Image processing system of smoke fire detector

Abstract

The invention relates to the technical field of fire detectors, and discloses a smoke-sensing fire detector image processing system, which is characterized in that background images and fire images are acquired, quality evaluation is carried out on the background images, and after screening, denoising processing is carried out and the images are transmitted to an identification module and an electric video monitoring center for storage; the feature extraction unit obtains features of the fire image and the background image, the features are compared through the image comparison unit, different parts of the fire image and the background image are screened out, and an identification result is generated and transmitted to the coding module; the coding module converts the identification result into a digital signal through an image coding method and transmits the digital signal to the electric video monitoring center, and the electric video monitoring center receives and decodes the digital signal and synthesizes the digital signal with a pre-stored background image to obtain a final output fire image. According to the method, the accurate recognition result of the fire image is coded, the large data transmission amount of large image coding is avoided, the transmission speed is improved, and a means is provided for remote and rapid fire confirmation in an unattended place.

Description

Technical Field

[0001] The invention relates to the technical field of fire detectors, in particular to an image processing system for smoke fire detectors. Background Art

[0002] Fire safety covers many aspects, including fire prevention, fire escape, fire emergency response and the correct use of firefighting equipment. Common preventive measures in firefighting include fire detectors, which detect physical or chemical changes before and after a fire to determine whether a fire has occurred. Specifically, common types of fire detectors include: heat detectors, smoke detectors, flame detectors, aspirating smoke detectors, etc.

[0003] Early detection and rapid confirmation of fires in unmanned base stations, 24-hour financial business halls, substations, underground pipe corridors and other unmanned places has become a pain point in fire safety. Traditional fire detectors have limitations in their application in such places and cannot provide intuitive fire scene information for fire confirmation. Once fire detection and confirmation are delayed, it may cause great losses. Summary of the invention

[0004] In view of the shortcomings of the existing methods, the present invention provides a smoke fire detector image processing system, which compresses and encodes the recognition results of accurate fire images through a coding method, avoids encoding the entire image, reduces the amount of transmitted data, and improves the transmission speed, providing a technical means for remote and rapid confirmation of fires in unmanned places.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A smoke fire detector image processing system, comprising a smoke fire detector for detecting smoke conditions at a fire scene, characterized in that it also comprises an acquisition module, a processing module, an identification module and an encoding module, wherein the acquisition module is provided with a camera for acquiring background images and fire images, the acquisition module acquires background images where no fire occurs by frequency through the camera and generates background image information, the acquisition module performs quality assessment on the background image, transmits the background image information that meets the image quality requirements to the processing module, and transmits it to the identification module and an electric video monitoring center for storage after denoising;

[0007] When a fire occurs, the smoke detector triggers the camera to collect fire images and generate fire image information, which is then transmitted to the recognition module after denoising;

[0008] The recognition module includes a feature extraction unit and an image comparison unit. The feature extraction unit obtains the features of the fire image and the background image. The image comparison unit compares the features of the background image with the fire image, screens out the parts that are different between the fire image and the background image, generates recognition results for the different parts and transmits them to the encoding module; the encoding module converts the recognition results into digital signals through an image encoding method and transmits them to the electronic video monitoring center. The electronic video monitoring center receives the digital signal, decodes it, and synthesizes it with the pre-stored background image to obtain the final output fire image.

[0009] The acquisition module sets the shooting frequency of the camera used to acquire the background image, and the camera captures the background image according to the shooting frequency.

[0010] The image information of the fire image and the background image generated by the acquisition module includes the ordinate value and pixel value, the abscissa value and pixel value, and the image data.

