Deicing identification system based on visual servo

Abstract

The invention discloses a deicing recognition system based on visual servo, and the system comprises an image obtaining system which is used for collecting an image, and employs a computer to carry out the analysis of the collected image; an image preprocessing system; the target detection system is used for identifying the ice piton and judging the characteristics of the ice piton, the target detection system comprises an ice piton area positioning module, an ice piton image binarization module and an image threshold segmentation module, and the ice piton area positioning module is used for extracting an icing target area in the collected image. It can be ensured that deicing operation covers the whole tunnel, no dead angle is left, meanwhile, manual intervention is reduced through an integrated automatic moving and recognizing system, unmanned operation can be achieved under the severe weather condition or in the dangerous environment, and operation safety is improved. The deicing device can stably operate under the condition of low visibility or extreme weather, ensures the continuity and reliability of deicing operation, and reduces the risk and cost of manual operation at the same time.

Description

Technical Field

[0001] The invention relates to the technical field of tunnel deicing, and in particular to a deicing identification system based on visual servoing. Background Art

[0002] There are many railway tunnels in the transportation sector of China Energy Group. The inner walls of the tunnels are affected by factors such as terrain, temperature, and construction quality, and cracks appear, causing water to seep out of the cracks. When the temperature is low in winter, the leaking water accelerates condensation and easily forms ice cones. Some ice cones located above the contact network gradually become longer as the infiltrating water freezes, which can easily cause a short circuit in the contact network, interrupt railway transportation, and even cause casualties. Completely solving the problem of tunnel ice damage can ensure unimpeded railway transportation, which is of great significance.

[0003] In order to deal with the problem of tunnel icing, railway operating units have adopted a variety of solutions, among which the most common is manual deicing. This method requires deicing personnel to patrol the tunnel several times each time and use long poles and other tools to remove icicles in the tunnel. However, this deicing method requires a lot of manpower and time. When the railway traffic is large, the deicing operation time is limited and the safety is also low. In addition, the low temperature and high dust in the tunnel make the working environment of the workers very harsh. Moreover, there is generally a high-voltage contact network under the icicles, so there is also a risk of electric shock in manual deicing.

[0004] In recent years, high-power laser technology has experienced rapid development. In laboratory environments, the output power of semiconductor laser linear arrays and arrays has reached kilowatts, and their photoelectric conversion efficiency is as high as 60% to 70%. The maximum power of fiber lasers has also reached the kilowatt level. These research advances provide a realistic basis for the use of laser technology in tunnel deicing.

[0005] Lasers have the characteristics of good monochromaticity, high energy, good directionality, and high long-distance transmission efficiency. When using laser technology to de-ice transmission lines and substation equipment, there is no need to add additional equipment inside the power grid. In addition, lasers themselves are not conductive, so ice cones can be effectively removed without disconnecting the power supply to the contact network. This provides huge advantages and application potential for laser de-icing technology. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a deicing recognition system based on visual servoing.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A deicing identification system based on visual servoing includes an image acquisition system, wherein the image acquisition device is used to acquire images and analyze the acquired images using a computer;

[0009] An image preprocessing system, wherein the image preprocessing system is used to enhance and denoise images in dim light conditions;

[0010] A target detection system, the target detection system is used to identify ice cones and determine their characteristics. The target detection system includes an ice cone area positioning module, an ice cone image binarization module and an image threshold segmentation module. The ice cone area positioning module is used to extract the ice-covered target area in the collected image. The ice cone image binarization module uses the HSI color space to perform color threshold segmentation on the ice cone image and realizes ice recognition by extracting the HSI component value of the ice-covered target area. The image threshold segmentation module is used to segment the ice-covered target area and the background area in the image;

[0011] A binocular camera spatial positioning system is used to spatially locate the ice cone using a binocular camera after acquiring the ice cone area;

[0012] Laser triangulation ranging system, used to dynamically adjust the parameters, timing and action position of the output laser beam according to the working distance between the laser emission area and the ice cone, so as to increase the ice melting speed;

[0013] Laser obstacle removal system for clearing ice picks.

