WO2019104780A1 - 激光雷达点云数据分类方法、装置、设备及存储介质 - Google Patents

激光雷达点云数据分类方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2019104780A1
WO2019104780A1 PCT/CN2017/117607 CN2017117607W WO2019104780A1 WO 2019104780 A1 WO2019104780 A1 WO 2019104780A1 CN 2017117607 W CN2017117607 W CN 2017117607W WO 2019104780 A1 WO2019104780 A1 WO 2019104780A1
Authority
WO
WIPO (PCT)
Prior art keywords
point cloud
cloud data
classification
neighborhood
sample point
Prior art date
Application number
PCT/CN2017/117607
Other languages
English (en)
French (fr)
Inventor
郭彦明
Original Assignee
北京数字绿土科技有限公司
深圳绿土智能科技有限公司
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 北京数字绿土科技有限公司, 深圳绿土智能科技有限公司 filed Critical 北京数字绿土科技有限公司
Priority to US16/304,673 priority Critical patent/US11636289B2/en
Publication of WO2019104780A1 publication Critical patent/WO2019104780A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
    • G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/2163—Partitioning the feature space
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/17—Terrestrial scenes taken from planes or by drones
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/182—Network patterns, e.g. roads or rivers

Definitions

  • the present invention relates to the field of power inspection technology, and in particular, to a laser radar point cloud data classification method, device, device and storage medium.
  • Transmission lines are an important part of the power grid. It is of far-reaching significance to quickly and efficiently inspect the conditions of vegetation and other features near the power lines for real-time monitoring, rapid assessment and scientific prediction of the power system.
  • the UAV is currently equipped with a laser radar scanning device for conducting transmission line inspections, and obtaining laser radar point cloud data corresponding to the transmission line.
  • a laser radar scanning device for conducting transmission line inspections, and obtaining laser radar point cloud data corresponding to the transmission line.
  • manual manual classification is usually used to classify power lines, towers, ground points and vegetation from the lidar point cloud data.
  • the lidar point cloud data Due to the large amount of data of the lidar point cloud data, it relies on manual classification, the workload is very large, the cost is high, and the efficiency is low. And the manual classification is low in automation, easy to make mistakes, and the classification accuracy is also low.
  • an object of the embodiments of the present invention is to provide a method, a device, a device, and a storage medium for a laser radar point cloud data classification, and a point cloud classifier is trained through the sample point cloud data, and the laser classified by the point cloud classifier is to be classified.
  • Radar point cloud data is automatically classified, which greatly reduces the manual intervention components in the classification process, with high automation, low cost, high efficiency and accuracy, and is not easy to make mistakes.
  • an embodiment of the present invention provides a method for classifying a lidar point cloud data, where the method includes:
  • the lidar point cloud data to be classified is classified by the point cloud classifier.
  • the embodiment of the present invention provides the first possible implementation manner of the foregoing first aspect, wherein the establishing a point cloud classifier according to the sample point cloud data includes:
  • the machine learning training is performed on the classification feature to obtain a point cloud classifier.
  • the embodiment of the present invention provides the second possible implementation manner of the foregoing first aspect, wherein the feature extraction of the sample point cloud data is performed to obtain a classification feature.
  • the feature extraction of the sample point cloud data is performed to obtain a classification feature.
  • the sample point cloud data is divided into spherical neighborhoods, and the spherical neighborhood classification features are obtained.
  • the embodiment of the present invention provides a third possible implementation manner of the foregoing first aspect, wherein the sample point cloud data is K-neighbor partitioned, and Obtain K neighborhood classification features, including:
  • the embodiment of the present invention provides the fourth possible implementation manner of the foregoing first aspect, wherein the performing the grid point neighborhood on the sample point cloud data, And obtain the grid neighborhood classification features, including:
  • the embodiment of the present invention provides the fifth possible implementation manner of the foregoing first aspect, wherein the sample point cloud data is subjected to cylindrical neighborhood division, And obtain cylindrical neighborhood classification features, including:
  • a cylindrical neighborhood centered on the first sample point and having a radius R height H is defined, and the first sample point is any point in the sample point cloud data.
