Cable trench cover plate positioning and internal cable scanning method based on binocular vision

Abstract

The invention relates to a cable trench cover plate positioning and internal cable scanning method based on binocular vision, and belongs to the field of power cables. Firstly, a binocular camera is used for collecting images of the cable trench and the surrounding environment of the cable trench, and preprocessing and stereo correction are carried out. Then, a disparity map between the cable trench cover plate and the surrounding building is calculated through an SGBM algorithm, and then accurate distance information of the cable trench cover plate and the surrounding building is obtained; and then, analyzing the corrected binocular image by using a YOLOv5 target detection algorithm, and identifying the position and distribution condition of the cable in the cable trench. And finally, according to an identification result and the disparity map, calculating accurate three-dimensional coordinate information of a cable in the cable trench, inputting the accurate three-dimensional coordinate information into a system to generate a three-dimensional model of the cable trench, and performing visual display. According to the invention, accurate positioning of the cable trench and detailed scanning of the internal cable can be realized, reliable data support is provided for construction, maintenance and management of the cable trench, and the working efficiency and the management precision are improved.

Description

Technical Field

[0001] The invention belongs to the field of power cables and relates to a cable trench cover positioning and internal cable scanning method based on binocular vision. Background Art

[0002] Electric energy transmission, usually the electric energy generated by power plants, is allocated by the dispatching center according to the electricity consumption in different regions. Electric energy is output to different regions through cables, so the transmission of electric energy is a bridge connecting the generation of electric energy and users. In the process of electric energy transmission, it can be simply divided into overhead cable transmission and underground cable transmission according to different transmission methods. Due to the low construction cost and easy fault detection and maintenance, overhead lines are the main transmission method in the power supply system. However, due to the advancement of my country's urbanization process, compared with underground transmission, overhead lines do not have an absolute advantage. For some lines close to buildings, their transmission failures will affect the safety of life and property of residents in the building. And the construction work area is wide and difficult during emergency repair operations. Therefore, the application of changing overhead lines to underground transmission came into being. With the increase in line mileage, underground laying has become the main method of urban power transmission.

[0003] As a form of underground power transmission and distribution network, underground cable trench is one of the indispensable facilities in power grid construction. Cables in underground cable trenches not only play the role of power transmission, but sometimes also involve metering, communication and relay protection functions. Long-term operation of power cables in underground cable trenches will lead to aging and discharge of cable insulation. In addition, the accumulation of flammable gases such as methane produced by microbial decomposition of some animal and plant corpses in the cable trench may eventually cause fire and explosion in the underground cable trench. Due to the narrow and complex internal environment of underground cable trenches, it is particularly difficult for staff to patrol. In the event of a cable failure in the cable trench, it may cause the relay protection to fail, resulting in an expansion of the range of obstacles, damage to the main equipment, and affect load production. Therefore, strengthening the inspection of underground cable trenches and discovering hidden dangers inside underground cable trenches as early as possible are of vital significance to improving the stability and reliability of power grid operation.

[0004] Existing cable trench inspection technologies mainly include manual inspection and automatic inspection based on monocular cameras. The manual inspection method relies on staff to inspect the cable trench section by section, record and report the problems found. The automatic inspection system based on monocular cameras collects cable trench images through the camera and uses image processing technology to perform cable detection and status assessment. These solutions rely on basic image processing and artificial intelligence technologies to improve inspection efficiency and accuracy to a certain extent.

[0005] Many traditional solutions rely on manual inspection or simple monocular camera systems, which are often inefficient and inaccurate when dealing with cable trenches in large-scale and complex environments. Manual inspection is not only time-consuming, but also easily affected by subjective factors and may miss important information. Monocular camera systems have accuracy limitations in ranging and target recognition, and are difficult to provide detailed three-dimensional information.

[0006] Traditional feature detection algorithms cannot achieve accurate target detection in complex environments, especially in cases of occlusion, lighting changes, etc., where the detection performance is significantly reduced. YOLOv5 has higher recognition accuracy and robustness when dealing with small targets and multi-scale targets caused by occlusion. Compared with traditional feature detection algorithms, YOLOv5 can maintain efficient and accurate detection performance in complex environments.

