WO2023174020A1 - 输电线的检测方法、装置、计算机设备和存储介质 - Google Patents

输电线的检测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2023174020A1
WO2023174020A1 PCT/CN2023/077537 CN2023077537W WO2023174020A1 WO 2023174020 A1 WO2023174020 A1 WO 2023174020A1 CN 2023077537 W CN2023077537 W CN 2023077537W WO 2023174020 A1 WO2023174020 A1 WO 2023174020A1
Authority
WO
WIPO (PCT)
Prior art keywords
point cloud
image data
candidate
data
target
Prior art date
Application number
PCT/CN2023/077537
Other languages
English (en)
French (fr)
Inventor
萧伟云
邓永成
李伟
陈志锐
王锦庆
邓春苗
Original Assignee
广东电网有限责任公司东莞供电局
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 广东电网有限责任公司东莞供电局 filed Critical 广东电网有限责任公司东莞供电局
Publication of WO2023174020A1 publication Critical patent/WO2023174020A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present application relates to the technical field of electric power, for example, to a detection method, device, computer equipment and storage medium for transmission lines.
  • Transmission lines in the power industry are mostly deployed outdoors and are directly exposed to various weather environments. In order to ensure normal power supply, technicians will regularly or irregularly (such as after typhoons and cold waves) use aircraft to collect image data along the transmission lines and use deep learning to detect them. Transmission lines, thereby screening transmission lines that may have abnormalities and submitting them for manual review.
  • This application proposes a method, device, computer equipment and storage medium for detecting transmission lines, which can solve the problem of low accuracy in detecting transmission lines using image data collected by aircraft.
  • an embodiment of the present application provides a method for detecting transmission lines, including:
  • Semantic recognition is performed on multiple frames of the target image data to detect power lines.
  • an embodiment of the present application also provides a detection device for a transmission line, including:
  • the tower determination module is used to determine multiple towers supporting transmission lines
  • a route planning module used to plan a non-linear route for the aircraft above between two adjacent towers, and the aircraft is equipped with a camera and a lidar;
  • a threshold generation module configured to generate a first threshold in depth for the transmission line along the route
  • a detection data receiving module configured to receive the original image data collected downward by the camera and the original point cloud data collected downward by the lidar when the aircraft flies along the route;
  • a point cloud filtering module is used to filter out the original point cloud data with a depth greater than or equal to the first threshold and obtain target point cloud data;
  • a point cloud projection module is used to project the target point cloud data into the original image data to obtain candidate image data
  • the image data screening module is used to screen multiple frames of candidate image data collected on the same object below at different angles as multi-frame target image data;
  • a semantic recognition module is used to perform semantic recognition on multiple frames of the target image data to detect power lines.
  • an embodiment of the present application further provides a computer device.
  • the computer device includes:
  • processors one or more processors
  • memory for storing one or more programs
  • the one or more processors When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the power transmission line detection method as described in the first aspect.
  • an embodiment of the present application further provides a computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium.
  • the computer program is executed by a processor, the power transmission as described in the first aspect is realized. Line detection method.
  • Figure 1 is a flow chart of a transmission line detection method provided in Embodiment 1 of the present application.
  • Figure 2 is an example diagram of a pole tower provided in Embodiment 1 of the present application.
  • Figure 3 is an example diagram of fitting the sagging of a transmission line provided in Embodiment 1 of the present application;
  • Figure 4 is a schematic structural diagram of a semantic recognition network provided by Embodiment 1 of the present application.
  • Figure 5 is a schematic structural diagram of a transmission line detection device provided in Embodiment 2 of the present application.
  • FIG. 6 is a schematic structural diagram of a computer device provided in Embodiment 3 of the present application.
  • Figure 1 is a flow chart of a transmission line detection method provided in Embodiment 1 of the present application.
  • This embodiment can be applied to the situation of transmitting point cloud data to image data to assist in detecting transmission lines.
  • This method can be based on transmission lines.
  • the detection device of the transmission line can be implemented by software and/or hardware, and can be configured in computer equipment, such as servers, workstations, personal computers, etc., specifically including the following steps:
  • Step 101 Determine multiple poles and towers supporting transmission lines.
  • Pole and Tower is a pole or tower-shaped structure that supports overhead transmission line conductors and overhead ground wires and maintains a certain distance between them and the earth.
  • the tower can adopt steel structure, wood structure and reinforced concrete structure.
  • wooden and reinforced concrete rod-shaped structures are called poles
  • tower-shaped steel structures and reinforced concrete chimney-shaped structures are called towers.
  • Towers without guy wires are called self-standing towers
  • towers with guy wires are called guy pole towers.
  • Pole towers can be classified into straight-line towers, straight-line corner towers, and tension-resistant corner towers. Among them, tension-resistant corner towers are divided into terminal towers, transposition towers, and divergence towers.
  • GIS Geographic Information System
  • Step 102 Plan a non-linear route for the aircraft above between two adjacent towers.
  • the order between the towers supporting the transmission line is determined according to the trend of the transmission line.
  • the order can be represented by a number, with the tower as the dividing point, and the transmission lines between each two adjacent towers are independently Perform testing.
  • the state of the transmission lines when looking down is also different, which may cause overlap in some cases.
  • the top and bottom are different.
  • Transmission lines are easy to overlap.
  • different transmission lines above and below are easy to overlap, etc. Therefore, when planning a flight route for an aircraft to fly above two adjacent towers, a non-linear route can be planned. route, so that transmission lines can be observed from multiple viewing angles to avoid false detections caused by overlap.
  • the original positioning data measured in advance on the root of each tower can be queried from GIS, where the original positioning data includes horizontal coordinates and the first vertical coordinate.
  • the height and the preset flight distance L2 are respectively added to obtain the second vertical coordinate, where the flight distance is the distance between the aircraft and the root of the tower.
  • the aircraft plans a route 241 that spans two adjacent target positioning data and deviates back and forth with the connection between the two adjacent target positioning data.
  • both sides of the connection line 242 between two adjacent target positioning data are the first side and the second side.
  • the so-called back-and-forth offset can mean that the route 241 is continuously planned from the first side to the second side, and then from The second side is planned to the first side, and then from the first side to the second side, and so on until the two towers are traversed.
  • Step 103 Generate a first threshold in depth for the transmission line along the route.
  • the distance between the transmission line and the route of the aircraft can be estimated, thereby generating a first threshold in depth of the transmission line along the flight route of the aircraft, that is, the transmission line is theoretically within the first threshold range .
  • the transmission line 230 between two adjacent towers is affected by gravity and may sag. Considering that it is outdoor The terrain is often uneven, especially in mountainous and hilly areas, so two adjacent towers are not on the same level, resulting in a more complicated situation of sagging transmission lines.
  • two adjacent towers can be placed on the same horizontal plane while maintaining the distance between the two adjacent towers.
  • the two adjacent towers are simplified to two points, namely , point M, point N, the two towers (i.e., point M, point N) are fitted with a polynomial curve 310 that conforms to the droop of the transmission line.
  • the curve 310 is relatively standardized. Easy to fit.
  • the maximum distance recorded between the support points on the tower and the top of the tower can be queried in GIS as the support distance L3.
  • the flight distance L2, the support distance L3, the separation distance L4 and the preset error distance are added to obtain the first threshold in depth.
  • Step 104 When the aircraft is flying along the route, the original image data collected by the camera and the original point cloud data collected by the lidar are called.
  • the aircraft is equipped with a camera and a Lidar.
  • the camera has a pan/tilt.