[0014] Preferably, the image preprocessing system uses a generator network algorithm to enhance dimly illuminated images, and uses a discriminator network algorithm to enhance detail clarity and quality of the images.

[0015] Preferably, the ice cone area positioning module can also perform human-computer interaction to selectively extract the rectangular pixel area in the ice-covered image to complete the extraction of the ice-covered target area.

[0016] Preferably, the target detection system uses an ice cone target recognition algorithm to identify the ice cone, and the ice cone target recognition algorithm includes the following process:

[0017] S1. Marking the connected domain of the binary image. After the ice cone image binarization module segments the image into a binary image by the color threshold, it divides the image area by specifying the relative position relationship between the foreground pixels, and defines the areas with adjacent positions and the same pixel values ​​as connected domains.

[0018] S2. Identify the ice cone. After connecting the connected domains, a more accurate target area of ​​the ice cone can be obtained. According to the gray or transparent color characteristics of the ice area, a one-to-one correspondence between the typical ice area and the HSI component is established to form a feature information library. The average values ​​and variances of the three color components H, S, and I are used as color feature parameters of the ice area to filter out the background area.

[0019] S3. Determine the edge of the ice cone and use the Canny operator to detect the edge of the ice cone.

[0020] Preferably, the laser triangulation ranging module adopts a triangulation laser ranging method to dynamically adjust the parameters of the output laser beam.

[0021] Preferably, the image acquisition system performs visual measurement and monitoring through an indicator laser, a PTZ camera, a control pan / tilt and a binocular camera.

[0022] The present invention has the following beneficial effects:

[0023] 1. It can fully cover the tunnel, and the integrated automatic movement and identification system reduces manual intervention, and can be operated unmanned in severe weather conditions or dangerous environments, improving operation safety. It can operate stably under low visibility or extreme weather conditions, ensuring the continuity and reliability of deicing operations, while reducing the risks and costs of manual operations.

[0024] 2. The high energy output of the laser can quickly melt the ice layer. Compared with the traditional physical deicing method, it greatly shortens the deicing time. In addition, the visual recognition technology can accurately identify the position and thickness of the ice cone, guide the laser instrument to perform targeted deicing, and improve the deicing efficiency;

[0025] 3. Automated deicing reduces labor costs and equipment wear, which helps reduce maintenance costs in the long run. The automated deicing system can operate without affecting the normal operation of the train, reducing train delays caused by deicing operations;

[0026] 4. It avoids the safety risks such as falling from high altitude and electric shock that may occur during manual de-icing, and reduces the risk of accidents caused by ice shedding;

[0027] 5. Laser deicing does not involve the use of chemical substances, is environmentally friendly and has no pollution;

[0028] 6. Operators can switch to manual de-icing, remotely monitor the de-icing process from a safe location, and adjust the de-icing strategy in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A system block diagram of a deicing recognition system based on visual servoing proposed by the present invention;

[0030] Figure 2 It is the HSI color space model diagram;

[0031] Figure 3 It is the connected domain graph of 4 domains and 8 domains;

[0032] Figure 4 is a process diagram of an ice cone target recognition algorithm in the present invention; DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0034] Embodiment 1:

[0035] Reference Figure 1 , a deicing recognition system based on visual servoing, includes an image acquisition system, an image preprocessing system, a target detection system, a binocular camera space positioning system, a laser triangulation ranging system and an optical obstacle removal system.

[0036] Image acquisition system: The image acquisition device is used to collect images and use a computer to analyze the collected images; the image acquisition system performs visual measurement and monitoring through the indicator laser, PTZ camera, pan / tilt control and binocular phase.

[0037] Image preprocessing system: The image preprocessing system is used to enhance and denoise images in low light conditions;

[0038] The target detection system is used to identify ice cones and determine their characteristics. The target detection system includes an ice cone area positioning module, an ice cone image binarization module and an image threshold segmentation module. The ice cone area positioning module is used to extract the ice-covered target area in the collected image.