[0007] In the depth map generated by the traditional matching algorithm, the target outline is fuzzy and the surface is rough, and mismatched areas are prone to appear between targets, resulting in unclear target segmentation, especially in dense scenes, where targets cannot be accurately distinguished due to mutual occlusion, resulting in large distance calculation errors. In contrast, the SGBM algorithm has a smoother effect in binocular camera image matching, with clear outlines between targets and good matching effects, which significantly improves the quality and accuracy of the depth map.

[0008] After GPS rough positioning, this solution combines binocular camera ranging, target detection and depth estimation technology to measure the distance between the cable trench cover and surrounding buildings to achieve accurate positioning of the cable trench cover. Secondly, by scanning and measuring the distance inside the cable trench, the detailed cable layout inside the cable trench is obtained. Compared with traditional manual inspection or monocular camera systems, this solution uses the stereoscopic vision information provided by the binocular camera to significantly improve the ranging accuracy and three-dimensional reconstruction capabilities.

[0009] Accurately positioning the cable trench makes it easier for subsequent scanning imaging data to accurately correspond to the actual cable trench location. This not only makes it easier to associate the scanning information with the specific cable trench location, but also clearly displays the connection status of each cable trench in the system, making it easier to add new cables during subsequent construction, thereby improving construction efficiency and accuracy. Summary of the invention

[0010] In view of this, the purpose of the present invention is to provide a method for positioning cable trench covers and scanning internal cables based on binocular vision. By combining binocular camera ranging, target detection and depth estimation technology, accurate positioning of the cable trench and imaging scanning of the internal cables can be achieved. First, the cable trench cover is positioned with high precision using a binocular camera to ensure that the scanned data in the cable trench is accurately connected to the actual position. Then, the distribution of cables inside the cable trench is identified through the binocular camera, and detailed cable distance information is obtained to facilitate subsequent measurement of cable diameters. Finally, these data are input into the system, which can comprehensively and accurately display the precise position of the cable trench and the layout of the internal cables, providing a reliable basis for subsequent construction and maintenance work, and improving work efficiency and management accuracy.

[0011] In order to achieve the above object, the present invention provides the following technical solutions:

[0012] A cable trench cover positioning and internal cable scanning method based on binocular vision comprises the following steps:

[0013] S1: Use a binocular camera to collect images of the cable trench and its surrounding environment;

[0014] S2: preprocessing the acquired images, including removing noise and adjusting brightness, to enhance image quality;

[0015] S3: Use the Bouguet algorithm to perform stereo correction on the binocular image to align the corresponding pixels of the left and right images;

[0016] S4: Use the YOLOv5 target detection algorithm to analyze the corrected binocular image to identify the location and distribution of cables inside the cable trench;

[0017] S5: Using the corrected binocular image, the disparity map between the cable trench cover and the surrounding buildings is calculated through the SGBM algorithm to obtain their precise distance information;

[0018] S6: Calculate the precise three-dimensional coordinate information of the cables inside the cable trench according to the recognition results and the disparity map;

[0019] S7: Input the cable trench location information and cable layout information into the system, generate a three-dimensional model of the cable trench, and perform a visual display.

[0020] Furthermore, the images collected in S1 include images of the cable trench cover and surrounding buildings.

[0021] Furthermore, the Bouguet algorithm used in S3 performs line alignment correction on the image to eliminate the parallax of the binocular camera imaging plane.

[0022] Furthermore, the YOLOv5 target detection algorithm used in S4 can accurately identify the cables inside the cable trench and identify the type and quantity information of the cables.

[0023] Furthermore, the SGBM algorithm used in S5 can process images with occlusion or illumination changes and generate a smooth disparity map.

[0024] Furthermore, the three-dimensional coordinate information of the cable inside the cable trench is calculated in S6, which can reflect the length and bending degree of the cable.