  • the pan/tilt can control the rotation of the camera. Since the distance between the aircraft and the power lines is relatively short, the distance between the Lidar and the power lines is smaller. Closer, therefore, the lidar is a single linear lidar, or a multi-linear lidar less than or equal to 8 lines. In this way, less sparse original point cloud data can be obtained, so that while maintaining accuracy, cut costs.
  • the aircraft While the aircraft is flying along the route, it can continuously control the rotation of the lidar, and collect original point cloud data downward during the rotation.
  • the original point cloud data and image data can be fused for perception.
  • the lidar scans into the visible range of the camera,
  • the camera is triggered by a specific synchronizer, the camera is called to collect original image data.
  • Step 105 Filter out original point cloud data whose depth is greater than or equal to the first threshold, and obtain target point cloud data.
  • the original point cloud data can be parsed to identify the depth, that is, the distance between the original point cloud data and the lidar, and the depth of the original point cloud data can be compared with the first threshold.
  • the depth of the original point cloud data is greater than or equal to the first threshold, it means that the original point cloud data is theoretically lower than the transmission line, such as point cloud data located on the ground. At this time, the original point cloud data can be filtered out.
  • the depth of the original point cloud data is less than or equal to the first threshold, it means that the original point cloud data may belong to a transmission line. At this time, the original point cloud data can be retained and recorded as target point cloud data for easy differentiation.
  • Step 106 Project the target point cloud data into the original image data to obtain candidate image data.
  • the camera and laser radar can be calibrated in advance, and the camera and laser can be identified.
  • the relationship between radars (such as rotation relationship, translation relationship), these relationships are used to project the target point cloud data into the original image data.
  • the original image data containing the target point cloud data is the candidate image data.
  • step 106 may include the following steps:
  • Step 1061 Cluster the target point cloud data to obtain multiple point cloud clusters.
  • methods such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise) can be used to cluster the target point cloud data to obtain multiple point cloud clusters.
  • DBSCAN Density-Based Spatial Clustering of Applications with Noise
  • Step 1062 Count the number of target point cloud data in the point cloud cluster.
  • For each point cloud cluster count the number of target point cloud data contained in it.
  • Step 1063 Filter out target point cloud data in part of the point cloud clusters based on quantity, project the target point cloud data in the remaining point cloud clusters into the original image data, and obtain candidate image data.
  • the transmission line is a long object with relatively dense point cloud data. Therefore, the number of each point cloud cluster can be used to analyze whether the point cloud cluster conforms to the theoretical point cloud distribution pattern of the transmission line, thereby identifying possible noise points and interference.
  • point cloud clusters of objects such as birds, floating garbage, etc.
  • filter out the target point cloud data in these point cloud clusters so that the remaining point cloud clusters that may be transmission lines are removed, and the targets in these point cloud clusters are
  • the point cloud data is projected into the original image data to obtain candidate image data.
  • the point cloud clusters can be divided into first candidate clusters, second candidate clusters, and third candidate clusters, where the number corresponding to the first candidate cluster is less than the second threshold, and the number corresponding to the second candidate cluster is greater than or Equal to the second threshold and less than the third threshold, the number corresponding to the third candidate cluster is greater than or equal to the third threshold, that is, the confidence that the first candidate cluster belongs to the transmission line is low enough, and the confidence that the third candidate cluster belongs to the transmission line is low enough. High enough, while the confidence level of the second candidate cluster belonging to the transmission line is average, and there is a certain probability of misjudgment.
  • the delay can be performed according to the trend of the third candidate cluster. If the third candidate cluster is extended and then passes through the second candidate cluster, and the distance between the third candidate cluster and the second candidate cluster is less than or equal to the fourth threshold, It means that the third candidate cluster and the second candidate cluster are close and the trend is consistent with belonging to a whole, then the new target point cloud data can be interpolated between the third candidate cluster and the second candidate cluster, so that the second candidate cluster and the new target Point cloud data are merged into the third candidate cluster.
  • the target point cloud data in the first candidate cluster and the target point cloud data in the second candidate cluster are filtered out, wherein the filtered out second candidate cluster is not merged into the third candidate cluster.
  • the target point cloud data in the third candidate cluster is projected into the original image data to obtain the candidate image data, which improves the recognition accuracy of the semantics of the target point cloud data, thereby improving the accuracy of detecting transmission lines.
  • Step 107 Screen multiple frames of candidate image data collected from the same object below at different angles as multi-frame target image data.
  • the candidate image data of the projected target point cloud data can be compared to filter out multiple frames of candidate image data collected for the same object below. Recorded as multi-frame target image data.
  • the scale-invariant feature transform operator SIFT Scale Invariant Feature Transform
  • SIFT Scale Invariant Feature Transform
  • the scale-invariant feature transform operator SIFT in the candidate image data is matched.
  • the candidate image data is determined to be the target image data, wherein the upper limit of the range is less than 1, and the difference between the upper limit and 1 is greater than the fifth threshold, indicating that the upper limit of the range The limit value is not close to 1.
  • the lower limit value of the range is greater than 0, and the difference between the lower limit value and 0 is greater than the sixth threshold, which means that the lower limit value of the range is not close to 0.
  • the two frames of candidate image data have a certain Similarity can ensure that it contains some identical objects, but it is not very similar and can ensure a certain angle difference.
  • Step 108 Perform semantic recognition on multiple frames of target image data to detect power lines.
  • the semantic recognition network can be loaded in the memory so that the semantic recognition network runs.
  • the semantic recognition network has multiple convolutional layers (Conv), reinforcement networks (Renforcing Block), multiple long short term memory networks (Long short term memory, LSTM), and three-dimensional generation networks (Integrating Block). ), the first fully connected layer (Fully connected Layers, FC) and the second fully connected layer FC.
  • each convolution layer a convolution operation is performed on each frame of target image data to obtain the first image feature.
  • the second image feature is obtained by marking the area where the successfully matched scale-invariant feature transform operator SIFT is located in the first image feature of each frame.
  • the second image features of each frame are processed to obtain the third image features.
  • multi-frame third image features are generated into three-dimensional fourth image features.
  • the fourth image feature is mapped to the fifth image feature.
  • the fifth image feature is mapped to the probability of belonging to the power transmission line.
  • the transmission line actually belongs to a 3D (three-dimensional) object. If some parts of the transmission line cannot be clearly understood from a specific viewpoint (for example, parts are reflected by other transmission lines, and parts are reflected), they can be viewed from other viewpoints. Find missing information. For a given view image, if there is a strategy for matching a region inside the given view image with corresponding regions in other views, the information of the given view can be enhanced by exploiting the relationship between the matched regions.
  • the assumption given in this embodiment is that connecting corresponding areas from different views and reasoning about the relationship between them can help the view better represent the 3D object.
  • the reinforcement network is responsible for exploring the relationship between regions to enhance the information of each individual view (first image feature), and the three-dimensional generation network is responsible for modeling the relationship from two-dimensional views to two-dimensional views in order to effectively integrate the information from Information about a single 3D view.
  • each spatial location in the feature map is a feature vector corresponding to a region in the image.