[0011] The acquisition module performs quality assessment on the background image data, including the following steps:

[0012] S1, set the signal-to-noise ratio threshold and calculate the root mean square error of the background image data;

[0013]

[0014] Wherein, α is the root mean square error of the background image data; m is the pixel value of the ordinate of the background image data, and n is the pixel value of the abscissa of the background image data; C1 is the initial background image data or the background image data collected and stored last time; C2 is the background image data with noise collected this time;

[0015] S2. Calculating a quality assessment value of the background image data according to a root mean square error of the background image data;

[0016]

[0017] Wherein, γ is the quality assessment value of the background image data; β is the maximum value of the point color of the background image data; log is the logarithmic function in the quality assessment of the background image data;

[0018] S3. Compare the quality assessment value γ of the background image data with the set signal-to-noise ratio threshold. When the quality assessment value γ of the background image data is less than the set signal-to-noise ratio threshold, the image quality requirement is not met, and the camera re-collects the background image. When the quality assessment value γ of the background image data is greater than or equal to the set signal-to-noise ratio threshold, the image quality requirement is met, and the image is transmitted to the recognition module and the video surveillance center for storage.

[0019] The feature extraction unit extracts the features of the fire image and the background image, comprising the following steps:

[0020] Q1. Calculate fire image features or background image features according to fire image data or background image data;

[0021] χ=δη(χ m , χ n )

[0022] Among them, χ is the fire image feature or background image feature; δ is the Gaussian function; χ m is the ordinate value of the fire image or background image, χ n is the horizontal coordinate value of the fire image or background image;

[0023] Q2. Perform extreme value detection according to the fire image features or the background image features to obtain the extreme values ​​of the fire image features or the background image features;

[0024]

[0025] in, is the extreme value of the fire image feature or the background image feature; λ is the Gaussian difference calculation function.

[0026] The calculation formula for the image contrast unit to screen out the parts that are different between the fire image and the background image is as follows:

[0027]

[0028] Among them, μ is the number of pixel value statistics of the fire image; is the extreme value of the background image feature; is the extreme value of the fire image;

[0029] When μ is greater than the set fire pixel threshold, it means that a fire has occurred at the scene. At this time, the part that is different from the background image is screened out and labeled. The label and the background image jointly generate the recognition result.

[0030] The encoding module converts the recognition result into a digital signal through an image encoding method and transmits it to the video surveillance center, specifically comprising the following steps:

[0031] T1, the encoding module receives the recognition result transmitted by the recognition module, and cuts the fire image of the recognition result into multiple area blocks;

[0032] T2. The encoder set in the encoding module receives multiple regional blocks and assigns corresponding block texture features to each regional block;

[0033] T3, dimensionally merge the texture features of multiple tiles to obtain a compressed texture feature set;

[0034] T4. The converter set in the encoding module converts the compressed texture feature set, converts the analog signal of the compressed texture feature set into a digital signal, and transmits the digital signal to the power transmission video monitoring center respectively.

[0035] The video surveillance center receives the digital signal and decodes it to obtain the final output image, which specifically includes the following steps:

[0036] L1, a preset decoder, which restores the digital signal to an original compressed texture feature set through the decoder, and finds out the row features closest to each row of texture features in the compressed texture feature set through a lookup table set in the encoder in the encoding module according to the compressed texture feature set, and obtains a nearest search result set according to the row features;

[0037] L2. Dimensionally splitting the nearest neighbor search result set to obtain the nearest neighbor texture feature that is closest to the block texture feature of each of the plurality of region blocks;

[0038] L3, inputting the nearest texture feature of each region block in the multiple region blocks into the current decoder to obtain multiple nearest region blocks;

[0039] L4, stitching multiple nearest area tiles;

[0040] L5. According to the label information generated by the image comparison unit, the decoded spliced ​​image is synthesized with the background image to generate the final fire image.