[0025] Furthermore, the three-dimensional model of the cable trench generated in S7 can be visualized in the form of a three-dimensional graph or a two-dimensional plane diagram, which is convenient for users to view and analyze.

[0026] A cable trench cover positioning and internal cable scanning system based on binocular vision, comprising:

[0027] Binocular camera to collect images of the cable trench and its surrounding environment;

[0028] An image preprocessing module preprocesses the collected images, including removing noise and adjusting brightness to enhance image quality;

[0029] The stereo correction module uses the Bouguet algorithm to perform stereo correction on the binocular image to align the corresponding pixels of the left and right images;

[0030] The disparity calculation module uses the corrected binocular image to calculate the disparity map between the cable trench cover and the surrounding buildings through the SGBM algorithm, thereby obtaining their precise distance information;

[0031] The target detection module uses the YOLOv5 target detection algorithm to analyze the calibrated binocular image and identify the location and distribution of cables inside the cable trench;

[0032] The 3D reconstruction module calculates the precise 3D coordinate information of the cables inside the cable trench based on the recognition results and disparity map;

[0033] The visualization display module inputs the cable trench positioning information and cable layout information into the system, generates a three-dimensional model of the cable trench, and displays it visually.

[0034] The beneficial effects of the present invention are as follows: by integrating binocular camera ranging, target detection and depth estimation technology, the accuracy and efficiency of cable trench management are significantly improved. The high-precision ranging and three-dimensional reconstruction capabilities provided by the binocular camera make the positioning of the cable trench cover more accurate and the layout of the internal cables clearer, thereby reducing the workload and potential errors of manual inspection. YOLOv5 Japanese standard detection technology can quickly identify targets in the cable trench, obtain cable distribution information, and overcome the impact of environmental changes on data quality. This solution not only improves the reliability and stability of data acquisition, but also speeds up the process of cable trench management and maintenance, reduces construction costs and time, and provides more accurate and comprehensive support for the construction, maintenance and subsequent management of the cable trench.

[0035] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0037] Figure 1 It is the technical roadmap of binocular ranging;

[0038] Figure 2 This is the effect picture before stereo correction;

[0039] Figure 3 This is the effect picture after stereo correction;

[0040] Figure 4 This is a schematic diagram of the Focus slicing operation;

[0041] Figure 5 This is the overall block diagram of the YOLOv5 target detection algorithm;

[0042] Figure 6 This is the flow chart of the SGBM algorithm;

[0043] Figure 7 This is the flow chart of binocular ranging. DETAILED DESCRIPTION

[0044] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0045] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0046] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0047] 1. Stereo correction based on Bouguet algorithm

[0048] The Bouguet algorithm decomposes the rotation matrix obtained by calibration into r l and r r The decomposed rotation matrix is ​​called the synthetic rotation matrix. Then, the translation matrix T obtained by binocular stereo calibration is used to construct the transformation matrix R rect , calibrate the image to a row-aligned state.

[0049] The pole of the left view is transformed to infinity through the transformation matrix. The translation vector between the projection centers of the left and right cameras is the direction of the left pole. The constructed e1 is:

[0050]

[0051] The direction of e2 is orthogonal to the main optical axis, along the direction of the image, and perpendicular to e1; the direction of e2 can be obtained by normalizing the cross product of e1 and the main optical axis.

[0052]

[0053] e3 is orthogonal to e1 and e2, that is, e3 is the cross product of e1 and e2.

[0054] e3=e1×e2 (3)

[0055] Then the transformation matrix R rect for:

[0056]

[0057] The rotation matrix for aligning the imaging planes of the left and right cameras is:

[0058] R L =R rect r l (5)

[0059] R R =R rect r r (6)

[0060] Finally, the intrinsic parameter matrix and reprojection matrix of the binocular camera after parallel correction are calculated, and the stereo correction is completed by the remapping method, such as Figure 2 and Figure 3 shown.

[0061] 2. Object detection based on YOLOv5

[0062] The YOLOv5 algorithm mainly includes four basic modules: network input, Backbone network module, Neck network module, and output.

