there is a fire based on the results integrated by the means.

(Function) As described above, according to the present invention, firstly, the occurrence of a fire is detected by determining whether it is caused by smoke or flames;
It is economical and highly accurate because it extracts changing areas from visible images obtained in chronological order, integrates these changing areas for each same event, and determines whether there is a fire or not based on the features of the integrated changing areas. fire can be detected.

(Example) Hereinafter, one embodiment of the present invention will be described in detail based on the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. A monitoring image captured by an imaging means 1, such as a camera fixed in advance at a high place, is input to a determining means 2, which determines whether to detect smoke or flame, and performs switching. This determination is made, for example, based on the average density of the entire input monitoring image.

The following policies are in place when detecting fires using surveillance footage: During the day, smoke is recognized for fire detection. That is, smoke always accompanies a fire and can be seen earlier than flames. However,
At night, smoke is harder to see and flames are easier to detect.

Therefore, at night, flames are recognized for fire detection.

The starting means 3 starts the smoke detecting means 4 or the flame detecting means 5 by giving different parameters depending on whether smoke or flame is to be detected, based on the judgment result of the judging means 2.

The smoke detection means 4 and the flame detection means 5 acquire monitoring images from the imaging means 1 in time series based on the parameters given by the activation means 3, and detect the occurrence of a fire by detecting smoke or flame areas through image processing. Detect.

FIG. 2 shows an example of the structure of the smoke detection means 4 and the flame detection means 5. The smoke detecting means 4 and the flame detecting means 5 are composed of common means, with the only difference being the parameter applied during execution. A monitoring image inputted from the imaging means 1 is A/D converted by the monitoring image capturing means 11 and captured at regular time intervals.

The monitoring images captured at regular time intervals are supplied in time series to the change area detection means 12, and change areas between the time series images are detected. The detected change areas are integrated by the integrating means 13 for each same event. The determining means 14 determines the feature amount of the integrated change area, and uses this feature amount to determine whether or not there is a fire. Here, the feature amount is the amount of expansion of the area of the temporally changing region.

Smoke and fire are characterized by temporal expansion. Therefore, fire detection is performed by analyzing temporal changes.

First, as candidates for fire areas, changing areas are extracted by differential processing between time-series images. The extracted change region contains noise due to confusion (disturbance) events.

Disruptive events include the shaking of trees due to the wind, animals such as people and cars during the day, and illuminated objects such as neon lights and car headlights during the night. In order to identify this disturbance event and the fire area, we perform a motion analysis based on tracking the changed area. In tracking, the change areas obtained in time series are integrated and the change areas caused by the same event are integrated into one group. Integration of change areas cannot be achieved by simply performing a logical sum between change areas obtained over time. Here, the integration uses a method to predict the expansion range of the changing area by using the location and attribute information of buildings, geographical information such as roads, etc.

It is thought that the area of change due to integrated fires will monotonically expand. On the other hand, the size of the region of change caused by a disruptive event is limited. Therefore, by examining the expandability of the integrated change area, we can identify whether or not it is a fire.

FIG. 3 shows an example of the structure of the change area detection means 12.

A difference means 22 calculates the difference between the current monitoring image supplied from the monitoring image capturing means 11 and the previous inspection image stored in the previous image storage means 21. This difference result is binarized by the binarization means 23 using a preset threshold value, and regions with large changes are extracted. The binarization result is given a different label to each connected region by the labeling means 24. Hereafter, this labeled image is labeled, the difference image is labeled as LDI, and each connected region is labeled as CR.

The CR is supplied as a fire area candidate to an integrating means 13, which will be described later.

Next, the processing procedure of the integrating means 13 will be explained based on FIGS. 4 to 9. Integration is performed by relating each CR on the LDI obtained in chronological order to the same event. This integration result is a differential image (ALD) with an integrated label.
I) and grouping information table.

ALDI is a label image in which CRs of LDI are accumulated.

These label areas (hereinafter referred to as LR) are integrated into one group for each same event. Each group has a CR (seed) that is initially generated on the LDI. In the case of a fire, this type refers to the point of origin. (See FIG. 4) The grouping information table relates to each group. It consists of the group ID, the area of the area integrated as a group and its boundary data, the elapsed time since the group was created, and the history information of the process of forming the group (proximity relationship between LRs). (Refer to Fig. 5) The integrating means 13 integrates the change area that overlaps with the integration result up to the previous point in time as the same animal body, and the integrating means 13 calculates the expansion range of the change area at the present time based on the integration result at the previous point in time. It consists of an integrating means (2) that predicts and integrates the change areas included in this expanded range as the same event.

First, the processing procedure of the integrating means (2) will be explained based on FIG. It is assumed that the previous ALDI and integrated information tables have been initialized at system startup.

First, the previous AL old is set as the current ALDI (step 101). The CRs on the LDI are taken out sequentially and the current ALDI
Check whether there is a duplicate LR above. Which LR is CR?
To determine if this is a duplicate, take out one by one Step 1
02), calculate the logical sum of this OCR and the current ALD l, label the result, check the inclusion relationship between each connected region and CR (step 103), and the area of the connected region to be included is included. If the area is different from the area of the CR, the CR is assumed to overlap the LR (Step 104).

The duplicate CR is added to the current ALD I as a new LR (step 105). Check whether CRs overlap in a single group or in multiple groups (step 10)
B). To determine whether the overlapping group is single or multiple, check the overlapping LR based on the inclusion relationship between the connected region that includes the CR and each LR, and check the group to which the overlapping LR belongs. (step 106). If there is only one overlapping group, update the integrated information regarding that group (step 107).
). If the information overlaps with a plurality of groups, select a representative group, integrate the integrated information regarding other overlapping groups into this representative group, and then update.