  • the reinforcement network can reuse information from filtering the target image data to find matching/related regions from other views and enhance the information in that region by leveraging cues from the matching regions. In this way, the view's information can be enhanced.
  • the 3D generative network employs a self-attention selection mechanism to generate an importance score for each view, which represents the relative discriminability of that view.
  • multiple towers supporting transmission lines are determined, and a non-linear route is planned for the aircraft above between two adjacent towers.
  • the aircraft is equipped with cameras and lidar, and the transmission lines are generated in depth along the route.
  • the receiving aircraft flies along the route, it calls the original image data collected downward by the camera, calls the original point cloud data collected downward by the lidar, and filters out the original point cloud data whose depth is greater than or equal to the first threshold, and obtains Target point cloud data, project the target point cloud data into the original image data, obtain candidate image data, screen the multi-frame candidate image data collected at different angles for the same object below, and use it as multi-frame target image data, for multi-frame target
  • the image data is subjected to semantic recognition to detect power lines.
  • the aircraft When the aircraft shoots transmission lines from the air, it filters out the point cloud data belonging to the ground background through depth, retains the point cloud data that may belong to the transmission lines with a high probability, and enhances the information of the image data, effectively alleviating the long tail effect. , reducing the generalization ability of deep learning Furthermore, combining the image data collected at multiple angles to detect wires can avoid the loss of image data caused by occlusion, reflection, etc., and improve the accuracy of detecting transmission lines.
  • FIG. 5 is a structural block diagram of a transmission line detection device provided in Embodiment 2 of the present application. Specifically, it may include the following modules:
  • the tower determination module 501 is used to determine multiple towers supporting transmission lines
  • the route planning module 502 is used to plan a non-linear route for the aircraft above between two adjacent towers.
  • the aircraft is equipped with a camera and a lidar;
  • a threshold generation module 503, configured to generate a first threshold in depth for the transmission line along the route;
  • the detection data receiving module 504 is used to receive the original image data collected downward by the camera and the original point cloud data collected downward by the lidar when the aircraft flies along the route;
  • the point cloud filtering module 505 is used to filter out the original point cloud data with a depth greater than or equal to the first threshold and obtain target point cloud data;
  • Point cloud projection module 506 is used to project the target point cloud data into the original image data to obtain candidate image data
  • the image data screening module 507 is used to screen multiple frames of the candidate image data collected from the same object below at different angles as multi-frame target image data;
  • the semantic recognition module 508 is used to perform semantic recognition on multiple frames of the target image data to detect power lines.
  • the route planning module 502 is also used to:
  • Plan a route for the aircraft that spans two adjacent target positioning data and deviates back and forth along a line between the two adjacent target positioning data.
  • the threshold generation module 503 is also used to:
  • the first threshold in depth is obtained by adding the flight distance, the support distance, the separation distance and the preset error distance.
  • the point cloud projection module 506 is also used to:
  • the point cloud projection module 506 is also used to:
  • the point cloud cluster is divided into a first candidate cluster, a second candidate cluster, and a third candidate cluster.
  • the number corresponding to the first candidate cluster is less than a second threshold, and the number corresponding to the second candidate cluster is greater than or equal to the second threshold and less than the third threshold, and the number corresponding to the third candidate cluster is greater than or equal to the third threshold;
  • the third candidate cluster Interpolate new target point cloud data between the cluster and the second candidate cluster, so that the second candidate cluster and the new target point cloud data are merged into the third candidate cluster;
  • the image data filtering module 507 is also used to:
  • the candidate image data is determined to be the target image data, wherein the upper limit of the range is less than 1, and the difference between the upper limit and 1 is greater than the fifth Threshold, the lower limit of the range is greater than 0, and the difference between the lower limit and 0 is greater than the sixth threshold.
  • the semantic recognition module 508 is also used to:
  • a semantic recognition network which has multiple convolutional layers, reinforcement networks, multiple long short-term memory networks, three-dimensional generation networks, a first fully connected layer and a second fully connected layer;
  • each convolution layer perform a convolution operation on the target image data of each frame to obtain the first image feature
  • the second image feature is obtained for the area where the scale-invariant feature transform operator SIFT that successfully matches the first image feature mark in each frame is located;
  • the second image features of each frame are processed to obtain third image features
  • the fifth image feature is mapped to a probability of belonging to a power transmission line.
  • the transmission line detection device provided by the embodiment of the present application can execute the transmission line detection method provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.
  • FIG. 6 is a schematic structural diagram of a computer device provided in Embodiment 3 of the present application. 6 illustrates a block diagram of an exemplary computer device 12 suitable for implementing embodiments of the present application.
  • the computer device 12 shown in FIG. 6 is only an example and should not bring any limitations to the functions and scope of use of the embodiments of the present application.
  • computer device 12 is embodied in the form of a general purpose computing device.
  • the components of computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and a bus 18 connecting various system components, including system memory 28 and processing unit 16.
  • Computer device 12 includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and nonvolatile media, removable and non-removable media.
  • System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32 .
  • Computer device 12 may further include other removable/non-removable, volatile/non-volatile computer system storage media.
  • storage system 34 may be used to read and write to non-removable, non-volatile magnetic media (not shown in Figure 6, commonly referred to as a "hard drive”).
  • a disk drive may be provided for reading and writing removable non-volatile disks (e.g., "floppy disks"), and for removable non-volatile optical disks (e.g., CD-ROM, DVD-ROM). or other optical media) that can read and write optical disc drives.
  • each drive may be connected to bus 18 through one or more data media interfaces.
  • System memory 28 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of each embodiment of the present application.
  • a program/utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28 Data, each of these examples or some combination may include an implementation of a network environment.
  • Program modules 42 generally perform functions and/or methods in the embodiments described herein.
  • Computer device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), with one or more devices that enable a user to interact with computer device 12, and/or with Any device (eg, network card, modem, etc.) that enables the computer device 12 to communicate with one or more other computing devices. This communication may occur through input/output (I/O) interface 22.
  • computer device 12 may also communicate with one or more networks (eg, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through network adapter 20. As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18 .
  • network adapter 20 communicates with other modules of computer device 12 via bus 18 .
  • the processing unit 16 executes a variety of functional applications and data processing by running programs stored in the system memory 28 , for example, implementing the transmission line detection method provided in the embodiment of the present application.
  • Embodiment 4 of the present application also provides a computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium.
  • the computer program is executed by a processor, each process of the above-mentioned transmission line detection method is implemented, and the same can be achieved. The technical effects will not be repeated here to avoid repetition.
  • the computer-readable storage medium may include, for example, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination thereof. More specific examples (non-exhaustive list) of computer readable storage media include: electrical connections having one or more conductors, portable computer disks, hard drives, random access memory (RAM), read only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • a computer-readable storage medium may be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

