CN212873683U - High-speed railway safety zone intrusion alarm device based on infrared image - Google Patents
High-speed railway safety zone intrusion alarm device based on infrared image Download PDFInfo
- Publication number
- CN212873683U CN212873683U CN202021346493.1U CN202021346493U CN212873683U CN 212873683 U CN212873683 U CN 212873683U CN 202021346493 U CN202021346493 U CN 202021346493U CN 212873683 U CN212873683 U CN 212873683U
- Authority
- CN
- China
- Prior art keywords
- image
- module
- monitoring
- speed railway
- railway safety
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Closed-Circuit Television Systems (AREA)
- Image Analysis (AREA)
Abstract
The utility model relates to a high-speed railway safety zone intrusion alarm device based on infrared images, which comprises a data acquisition module, a ground analysis module, an early warning module, a remote communication module, a power supply module and a terminal monitoring platform; the data acquisition module is used for acquiring a monitoring image of a high-speed railway safety area in real time; the ground analysis module is used for analyzing and processing the acquired monitoring image; the early warning module is used for sending out a warning signal; the remote communication module is used for transmitting the identification result and the monitoring image; the power module is used for supplying power to the whole set of device; and the terminal monitoring platform is used for the staff to check the identification result and the field image. The utility model discloses can realize that intrusion alert device independently carries out image processing, shape of a person discernment, result feedback, avoid the adverse effect that factors such as illumination condition, weather condition produced intrusion alert device, improve intrusion alert device's work efficiency and rate of accuracy.
Description
Technical Field
The utility model belongs to the technical field of high-speed railway safety zone state monitoring, a high-speed railway safety zone intrusion alert device based on infrared image is related to.
Background
The general target of the 'thirteen-five' railway planning is that the railway mileage reaches 15 kilometers in 2020, the high-speed railway reaches 3 kilometers, the total railway scale is increased by 2.9 kilometers, wherein the high-speed railway is increased by 1.1 kilometer, and the investment of the 'thirteen-five' railway is about to reach 3.8 trillion yuan. Along with the development of the high-speed railway in China, the mileage of the high-speed railway is longer and longer, and the economic development and the efficiency of people in various places are greatly promoted. But also bring a lot of potential safety hazards, railway facilities are damaged, and the situation that the protective net directly passes through a railway line is ignored, so that the railway line protective net has great damage to railway safety production, economic loss is caused, and personal injury is also caused. If the intrusion alarm device is installed in the railway safety area, the dangerous behaviors that pedestrians pass through the high-speed railway line due to the fact that the pedestrians cannot see the railway safety area protective net or safely warn can be avoided to the greatest extent, and accidents that endanger personal safety are avoided.
In order to avoid the phenomenon that pedestrians invade a railway safety area, an invasion monitoring range of a high-speed railway safety area invasion alarm system firstly covers the range meeting the safety production requirement, and railway management departments mainly adopt an electronic fence type invasion monitoring device to cover the whole safety area to solve the problem, wherein the electronic fence type invasion monitoring device mainly comprises a pulse electronic fence, an infrared detection beam net, a tension type electronic fence, a vibration monitoring type electronic fence, an optical fiber type electronic fence and the like. Although these electronic fence type intrusion monitoring devices perform intrusion monitoring based on different principles, essentially, the intrusion behavior of pedestrians is determined by monitoring the changes of photoelectric signals generated by factors such as stress, vibration, sound and the like when the pedestrians pass through the electronic fence in the intrusion process. However, in order to cover the monitoring range of the whole safety area, the electronic fence must adopt a plurality of monitoring devices to form a monitoring system, which increases the cost of equipment investment in a certain sense; meanwhile, as the environment of the high-speed railway line is complex and variable, the electronic fence formed by a plurality of monitoring devices also needs to spend excessive time and labor in equipment maintenance.
Another important problem to be solved by the high-speed railway safety area intrusion alarm system is accurate human-shaped target recognition of an intruding object. The electronic fence type intrusion monitoring device indirectly judges the intrusion behavior of the safety area by monitoring the change of the photoelectric signal of the internal sensor, but the electronic fence type intrusion monitoring device cannot effectively identify whether an intruding object is a human, and meanwhile, a high-speed train running on a high-speed railway can bring huge noise, vibration, air pressure difference and other interference factors, so that the identification of the electronic fence type intrusion monitoring device is influenced. In order to solve the human-shaped target recognition problem of an invading object, the railway department utilizes a high-definition optical camera to monitor a safe area in real time, and utilizes an image recognition algorithm to carry out human-shaped classification recognition on a moving target in a monitored image. However, the video monitoring device for monitoring the safety area in real time is generally based on a visible light imaging sensor, so that the video monitoring device cannot obtain a clearly imaged monitoring image under the conditions of poor illumination effect, poor weather condition and the like, and the identification accuracy of the intrusion alarm system is greatly reduced. On the other hand, in order to complete the image recognition process of the monitoring picture of the high-speed railway safety area, the monitoring device must transmit the monitoring data to a terminal processing platform of a designated station monitoring center after acquiring the monitoring image of the high-speed railway safety area, and the condition of transmission data loss or delay may occur in the transmission process, so that more time is required for intrusion monitoring.
SUMMERY OF THE UTILITY MODEL
The utility model aims at solving the above-mentioned problem that exists among the prior art, provide a high-speed railway safe area intrusion alert device based on infrared image.