  • the embodiment of the present invention provides the sixth possible implementation manner of the foregoing first aspect, wherein the sample point cloud data is spherically partitioned, and Obtain spherical neighborhood classification features, including:
  • a spherical neighborhood centered on the first sample point and having a radius r is defined, and the first sample point is any point in the sample point cloud data;
  • the embodiment of the present invention provides the seventh possible implementation manner of the foregoing first aspect, wherein the The laser radar point cloud data to be classified is classified, including:
  • the lidar point cloud data to be classified into the point cloud classifier to obtain a point cloud classification result, where the point cloud classification result includes a ground point, a power line, and a pole tower.
  • the embodiment of the present invention provides the eighth possible implementation manner of the foregoing first aspect, wherein the laser to be classified is used by the point cloud classifier After the radar point cloud data is classified, it also includes:
  • the point cloud classification result is subjected to speckle combination optimization, and the point cloud classification result is classified and optimized according to the tower position file and the preset optimization rule.
  • the embodiment of the present invention provides the ninth possible implementation manner of the foregoing first aspect, wherein the sample point cloud data includes Tower point cloud data, power line point cloud data and ground point cloud data.
  • an embodiment of the present invention provides a laser radar point cloud data classification device, where the device includes:
  • An acquisition module configured to acquire sample point cloud data and lidar point cloud data to be classified
  • a classification module configured to classify the lidar point cloud data to be classified by the point cloud classifier.
  • the embodiment of the present invention provides the first possible implementation manner of the foregoing second aspect, wherein the establishing module includes:
  • a feature extraction unit configured to perform feature extraction on the sample point cloud data to obtain a classification feature
  • the training unit is configured to perform machine learning training on the classification feature to obtain a point cloud classifier.
  • the embodiment of the present invention provides the second possible implementation manner of the foregoing second aspect, wherein the feature extraction unit includes:
  • the K-neighbor feature extraction sub-unit is configured to perform K-neighbor partitioning on the sample point cloud data, and acquire a K-neighbor classification feature;
  • the grid neighborhood feature extraction sub-unit is configured to perform grid neighborhood classification on the sample point cloud data, and obtain a grid neighborhood classification feature
  • the cylindrical neighborhood feature extraction sub-unit is configured to perform cylindrical neighborhood division on the sample point cloud data, and obtain a cylindrical neighborhood classification feature
  • the spherical neighborhood feature extraction sub-unit is configured to perform spherical neighborhood partitioning on the sample point cloud data, and obtain a spherical neighborhood classification feature.
  • the embodiment of the present invention provides the third possible implementation manner of the foregoing second aspect, wherein the classification module is configured to:
  • the lidar point cloud data to be classified is input to the point cloud classifier to obtain a point cloud classification result, and the point cloud classification result includes a ground point, a power line, and a pole tower.
  • the embodiment of the present invention provides the fourth possible implementation manner of the foregoing second aspect, where the apparatus further includes: an optimization module configured to classify the point cloud result The speckle combination optimization is performed, and the point cloud classification result is classified and optimized according to the tower position file and the preset optimization rule.
  • an embodiment of the present invention provides a laser radar point cloud data classification device, including a memory and a processor, where the memory is configured to store a program supporting a processor to perform the method according to any one of the first aspects, The processor is configured to execute a program stored in the memory.
  • an embodiment of the present invention provides a computer storage medium, configured to store computer software instructions for use in the method of any of the first aspects.
  • the sample point cloud data and the lidar point cloud data to be classified are acquired; the point cloud classifier is established according to the sample point cloud data; and the point cloud classifier is used to classify The lidar point cloud data is classified.
  • the invention trains the point cloud classifier through the sample point cloud data, and automatically classifies the classified lidar point cloud data by the point cloud classifier, thereby greatly reducing the manual intervention component in the classification process, with high automation degree and low cost.
  • the sample point cloud data used to train the point cloud classifier covers various tower type tower data and various line type power line data, and the data is comprehensive.
  • the point cloud classifier trained by the sample point cloud data has high accuracy. , not easy to make mistakes.
  • the speckle combination optimization is performed, and the optimization is performed according to the tower position file and the preset optimization rule, thereby further improving the classification accuracy.
  • Embodiment 1 is a flowchart of a method for classifying a lidar point cloud data provided by Embodiment 1 of the present invention
  • FIG. 2 is a schematic diagram showing classification of a lidar point cloud data provided by Embodiment 1 of the present invention
  • FIG. 3 is a schematic structural diagram of a laser radar point cloud data classification device according to Embodiment 2 of the present invention.