一种输电线的检测方法、装置、计算机设备和存储介质，该方法包括：确定多个支撑输电线的杆塔，在相邻两个杆塔之间的上方对飞行器规划非直线的路线，沿路线对输电线生成在深度上的第一阈值，接收飞行器沿路线飞行时，调用摄像头向下方采集的原始图像数据、调用激光雷达向下方采集的原始点云数据，滤除深度大于或等于第一阈值的原始点云数据，获得目标点云数据，对目标点云数据投影至原始图像数据中，获得候选图像数据，筛选在不同角度下对下方相同物体采集的多帧候选图像数据，作为多帧目标图像数据，对多帧目标图像数据进行语义识别，以检测出输电线。

Description

输电线的检测方法、装置、计算机设备和存储介质
本申请要求申请日为2022年03月18日、申请号为202210267043.0的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及电力的技术领域，例如涉及一种输电线的检测方法、装置、计算机设备和存储介质。
背景技术
电力行业的输电线多部署在户外，直接面临各种天气环境，为保证电力正常供应，技术人员会定期或不定期(如台风、寒潮过后)使用飞行器沿输电线采集图像数据，使用深度学习检测输电线，从而筛选可能发生异常的输电线提交人工复核。
但是，飞行器从空中往下拍摄输电线，地面的背景很复杂，树木、草丛、砂砾、河流、道路等等，长尾效应尤为明显，这对深度学习的泛化能力要求很高，但实际上很难将户外所有的场景都采集到作为样本，这导致检测输电线的精确度低。
发明内容
本申请提出了一种输电线的检测方法、装置、计算机设备和存储介质，能够解决使用飞行器采集的图像数据检测输电线的精确度低的问题。
第一方面，本申请一实施例提供了一种输电线的检测方法，包括：
确定多个支撑输电线的杆塔；
在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，所述飞行器搭载有摄像头和激光雷达；
沿所述路线对所述输电线生成在深度上的第一阈值；
接收所述飞行器沿所述路线飞行时，调用所述摄像头向下方采集的原始图像数据、调用所述激光雷达向下方采集的原始点云数据；
滤除深度大于或等于所述第一阈值的所述原始点云数据，获得目标点云数据；
对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据；
筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据；
对多帧所述目标图像数据进行语义识别，以检测出输电线。
第二方面，本申请一实施例还提供了一种输电线的检测装置，包括：
杆塔确定模块，用于确定多个支撑输电线的杆塔；
路线规划模块，用于在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，所述飞行器搭载有摄像头和激光雷达；
阈值生成模块，用于沿所述路线对所述输电线生成在深度上的第一阈值；
检测数据接收模块，用于接收所述飞行器沿所述路线飞行时，调用所述摄像头向下方采集的原始图像数据、调用所述激光雷达向下方采集的原始点云数据；
点云滤除模块，用于滤除深度大于或等于所述第一阈值的所述原始点云数据，获得目标点云数据；
点云投影模块，用于对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据；
图像数据筛选模块，用于筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据；
语义识别模块，用于对多帧所述目标图像数据进行语义识别，以检测出输电线。
第三方面，本申请一实施例还提供了一种计算机设备，所述计算机设备包括：
一个或多个处理器；
存储器，用于存储一个或多个程序，
当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如第一方面所述的输电线的检测方法。
第四方面，本申请一实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质上存储计算机程序，所述计算机程序被处理器执行时实现如第一方面所述的输电线的检测方法。
附图说明
图1为本申请实施例一提供的一种输电线的检测方法的流程图；
图2为本申请实施例一提供的一种杆塔的示例图；
图3为本申请实施例一提供的一种拟合输电线的下垂的示例图；
图4是本申请实施例一提供的一种语义识别网络的结构示意图；
图5为本申请实施例二提供的一种输电线的检测装置的结构示意图；
图6为本申请实施例三提供的一种计算机设备的结构示意图。
具体实施方式
实施例一
图1为本申请实施例一提供的一种输电线的检测方法的流程图，本实施例可适用于将点云数据透射到图像数据上、辅助检测输电线的情况，该方法可以由输电线的检测装置来执行，该输电线的检测装置可以由软件和/或硬件实现，可配置在计算机设备中，例如，服务器、工作站、个人电脑，等等，具体包括如下步骤：
步骤101、确定多个支撑输电线的杆塔。
杆塔(Pole and Tower)是支承架空输电线导线和架空地线并使它们之间以及与大地之间保持一定距离的杆形或塔形构筑物。杆塔可采用钢结构、木结构和钢筋混凝土结构。通常对木和钢筋混凝土的杆形结构称为杆，塔形的钢结构和钢筋混凝土烟囱形结构称为塔。不带拉线的杆塔称为自立式杆塔，带拉线的杆塔称为拉线杆塔。
对杆塔进行分类，可以划分为直线塔、直线转角塔、耐张转角塔，其中，耐张转角塔又分为终端塔、换位塔、分歧塔等形式。
这些杆塔的信息已在建造、检修的过程中，记录在GIS(Geographic Information System，地理信息系统)中，在检测输电线时，可以从GIS中读取杆塔的相关信息。
步骤102、在相邻两个杆塔之间的上方对飞行器规划非直线的路线。
在本实施例中，按照输电线的走势确定支撑该输电线的杆塔之间的顺序，该顺序可以以编号表示，以杆塔作为切分点，独立对每相邻两个杆塔之间的输电线进行检测。
考虑到不同类型的杆塔支撑输电线的方式有所不同，俯视输电线时的状态也有所不同，部分情况会造成重叠，例如，俯视带回路的换位塔时，上下不同 的输电线容易重叠，俯视带回路的直线塔时，上下不同的输电线容易重叠，等等，因此，对飞行器规划在相邻两个杆塔之间的上方飞行的路线时，可以规划非直线的路线，从而可以在多个视角观察到输电线，避免重叠造成误检测。
在具体实现中，如图2所示，可以从GIS中查询预先对每个杆塔(如图2中的塔杆210、塔杆220)的根部测量的原始定位数据，其中，原始定位数据包括水平坐标和第一垂直坐标。
从GIS中查询对每个杆塔记录的杆塔高度L1，即从根部到顶部之间的距离。
在第一垂直坐标L1的基础上分别加上高度与预设的飞行距离L2，获得第二垂直坐标，其中，飞行距离为飞行器与杆塔的根部之间的距离。
将第二垂直坐标替代第一垂直坐标，获得目标定位数据(如图2中的点I、点J)。
通过Dijkstra算法、A*算法等方法对飞行器规划跨越相邻两个目标定位数据、且以相邻两个目标定位数据之间的连线来回偏离的路线241。
其中，设相邻两个目标定位数据之间的连线242两侧为第一侧、第二侧，所谓来回偏移，可以指路线241不断地从第一侧规划到第二侧，然后从第二侧规划到第一侧，再从第一侧规划到第二侧，如此循环，直至遍历完两个杆塔。
步骤103、沿路线对输电线生成在深度上的第一阈值。
在本实施例中，可以预估输电线与飞行器的路线之间的距离，从而沿飞行器飞行的路线生成输电线在深度上的第一阈值，即，输电线理论上在该第一阈值范围内。
在具体实现中，如图2所示，相邻两个杆塔(如图2中的塔杆210、塔杆220)之间的输电线230受重力影响，会出现下垂的情况，考虑到在户外地势多不平坦，尤其是山区、丘陵地区，因而相邻两个杆塔并不在同一水平面，导致输电线下垂的情况较为复杂。
针对这种情况，经过实验，相邻两个杆塔在同一水平面时输电线下垂的情况与相邻两个杆塔不在同一水平面时输电线下垂的情况是相近的，因此，为了化简计算，如图3所示，可以在维持相邻两个杆塔之间的距离不变的条件下，将相邻两个杆塔置于同一水平面，为便于理解，相邻两个杆塔化简为2个点，即，点M、点N，通过多项式对两个杆塔(即点M、点N)拟合符合输电线下垂的曲线310，在相邻两个杆塔置于同一水平面的情况下，曲线310较为规范，容易拟合。
计算两个杆塔(即点M、点N)之间连线与曲线相隔最大的距离，作为间隔 距离L4，这个间隔距离与真实的间隔距离存在一定的误差，这个间隔距离可能小于真实的间隔距离、也可能大于真实的间隔距离，对该误差可以设置误差距离进行覆盖，即，通过误差距离对间隔距离适当增加，保证添加误差距离后的间隔距离可以覆盖真实的间隔距离，额外增加少许的深度，对于检测悬挂较高的输电线并不造成明显的影响，依然可以保证检测输电线的准确度。