In order to achieve the above purpose, the utility model adopts the following technical scheme:
an intrusion alarm device for a high-speed railway safety area based on infrared images comprises a data acquisition module, a ground analysis module, an early warning module, a power supply module, a remote communication module and a terminal monitoring platform;
the data acquisition module is connected with the ground analysis module through a data transmission line, so that the monitoring image acquired by the data acquisition module is transmitted to the ground analysis module in real time; the ground analysis module is used for analyzing and processing the acquired monitoring image; the early warning module is connected with the ground analysis module through a data transmission line and carries out early warning according to an analysis processing result of the ground analysis module; the remote communication module is connected with the ground analysis module through a data transmission line, and is used for storing the monitoring image and the analysis processing result (human shape recognition result) of the ground analysis module and transmitting the monitoring image and the analysis processing result to the terminal monitoring platform; the power supply module is used for supplying power to the data acquisition module, the ground analysis module, the early warning module and the remote communication module, and the specific connection mode is not limited as long as the function can be realized;
the data acquisition module comprises multispectral security monitoring equipment which is arranged on a tower along the high-speed railway;
the multispectral security monitoring equipment consists of a high-definition visible light camera and a thermal infrared imager, and the high-definition visible light camera and the thermal infrared imager are provided with corresponding local area network ports so as to be convenient for transmitting monitoring image data with the ground analysis module; the data acquisition module simultaneously acquires visible light monitoring images and infrared monitoring images of the same high-speed railway safety area by using a high-definition visible light camera and a thermal infrared imager;
the ground analysis module comprises a hardware layer, an operating system, a bottom driver and an image processing software layer, wherein the operating system, the bottom driver and the image processing software layer are stored in the hardware layer; the image processing software layer consists of an image preprocessing unit, an image fusion unit, a foreground target extraction unit and a human-shaped target identification unit; the hardware layer consists of a CPU (central processing unit) processor (used for executing a corresponding image processing algorithm), a memory, a mainboard, a power supply input port, a local area network port and an input/output interface;
the image preprocessing unit is used for denoising the infrared monitoring image and the visible light monitoring image of the high-speed railway safety area to remove or reduce noise points and clutter in the image and improve the quality of the monitoring image of the high-speed railway safety area;
the image fusion unit is used for carrying out weighted-based image fusion processing on the denoised infrared monitoring image and the denoised visible light monitoring image to obtain a fused monitoring image of the safety area of the high-speed railway;
the foreground target extraction unit is used for carrying out background segmentation on the fused high-speed railway safety area monitoring image, accurately extracting a moving target in the fused high-speed railway safety area monitoring image and realizing the separation of a foreground and a background;
the human-shaped target recognition unit is used for extracting human-shaped features of the extracted moving target and recognizing whether the moving target is a human-shaped target or not by taking the extracted human-shaped features as the basis.
The infrared thermal imager in the data acquisition module performs imaging display by the temperature difference or radiation difference between a detected target and a background, so that the infrared image has more obvious moving target information no matter under severe weather conditions such as day, night, haze, rainstorm and the like, but the edge details are lacked; a high-definition visible light camera in the data acquisition module performs imaging by utilizing the reflection capability difference of the surfaces of different objects, so that a visible light image has rich texture and color information, the performance capability of visible light on all objects in a monitored scene can be weakened under the influence of external conditions such as illumination, weather and the like, after an image fusion unit performs image fusion on an infrared image and the visible light image, the insensitivity of an infrared imaging technology to an external environment and the good detail information of the visible light image in a night and low-illumination space are fully utilized, and the two characteristics are complemented to obtain the monitored image with clear moving target contour characteristics.
As a preferred technical scheme:
according to the high-speed railway safety zone intrusion alarm device based on the infrared images, the data acquisition module and the ground analysis module are connected with respective local area network ports through data transmission lines, so that the monitoring images acquired by the data acquisition module are transmitted to the ground analysis module in real time.
According to the high-speed railway safety zone intrusion alarm device based on the infrared images, the early warning module is composed of a buzzer, an indicator light, a power input port and an input/output interface.
The high-speed railway safety zone intrusion alarm device based on the infrared image comprises a remote communication module, a wireless network transmission unit, a memory and an input/output interface.
According to the high-speed railway safety zone intrusion alarm device based on the infrared images, the monitoring images and the analysis processing results of the ground analysis module are stored in the memory of the remote communication module, and the monitoring images and the analysis processing results in the memory are transmitted to the terminal monitoring platform by the wireless network transmission unit.
According to the high-speed railway safety zone intrusion alarm device based on the infrared images, the terminal monitoring platform is a PC (personal computer) located in a high-speed railway station monitoring center, and can receive the identification result and the field monitoring image transmitted by the remote communication module in real time.
According to the high-speed railway safety zone intrusion alarm device based on the infrared images, the power module is a lithium battery.
According to the high-speed railway safety area intrusion alarm device based on the infrared image, the image preprocessing unit carries out denoising on the monitored image by using a median filtering method, the method slides the determined filtering window in the collected monitored image, the center of the filtering window is overlapped with the position of the current pixel point in the monitored image, the gray values of pixels contained in the adjacent area of the window are arranged in an ascending order, the gray value of the current pixel point is replaced by the ordered intermediate value, and the result that the noise point and the clutter generated in the collecting process of the data collecting module can be effectively inhibited is achieved; the median filtering method comprises the following specific steps: z is a radical of1，z2，z3…zkFor the gray values of the pixels in the filter window, R (z) is used1，z2，z3…zk) Representing the result function of median filtering of the pixels in the window, k being the number of pixels in the filtering window, R (z)1，z2，z3…zk)＝Med(z1，z2，z3…zk)，Med(z1，z2，z3…zk) Denotes z1，z2，z3…zkSorting the median values according to the sizes, and selecting a 3 multiplied by 3 filtering window from the filtering window; the image preprocessing unit calculates the gray average value of all pixel points in a 3 multiplied by 3 filtering window, the gray average value of the pixel points in the horizontal and vertical directions of the pixel point to be processed (the central pixel point of the filtering window) and the gray average value of the pixel point on the major-minor diagonal line of the filtering window, selects the maximum value and the minimum value of the 5 average values as the maximum threshold value and the minimum threshold value, and if the gray value of the pixel point to be processed is not in the range of the maximum threshold value and the minimum threshold value, the pixel point to be processed is judged as a noise point and is replaced by the gray median value of the pixel point in the. Whether the pixel to be processed is noise is judged by selecting a self-adaptive threshold range for the filtering window, and the loss of edges and details of the monitoring image of the high-speed railway safety area caused by a median filtering algorithm is avoided.