[0063] (1) Input terminal

[0064] YOLOv5 has made the following improvements in the model training phase:

[0065] 1) Use Mosaic data enhancement. In YOLOv4, it was proposed to use Mosaic data enhancement to expand the data set, and YOLOv5 continued this method. This method is to splice four images that have been randomly cropped, scaled, and arbitrarily arranged into a new image, and then input the new image into the network for training. This data enhancement method is conducive to small target detection.

[0066] 2) Adaptive anchor box calculation. YOLOv5 embeds the calculation of anchor box size into the model training process, so that the model can automatically calculate the optimal anchor box. There is no need to calculate the anchor box size separately when training on different data sets.

[0067] 3) Adaptive image scaling: This method can be used to scale the original image to a standard size by filling the original image with the least amount of pixels, which can reduce redundant information and help improve the network inference speed.

[0068] (2) Backbone

[0069] Backbone is the backbone network, which is used to extract image features for use by subsequent network layers.

[0070] 1) Focus module

[0071] Before the image is input into the backbone network, the Focus module first slices it to split the high-resolution image into multiple low-resolution images. Figure 4 As shown in the figure, taking a 4x4x3 input image as an example, it is sampled at intervals and the sampling results are concatenated in the channel dimension, which can expand the channel to 4 times the original size and generate a 2x2x12 feature map. The network then performs convolution operations on the new feature map.

[0072] The image with a scale of 608x608x3 is input into YOLOv5s, and after the Focus layer is sliced, it becomes a feature map of 304x304x12, and then input into the 32-channel convolution layer for convolution operation, and finally obtains a feature map of 304x304x32. Using the Focus slicing operation to convert the plane information of the image into the channel dimension can achieve a double downsampling operation while ensuring that the image information is lossless, which is conducive to improving the network speed.

[0073] 2) CSP module

[0074] Due to the problem of repeated gradient calculations during network optimization, the inference process has too much computation. In order to solve this problem, the CSP module was proposed in YOLOv4 by referring to the idea of ​​CSPNet. The introduction of the CSP module is conducive to enhancing the learning ability of the network, which can reduce the amount of calculation while ensuring accuracy, and also reduce memory consumption.

[0075] (3) Neck

[0076] As feature extraction continues to deepen, some local information of the image will disappear. By using the Neck network to fuse feature maps of different network layers, richer feature information of the image can be obtained, and these processed features can be input into the Head layer for better classification and positioning.

[0077] YOLOv5 uses the FPN structure to upsample the deep feature map and fuse it with the shallow feature map, and then predicts each layer of the feature map in the feature pyramid. On this basis, a bottom-up feature pyramid is added. The feature pyramid contains two PAN structures, which can downsample the shallow feature map and fuse it with the deep feature map. This two-pronged design can not only transmit strong semantic information from top to bottom, but also convey strong positioning information from bottom to top, effectively improving the performance of the algorithm.

[0078] (4)Head

[0079] When using YOLOv5 to handle target detection tasks, the image is first input into the backbone network to extract features, and then the features extracted by the backbone network are input into the Neck network for processing. Finally, the Head layer outputs the target category and corrects the position of the candidate box according to the position offset, thereby obtaining a more accurate detection result.

[0080] 1) Loss Function

[0081] The loss function of YOLOv5 consists of three parts: bounding box loss, confidence loss, and classification loss. The bounding box regression loss is calculated using CIoU-Loss, which combines the overlapping area, center point distance, and aspect ratio information between the prediction box and the target box to improve the regression speed and accuracy of the detection box. The calculation method of CIoU-Loss is shown below.

[0082]

[0083]

[0084]

[0085] By default, both confidence loss and classification loss use the BCEWithLogitsLoss loss function, which combines the binary cross entropy (BCE) loss function and the Sigmoid activation function.