[0041] The present invention has the following beneficial effects and advantages:

[0042] 1. The present invention collects the background image by setting a collection module, and performs quality assessment on the background image, compares the background image quality assessment value with the set signal-to-noise ratio threshold, and decides whether to re-collect the image as the background image, thereby ensuring the accuracy of background image collection;

[0043] 2. The present invention uses a Gaussian function to calculate the background image and fire features by setting a feature extraction unit; the image comparison unit compares the extreme values ​​of the background image and the fire image features to screen out an accurate fire image and provide intuitive fire scene information;

[0044] 3. The present invention compresses and encodes the recognition results of accurate fire images through the decoding module set up by the coding method, thereby avoiding the encoding of the entire image, reducing the amount of transmitted data, and improving the transmission speed, thereby providing a technical means for remote and rapid confirmation of fires in unmanned places. DETAILED DESCRIPTION

[0045] The present invention is described in detail below with reference to specific examples.

[0046] The present invention provides a smoke fire detector image processing system and an image coding method, including a smoke fire detector for detecting smoke conditions at a fire scene, characterized in that it also includes an acquisition module, a processing module, an identification module and an encoding module, wherein the acquisition module is provided with a camera for acquiring background images and fire images; the acquisition module generates image information of the fire image and the background image including the image's ordinate value and pixel value, abscissa value and pixel value, and image data.

[0047] The acquisition module is set to acquire a camera's shooting frequency for background images. The camera takes a background image at regular intervals according to the shooting frequency and generates background image information. The acquisition module evaluates the quality of the background image and transmits the background image information that meets the image quality requirements to the processing module, and then transmits it to the recognition module and the video surveillance center for storage after denoising. Specifically, the steps include:

[0048] S1, set the signal-to-noise ratio threshold and calculate the root mean square error of the background image data;

[0049]

[0050] Wherein, α is the root mean square error of the background image data; m is the pixel value of the ordinate of the background image data, and n is the pixel value of the abscissa of the background image data; C1 is the initial background image data or the background image data collected and stored last time; C2 is the background image data with noise collected this time;

[0051] S2. Calculating a quality assessment value of the background image data according to a root mean square error of the background image data;

[0052]

[0053] Wherein, γ is the quality assessment value of the background image data; β is the maximum value of the point color of the background image data; log is the logarithmic function in the quality assessment of the background image data;

[0054] S3. Compare the quality assessment value γ of the background image data with the set signal-to-noise ratio threshold. When the quality assessment value γ of the background image data is less than the set signal-to-noise ratio threshold, the image quality requirement is not met, and the camera re-collects the background image. When the quality assessment value γ of the background image data is greater than or equal to the set signal-to-noise ratio threshold, the image quality requirement is met, and the image is transmitted to the recognition module and the video surveillance center for storage.

[0055] When a fire occurs, the smoke detector triggers the camera to collect fire images and generate fire image information, which is then transmitted to the recognition module after denoising;

[0056] Preferably, the noise of the background image and the fire image is eliminated by using a Kalman filter. On the basis of detecting the image, an optimal filtering method under minimum variance is used according to corresponding filtering rules to filter and repair the image.

[0057] The recognition module includes a feature extraction unit and an image comparison unit. The feature extraction unit obtains the features of the fire image and the background image. The image comparison unit compares the features of the background image with the fire image. Specifically, the recognition module includes the following steps:

[0058] Q1. Calculate fire image features or background image features according to fire image data or background image data;

[0059] χ=δη(χ m , χ n )

[0060] Among them, χ is the fire image feature or background image feature; δ is the Gaussian function; χ m is the ordinate value of the fire image or background image, χ n is the horizontal coordinate value of the fire image or background image;

[0061] Q2. Perform extreme value detection according to the fire image features or the background image features to obtain the extreme values ​​of the fire image features or the background image features;

[0062]

[0063] in, is the extreme value of the fire image feature or the background image feature; λ is the Gaussian difference calculation function.

[0064] Filter out the different parts of the fire image and the background image, generate recognition results for the different parts and transmit them to the encoding module; the calculation formula is as follows;

[0065]

[0066] Among them, μ is the number of pixel value statistics of the fire image; is the extreme value of the background image feature; is the extreme value of the fire image;

[0067] When μ is greater than the set fire pixel threshold, it means that a fire has occurred at the scene. At this time, the part that is different from the background image is screened out and labeled. The label and the background image jointly generate the recognition result.