According to the high-speed railway safety zone intrusion alarm device based on the infrared image, the infrared monitoring image and the visible light monitoring image which are denoised by the image preprocessing unit are respectively recorded as a source image A and a source image B, and the image fusion unit carries out fusion processing on the source image A and the source image B by using an image fusion algorithm based on weighted fusion;
for the weighted fusion of the infrared image and the visible light image in the safety zone of the high-speed railway, the weighted fusion based on the pixel values of the two source images is not directly adopted, but the high-frequency coefficient and the low-frequency coefficient of the source images are decomposed by utilizing wavelet change, the high-frequency coefficient and the low-frequency coefficient of the fused image are obtained by adopting different fusion rules for the high-frequency coefficient and the low-frequency coefficient of the source images, the fused image is obtained by utilizing inverse wavelet transformation, and the advantages of the two source images can be fully reserved in the fused image;
the image fusion algorithm based on weighted fusion is as follows:
step 1: respectively carrying out wavelet transformation on the source image A and the source image B to obtain a low-frequency coefficient C of the source image A after the wavelet transformationAAnd a high frequency coefficient FAAnd low frequency coefficient C of source image BBAnd a high frequency coefficient FB；
Step 2: low frequency coefficient C of source image AAAnd low frequency coefficient C of source image BBObtaining a low-frequency coefficient C of the fused image by adopting a low-frequency coefficient fusion rule based on weightingF(ii) a The low-frequency coefficient obtained by wavelet transformation of the source image retains the basic general picture information of the original image, and the low-frequency coefficient C of the source image A is subjected toAAnd low frequency coefficient C of source image BBGiving corresponding weight to obtain low-frequency coefficient C of fused imageFThe low-frequency coefficient of the fused image can fully retain effective basic summary information in the two source images;
and step 3: high frequency coefficient F to source image AAAnd the high frequency coefficient F of the source image BBObtaining a high-frequency coefficient F of the fused image by adopting a high-frequency coefficient fusion rule based on the maximum high-frequency coefficientF(ii) a The high-frequency coefficient obtained by wavelet transformation of the source image reflects the edge and boundary information of the source image and the detail information of the brightness abrupt change part of the pixel, and the high-frequency coefficients of the source image A and the source image B are acquiredObtaining the high-frequency coefficient F of the fused image by using the high-frequency coefficient fusion rule based on the maximum high-frequency coefficientFThe high-frequency coefficient of the fused image can fully retain the image edge characteristics of the source image;
and 4, step 4: low frequency coefficient C of fused imageFAnd a high frequency coefficient FFAnd performing inverse wavelet transformation to obtain a fused monitoring image of the high-speed railway safety area.
In the high-speed railway safety zone intrusion alarm device based on the infrared image, the low-frequency coefficient C of the source image A in the step 2ALow frequency coefficient C with source image BBThe low-frequency coefficient fusion rule based on weighting is as follows:
wherein the low frequency coefficient CAAnd a low frequency coefficient CBWeight value of alpha1、α2Comprises the following steps:
high-frequency coefficient F of source image A in step 3AHigh frequency coefficient F with source image BBThe high-frequency coefficient fusion rule based on the maximum high-frequency coefficient is as follows:
in the infrared image-based high-speed railway security zone intrusion alarm device, the foreground object extraction unit adopts a background difference method to realize separation of a background and a foreground in the fused monitoring image, and further extracts a moving object foreground in the monitoring image; the background subtraction method comprises the following specific steps:
step 1: acquiring a current frame image; acquiring 50 frames of images before the current frame for background construction, and gray at the same coordinate of the 50 frames of imagesAccumulating the values and calculating the average value, using the calculated average value as the background gray value,
RG (x, y) represents a background gray value at coordinates (x, y), I (x, y, I) represents a gray value at coordinates (x, y) of an ith frame image in a 50 frame image before the current frame;
step 2: subtracting the gray value of the current frame and the constructed background model at the same coordinate to obtain a gray value difference | BG (x, y) -I (x, y) |, wherein I (x, y) represents the gray value of the current frame image at the coordinate (x, y);
and step 3: constructing an adaptive threshold T by a maximum inter-class variance method, wherein pixel points with the gray value difference larger than the adaptive threshold T are foreground points, pixel points with the gray value difference smaller than the adaptive threshold T are background points, and the gray value difference is | BG (x, y) -I (x, y) |; the self-adaptive threshold T can be automatically adjusted according to the specific condition of the current frame, so that the condition that the manually set threshold cannot achieve the best segmentation effect is avoided; the self-adaptive threshold value T is a certain gray value of a differential image of the current frame and the background model, and the differential image is divided into two types C by taking the gray value as a boundary00 to T and C1- { T to L-1} (L is a differential image gray level), C is calculated0Probability of (2)
Mean value of
And C1Probability w of1-1-w0Average value of
(piIs the probability of occurrence of a grey value i, pi-ni/N，niThe number of pixels with the gray value of i and the total number of pixel points of the differential image N); the adaptive threshold T is such that the between-class variance 8 is w0w1(u0-u1)2Reaches a maximum value, and when the threshold value is T, the two parts C of the image are differentiated0And C1The gray value difference is maximum, and the segmentation effect of the foreground and the background of the monitored image is best;
and 4, step 4: and binarizing the foreground points and the background points, and separating the moving target and the background in the fused monitoring image.