[0086] 2) NMS non-maximum suppression

[0087] In the target detection task, the network will output a large number of detection frames, and many detection frames contain the same target, but only one detection result is needed for one target in the end. Usually, in the post-processing process, the NMS algorithm is used to filter out the best result from multiple prediction frames. The NMS operation obtains all local maximum elements by searching, and suppresses non-local maximum elements to achieve the screening purpose. The specific operation steps are as follows: Sort the scores of all detection frames, and then select the frame with the highest score. Calculate the IoU value between the remaining detection frames and the detection frame. If it is greater than the set value, delete it from the remaining detection frames. Select the one with the highest score from the unprocessed detection frames, and repeat this cycle. The IoU values ​​between the remaining detection frames are very low, and basically only one detection result can be achieved for one target.

[0088] 3. Depth calculation based on SGBM algorithm

[0089] The SGBM (Semi-Global-Block Matching) algorithm is mainly based on the OpenCV library to achieve matching functions. The algorithm sets an energy function related to the disparity map. The disparity is mainly formed by each pixel in the image information. The purpose is to make its energy function as low as possible. Its expression is as follows:

[0090]

[0091] In formula 10, D represents the disparity map of the image, p and q represent any pixel in the image, where the pixel p is the center and the surrounding eight pixels form a 3x3 size neighboring pixel range, N p That is, the nearest pixel of the pixel p, D p is the depth value of the pixel, C(P,D p ) is the matching cost, P1 and P2 are penalty coefficients, and the implementation of the SGBM algorithm is mainly divided into five steps:

[0092] 1) Preprocessing

[0093] The image is collected in real time by the binocular camera, and then the image is preprocessed to obtain the gradient information of the image. The purpose is to use the new image obtained after the image processing for the next cost calculation. The preprocessing process first introduces the horizontal operator (Sobel) to process the image data, and then uses the mapping function to map each pixel in the image processed by the horizontal operator into a new image. p represents the new image value. The mapping function is shown as follows:

[0094]

[0095] Wherein preFilterCap represents a constant.

[0096] 2) Matching cost calculation

[0097] The matching cost mainly consists of two parts: first, the gradient cost of the image is obtained, which is mainly obtained by sampling the gradient information of the preprocessed image; then the SAD cost of the original image is obtained.

[0098] Let the function D(x i ,y i ,I L ,I R ), this function is mainly used to determine the possibility that the corresponding images are the same pixel point in two images, as shown in the following formula (12):

[0099]

[0100] In formula (12), x, y are any pixel points of the image captured by the left and right binocular cameras, and I L is the gray value of the pixel at x after preprocessing, I R It is obtained by linear interpolation of the sampling points on the collected right image.

[0101] Usually, the matching cost is measured by the SSD window. Each pixel in the window is grouped and subtracted by SAD. Then, the absolute value of the difference is calculated for each pixel in the window. All the absolute values ​​are added together to get the matching cost of the window.

[0102] 3) Dynamic programming

[0103] The 360° range around any pixel point p in the image is divided into eight paths on average. The minimum cost path L is calculated through the eight paths. r (p, d), thereby realizing the constraint matching calculation of two dimensions approximated by one-dimensional constraints. In this process, two important quantities P1 and P2 are included, and their mathematical expressions are shown in the following formula (13):

[0104]

[0105] In the above formula, cn is the number of channels for processing the image collected by the camera, SADWindowSize represents the window size of the matching cost. If cn and SAD window remain unchanged, the planning parameters P1 and P2 of SGBM are constants.

[0106] 4) Post-processing

[0107] (a) Uniqueness test:

[0108] When the lowest cost within the disparity window is (1+uiquenessRatio / 100) times the second lowest cost, the disparity value corresponding to the lowest cost is the disparity of the pixel point, otherwise the disparity of the pixel point is 0, where uiquenessRatio is a constant parameter.

[0109] Sub-pixel difference:

[0110]

[0111]

[0112] (b) Left and right pixel consistency detection:

[0113] Suppose the disparity of the image pixel p collected by the left camera is d, and the disparity of the image pixel collected by the right camera is d * , as shown below:

[0114] |dd * |<threshold (16)

[0115] In formula (16), threshold is a constant, generally threshold=1 or threshold=2. If the above formula (16) holds, d is defined as the correct disparity.