[0068] The encoding module converts the recognition result into a digital signal through an image encoding method and transmits it to the video surveillance center; specifically, the following steps are included:

[0069] T1, the encoding module receives the recognition result transmitted by the recognition module, and cuts the fire image of the recognition result into multiple area blocks;

[0070] More specifically, each image in the fire scene image of the recognition result is cut into multiple training blocks. The size of the training block can be 16×16, 32×32 or 64×64. The specific selection needs to be adjusted according to the test results obtained after the training is completed. For example, assuming that there are 100 images of the fire scene image of the recognition result, the resolution of each image is 1920×1080, and the resolution of the training block is selected as 32×32, then each image can be divided into 2025 training blocks, and the entire image data set generates a total of 202,500 regional blocks;

[0071] T2. The encoder set in the encoding module receives multiple regional blocks and assigns corresponding block texture features to each regional block;

[0072] The network in the encoder includes multiple cascaded convolutional neural networks, each convolutional neural network module may include a combination of one or more layers of a two-dimensional convolution layer, an activation layer, a normalization layer, and a residual structure layer. The input of the encoder is a training block, and the pixel data of each training block is used as a set of training data;

[0073] T3, dimensionally merge the texture features of multiple tiles to obtain a compressed texture feature set;

[0074] T4. The converter set in the encoding module converts the compressed texture feature set, converts the analog signal of the compressed texture feature set into a digital signal, and transmits the digital signal to the power transmission video monitoring center respectively.

[0075] The video surveillance center receives the digital signal, decodes it, and synthesizes it with the pre-stored background image to obtain the final output fire image; specifically, the following steps are included:

[0076] L1, a preset decoder, which restores the digital signal to an original compressed texture feature set through the decoder, and finds out the row features closest to each row of texture features in the compressed texture feature set through a lookup table set in the encoder in the encoding module according to the compressed texture feature set, and obtains a nearest search result set according to the row features;

[0077] L2. Dimensionally splitting the nearest neighbor search result set to obtain the nearest neighbor texture feature that is closest to the block texture feature of each of the plurality of region blocks;

[0078] L3, inputting the nearest texture feature of each region block in the multiple region blocks into the current decoder to obtain multiple nearest region blocks;

[0079] L4, stitching multiple nearest area tiles;

[0080] L5. According to the label information generated by the image comparison unit, the decoded spliced ​​image is synthesized with the background image to generate the final fire image.

Claims

1. A smoke fire detector image processing system, comprising a smoke fire detector for detecting smoke conditions at a fire scene, characterized in that: It also includes an acquisition module, a processing module, an identification module and an encoding module. The acquisition module is provided with a camera for acquiring background images and fire images. The acquisition module uses the camera to acquire background images without fire according to frequency and generates background image information. The acquisition module evaluates the quality of the background image and transmits the background image information that meets the image quality requirements to the processing module. After denoising, the background image information is transmitted to the identification module and the video surveillance center for storage. When a fire occurs, the smoke detector triggers the camera to collect fire images and generate fire image information, which is then transmitted to the recognition module after denoising; The recognition module includes a feature extraction unit and an image comparison unit. The feature extraction unit obtains the features of the fire image and the background image. The image comparison unit compares the features of the background image with the fire image, screens out the parts that are different from the fire image and the background image, and generates recognition results for the different parts and transmits them to the encoding module; the encoding module converts the recognition results into digital signals through an image encoding method and transmits them to the electronic video monitoring center. The electronic video monitoring center receives the digital signal, decodes it, and synthesizes it with the pre-stored background image to obtain the final output fire image.

2. The smoke fire detector image processing system according to claim 1, characterized in that: The acquisition module sets the shooting frequency of the camera used to acquire the background image, and the camera captures the background image according to the shooting frequency.

3. The smoke fire detector image processing system according to claim 1, characterized in that: The image information of the fire image and the background image generated by the acquisition module includes the ordinate value and pixel value, the abscissa value and pixel value, and the image data.