According to the high-speed railway safety zone intrusion alarm device based on the infrared image, for the contour region of the moving object extracted by the foreground object extraction unit, the contour edge of the moving object can well show the human-shaped feature of the moving object, and the pixel points at the contour edge (local) of the moving object have sudden change in the gradient size and direction, so that the gradient direction histogram (HOG feature) of the monitoring image can be used as the human-shaped feature description of the moving object; for the human-shaped feature description of the moving target in the monitored image, the calculation formula of the gradient distribution (gradient direction and size) of each pixel point of the monitored image is as follows:
gradient in horizontal direction: gx(x，y)＝[-1 0 1]*I(x，y)；
Gradient in vertical direction: gy(x，y)＝[-1 0 1]T*I(x，y)；
Gradient size of pixel point at image (x, y) position:
the gradient direction of the pixel point at the image (x, y) position is:
in the formula, F (x, y) represents the gray value of a pixel point at the position of the image (x, y);
the specific steps of the human-shaped target recognition unit for extracting the HOG characteristic value of the monitoring image are as follows:
step 1: correcting the image by using a gamma algorithm, and reducing the influence of illumination and noise on feature extraction, wherein F (x, y) is F (x, y)gammaThe gamma value is 1/2;
step 2: calculating the gradient direction and the size of each pixel point in the monitoring image;
and step 3: dividing an image into cell blocks (cells) with equal sizes, wherein the pixel size of each cell is gamma multiplied by gamma (gamma depends on the pixel size delta multiplied by beta of the fused image, the gamma can be divided by the delta and the beta, and the gamma is smaller than 10), and counting the gradient size and the gradient direction of all pixel points in the cell into a gradient direction histogram;
and 4, step 4: histogram normalization of the direction of cell gradient was performed in blocks (each block consisting of 2 × 2 cells), the normalization function being an L2-norm function:
v*the normalization result of the histogram of gradient direction is shown, v is the histogram of gradient direction of the cell, epsilon is a very small constant, and the denominator is avoided
Is 0;
and 5: connecting the gradient direction histograms of all the blocks in series to generate HOG characteristics of the monitoring image as human shape characteristic description of the moving target;
the human-shaped target recognition unit forms an effective human-shaped recognition classifier by utilizing an AdaBoost algorithm to recognize the human shape of the moving target; the human-shaped target recognition unit inputs the extracted human-shaped feature description of the moving target into a human-shaped recognition classifier, and a discrimination function in the human-shaped recognition classifier carries out classification decision calculation on the moving target to finally recognize whether the moving target is a human-shaped target; considering the problems of different human figures and postures and more types of human figures and non-human figures existing in the distinction, in order to improve the identification accuracy of a discrimination function in an identification classifier, a human figure target unit carries out multi-sample iterative training by utilizing a machine learning algorithm (AdaBoost algorithm) to obtain a human figure identification classifier; the training process of the AdaBoost algorithm is as follows:
step 1: selecting an existing infrared human-shaped Database OTCBVS OSU Thermal Database in the field of video monitoring as a training data set, manually classifying training samples, and classifying positive and negative samples in the training data set; wherein the training data set S { (x)1，y1)，(x2，y2)，…(xn，yn)}，xiFeature vector (HOG feature), y, for each training sample in the training dataseti∈{-1，+1}；
Step 2: during the first training, all training samples are given the same distribution weight D1＝(w11，w12，…w1n) Wherein
D1The distribution weight of all training samples in the 1 st training process is obtained;
and step 3: performing a multi-sample iterative training process, wherein the training process is as follows:
(a) using a weight distribution DmLearning the training data set to obtain a basic classifier Gm(x)；
Gm(x)：x→{-1，+1}；Dm＝(wm1，wm2，…wmm)；
Gm(x) Is the basic classifier in the mth training process, x is the input value of the basic classifier, and the input value x of the basic classifier in the training process is the feature vector x of the training samples in the training data seti；DmThe distribution weight of all training samples in the mth training process;
(b) calculation of Gm(x) Classification error rate on training set em；
I(Gm(xi)≠yi) Represents: when G ism(xi) And yiWhen equal, the value is 0; when G ism(xi) And yiWhen the values are not equal, the value is 1; x is the number ofiFor each training sample's feature vector, y, in the training dataseti∈{-1，+1}；wmiRepresenting the distribution weight of the ith training sample in the mth training process;
(c) computing basic classifier Gm(x) Coefficient alpha in human shape recognition classifier G (x) in the human shape target recognition unitm；
(d) Updating the weight distribution of the training data set for the next iteration;
Dm+1＝(wm+1，1，wm+1，2，…wm+1，n)；
wherein
wm+1，iRepresents the distribution weight of the ith training sample in the (m + 1) th training process, Dm+1The distribution weight of all training samples in the (m + 1) th training process is calculated;
(e)emwhen the value is less than 0.5, ending the iterative training process;
and 4, step 4: combining all basic classifiers in the training process in the multi-sample iterative training process to form a human shape recognition classifier in the human shape target recognition unit as follows:
basic classifier G in the mth training processm(x) Is constructed as follows:
wherein H is all positive sample feature vectors x 'of the training data set'i(HOG feature) of the mean vector,
p is the total number of all positive samples in the training dataset, xiA feature vector (HOG feature) for each training sample in the training dataset; d (x)iH) is xiAnd H; thetamAs a basic classifier Gm(x) Is measured.
A high speed railway safety area intrusion alert method adapted to the high speed railway safety area intrusion alert apparatus based on infrared images as described above, comprising the steps of:
(1) the data acquisition module acquires an infrared monitoring image and a visible light monitoring image of a high-speed railway safety area in real time by using multispectral security monitoring equipment;
(2) transmitting the monitoring image acquired by the data acquisition module to a ground analysis module in real time;
(3) the image preprocessing unit is used for preprocessing and denoising the infrared monitoring image and the visible light monitoring image;
(4) the image fusion unit carries out image fusion processing on the infrared monitoring image and the visible light image which are subjected to preprocessing and denoising to obtain a fused monitoring image;
(5) the foreground object extraction unit extracts a moving object in the fused monitoring image;
(6) the human-shaped target recognition unit extracts human-shaped features of the moving target and recognizes the human-shaped target of the moving target according to the extracted human-shaped features;
(7) if the ground analysis module judges that the moving target in the monitoring image of the high-speed railway safety area is a human-shaped target, the early warning module warns an intruder according to the recognition result;
(8) the remote communication module sends the identification result and the monitoring image to the terminal monitoring platform in time, and the staff takes necessary measures according to the identification result and the monitoring image received by the terminal monitoring platform.