[0116] 5) Depth map calculation

[0117] After obtaining the disparity map, the depth map of the pixel can be calculated by the following formula, as shown in the following formula:

[0118]

[0119] In the above formula (17), f is the focal length of the binocular camera, b is the baseline, which refers to the distance between the optical centers of the left and right cameras; d is the parallax. When the object to be measured is far away from the camera, the parallax is smaller, and vice versa; c is xr With c xl is the column coordinate of the principal point of the left and right cameras. By calculating the parallax of the image captured by the binocular camera, the depth map of a certain area directly measured in the captured image can be obtained, that is, the value of depth.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for positioning cable trench cover and scanning internal cables based on binocular vision, characterized in that: The following steps are involved: S1: Use a binocular camera to collect images of the cable trench and its surrounding environment; S2: preprocessing the acquired images, including removing noise and adjusting brightness, to enhance image quality; S3: Use the Bouguet algorithm to perform stereo correction on the binocular image to align the corresponding pixels of the left and right images; S4: Use the YOLOv5 target detection algorithm to analyze the corrected binocular image to identify the location and distribution of cables inside the cable trench; S5: Using the corrected binocular image, the disparity map between the cable trench cover and the surrounding buildings is calculated through the SGBM algorithm to obtain their precise distance information; S6: Calculate the precise three-dimensional coordinate information of the cables inside the cable trench according to the recognition results and the disparity map; S7: Input the cable trench location information and cable layout information into the system, generate a three-dimensional model of the cable trench, and perform a visual display.

2. According to claim 1, a method for positioning cable trench covers and scanning internal cables based on binocular vision is characterized in that: In S1, the collected images include images of the cable trench cover and surrounding buildings.

3. According to claim 1, a method for positioning cable trench covers and scanning internal cables based on binocular vision is characterized in that: In S3, the Bouguet algorithm is used to perform line alignment correction on the image to eliminate the parallax of the binocular camera imaging plane.

4. According to claim 1, a method for positioning cable trench covers and scanning internal cables based on binocular vision is characterized in that: In S4, the YOLOv5 target detection algorithm used can accurately identify the cables inside the cable trench and identify the type and quantity information of the cables.

5. The method for positioning cable trench covers and scanning internal cables based on binocular vision according to claim 1 is characterized in that: In S5, the SGBM algorithm used can process images with occlusion or illumination changes and generate a smooth disparity map.

6. The method for positioning cable trench covers and scanning internal cables based on binocular vision according to claim 1 is characterized in that: In S6, the three-dimensional coordinate information of the cable inside the cable trench is calculated, which can reflect the length and bending degree of the cable.

7. The method for positioning cable trench covers and scanning internal cables based on binocular vision according to claim 1 is characterized in that: In the above S7, the generated three-dimensional model of the cable trench can be visualized in the form of a three-dimensional graph or a two-dimensional plane diagram, which is convenient for users to view and analyze.

8. A cable trench cover positioning and internal cable scanning system based on binocular vision, characterized in that: include: Binocular camera to collect images of the cable trench and its surrounding environment; An image preprocessing module preprocesses the collected images, including removing noise and adjusting brightness to enhance image quality; The stereo correction module uses the Bouguet algorithm to perform stereo correction on the binocular image to align the corresponding pixels of the left and right images; The disparity calculation module uses the corrected binocular image to calculate the disparity map between the cable trench cover and the surrounding buildings through the SGBM algorithm, thereby obtaining their precise distance information; The target detection module uses the YOLOv5 target detection algorithm to analyze the calibrated binocular image and identify the location and distribution of cables inside the cable trench; The 3D reconstruction module calculates the precise 3D coordinate information of the cables inside the cable trench based on the recognition results and disparity map; The visualization display module inputs the cable trench positioning information and cable layout information into the system, generates a three-dimensional model of the cable trench, and displays it visually.

Images