此时，如图2所示，可以在GIS中查询对杆塔上的支撑点记录的、与塔杆顶部相隔最大的距离，作为支撑距离L3。
如图2与图3所示，将飞行距离L2、支撑距离L3、间隔距离L4与预设的误差距离相加，获得深度上的第一阈值。
步骤104、接收飞行器沿路线飞行时，调用摄像头向下方采集的原始图像数据、调用激光雷达向下方采集的原始点云数据。
在本实施例中，飞行器搭载有摄像头和激光雷达Lidar，摄像头具有云台，云台可以控制摄像头旋转，而由于飞行器与输电线之间的距离较近，使得激光雷达与输电线之间的距离较近，因此，激光雷达为单线性激光雷达，或者，小于或等于8线的多线性激光雷达，这样子也能获得不太稀疏的原始点云数据，使得在保持精确性的情况下，可以降低成本。
飞行器在沿路线飞行的过程中，可以持续控制激光雷达旋转，在旋转的过程中向下采集原始点云数据，原始点云数据与图像数据可融合感知，当激光雷达扫描到摄像头的可视范围时，由特定的同步器触发调用该摄像头采集原始图像数据。
步骤105、滤除深度大于或等于第一阈值的原始点云数据，获得目标点云数据。
在本实施例中，可以对原始点云数据进行解析，识别深度，即，原始点云数据与激光雷达之间的距离，将这些原始点云数据的深度与第一阈值进行比较。
如果原始点云数据的深度大于或等于第一阈值，表示该原始点云数据理论上低于输电线，如位于地面上的点云数据，此时，可以滤除该原始点云数据。
如果原始点云数据的深度小于或等于第一阈值，表示该原始点云数据可能属于输电线，此时，可以保留该原始点云数据，为便于区分，记为目标点云数据。
步骤106、对目标点云数据投影至原始图像数据中，获得候选图像数据。
在实施例中，可以预先对摄像头与激光雷达进行标定，识别摄像头与激光 雷达之间的关系(如旋转关系、平移关系)，利用这些关系将对目标点云数据投影至原始图像数据中，为便于区分，包含目标点云数据的原始图像数据为候选图像数据。
在本申请的一个实施例中，步骤106可以包括如下步骤：
步骤1061、对目标点云数据进行聚类，获得多个点云簇。
在本实施例中，可以使用DBSCAN(Density-Based Spatial Clustering of Applications with Noise，具有噪声的基于密度的聚类方法)等方法对目标点云数据进行聚类，得到多个点云簇。
步骤1062、统计点云簇中目标点云数据的数量。
对每个点云簇，统计其中包含的目标点云数据的数量。
步骤1063、基于数量滤除部分点云簇中的目标点云数据，将剩余的点云簇中的目标点云数据投影至原始图像数据中，获得候选图像数据。
输电线属于一个长条形的物体，点云数据较为密集，因此，通过各个点云簇的数量可以分析点云簇是否符合输电线理论上的点云分布的规律，从而识别可能属于噪点、干扰物(如鸟类、漂浮的垃圾等)的点云簇，并滤除这些点云簇中的目标点云数据，从而剩余可能为输电线的点云簇，从而将这些点云簇中的目标点云数据投影至原始图像数据中，获得候选图像数据。
在一个示例中，可以将点云簇划分为第一候选簇、第二候选簇、第三候选簇，其中，第一候选簇对应的数量小于第二阈值，第二候选簇对应的数量大于或等于第二阈值、且小于第三阈值，第三候选簇对应的数量大于或等于第三阈值，即，第一候选簇属于输电线的置信度足够低，第三候选簇属于输电线的置信度足够高，而第二候选簇属于输电线的置信度一般，存在一定误判的几率。
此时，可以按照第三候选簇的走势进行延时，若延伸第三候选簇之后经过第二候选簇、且第三候选簇与第二候选簇之间间隔的距离小于或等于第四阈值，表示第三候选簇与第二候选簇相近且走势符合属于一个整体，则可以对第三候选簇与第二候选簇之间插值新的目标点云数据，以使第二候选簇与新的目标点云数据合并至第三候选簇中。
滤除第一候选簇中的目标点云数据和第二候选簇中的目标点云数据，其中，被滤除的第二候选簇未合并至第三候选簇。
将第三候选簇中的目标点云数据投影至原始图像数据中，获得候选图像数据，提高了目标点云数据语义的识别精确度，从而提高了检测输电线的精确度。
步骤107、筛选在不同角度下对下方相同物体采集的多帧候选图像数据，作为多帧目标图像数据。
由于飞行器是按照非直线的路线飞行，针对输电线可以采集不同角度的图像数据，因此，可以对比已投影目标点云数据的候选图像数据，从中筛选对下方相同物体采集的多帧候选图像数据，记为多帧目标图像数据。
在具体实现中，可以在每帧候选图像数据中提取表征物体的尺度不变特征变换算子SIFT(Scale Invariant Feature Transform)。
对多帧候选图像数据添加时间的窗口，如5秒。
在时间的窗口内，对候选图像数据中的尺度不变特征变换算子SIFT进行匹配。
统计匹配成功的尺度不变特征变换算子SIFT的占比。
若占比位于预设的范围内，则确定候选图像数据为目标图像数据，其中，该范围的上限值小于1、且上限值与1的差值大于第五阈值，表示该范围的上限值并非接近1，范围的下限值大于0、且下限值与0的差值大于第六阈值，表示该范围的下限值并非接近0，这样子，两帧候选图像数据具有一定的相似性，可以保证包含了一些相同的物体，但又并非十分相似，可以保证一定的角度差异。
步骤108、对多帧目标图像数据进行语义识别，以检测出输电线。
多帧目标图像数据之间存在一定的相同之处、也存在一定的差异之处，可以相互对照进行语义识别，从而检测出输电线。
在具体实现中，可以在内存中加载语义识别网络，使得语义识别网络运行。
如图4所示，语义识别网络具有多个卷积层(Convolutional layer，Conv)、加强网络(Renforcing Block)、多个长短期记忆网络(Long short term memory，LSTM)、三维生成网络(Integrating Block)、第一全连接层(Fully connected Layers，FC)和第二全连接层FC。
在每个卷积层中，对每帧目标图像数据执行卷积操作，获得第一图像特征。
在加强网络中，对每帧第一图像特征标记匹配成功的尺度不变特征变换算子SIFT所处的区域，获得第二图像特征。
在每个长短期记忆网络中，对每帧第二图像特征进行处理，获得第三图像特征。
在三维生成网络中，将多帧第三图像特征生成三维的第四图像特征。
在第一全连接层中，将第四图像特征映射为第五图像特征。
在第二全连接层中，将第五图像特征映射为属于输电线的概率。
其中，输电线在实际上属于一个3D(三维)的对象，输电线的某些部分不能从一个特定视点清晰理解的情况下(例如，部分被其他输电线、部分存在反射)，可以从其他视点找到缺失的信息。对于给定的视图图像，如果存在匹配给定的视图图像内部的区域和其他视图中的对应区域的策略，则可以通过利用匹配区域之间的关系来增强该给定视图的信息。
在3D对象的几个部分从某些视点完全不可见的情况下，仅仅建模区域到区域的关系不能进一步帮助那些视图获得关于不可见部分的信息。因而对视图到视图的关系进行建模，以确定每个视图的辨别能力，并进一步整合这些视图以获得最终的3D对象的描述符。
对此，本实施例给定的假设，是连接来自不同视图的对应区域并推理它们之间的关系可以帮助视图更好地表征3D对象。
加强网络负责探索区域与区域之间的关系，以强化每个单独视图(第一图像特征)的信息，三维生成网络负责建模二维的视图到二维的视图的关系，以便有效地集成来自单个三维视图的信息。
具体而言，对于给定视图图像的特征图，特征图中的每个空间位置是对应于图像中一个区域的特征向量。对于给定视图图像中的每个区域，加强网络可以复用筛选目标图像数据时的信息从其他视图中找到匹配/相关区域，并通过利用来自匹配区域的线索来增强该区域的信息。这样，视图的信息可以得到加强。之后，三维生成网络采用自关注选择机制来为每个视图生成重要性分数，这表示该视图的相对辨别能力。
在本实施例中，确定多个支撑输电线的杆塔，在相邻两个杆塔之间的上方对飞行器规划非直线的路线，飞行器搭载有摄像头和激光雷达，沿路线对输电线生成在深度上的第一阈值，接收飞行器沿路线飞行时，调用摄像头向下方采集的原始图像数据、调用激光雷达向下方采集的原始点云数据，滤除深度大于或等于第一阈值的原始点云数据，获得目标点云数据，对目标点云数据投影至原始图像数据中，获得候选图像数据，筛选在不同角度下对下方相同物体采集的多帧候选图像数据，作为多帧目标图像数据，对多帧目标图像数据进行语义识别，以检测出输电线。飞行器从空中往下拍摄输电线时，通过深度滤除了属于地面背景的点云数据，保留了较大概率可能属于输电线的点云数据，对图像数据的信息进行增强，有效缓解了长尾效应，降低了对深度学习的泛化能力的 要求，并且，联合多个角度下采集的图像数据检测电线，可以避免遮挡、反光等情况造成图像数据的信息丢失，提高了检测输电线的精确度。
实施例二