4. The smoke fire detector image processing system according to claim 1, characterized in that: The acquisition module performs quality assessment on the background image data, including the following steps: S1, set the signal-to-noise ratio threshold and calculate the root mean square error of the background image data; Wherein, α is the root mean square error of the background image data; m is the pixel value of the ordinate of the background image data, and n is the pixel value of the abscissa of the background image data; C1 is the initial background image data or the background image data collected and stored last time; C2 is the background image data with noise collected this time; S2. Calculating a quality assessment value of the background image data according to a root mean square error of the background image data; Wherein, γ is the quality assessment value of the background image data; β is the maximum value of the point color of the background image data; log is the logarithmic function in the quality assessment of the background image data; S3. Compare the quality assessment value γ of the background image data with the set signal-to-noise ratio threshold. When the quality assessment value γ of the background image data is less than the set signal-to-noise ratio threshold, the image quality requirement is not met, and the camera re-collects the background image. When the quality assessment value γ of the background image data is greater than or equal to the set signal-to-noise ratio threshold, the image quality requirement is met, and the image is transmitted to the recognition module and the video surveillance center for storage.

5. The smoke fire detector image processing system according to claim 1, characterized in that: The feature extraction unit extracts the features of the fire image and the background image, comprising the following steps: Q1. Calculate fire image features or background image features according to fire image data or background image data; x=δη(x m ,x n ) Among them, χ is the fire image feature or background image feature; δ is the Gaussian function; χ m is the ordinate value of the fire image or background image, χ n is the horizontal coordinate value of the fire image or background image; Q2. Perform extreme value detection according to the fire image features or the background image features to obtain the extreme values ​​of the fire image features or the background image features; in, is the extreme value of the fire image feature or the background image feature; λ is the Gaussian difference calculation function.

6. A smoke fire detector image processing system according to claim 5, characterized in that: The calculation formula for the image contrast unit to screen out the parts that are different between the fire image and the background image is as follows: Among them, μ is the number of pixel value statistics of the fire image; is the extreme value of the background image feature; is the extreme value of the fire image; When μ is greater than the set fire pixel threshold, it means that a fire has occurred at the scene. At this time, the part that is different from the background image is screened out and labeled. The label and the background image jointly generate the recognition result.

7. The smoke fire detector image processing system according to claim 1, characterized in that: The encoding module converts the recognition result into a digital signal through an image encoding method and transmits it to the video surveillance center, specifically comprising the following steps: T1, the encoding module receives the recognition result transmitted by the recognition module, and cuts the fire image of the recognition result into multiple area blocks; T2. The encoder set in the encoding module receives multiple regional blocks and assigns corresponding block texture features to each regional block; T3, dimensionally merge the texture features of multiple tiles to obtain a compressed texture feature set; T4. The converter set in the encoding module converts the compressed texture feature set, converts the analog signal of the compressed texture feature set into a digital signal, and transmits the digital signal to the power transmission video monitoring center respectively.

8. The smoke fire detector image processing system according to claim 1, characterized in that: The video surveillance center receives the digital signal and decodes it to obtain the final output image, which specifically includes the following steps: L1, a preset decoder, which restores the digital signal to an original compressed texture feature set through the decoder, and finds out the row features closest to each row of texture features in the compressed texture feature set through a lookup table set in the encoder in the encoding module according to the compressed texture feature set, and obtains a nearest search result set according to the row features; L2. Dimensionally splitting the nearest neighbor search result set to obtain the nearest neighbor texture feature that is closest to the block texture feature of each of the plurality of region blocks; L3, inputting the nearest texture feature of each region block in the multiple region blocks into the current decoder to obtain multiple nearest region blocks; L4, stitching multiple nearest-neighborhood tiles; L5. According to the label information generated by the image comparison unit, the decoded spliced ​​image is synthesized with the background image to generate the final fire image.

Images