[0039] For the ice cone area positioning module, considering the design requirements of the visual measurement and monitoring system in the laser deicing device, the present invention focuses more on the ice-covered target area, that is, the region of interest (ROI), compared with the entire image. Since the working distance of the laser deicing device is usually dynamically changing, there may be a large background area in the camera image, which will weaken the characteristics of the ice-covered target in the image, interfere with the subsequent ice target recognition and visual measurement, and increase the difficulty of visual algorithm design.

[0040] The ice cone image binarization module uses the HSI color space to perform color threshold segmentation on the ice cone image, and realizes ice recognition by extracting the HSI component value of the ice-covered target area. The image threshold segmentation module is used to segment the ice-covered target area and the background area in the image. Specifically, the HSI (hue / saturation / intensity) color space model is designed based on the human eye's visual perception of color information. Figure 2As shown, in this model, the hue H component is represented by an angle, and its value range is [0,2Π]; the saturation S component is represented by the radius length, and its value range is [0,1]; the intensity I component is represented along the central axis, and its value range is [0,1]. The HSI color space model is a color space model that separates chromaticity and brightness, in which the image color information is described by the H component and the S component respectively, and the brightness information is described by the I component. This description method is extremely similar to the human perception of color. This application uses the HSI color space to perform color threshold segmentation on the ice cone image, and realizes the recognition of ice by extracting the HSI component value of the ice-covered target area.

[0041] For the image threshold segmentation module, color is a very powerful descriptive factor in image processing. Unlike grayscale images, color images can usually express image information more completely. Extract a row of pixels containing the ice cone target from the RGB image and convert it to the HSI color space. Normalize the value range of the H, S, and I components to [0, 255] to obtain the distribution of the H, S, and I color components.

[0042] The color characteristics of the ice cone area are obvious. The H and I components maintain higher values ​​in the ice cone area, and the H component changes more slowly. The S component maintains a lower value in the ice area, and the difference between the H and S values ​​is large in the ice area, while in the background area, this difference is small.

[0043] This paper uses the H component and S component as the color segmentation features of the ice cone area and the background area, while the background area outside the ice cone area does not have such obvious distribution features. On this basis, a threshold segmentation algorithm based on the HSI color space is designed, that is, the segmentation of the ice target and the background is achieved by setting the difference size of the HS component in the image.

[0044] Since there may be small holes in the ice cone image area after threshold segmentation, and there will also be many isolated black spots around it. In order to remove noise and fill holes, it is necessary to optimize the binary image after threshold segmentation, that is, perform morphological processing. The set of all pixel points is regarded as a matrix, and with the help of mathematical morphological analysis methods, the image shape and structural features are analyzed by defining structural elements and performing logical operations on them. Mathematical morphology [8] is a mathematical method for analyzing geometric shapes and structures. It can implement morphological analysis and processing algorithms in parallel, greatly improving the speed of image analysis and processing. The basic operations of mathematical morphology are dilation and erosion, which are combined to form opening and closing operations. Dilation will make the target area larger and is mainly used to fill certain holes in the target image. Erosion is the dual operation of dilation and can eliminate small and meaningless targets. Opening operation refers to erosion followed by dilation. It can be used to eliminate isolated small points and burrs, separate objects at thin points, and smooth the boundaries of larger objects, but the overall position and shape remain unchanged. The closing operation refers to dilation followed by erosion. It can fill tiny holes inside image objects, connect adjacent objects, merge subtle connections, and can be used to repair disconnected objects.

[0045] A binocular camera spatial positioning system is used to spatially locate the ice cone using a binocular camera after acquiring the ice cone area;

[0046] Specifically, after obtaining the positions of the ice cone and the top of the tunnel, a binocular camera is used to spatially locate their positions.