图5为本申请实施例二提供的一种输电线的检测装置的结构框图，具体可以包括如下模块：
杆塔确定模块501，用于确定多个支撑输电线的杆塔；
路线规划模块502，用于在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，所述飞行器搭载有摄像头和激光雷达；
阈值生成模块503，用于沿所述路线对所述输电线生成在深度上的第一阈值；
检测数据接收模块504，用于接收所述飞行器沿所述路线飞行时，调用所述摄像头向下方采集的原始图像数据、调用所述激光雷达向下方采集的原始点云数据；
点云滤除模块505，用于滤除深度大于或等于所述第一阈值的所述原始点云数据，获得目标点云数据；
点云投影模块506，用于对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据；
图像数据筛选模块507，用于筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据；
语义识别模块508，用于对多帧所述目标图像数据进行语义识别，以检测出输电线。
在本申请的一个实施例中，所述路线规划模块502还用于：
查询预先对每个所述杆塔的根部测量的原始定位数据，所述原始定位数据包括水平坐标和第一垂直坐标；
查询对每个所述杆塔记录的所述杆塔高度；
在所述第一垂直坐标的基础上分别加上所述高度与预设的飞行距离，获得第二垂直坐标；
将所述第二垂直坐标替代所述第一垂直坐标，获得目标定位数据；
对飞行器规划跨越相邻两个所述目标定位数据、且以相邻两个所述目标定位数据之间的连线来回偏离的路线。
在本申请的一个实施例中，所述阈值生成模块503还用于：
在维持相邻两个所述杆塔之间的距离不变的条件下，将相邻两个所述杆塔置于同一水平面；
对两个所述杆塔拟合符合所述输电线下垂的曲线；
计算两个所述杆塔之间连线与所述曲线相隔最大的距离，作为间隔距离；
查询对所述杆塔上的支撑点记录的、与所述塔杆顶部相隔最大的距离，作为支撑距离；
将所述飞行距离、所述支撑距离、所述间隔距离与预设的误差距离相加，获得深度上的第一阈值。
在本申请的一个实施例中，所述点云投影模块506还用于：
对所述目标点云数据进行聚类，获得多个点云簇；
统计所述点云簇中所述目标点云数据的数量；
基于所述数量滤除部分所述点云簇中的所述目标点云数据，将剩余的所述点云簇中的所述目标点云数据投影至所述原始图像数据中，获得候选图像数据。
在本申请的一个实施例中，所述点云投影模块506还用于：
将所述点云簇划分为第一候选簇、第二候选簇、第三候选簇，所述第一候选簇对应的所述数量小于第二阈值，所述第二候选簇对应的所述数量大于或等于第二阈值、且小于第三阈值，所述第三候选簇对应的所述数量大于或等于第三阈值；
若延伸所述第三候选簇之后经过所述第二候选簇、且所述第三候选簇与所述第二候选簇之间间隔的距离小于或等于第四阈值，则对所述第三候选簇与所述第二候选簇之间插值新的目标点云数据，以使所述第二候选簇与新的所述目标点云数据合并至所述第三候选簇中；
滤除所述第一候选簇中的所述目标点云数据和所述第二候选簇中的所述目标点云数据；
将所述第三候选簇中的所述目标点云数据投影至所述原始图像数据中，获得候选图像数据。
在本申请的一个实施例中，所述图像数据筛选模块507还用于：
在每帧所述候选图像数据中提取表征物体的尺度不变特征变换算子SIFT；
对多帧所述候选图像数据添加时间的窗口；
在所述时间的窗口内，对所述候选图像数据中的所述尺度不变特征变换算子SIFT进行匹配；
统计匹配成功的所述尺度不变特征变换算子SIFT的占比；
若所述占比位于预设的范围内，则确定所述候选图像数据为目标图像数据，其中，所述范围的上限值小于1、且所述上限值与1的差值大于第五阈值，所述范围的下限值大于0、且所述下限值与0的差值大于第六阈值。
在本申请的一个实施例中，所述语义识别模块508还用于：
加载语义识别网络，所述语义识别网络具有多个卷积层、加强网络、多个长短期记忆网络、三维生成网络、第一全连接层和第二全连接层；
在每个所述卷积层中，对每帧所述目标图像数据执行卷积操作，获得第一图像特征；
在所述加强网络中，对每帧所述第一图像特征标记匹配成功的所述尺度不变特征变换算子SIFT所处的区域，获得第二图像特征；
在每个所述长短期记忆网络中，对每帧所述第二图像特征进行处理，获得第三图像特征；
在所述三维生成网络中，将多帧所述第三图像特征生成三维的第四图像特征；
在所述第一全连接层中，将所述第四图像特征映射为第五图像特征；
在所述第二全连接层中，将所述第五图像特征映射为属于输电线的概率。
本申请实施例所提供的输电线的检测装置可执行本发明任意实施例所提供的输电线的检测方法，具备执行方法相应的功能模块和有益效果。
实施例三
图6为本申请实施例三提供的一种计算机设备的结构示意图。图6示出了适于用来实现本申请实施方式的示例性计算机设备12的框图。图6显示的计算机设备12仅仅是一个示例，不应对本申请实施例的功能和使用范围带来任何限制。
如图6所示，计算机设备12以通用计算设备的形式表现。计算机设备12的组件可以包括但不限于：一个或者多个处理器或者处理单元16，系统存储器28，连接不同系统组件(包括系统存储器28和处理单元16)的总线18。
总线18表示几类总线结构中的一种或多种，包括存储器总线或者存储器控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构包括但不限于工业标准体系结构(ISA)总线，微通道体系结构(MAC)总线，增强型ISA总线、视频电子标准协会(VESA) 局域总线以及外围组件互连(PCI)总线。
计算机设备12包括多种计算机系统可读介质。这些介质可以是任何能够被计算机设备12访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。
系统存储器28可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(RAM)30和/或高速缓存存储器32。计算机设备12可以进一步包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储系统34可以用于读写不可移动的、非易失性磁介质(图6未显示，通常称为“硬盘驱动器”)。尽管图6中未示出，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘(例如CD-ROM,DVD-ROM或者其它光介质)读写的光盘驱动器。在这些情况下，每个驱动器可以通过一个或者多个数据介质接口与总线18相连。系统存储器28可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程序模块被配置以执行本申请每一实施例的功能。
具有一组(至少一个)程序模块42的程序/实用工具40，可以存储在例如系统存储器28中，这样的程序模块42包括但不限于操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块42通常执行本申请所描述的实施例中的功能和/或方法。
计算机设备12也可以与一个或多个外部设备14(例如键盘、指向设备、显示器24等)通信，还可与一个或者多个使得用户能与该计算机设备12交互的设备通信，和/或与使得该计算机设备12能与一个或多个其它计算设备进行通信的任何设备(例如网卡，调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口22进行。并且，计算机设备12还可以通过网络适配器20与一个或者多个网络(例如局域网(LAN)，广域网(WAN)和/或公共网络，例如因特网)通信。如图所示，网络适配器20通过总线18与计算机设备12的其它模块通信。应当明白，尽管图中未示出，可以结合计算机设备12使用其它硬件和/或软件模块，包括但不限于：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAI D系统、磁带驱动器以及数据备份存储系统等。
处理单元16通过运行存储在系统存储器28中的程序，从而执行多种功能应用以及数据处理，例如实现本申请实施例所提供的输电线的检测方法。
实施例四
本申请实施例四还提供一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时实现上述输电线的检测方法的每一个过程，且能达到相同的技术效果，为避免重复，这里不再赘述。
其中，计算机可读存储介质例如可以包括但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。