Has the advantages that:
compared with the prior art, the beneficial effects of the utility model reside in that:
(1) through data acquisition device's thermal infrared imager, the utility model discloses can effectively adopt the thermal infrared imaging technique to realize all-weather video monitoring. The infrared thermal imaging technology is characterized in that a thermal image picture is generated according to infrared radiation emitted by a scene, and the state information of a moving target in the scene can be still acquired under severe weather conditions such as heavy rain, strong sunlight, strong wind, haze, snowfall and the like or under the condition that a large number of objects are shielded on site, so that the accuracy of human-shaped target identification of the moving object is improved;
(2) considering that the infrared pattern can produce the weak, motion target detail of space correlation because of factors such as moving object self radiation characteristic, transmission distance, atmospheric attenuation, contrast low grade shortcoming, the utility model discloses rely on the visible light image that high definition visible light camera generated and place in the image fusion unit of ground analysis module in, through fusing infrared and visible light monitoring image, the image after the fusion can increase the definition of image, reflect the whole information of target and background comprehensively, improve the stability and the reliability of image, improve intrusion monitoring device's monitoring accuracy;
(3) the utility model discloses utilize ground analysis module can independently accomplish separation, the humanoid characteristic of moving target that monitoring image preliminary treatment, monitoring image fuse, background and moving target prospect draw and a series of steps of humanoid target recognition, improved the efficiency and the accuracy of invasion monitoring, reduced staff's working strength. In addition, the staff can transplant different image processing algorithms to the application software layer in the ground analysis module, so that the utility model can monitor the invasion of various objects such as animals and vehicles;
(4) before whole invasion alarm device operation, the utility model provides a ground analysis module has just built-in hardware layer, operating system, bottom driver and corresponding image processing software layer in advance, and this makes the invasion alarm device can independently carry out the invasion monitoring step through ground analysis module, need not to carry out relevant algorithm step through station yard surveillance center's terminal monitoring platform, has improved invasion monitoring efficiency.
Drawings
Fig. 1 is a schematic layout diagram of an intrusion alarm device for a high-speed railway safety area based on infrared images according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for alarming intrusion in a safe area of a high speed railway based on infrared images according to an embodiment of the present invention;
FIG. 3 is a flowchart of a multi-sample iterative training algorithm of a machine learning algorithm used by a human shape recognition unit in an intrusion alarm device for a high speed railway safety zone based on infrared images according to an embodiment of the present invention;
FIG. 4 is an image processing platform system of a ground analysis module in the intrusion alarm device for high-speed railway safety area based on infrared image according to the embodiment of the present invention;
the system comprises a data acquisition module, a ground analysis module, a warning module, a power supply module and a remote communication module, wherein the data acquisition module is 1, the ground analysis module is 2, the warning module is 3, and the remote communication module is 5.
Detailed Description
The present invention will be further described with reference to the following detailed description. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Furthermore, it should be understood that various changes and modifications of the present invention may be made by those skilled in the art after reading the teachings of the present invention, and these equivalents also fall within the scope of the appended claims.
An intrusion alarm device for a high-speed railway safety zone based on infrared images is shown in figure 1 and comprises a data acquisition module 1, a ground analysis module 2, an early warning module 3, a power supply module 4, a remote communication module 5 and a terminal monitoring platform;
the data acquisition module 1 is connected with the ground analysis module 2 through a data transmission line, so that the monitoring image acquired by the data acquisition module 1 is transmitted to the ground analysis module 2 in real time; the ground analysis module 2 is used for analyzing and processing the collected monitoring image; the early warning module 3 is connected with the ground analysis module 2 through a data transmission line and carries out early warning according to the analysis processing result of the ground analysis module 2; the remote communication module 5 is connected with the ground analysis module 2 through a data transmission line, and the remote communication module 5 is used for storing the monitoring image and the analysis processing result (human shape recognition result) of the ground analysis module 2 and transmitting the monitoring image and the analysis processing result to the terminal monitoring platform; the power supply module 4 is used for supplying power to the data acquisition module 1, the ground analysis module 2, the early warning module 3 and the remote communication module 5;
the data acquisition module 1 comprises multispectral security monitoring equipment which is installed on towers along the high-speed railway and consists of a high-definition visible light camera and a thermal infrared imager, the high-definition visible light camera and the thermal infrared imager are provided with corresponding local area network ports, and the data acquisition module 1 is used for simultaneously acquiring visible light monitoring images and infrared monitoring images of the same high-speed railway safety area by using the high-definition visible light camera and the thermal infrared imager; as shown in fig. 4, the ground analysis module 2 includes a hardware layer, an operating system, a bottom driver, and an image processing software layer, where the operating system, the bottom driver, and the image processing software layer are stored in the hardware layer; the image processing software layer consists of an image preprocessing unit, an image fusion unit, a foreground target extraction unit and a human-shaped target identification unit; the hardware layer consists of a CPU processor, a memory, a storage, a mainboard, a power input port, a local area network port and an input/output interface; the early warning module 3 consists of a buzzer, an indicator light, a power input port and an input/output interface; the remote communication module 5 consists of a wireless network transmission unit, a memory and an input/output interface, the monitoring image and the analysis processing result of the ground analysis module 2 are stored in the memory of the remote communication module 5, and the monitoring image and the analysis processing result in the memory are transmitted to the terminal monitoring platform by utilizing the wireless network transmission unit; the power module 4 is a lithium battery; the terminal monitoring platform is a PC (personal computer) machine positioned in a monitoring center of a high-speed railway station yard;