[0047] The structure of the binocular camera and the process of calculating the target depth are similar to those of the human eye. The two cameras shoot the same target object from different angles. The binocular camera collects the light generated by the reflection or refraction of the background light source to other objects to form an image. The binocular camera is used to capture spatial objects within the visible range, and the position of any point in the space can be mapped to the captured image. By connecting the coordinate system origin, pixel points, a point in space, and the optical axis to form multiple similar triangles, the basic principle of similar triangles can be used to calculate the position of a point in space projected to the pixel points in the left and right images of the camera; the parallax image is obtained through the stereo matching algorithm, and the three-dimensional coordinates of the point in space can be calculated. This application needs to obtain the three-dimensional coordinates of the ice cone in three-dimensional space, and determine the impact position of the laser through the three-dimensional coordinates to achieve the removal of the ice cone.

[0048] Laser triangulation ranging system, used to dynamically adjust the parameters, timing and action position of the output laser beam according to the working distance between the laser emission area and the ice cone, so as to increase the ice melting speed;

[0049] Specifically, during the laser remote deicing process, the laser power density will be greatly reduced as the working distance increases. Therefore, dynamically adjusting the parameters, timing and action position of the output laser beam based on the working distance can effectively increase the ice melting speed.

[0050] Triangulation laser ranging mainly achieves the purpose of ranging by measuring the phase displacement of the object. When the laser beam emitted by the laser is irradiated on the target surface, it is reflected. When the reflected light is gathered, a light spot is formed at the sensor. When the position of the target changes, the reflected light also changes accordingly. The incident light and the reflected normal of the target form a geometric triangle. The triangulation laser ranging method is divided into oblique ranging method and direct ranging method according to whether the incident light is parallel to the normal of the target. Its operation process mainly uses the laser emitted by the laser to focus on the surface of the object to be measured. The receiving lens images the scattered light at the incident point on the photosensitive surface of the detector. When the object moves, the light spot moves with it. By calculating the displacement of the light spot at the center of mass of the photosensitive surface and then using algorithm design, the relative distance of the object's movement can be calculated. The direct ranging formula is as follows:

[0051]

[0052] Where y represents the displacement of the object, a is the distance from the measured surface of the object to the focusing lens, b is the image distance, Δx represents the movement of the photosensitive surface of the light spot, α is the incident angle, and β is the angle between the optical axis of the focusing lens and the photosensitive surface.

[0053] The laser obstacle removal system is used to remove ice cones. Specifically, after obtaining the spatial coordinates of the area where the ice cone is connected to the top of the tunnel and the system longitudinal working distance between the laser emitter and the ice cone, the de-icing cone system is moved on the track to a suitable position to perform de-icing work. The basic principle is as follows.

[0054] The essence of laser is a light source with highly consistent photon optical properties and highly concentrated energy density. According to the optical properties of ice, when the laser is irradiated to the surface of ice, only part of the laser energy is absorbed by the surface, while the rest of the energy passes through the ice layer and continues to propagate. As the incident depth increases, the laser energy gradually decreases. When the ice layer is thick enough, the laser energy can be regarded as completely absorbed. According to the Bouguer-Lambert law, the laser intensity decays exponentially with the increase of propagation distance:

[0055] I (Z) =I(0) -az

[0056] Where I is the laser intensity (W / m 2 ), z is the incident depth of the laser (m), and α is the linear absorption coefficient of ice to light.

[0057] The basic principle of laser obstacle removal technology is the interaction between laser and material. The effect of laser on material is caused by the interaction between the high-frequency electromagnetic field of light and the electrons in the material. The wavelength, power and other parameters of the laser are different from the characteristics of the material itself, which will lead to different effects. When the laser wavelength is in the infrared region, laser irradiation of obstacles usually causes laser thermal evaporation effect, causing the temperature of the obstacle to rise to melting or evaporation. If a large temperature difference occurs inside the obstacle, thermal stress may also be generated. Laser obstacle removal technology takes advantage of this characteristic to achieve the effect of melting through, cutting foreign objects or melting ice.