Claims (10)

  1. 一种输电线的检测方法，包括：
    确定多个支撑输电线的杆塔；
    在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，所述飞行器搭载有摄像头和激光雷达；
    沿所述路线对所述输电线生成在深度上的第一阈值；
    接收所述飞行器沿所述路线飞行时，调用所述摄像头向下方采集的原始图像数据、调用所述激光雷达向下方采集的原始点云数据；
    滤除深度大于或等于所述第一阈值的所述原始点云数据，获得目标点云数据；
    对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据；及
    筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据；
    对多帧所述目标图像数据进行语义识别，以检测出输电线。
  2. 根据权利要求1所述的方法，其中，所述在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，包括：
    查询预先对每个所述杆塔的根部测量的原始定位数据，所述原始定位数据包括水平坐标和第一垂直坐标；
    查询对每个所述杆塔记录的所述杆塔高度；
    在所述第一垂直坐标的基础上分别加上所述高度与预设的飞行距离，获得第二垂直坐标；
    将所述第二垂直坐标替代所述第一垂直坐标，获得目标定位数据；及
    对飞行器规划跨越相邻两个所述目标定位数据、且以相邻两个所述目标定位数据之间的连线来回偏离的路线。
  3. 根据权利要求2所述的方法，其中，所述沿所述路线对所述输电线生成在深度上的第一阈值，包括：
    在维持相邻两个所述杆塔之间的距离不变的条件下，将相邻两个所述杆塔置于同一水平面；
    对两个所述杆塔拟合符合所述输电线下垂的曲线；
    计算两个所述杆塔之间连线与所述曲线相隔最大的距离，作为间隔距离；
    查询对所述杆塔上的支撑点记录的、与所述塔杆顶部相隔最大的距离，作为支撑距离；及
    将所述飞行距离、所述支撑距离、所述间隔距离与预设的误差距离相加，获得深度上的第一阈值。
  4. 根据权利要求1所述的方法，其中，所述对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据，包括：
    对所述目标点云数据进行聚类，获得多个点云簇；
    统计所述点云簇中所述目标点云数据的数量；及
    基于所述数量滤除部分所述点云簇中的所述目标点云数据，将剩余的所述点云簇中的所述目标点云数据投影至所述原始图像数据中，获得候选图像数据。
  5. 根据权利要求4所述的方法，其中，所述基于所述数量滤除部分所述点云簇中的所述目标点云数据，将剩余的所述点云簇中的所述目标点云数据投影至所述原始图像数据中，获得候选图像数据，包括：
    将所述点云簇划分为第一候选簇、第二候选簇、第三候选簇，所述第一候选簇对应的所述数量小于第二阈值，所述第二候选簇对应的所述数量大于或等于第二阈值、且小于第三阈值，所述第三候选簇对应的所述数量大于或等于第三阈值；
    若延伸所述第三候选簇之后经过所述第二候选簇、且所述第三候选簇与所述第二候选簇之间间隔的距离小于或等于第四阈值，则对所述第三候选簇与所述第二候选簇之间插值新的目标点云数据，以使所述第二候选簇与新的所述目标点云数据合并至所述第三候选簇中；
    滤除所述第一候选簇中的所述目标点云数据和所述第二候选簇中的所述目标点云数据；
    将所述第三候选簇中的所述目标点云数据投影至所述原始图像数据中，获得候选图像数据。
  6. 根据权利要求1-5中任一项所述的方法，其中，所述筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据，包括：
    在每帧所述候选图像数据中提取表征物体的尺度不变特征变换算子SIFT；
    对多帧所述候选图像数据添加时间的窗口；
    在所述时间的窗口内，对所述候选图像数据中的所述尺度不变特征变换算子SIFT进行匹配；
    统计匹配成功的所述尺度不变特征变换算子SIFT的占比；
    若所述占比位于预设的范围内，则确定所述候选图像数据为目标图像数据， 其中，所述范围的上限值小于1、且所述上限值与1的差值大于第五阈值，所述范围的下限值大于0、且所述下限值与0的差值大于第六阈值。
  7. 根据权利要求6所述的方法，其中，所述对多帧所述目标图像数据进行语义识别，以检测出输电线，包括：
    加载语义识别网络，所述语义识别网络具有多个卷积层、加强网络、多个长短期记忆网络、三维生成网络、第一全连接层和第二全连接层；
    在每个所述卷积层中，对每帧所述目标图像数据执行卷积操作，获得第一图像特征；
    在所述加强网络中，对每帧所述第一图像特征标记匹配成功的所述尺度不变特征变换算子SIFT所处的区域，获得第二图像特征；
    在每个所述长短期记忆网络中，对每帧所述第二图像特征进行处理，获得第三图像特征；
    在所述三维生成网络中，将多帧所述第三图像特征生成三维的第四图像特征；
    在所述第一全连接层中，将所述第四图像特征映射为第五图像特征；
    在所述第二全连接层中，将所述第五图像特征映射为属于输电线的概率。
  8. 一种输电线的检测装置，包括：
    杆塔确定模块，用于确定多个支撑输电线的杆塔；
    路线规划模块，用于在相邻两个所述杆塔之间的上方对飞行器规划非直线的路线，所述飞行器搭载有摄像头和激光雷达；
    阈值生成模块，用于沿所述路线对所述输电线生成在深度上的第一阈值；
    检测数据接收模块，用于接收所述飞行器沿所述路线飞行时，调用所述摄像头向下方采集的原始图像数据、调用所述激光雷达向下方采集的原始点云数据；
    点云滤除模块，用于滤除深度大于或等于所述第一阈值的所述原始点云数据，获得目标点云数据；
    点云投影模块，用于对所述目标点云数据投影至所述原始图像数据中，获得候选图像数据；
    图像数据筛选模块，用于筛选在不同角度下对下方相同物体采集的多帧所述候选图像数据，作为多帧目标图像数据；及
    语义识别模块，用于对多帧所述目标图像数据进行语义识别，以检测出输电线。
  9. 一种计算机设备，包括：
    一个或多个处理器；
    存储器，用于存储一个或多个程序，
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-7中任一项所述的输电线的检测方法。
  10. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储计算机程序，所述计算机程序被处理器执行时实现如权利要求1-7中任一项所述的输电线的检测方法。
PCT/CN2023/077537 2022-03-18 2023-02-22 输电线的检测方法、装置、计算机设备和存储介质 WO2023174020A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210267043.0 2022-03-18
CN202210267043.0A CN114359758B (zh) 2022-03-18 2022-03-18 一种输电线的检测方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2023174020A1 true WO2023174020A1 (zh) 2023-09-21