the image preprocessing unit is used for denoising the monitoring image of the high-speed railway safety area, specifically, a median filtering method is used for denoising the monitoring image, the method slides a determined filtering window in the collected monitoring image, the center of the filtering window is overlapped with the position of a current pixel point in the monitoring image, gray values of pixels contained in the window neighborhood are arranged in an ascending order, the sorted middle value replaces the gray value of the current pixel point, and the purpose of effectively inhibiting the generation of the gray value of the current pixel point in the collection process of the data collection module is achievedNoise and clutter results; the median filtering method comprises the following specific steps: z is a radical of1，z2，z3…zkFor the gray values of the pixels in the filter window, R (z) is used1，z2，z3…zk) Representing the result function of median filtering of the pixels in the window, k being the number of pixels in the filtering window, R (z)1，z2，z3…zk)＝Med(z1，z2，z3…zk)，Med(z1，z2，z3…zk) Denotes z1，z2，z3…zkSorting the median values according to the sizes, and selecting a 3 multiplied by 3 filtering window from the filtering window; the image preprocessing unit calculates the gray average value of all pixel points in a 3 multiplied by 3 filtering window, the gray average value of the pixel points in the horizontal and vertical directions of the pixel point to be processed (the central pixel point of the filtering window) and the gray average value of the pixel point on the major-minor diagonal line of the filtering window, selects the maximum value and the minimum value of the 5 average values as the maximum threshold value and the minimum threshold value, and if the gray value of the pixel point to be processed is not in the range of the maximum threshold value and the minimum threshold value, the pixel point to be processed is judged as a noise point and is replaced by the gray median value of the pixel point in the. Whether a pixel to be processed is noise is judged by selecting a self-adaptive threshold range for a filtering window, so that the loss of edges and details of a monitoring image of a high-speed railway safety area caused by a median filtering algorithm is avoided;
the method comprises the following steps of recording an infrared monitoring image and a visible light monitoring image which are subjected to preprocessing and denoising as a source image A and a source image B respectively, carrying out fusion processing on the source image A and the source image B by using an image fusion algorithm based on weighted fusion to obtain a fused high-speed railway safety area monitoring image, wherein the image fusion algorithm based on weighted fusion is as follows:
step 1: respectively carrying out wavelet transformation on the source image A and the source image B to obtain a low-frequency coefficient C of the source image A after the wavelet transformationAHigh frequency coefficient FAAnd low frequency coefficient C of source image BBHigh frequency coefficient FB；
Step 2: to the sourceLow frequency coefficient C of image aAAnd low frequency coefficient C of source image BBObtaining a low-frequency coefficient C of the fused image by adopting a low-frequency coefficient fusion rule based on weightingF(ii) a The low-frequency coefficient fusion rule based on weighting of the source image A and the source image B is as follows:
wherein the low frequency coefficient CAAnd a low frequency coefficient CBWeight of a1、α2Is composed of
And step 3: high frequency coefficient F to source image AAAnd the high frequency coefficient F of the source image BBObtaining a high-frequency coefficient F of the fused image by adopting a high-frequency coefficient fusion rule based on the maximum high-frequency coefficientF(ii) a The high-frequency coefficient fusion rule based on the maximum high-frequency coefficient of the source image A and the source image B is as follows:
and 4, step 4: low frequency coefficient C of fused imageFAnd a high frequency coefficient FFPerforming inverse wavelet transform to obtain a fused monitoring image of the high-speed railway safety area;
the foreground object extraction unit is used for performing background segmentation on the monitoring image subjected to image fusion processing, accurately extracting a moving object in the monitoring image of the high-speed railway safety area, and realizing the separation of the foreground from the background, and particularly, a background difference method is adopted to quickly realize the separation of the background from the foreground in the fused monitoring image, wherein the background difference method specifically comprises the following steps:
step 1: acquiring a current frame image; acquiring 50 frames of images before the current frame for background construction, and gray at the same coordinate of the 50 frames of imagesAccumulating the values and calculating the average value, using the calculated average value as the background gray value,
RG (x, y) represents a background gradation value at coordinates (x, y), J (x, y, i) represents a gradation value at coordinates (x, y) of an ith frame image in 50 frame images before the current frame;
step 2: subtracting the gray value of the current frame and the constructed background model at the same coordinate to obtain a gray value difference | RG (x, y) -I (x, y) |, wherein I (x, y) represents the gray value of the current frame image at the coordinate (x, y);
and step 3: constructing an adaptive threshold T by a maximum inter-class variance method, wherein pixel points with the gray value difference larger than the adaptive threshold T are foreground points, pixel points with the gray value difference smaller than the adaptive threshold T are background points, and the gray value difference is | RG (x, y) -I (x, y) |; the self-adaptive threshold T can be automatically adjusted according to the specific condition of the current frame, so that the condition that the manually set threshold cannot achieve the best segmentation effect is avoided; the self-adaptive threshold value T is a certain gray value of a differential image of the current frame and the background model, and the differential image is divided into two types C by taking the gray value as a boundary00 to T and C1- { T to L-1} (L is a differential image gray level), C is calculated0Probability of (2)
Mean value of
And C1Probability w of1-1-w0Average value of
(piIs the probability of occurrence of a grey value i, pi-ni/N，niThe number of pixels with the gray value of i and the total number of pixel points of the differential image N); the adaptive threshold T is such that the inter-class variance g is w0w1(u0-u1)2Reaches a maximum value, and when the threshold value is T, the two parts C of the image are differentiated0And C1The gray value difference is maximum, and the segmentation effect of the foreground and the background of the monitored image is best;
and 4, step 4: binarizing foreground points and background points, and separating a moving target and a background of the fused monitoring image;
for the contour region of the moving target extracted by the foreground target extraction unit, the contour edge of the moving target can well show the human-shaped feature of the moving target, and the pixel points at the contour edge (local) of the moving target can be suddenly changed in the size and direction of the gradient, so that the gradient direction histogram (HOG feature) of the monitoring image can be used as the human-shaped feature description of the moving target; for the human-shaped feature description of the moving target in the monitored image, the calculation formula of the gradient distribution (gradient direction and size) of each pixel point of the monitored image is as follows:
gradient in horizontal direction: gx(x，y)＝[-1 0 1]*I(x，y)；
Gradient in vertical direction: gy(x，y)＝[-1 0 1]T*I(x，y)；
Gradient size of pixel point at image (x, y) position:
the gradient direction of the pixel point at the image (x, y) position is:
in the formula, I (x, y) represents the gray value of a pixel point at the position of the image (x, y);
the specific steps of the human-shaped target recognition unit for extracting the HOG characteristic value of the monitoring image are as follows:
step 1: correcting the image by using a gamma algorithm, and reducing the influence of illumination and noise on feature extraction, wherein I (x, y) is I (x, y)gammaThe gamma value is 1/2;
step 2: calculating the gradient direction and the size of each pixel point in the monitoring image;
and step 3: dividing an image into cell blocks (cells) with equal sizes, wherein the pixel size of each cell is gamma multiplied by gamma (gamma depends on the pixel size delta multiplied by beta of the fused image, the gamma can be divided by the delta and the beta, and the gamma is smaller than 10), and counting the gradient size and the gradient direction of all pixel points in the cell into a gradient direction histogram;
and 4, step 4: histogram normalization of the direction of cell gradient was performed in blocks (each block consisting of 2 × 2 cells), the normalization function being an L2-norm function:
v*the normalization result of the histogram of gradient direction is shown, v is the histogram of gradient direction of the cell, epsilon is a very small constant, and the denominator is avoided
Is 0;
and 5: connecting the gradient direction histograms of all the blocks in series to generate HOG characteristics of the monitoring image as human shape characteristic description of the moving target;
the human-shaped target recognition unit forms a human-shaped recognition classifier by using an AdaBoost algorithm to recognize the human shape of the moving target, and as shown in FIG. 3, the training process of the AdaBoost algorithm is as follows:
step 1: selecting an existing infrared human-shaped Database OTCBVS OSU Thermal Database in the field of video monitoring as a training sample set, manually classifying the training samples, and classifying positive and negative samples in the training data set, wherein the training data set S { (x)1，y1)，(x2，y2)，…(xN，yN)}，xiFeature vector (HOG feature), y, for each training sample in the training dataseti∈{-1，+1}；
Step 2: during the first training, all training samples are given the same distribution weight D1＝(w11，w12，…w1n) Wherein
D1For all training samples in the 1 st training processThe distribution weight of the book;
and step 3: performing a multi-sample iterative training process, wherein the training process is as follows:
(a) using a weight distribution DmLearning the training data set to obtain a basic classifier Gm(x)；
Gm(x)：x→{-1，+1}；Dm＝(wm1，wm2，…wmm)；
Gm(x) Is the basic classifier in the mth training process, x is the input value of the basic classifier, and the input value x of the basic classifier in the training process is the feature vector x of the training samples in the training data seti；DmThe distribution weight of all training samples in the mth training process;
(b) calculation of Gm(x) Classification error rate on training set em；
Wherein, I (G)m(xi)≠yi) Represents: when G ism(xi) And yiWhen equal, the value is 0; when G ism(xi) And yiWhen the values are not equal, the value is 1; x is the number ofiFor each training sample's feature vector, y, in the training dataseti{-1，+1}；wmiRepresenting the distribution weight of the ith training sample in the mth training process;
(c) computing basic classifier Gm(x) Coefficient alpha in human shape recognition classifier G (x) in the human shape target recognition unitm；
(d) Updating the weight distribution of the training data set for the next iteration;
Dm+1＝(wm+1，1，wm+1，2，…wm+1，n)；
wherein
wm+1，iRepresents the distribution weight of the ith training sample in the (m + 1) th training process, Dm+1The distribution weight of all training samples in the (m + 1) th training process is calculated;
(e)emwhen the value is less than 0.5, ending the iterative training process;
and 4, step 4: combining all basic classifiers in the training process in the multi-sample iterative training process to form a human shape recognition classifier in the human shape target recognition unit
Basic classifier G in the mth training processm(x) Is constructed as follows:
wherein H is all positive sample feature vectors x 'of the training data set'i(HOG feature) of the mean vector,
p is the total number of all positive samples in the training dataset, xiA feature vector (HOG feature) for each training sample in the training dataset; d (x)iH) is xiAnd H; thetamAs a basic classifier Gm(x) Is measured.
The intrusion alarm method for the high-speed railway safety area, which is suitable for the intrusion alarm device for the high-speed railway safety area based on the infrared image, as shown in fig. 2, comprises the following steps:
(1) the data acquisition module acquires an infrared monitoring image and a visible light monitoring image of a high-speed railway safety area in real time by using multispectral security monitoring equipment;
(2) transmitting the monitoring image acquired by the data acquisition module to a ground analysis module in real time;
(3) the image preprocessing unit is used for preprocessing and denoising the infrared monitoring image and the visible light monitoring image;
(4) the image fusion unit carries out image fusion processing on the infrared monitoring image and the visible light image which are subjected to preprocessing and denoising to obtain a fused monitoring image;
(5) the foreground object extraction unit extracts a moving object in the fused monitoring image;
(6) the human-shaped target recognition unit extracts human-shaped features of the moving target and recognizes the human-shaped target of the moving target according to the extracted human-shaped features;
(7) if the ground analysis module judges that the moving target in the monitoring image of the high-speed railway safety area is a human-shaped target, the early warning module warns an intruder according to the recognition result;
(8) the remote communication module sends the identification result and the monitoring image to the terminal monitoring platform in time, and the staff takes necessary measures according to the identification result and the monitoring image received by the terminal monitoring platform.
The utility model can realize that the all-weather 24-hour alarm device autonomously monitors the invasion of humanoid targets in the high-speed railway safety area, prompts and warns the invasion targets, and can timely send identification and on-site monitoring images to station monitoring personnel;
the utility model discloses utilize ground analysis module can independently accomplish separation, the humanoid characteristic of moving target that monitoring image preliminary treatment, monitoring image fuse, background and moving target prospect draw and a series of steps of humanoid target recognition, improved the efficiency and the accuracy of invasion monitoring, reduced staff's working strength. In addition, the working personnel can transplant different image processing algorithms to the application software layer in the ground analysis module, so that the high-speed railway intrusion alarm device provided by the utility model can monitor the intrusion of various objects such as animals and vehicles;
the utility model discloses a data acquisition module gathers infrared monitoring image and visible light image under the same time point in high-speed railway safe area to fuse two kinds of monitoring images through the image fusion algorithm, the image after the fusion is when the profile state information of moving target in having kept infrared image, has also kept the scene information at clear moving target place in the visible light image, has improved the rate of accuracy of intrusion monitoring.