[0058] Normally, the laser beam emitted by the laser is a Gaussian beam, and the amplitude distribution of the beam cross section obeys the Gaussian function, that is:

[0059]

[0060] Where: r is the distance between a point in the beam and the central axis; ω is the beam radius; P 0 is the laser power.

[0061] Heat transfer can be divided into three main forms: heat conduction, heat convection and heat radiation. The main form of heat transfer inside a solid is heat conduction. The heat conduction process inside an icicle follows Fourier's law, that is:

[0062] q=-k▽T

[0063] Where: q is the heat flux density; k is the thermal conductivity of the object. Considering the laser as a heat source, the heat conduction equation in an isotropic object in a three-dimensional rectangular coordinate system is as follows:

[0064]

[0065] Where: ρ is the density of the object; C is the heat capacity of the object; Q is the heat generation rate of the object absorbing laser energy per unit volume. The boundary conditions are:

[0066]

[0067] Where: h is the convective heat transfer coefficient; σ is the Boltzmann constant; ε is the thermal radiation coefficient.

[0068] In this embodiment, an indicator laser coaxial with its emitted light beam is used to aim at the target area to be cleared, and the on-site ice cone image is collected by a PTZ camera and sent to a computer system. The computer locates and identifies the ice cone target area in the image through system software, and obtains the longitudinal working distance of the system in combination with the laser triangulation function. A binocular camera is used to obtain the stereo coordinates of the connection position of the ice cone and the tunnel top in three dimensions, and then laser clearing is performed.

[0069] Embodiment 2:

[0070] Compared with the first embodiment, further, the image preprocessing system in this embodiment adopts a generator network algorithm to enhance dim light images, and adopts a discriminator network algorithm to enhance the detail clarity and quality of the image.

[0071] Due to the possibility of insufficient lighting and limited exposure time in tunnels, problems such as low image brightness, contrast, signal-to-noise ratio, and high noise pollution may occur in low-light conditions, which will inevitably cause serious impact on the ice cone detection task. Therefore, before the computer system locates and identifies the ice cone target area, a low-light image enhancement algorithm LEGAN[5] can be used to restore image details and avoid the problem of poor local brightness in the enhanced image.

[0072] Since Goodfellow et al. proposed the generative adversarial network (GAN) in 2014[6], researchers have continued to deepen their research on GAN networks, and have been widely used in image generation, image translation, super-resolution, and image motion transfer. Low-light image enhancement can be regarded as a style transfer transformation from low-light images to normal-brightness images. Therefore, algorithms based on GAN networks can be used to achieve the purpose of enhancing low-light images.

[0073] The LEGAN method mainly includes a generator network and a discriminator network. The generator network includes a gamma curve estimation network (GCE-Net) and a convolutional block attention module (CBAM). The gamma curve estimation network extracts features through a convolutional module and uses jump connections to fuse information from different depth feature layers. The convolutional attention module is inserted in the jump connection process to increase the model's attention to important features at one time, and finally generate a feature map containing gamma parameters. After that, the LEGAN model separates the feature map output by GCE-Net into 8 groups of RGB channels, uses a light enhancement block (LEB) to perform preliminary image enhancement, and continuously iterates by cascading LEBs to improve the enhancement effect of the model and obtain the final enhancement result.

[0074] If only the generator network is used for model training, the overall edge and color restoration of the enhanced image can be ensured, but the details of the image will be seriously lost, resulting in a relatively blurred result. In order to make the image clearer, that is, to build the high-frequency part of the image, a discriminator is needed to retain the high-frequency information of the image. In order to improve the global illumination effect of the image, LEGAN introduces a global-local discriminator structure into the generator network and uses PatchGAN as the backbone structure of the discriminator. The PatchGAN network usually consists of multiple convolutional layers and pooling layers to gradually reduce the spatial resolution of the input image and extract higher-level features. The output of the last convolutional layer is a matrix composed of many small blocks, each of which corresponds to a small area of ​​the input image and outputs a judgment value. These judgment values ​​are used to calculate the loss function of GAN to optimize the parameters of the generator and discriminator to better generate images. In order to achieve more detailed inspection, an additional local discriminator is added to the discriminator. By randomly cropping patches of the image and identifying them separately, it can not only ensure the clarity and quality of the details of the enhanced image, but also effectively avoid overexposure or underexposure problems.