Family

ID=81094987

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/077537 WO2023174020A1 (zh) 2022-03-18 2023-02-22 输电线的检测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN114359758B (zh)
WO (1) WO2023174020A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117092641A (zh) * 2023-10-20 2023-11-21 江苏翰林正川工程技术有限公司 一种基于雷达测距与图像识别的信息融合方法
CN117314022A (zh) * 2023-11-29 2023-12-29 深圳金三立视频科技股份有限公司 一种输电线路隐患分析方法及装置
CN117634968A (zh) * 2023-12-04 2024-03-01 国网江苏省电力有限公司建设分公司 基于卫星互联网与云图数据集成的输变电工程定验方法
CN117314022B (zh) * 2023-11-29 2024-05-28 深圳金三立视频科技股份有限公司 一种输电线路隐患分析方法及装置

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114359758B (zh) * 2022-03-18 2022-06-14 广东电网有限责任公司东莞供电局 一种输电线的检测方法、装置、计算机设备和存储介质
CN114972824B (zh) * 2022-06-24 2023-07-14 小米汽车科技有限公司 杆件检测方法、装置、车辆和存储介质
CN115830010B (zh) * 2023-02-07 2023-06-27 南方电网数字电网研究院有限公司 输电导线交叉跨越距离检测方法、装置和计算机设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107767389A (zh) * 2017-10-27 2018-03-06 云南电网有限责任公司电力科学研究院 一种基于机载激光点云数据的输电线自动提取方法
CN109146836A (zh) * 2017-09-19 2019-01-04 上海华测导航技术股份有限公司 一种基于点云数据进行输电塔及线路检测的系统及方法
CN111242080A (zh) * 2020-01-21 2020-06-05 南京航空航天大学 一种基于双目相机与深度相机的输电线识别与定位方法
CN111709343A (zh) * 2020-06-09 2020-09-25 广州文远知行科技有限公司 一种点云检测方法、装置、计算机设备和存储介质
US20210224613A1 (en) * 2017-11-29 2021-07-22 Beijing Greenvalley Technology Co., Ltd. Method, Apparatus, and Device for Classifying LiDAR Point Cloud Data, and Storage Medium
CN114359758A (zh) * 2022-03-18 2022-04-15 广东电网有限责任公司东莞供电局 一种输电线的检测方法、装置、计算机设备和存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102590823B (zh) * 2012-01-06 2014-05-07 中国测绘科学研究院 一种机载lidar数据电力线快速提取及重构方法
CN109254586A (zh) * 2018-09-19 2019-01-22 绵阳紫蝶科技有限公司 线上点与非线上点分类、电力线提取及航迹规划方法
CN112883820B (zh) * 2021-01-26 2024-04-19 上海应用技术大学 基于激光雷达点云的道路目标3d检测方法及系统
CN113189615B (zh) * 2021-03-26 2023-06-30 国家电网有限公司 一种使用垂直起降固定翼无人机对输电线路进行巡检的方法
CN113494913A (zh) * 2021-09-07 2021-10-12 中国南方电网有限责任公司超高压输电公司检修试验中心 无人机电力巡检规划方法、装置、计算机设备和存储介质
CN114119917A (zh) * 2021-10-27 2022-03-01 国网浙江省电力有限公司温州供电公司 基于输电线路高精度三维模型的可视化隐患测距方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109146836A (zh) * 2017-09-19 2019-01-04 上海华测导航技术股份有限公司 一种基于点云数据进行输电塔及线路检测的系统及方法
CN107767389A (zh) * 2017-10-27 2018-03-06 云南电网有限责任公司电力科学研究院 一种基于机载激光点云数据的输电线自动提取方法
US20210224613A1 (en) * 2017-11-29 2021-07-22 Beijing Greenvalley Technology Co., Ltd. Method, Apparatus, and Device for Classifying LiDAR Point Cloud Data, and Storage Medium
CN111242080A (zh) * 2020-01-21 2020-06-05 南京航空航天大学 一种基于双目相机与深度相机的输电线识别与定位方法
CN111709343A (zh) * 2020-06-09 2020-09-25 广州文远知行科技有限公司 一种点云检测方法、装置、计算机设备和存储介质
CN114359758A (zh) * 2022-03-18 2022-04-15 广东电网有限责任公司东莞供电局 一种输电线的检测方法、装置、计算机设备和存储介质

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117092641A (zh) * 2023-10-20 2023-11-21 江苏翰林正川工程技术有限公司 一种基于雷达测距与图像识别的信息融合方法
CN117092641B (zh) * 2023-10-20 2023-12-15 江苏翰林正川工程技术有限公司 一种基于雷达测距与图像识别的信息融合方法
CN117314022A (zh) * 2023-11-29 2023-12-29 深圳金三立视频科技股份有限公司 一种输电线路隐患分析方法及装置
CN117314022B (zh) * 2023-11-29 2024-05-28 深圳金三立视频科技股份有限公司 一种输电线路隐患分析方法及装置
CN117634968A (zh) * 2023-12-04 2024-03-01 国网江苏省电力有限公司建设分公司 基于卫星互联网与云图数据集成的输变电工程定验方法
CN117634968B (zh) * 2023-12-04 2024-05-17 国网江苏省电力有限公司建设分公司 基于卫星互联网与云图数据集成的输变电工程定验方法

Also Published As

Publication number Publication date
CN114359758A (zh) 2022-04-15
CN114359758B (zh) 2022-06-14

Similar Documents

Publication Publication Date Title
WO2023174020A1 (zh) 输电线的检测方法、装置、计算机设备和存储介质
US9911340B2 (en) Real-time system for multi-modal 3D geospatial mapping, object recognition, scene annotation and analytics
US10964054B2 (en) Method and device for positioning
US11416644B2 (en) Supervised automatic roof modeling
US10628712B2 (en) Method and apparatus for processing high-precision map data, storage medium and device
US9424373B2 (en) Site modeling using image data fusion
CN112347999B (zh) 障碍物识别模型训练方法、障碍物识别方法、装置及系统
CN102804231A (zh) 三维场景的分段平面重建
CN108470174A (zh) 障碍物分割方法及装置、计算机设备及可读介质
CN112305559A (zh) 基于地面定点激光雷达扫描的输电线距离测量方法、装置、系统和电子设备
US11915478B2 (en) Bayesian methodology for geospatial object/characteristic detection
CN106292656A (zh) 一种环境建模方法及装置
CN110909713A (zh) 点云数据的轨道提取方法、系统及介质
CN112023400A (zh) 一种高度图生成方法、装置、设备及存储介质
CN115797736A (zh) 目标检测模型的训练和目标检测方法、装置、设备和介质
CN111611900A (zh) 一种目标点云识别方法、装置、电子设备和存储介质
CN112509135A (zh) 元素标注方法、装置、设备、存储介质及计算机程序产品
CN115565072A (zh) 一种道路垃圾识别和定位方法、装置、电子设备及介质
CN115527028A (zh) 地图数据处理方法及装置
CN113096104A (zh) 目标分割模型的训练方法及装置和目标分割方法及装置
US20240127475A1 (en) Information processing apparatus, information processing method, and program
Tarko et al. Guaranteed lidar-aided multi-object tracking at road intersections
CN113850907A (zh) 一种路网信息提取方法、装置、电子设备和存储介质
Semaan et al. A geographic data gathering system for image geolocalization refining
Roozenbeek Dutch Open Topographic Data Sets as Georeferenced Markers in Augmented Reality

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23769528

Country of ref document: EP

Kind code of ref document: A1