Claims (7)
1. A high-speed railway safety zone intrusion alarm device based on infrared images is characterized in that: the system comprises a data acquisition module, a ground analysis module, an early warning module, a power supply module, a remote communication module and a terminal monitoring platform;
the data acquisition module is connected with the ground analysis module through a data transmission line, so that the monitoring image acquired by the data acquisition module is transmitted to the ground analysis module in real time; the ground analysis module is used for analyzing and processing the acquired monitoring image; the early warning module is connected with the ground analysis module through a data transmission line and carries out early warning according to the analysis processing result of the ground analysis module; the remote communication module is connected with the ground analysis module through a data transmission line and is used for storing the monitoring image and the analysis processing result of the ground analysis module and transmitting the monitoring image and the analysis processing result to the terminal monitoring platform; the power supply module is used for supplying power to the data acquisition module, the ground analysis module, the early warning module and the remote communication module;
the data acquisition module comprises multispectral security monitoring equipment which is installed on towers along the high-speed railway and consists of a high-definition visible light camera and a thermal infrared imager, the high-definition visible light camera and the thermal infrared imager are provided with corresponding local area network ports, and the data acquisition module acquires visible light monitoring images and infrared monitoring images of the same high-speed railway safety area simultaneously by using the high-definition visible light camera and the thermal infrared imager.
2. The high-speed railway safety zone intrusion alarm device based on the infrared image as claimed in claim 1, wherein the data acquisition module and the ground analysis module are connected with respective local area network ports through data transmission lines, so that the monitoring image acquired by the data acquisition module is transmitted to the ground analysis module in real time.
3. The high-speed railway safety zone intrusion alarm device based on the infrared image is characterized in that the early warning module consists of a buzzer, an indicator light, a power supply input port and an input/output interface.
4. The high-speed railway safety zone intrusion alarm device based on the infrared image is characterized in that the remote communication module consists of a wireless network transmission unit, a memory and an input/output interface.
5. The high-speed railway safety zone intrusion alarm device based on the infrared image is characterized in that the monitoring image and the analysis processing result of the ground analysis module are stored in the memory of the remote communication module, and the monitoring image and the analysis processing result in the memory are transmitted to the terminal monitoring platform by using a wireless network transmission unit.
6. The high-speed railway safety zone intrusion alarm device based on the infrared image as claimed in claim 1, wherein the terminal monitoring platform is a PC located in a monitoring center of a high-speed railway yard.
7. The high-speed railway safety zone intrusion alarm device based on the infrared image as claimed in claim 1, wherein the power module is a lithium battery.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202021346493.1U CN212873683U (en) | 2020-07-10 | 2020-07-10 | High-speed railway safety zone intrusion alarm device based on infrared image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202021346493.1U CN212873683U (en) | 2020-07-10 | 2020-07-10 | High-speed railway safety zone intrusion alarm device based on infrared image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN212873683U true CN212873683U (en) | 2021-04-02 |
Family
ID=75214061
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202021346493.1U Active CN212873683U (en) | 2020-07-10 | 2020-07-10 | High-speed railway safety zone intrusion alarm device based on infrared image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN212873683U (en) |
-
2020
- 2020-07-10 CN CN202021346493.1U patent/CN212873683U/en active Active
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111667655A (en) | Infrared image-based high-speed railway safety area intrusion alarm device and method | |
| CN105809679B (en) | Mountain railway side slope rockfall detection method based on visual analysis | |
| CN111738342B (en) | Pantograph foreign matter detection method, storage medium and computer equipment | |
| CN111260629A (en) | Pantograph structure abnormity detection algorithm based on image processing | |
| CN101715111B (en) | Method for automatically searching abandoned object in video monitoring | |
| CN110852179B (en) | Suspicious personnel invasion detection method based on video monitoring platform | |
| CN105574488A (en) | Low-altitude aerial infrared image based pedestrian detection method | |
| CN112183472A (en) | Method for detecting whether test field personnel wear work clothes or not based on improved RetinaNet | |
| CN113192038B (en) | Method for recognizing and monitoring abnormal smoke and fire in existing flame environment based on deep learning | |
| CN112699267B (en) | Vehicle type recognition method | |
| CN109492700A (en) | A kind of Target under Complicated Background recognition methods based on multidimensional information fusion | |
| CN113947555A (en) | Infrared and visible light fused visual system and method based on deep neural network | |
| CN110321855A (en) | A kind of greasy weather detection prior-warning device | |
| CN112861617A (en) | Slope disaster identification system based on monitoring image | |
| Kapoor et al. | Deep learning based object and railway track recognition using train mounted thermal imaging system | |
| CN115311241A (en) | Coal mine down-hole person detection method based on image fusion and feature enhancement | |
| CN114155472A (en) | Method, device and equipment for detecting abnormal state of factory scene empty face protection equipment | |
| CN112508893B (en) | Method and system for detecting tiny foreign matters between double rails of railway based on machine vision | |
| CN109815784A (en) | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium | |
| CN212873683U (en) | High-speed railway safety zone intrusion alarm device based on infrared image | |
| CN113066041A (en) | Pavement crack detection method based on stack sparse self-coding deep learning | |
| CN115410114A (en) | Urban rail flood prevention early warning method and system based on multiple characteristics | |
| CN115546738A (en) | Rail foreign matter detection method | |
| CN113223081A (en) | High-altitude parabolic detection method and system based on background modeling and deep learning | |
| CN113327245B (en) | Tunnel protection door state detection method based on image recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230512 Address after: Room 1507-59, No. 500 Caoyang Road, Putuo District, Shanghai, 200063 Patentee after: SHANGHAI RAILWAY INSTITUTE RAIL TRANSIT TECHNOLOGY CO.,LTD. Address before: 201620 No. 333, Longteng Road, Shanghai, Songjiang District Patentee before: SHANGHAI University OF ENGINEERING SCIENCE |
|
| TR01 | Transfer of patent right |