[0075] Embodiment three:

[0076] Compared with the first embodiment, further, the ice cone area positioning module in this embodiment can also perform human-computer interaction to selectively extract the rectangular pixel area in the ice-covered image to complete the extraction of the ice-covered target area.

[0077] Furthermore, the target detection system in the first embodiment uses an ice cone target recognition algorithm to recognize ice cones, and the ice cone target recognition algorithm includes the following process:

[0078] S1. Marking the connected domain of the binary image. After the ice cone image binarization module segments the image into a binary image by the color threshold, it divides the image area by specifying the relative position relationship between the foreground pixels, and defines the areas with adjacent positions and the same pixel values ​​as connected domains.

[0079] Specifically, the process of dividing the image area by specifying the relative position relationship between foreground pixels in a binary image is called image connected domain labeling, and the areas with adjacent positions and the same pixel values ​​are defined as connected domains. Figure 3 Shown are two common ways of dividing connected domains.

[0080] After the image is processed by morphology, the areas where each object in the image is located are independent of each other, so the ice area must be in the same connected domain. This application uses 8 connected domains to mark the binary image and uses the itinerary method to traverse the image. The specific marking process is: start traversing from the upper left corner of the image, if a pixel with a pixel value of 0 is encountered, continue scanning; if a pixel with a pixel value of 255 is encountered, labeling is performed at this time; after traversing all the pixels of the entire image in turn, determine again whether the labels of the same area are consistent. If not, traverse the image again to ensure that the labels in the same connected domain are unified. The labeling rules are as follows: first, compare the pixel values ​​of the scanned pixel and its upper left corner pixel to see if they are equal. If they are equal, the label is the same as the label of the upper left corner. If they are not equal, proceed to the next step. Then, compare the pixel values ​​of the scanned pixel and the pixel directly above it to see if they are equal. If they are equal, the label is the same as the label directly above it. If they are not equal, proceed to the next step. Then, when the pixel values ​​of the upper left corner and the left pixel are equal and 255, and the pixel value of the upper right corner pixel is 0, the label is incremented by one. Otherwise, if the pixel value of the upper right corner pixel is 255, the label is the same as the label of the upper right corner. The initial labels of all pixels in the entire image are 0.

[0081] S2. Identify the ice cone. After connecting the connected domains, a more accurate target area of ​​the ice cone can be obtained. According to the gray or transparent color characteristics of the ice area, a one-to-one correspondence between the typical ice area and the HSI component is established to form a feature information library. The average values ​​and variances of the three color components H, S, and I are used as color feature parameters of the ice area to filter out the background area.

[0082] Specifically, a relatively accurate target area of ​​the ice cone can be obtained by marking the connected domain, but many interference areas are also returned as independent areas. Therefore, according to the gray or transparent color characteristics of the ice area, a method for determining ice targets based on the HSI component of the image is proposed. By establishing a one-to-one correspondence between typical ice areas and HSI components, a feature information library is formed, and the average and variance of the three color components H, S, and I are used as the color feature parameters of the ice area, thereby filtering out the background area. The calculation formula is as follows:

[0083]

[0084] Among them, the value of i represents the H, S, and I components respectively, P_ij is the jth pixel value of the i-th color component in the ice area image, and N is the total number of pixels in the image; u_i is the average value of the i-th component of the image, which represents the variance of the i-th component of the image. By mathematically counting ice-covered images under different backgrounds and weather conditions and based on experimental verification, the discrimination interval of ice targets is obtained as shown in the following table. According to the distribution range of the color feature values ​​in the table, the connected domain is marked and the ice cone is judged on the binary result map.