Assume that the representative group is the one for which the longest elapsed time has passed since the group was created (step 108).

Execute the above process for all CRs on LDI (
Step 109).

Next, the processing procedure of the integrating means (2) will be explained based on FIG. The integration means (2) is executed for the CRs that were not integrated by the above-mentioned integration means (2).

First, the integrated information of each group is sequentially retrieved from the integrated information table (step 210), and the processing mode of smoke detection or flame detection is checked in order to predict the current expansion area of each group (step 202). In the case of smoke detection, the expansion area is calculated based on group boundary information and a smoke diffusion model, as shown in FIG. Calculation of group boundary information is based on Yr of the position information of each LR making up the group.
Ain and Ymax.

And Xm1n and XI of LR included in each divided area when the area between Ywin and Ymax is divided into several parts in the Y direction.
Determine from the value of aax. Based on this boundary information, a smoke diffusion model (diffusion width) for each direction: left, right, and upward direction.
Find the extended area from. The diffusion model is experimentally set in advance so that it expands greatly in the rising direction of smoke and in the direction of the wind (step 203). In the case of flame detection, the expansion area is determined using a fire spread model (for example, the Hamada model is a typical model) as follows. This model is
The method estimates the area where fire will spread from a certain building to nearby buildings at each hour based on wind direction, wind strength, building shape, etc.

This situation is shown in FIG. Therefore, the current fire spread range is predicted based on the group boundary information at the previous point in time, the building form information near the group boundary, and the wind direction and wind force (step 204). Next step 203 or step 2
It is checked whether or not there is a CR included in the extended area calculated in step 04 (step 205), and the included CR is the current A
Add it to LDI as a new LR (step 20B). Also, the integrated information regarding the included group is updated (step 207). The above processing is executed for all groups registered in the integrated information processing table (step 208). After performing the above processing for all groups, check the integration information of each group, and if the LRs overlap between groups, merge the overlapping groups into one representative using the same method as the integration method ■. group - integrate into one (Ste Nobu 209). If there is an unintegrated CR at this point, execute the following process (
Ste Nobu 210). First, unintegrated CRs are sequentially taken out (step knob 211), and it is checked whether they are included in the mask area (step knob 212). If it is not included in the mask, select that CR as a new fire area candidate.
It is added to the old (step 214). Then, information related to CR is newly registered in the integrated information table (step
Whelk 215). It is assumed that fires do not start from areas that have been masked in advance, such as roads, trees, and neon lights, and that CR included within the masked area is determined to be noise. It also has no effect on the DI and integrated information table. The above process is executed for all unintegrated CRs (step 216). The information of the current LD+ and the integrated information table is supplied to the determination means 14.

FIG. 10 shows the processing procedure of the determining means 14. Whether or not there is a fire is determined based on the area of each group and the elapsed time. First, the integrated information of each group is sequentially extracted from the integrated information table (step 501), and the area and elapsed time of the integrated area of each group are checked (step 502).

If the area of the integrated area is greater than a certain value, it is determined that it is a fire area, and the location of the fire is output, an alarm is issued, and instructions are given by the administrator (step 504). If the area of the integrated area is less than a certain value and a certain period of time has elapsed, it is considered noise and the Li2 on the current AL and the integrated information on the integrated information table for that group are deleted (
Step 503). The above processing is executed for all groups (step 505).

After processing all groups, change the current ALDI to the previous AL
The process ends as old (step 506).

By repeatedly performing the above processing on monitoring images obtained in chronological order, the occurrence of a fire is automatically detected.

〔Effect of the invention〕

As described above in detail based on the embodiments, according to the present invention, changing regions are extracted from visible images obtained in time series, these changing regions are integrated for each same event, and the integrated changing regions are By determining whether or not there is a fire based on the feature amount, it is possible to realize a practical fire detection device that is economical and has high detection accuracy.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram of the embodiment, FIG. 2 is a diagram of the smoke and flame detection means, FIG. 3 is a diagram of the change area detection means, and FIG.
The figure shows an example of an image with an integrated label, Figure 5 shows an example of an integrated information table, Figure 6 is a flowchart of the process of the integrating means ■, and Figure 7 shows the process of the integrating means ■. FIG. 8 is an explanatory diagram of smoke diffusion area prediction, FIG. 9 is an explanatory diagram of fire spread prediction, and FIG. 1O is a flowchart of the determination means. 1... Imaging means 2... Determination means 3... Starting means 4... Smoke detection means 5... Flame detection means 11... Surveillance image capturing means 12... Change area detection means 13. ... Integration means 14 ... Judgment means 21 ... Previous image storage means 22 ... Differentiation means 23 ... Binarization means 24 ... Labeling means

Claims (3)

[Claims] (1) In a fire detection device that captures monitoring images from a camera in chronological order and detects the occurrence of a fire from images that have been subjected to predetermined image processing, the first mode detects the occurrence of a fire using smoke; or means for determining and switching to a second mode of detection based on flame; detecting a smoke area from the image when the first mode is selected by the means; and when the second mode is selected; A fire detection device comprising a detection means for detecting a flame area from the image, and detects a fire based on generated smoke or flames. (2) means for capturing monitoring images from the camera in chronological order; means for detecting changed areas in the images captured by this means; and means for integratedly storing changed areas detected by this means for each same event; , a fire detection device characterized by comprising means for determining whether or not there is a fire based on the results integrated by this means. (3) The integration means includes a first integration means that integrates change areas that overlap with the integration results up to the previous point in time as the same event, and a first integration means that predicts the expansion range of the change area at the current time based on the integration results up to the previous point in time. 3. The fire detection device according to claim 2, further comprising second integrating means for integrating the change areas included in the expanded range as the same event.