[0085]

[0086] S3. Determine the edge of the ice cone and use the Canny operator to detect the edge of the ice cone. Specifically, the Canny operator is a common non-differential edge detection operator, and its detailed processing process is as follows:

[0087] ① Image denoising. A Gaussian filter template is constructed through a one-dimensional zero-mean Gaussian function, and then the original binary image is Gaussian filtered to eliminate image noise. The signal-to-noise ratio of the resulting image can be maximized by selecting an appropriate filter template size. This application uses a 3×3 size template.

[0088] ② Calculate the magnitude and direction of the gradient. Obtain the gradient magnitude graph under different gradient directions (0°, 45°, 90°, 135°).

[0089] ③ Non-maximum suppression in gradient direction. The gradient amplitude image is further refined to obtain the image edge of a single pixel. The main idea is: traverse the gradient amplitude image in all directions. If the grayscale value corresponding to a pixel point is less than the gradient amplitude of a point in its neighborhood (usually defined as a 3×3 size neighborhood), the gradient amplitude corresponding to the pixel point is set to 0, which is a non-edge point.

[0090] ④ Double threshold screening. By setting high and low thresholds to remove false edges in the image, first initialize the values ​​of the high threshold x1 and the low threshold x2. Then traverse the gradient amplitude image. If the pixel value of a point is greater than or equal to x1, the pixel point is marked as a strong edge; if the pixel value of the point is less than or equal to x2, the pixel value of the point is marked as a weak edge; in other cases, detect whether there is a strong edge in its neighborhood. If so, it is marked as a strong edge, otherwise it is marked as a weak edge. Finally, connect the strong edges of the image.

[0091] Since trivial edges and internal interference edges will be generated around the ice target contour detected by the Canny operator, it cannot be directly used for subsequent visual measurement. In order to obtain accurate and clear ice cone target contour features, an additional ice cone target contour extraction algorithm is used. The main idea is: first, the ice cone area is filled using the seed filling method; then the morphological opening operation is used to smooth the edges to eliminate the isolated areas around the ice target; finally, the Canny operator is used again to extract the edges of the seed-filled image. Finally, the detected edges are attached to the original image.

[0092] In summary, the deicing process of the present invention is as follows: First, the area where ice cones may exist is detected and photographed in real time by a camera to obtain image data. These image data are pre-processed through technologies including enhancement and denoising to ensure that subsequent recognition work is accurate and reliable. Next, the captured images are analyzed and processed through advanced image recognition technology to identify possible ice cones. The results need to be verified during the recognition process to ensure that there will be no misjudgment. Once the recognition results are confirmed to be accurate, the subsequent workflow is entered. After identifying the position of the ice cone, it is necessary to accurately locate the position of the ice cone through image positioning technology. Subsequently, the laser transmitter emits a laser to the tail of the ice cone for clearing. If the ice cone still exists after laser irradiation, the emission operation will be performed again until the ice cone is completely removed.

[0093] Considering that the railway tunnel is equipped with power supply contact network, various communication equipment and other necessary equipment to ensure the operation of the railway, in order to ensure the safe operation of the above equipment, the laser power used for deicing must be as small as possible, so as not to affect the normal operation of various equipment. Therefore, the deicing model of electrified railway tunnels cannot only consider the temperature change at the heated position at the moment of heating, like high-power laser cutting of metal materials; but it is necessary to set reasonable calculation conditions from the perspective of time domain to gradually melt the icicle within a certain time range (tens of seconds to minutes). This model needs to consider multiple factors such as the ice-water phase change process, the heat conduction process caused by the low temperature in the tunnel environment, and the heat conduction process of the ice itself. On the other hand, the hanging ice in the tunnel is a long column hanging on the top wall of the tunnel. When deicing, it is not necessary to melt the entire icicle, but only to melt the root of the icicle, so that the icicle loses adhesion and falls naturally due to gravity. In this process, the laser is heated from the oblique bottom, and the liquid water produced by heating and melting will naturally detach from the icicle under the action of gravity. The model also needs to consider the impact of this phenomenon, so the material properties of the melted part of the icicle are replaced with air.

[0094] In the entire process of removing ice cones, in addition to ice cone detection and de-icing, other possible problems need to be considered. Sound sensors and image sensors are integrated with lasers and cameras to monitor whether there are workers or trains passing through the tunnel, and stop working and issue an alarm in time when relevant information is detected. In addition, cameras working in tunnels may also be affected by mud spots and fog, and need to be cleaned and maintained regularly to ensure the clarity and accuracy of image data. For this point, the system integrates fog and mud spot detection functions. When relevant information is detected, the equipment will stop working and automatically return to the maintenance room for cleaning by staff.

[0095] In order to ensure data security and work continuity, the collected data needs to be backed up and archived every day to prevent accidental data loss or damage. In addition, in order to deal with possible unexpected situations, abnormal detection components, such as gas monitoring systems, can be installed on the equipment. Once an abnormal situation is detected, an alarm will be automatically triggered to ensure the safety and stability of work.

[0096] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A deicing recognition system based on visual servoing, characterized in that: include: An image acquisition system, wherein the image acquisition device is used to acquire images and analyze the acquired images using a computer; An image preprocessing system, wherein the image preprocessing system is used to enhance and denoise images in dim light conditions; A target detection system, the target detection system is used to identify ice cones and determine their characteristics. The target detection system includes an ice cone area positioning module, an ice cone image binarization module and an image threshold segmentation module. The ice cone area positioning module is used to extract the ice-covered target area in the collected image. The ice cone image binarization module uses the HSI color space to perform color threshold segmentation on the ice cone image and realizes ice recognition by extracting the HSI component value of the ice-covered target area. The image threshold segmentation module is used to segment the ice-covered target area and the background area in the image; A binocular camera spatial positioning system is used to spatially locate the ice cone using a binocular camera after acquiring the ice cone area; Laser triangulation ranging system, used to dynamically adjust the parameters, timing and action position of the output laser beam according to the working distance between the laser emission area and the ice cone, so as to increase the ice melting speed; Laser obstacle removal system for clearing ice picks.

2. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The image preprocessing system adopts a generator network algorithm to enhance dark and weakly illuminated images, and adopts a discriminator network algorithm to enhance the detail clarity and quality of the images.

3. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The ice cone area positioning module can also perform human-computer interaction to selectively extract the rectangular pixel area in the ice-covered image to complete the extraction of the ice-covered target area.

4. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The target detection system uses an ice cone target recognition algorithm to identify ice cones. The ice cone target recognition algorithm includes the following processes: S1. Marking the connected domain of the binary image. After the ice cone image binarization module segments the image into a binary image by the color threshold, it divides the image area by specifying the relative position relationship between the foreground pixels, and defines the areas with adjacent positions and the same pixel values ​​as connected domains. S2. Identify the ice cone. After connecting the connected domains, a more accurate target area of ​​the ice cone can be obtained. According to the gray or transparent color characteristics of the ice area, a one-to-one correspondence between the typical ice area and the HSI component is established to form a feature information library. The average values ​​and variances of the three color components H, S, and I are used as color feature parameters of the ice area to filter out the background area. S3. Determine the edge of the ice cone and use the Canny operator to detect the edge of the ice cone.

5. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The binocular camera spatial positioning system calculates the stereo coordinates of the ice cone in three-dimensional space through the principle of similar triangles and a stereo matching algorithm.

6. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The laser triangulation distance measurement module adopts a triangulation laser distance measurement method to dynamically adjust the parameters of the output laser beam.

7. The deicing identification system based on visual servoing according to claim 1 is characterized in that: The image acquisition system performs visual measurement and monitoring through an indicator laser, a PTZ camera, a control pan / tilt and a binocular camera.

Images