  • FIG. 4 is a schematic structural diagram of another laser radar point cloud data classification device according to Embodiment 2 of the present invention.
  • an embodiment of the present invention provides a method, a device, a device, and a storage medium for a laser radar point cloud data classification, which are described below by using an embodiment.
  • an embodiment of the present invention provides a method for classifying a point cloud data of a laser radar, and the method specifically includes the following steps:
  • Step 101 Acquire sample point cloud data and lidar point cloud data to be classified.
  • a laser radar (Light Detection and Ranging, LiDAR) device is mounted on a flight platform such as a helicopter or a drone, and then the transmission line is inspected by a flight platform such as a helicopter or a drone, and the inspection process is performed during the inspection process.
  • the laser radar device mounted on the flight platform collects data on the transmission line and obtains the lidar point cloud data corresponding to the transmission line.
  • the above sample point cloud data is a training sample manually selected from the original lidar point cloud data corresponding to the transmission line.
  • the transmission line includes power lines, towers, vegetation, etc., so the lidar point cloud data collected by the laser radar equipment includes point cloud data corresponding to power lines, towers, vegetation, and the like.
  • the tower type of different towers should be considered when manually selecting training samples, such as cat head tower, wine glass tower, dry type tower, door type tower, etc., and the line type of different power lines should be considered comprehensively, such as single Wires, split wires, etc., and manually separate the power line and tower categories, and separate the ground points by a filtering algorithm.
  • the sample point cloud data includes selected tower point cloud data, power line point cloud data, and ground point cloud data. That is, the point cloud data of the towers of different tower types manually selected, the point cloud data of the power lines of different line types, and the point cloud data corresponding to the ground points are formed into the sample point cloud data.
  • the lidar point cloud data to be classified is the point cloud data collected by the lidar device except the sample point cloud data.
  • Step 102 Establish a point cloud classifier according to the sample point cloud data.
  • Feature extraction is performed on the sample point cloud data to obtain classification features; machine learning training is performed on the classification features to obtain a point cloud classifier.
  • the embodiment of the present invention obtains the classification feature by the operations of the following steps A1 - A4, and specifically includes:
  • A1 K-neighbor partitioning is performed on the sample point cloud data, and the K-neighbor classification feature is obtained.
  • any point in the sample point cloud data is referred to as a first sample point in the embodiment of the present invention.
  • K neighbor points adjacent to the first sample point are selected from the sample point cloud data. Construct a covariance matrix of the first sample point and the K neighborhood points. According to the covariance matrix, the K neighborhood classification feature corresponding to the first sample point is calculated.
  • the eigenvalues ⁇ 1 , ⁇ 2 , and ⁇ 3 are calculated based on the covariance matrix.
  • ⁇ 1 ⁇ 2 ⁇ 3 ⁇ 0 based on the feature values of ⁇ 1, ⁇ 2, ⁇ K 3 neighborhood classification characteristic point corresponding to the first sample is calculated:
  • Sum is the sum of the eigenvalues
  • Omnivariance is the eigenvalue full variance
  • Eigenentropy is the characteristic entropy
  • Anisotropy is the anisotropy
  • Planarity is the flatness
  • Linearity is the linearity.
  • the K neighborhood corresponding to each of the other sample points is divided according to the above manner, and the K neighborhood corresponding to each of the other sample points is calculated. Classification features.
  • A2 Perform grid neighborhood classification on the sample point cloud data, and obtain the grid neighborhood classification feature.
  • any grid that is divided in the embodiment of the present invention is referred to as a first grid.
  • the difference between the maximum point cloud elevation value and the minimum point cloud elevation value is calculated, and the difference is determined as the grid neighborhood classification feature corresponding to the first grid.
  • the same grid is used, and the grid neighborhood classification features corresponding to each grid are calculated in the above manner.
  • A3 Perform cylindrical neighborhood division on sample point cloud data, and obtain cylindrical neighborhood classification features
  • the first sample point is any point in the sample point cloud data.
  • a cylindrical neighborhood centered on the first sample point and having a radius R height of H is divided.
  • the point cloud layer is layered according to the preset layering rule for the cylindrical neighborhood corresponding to the first sample point.
  • the number of points included in each layer, the elevation value of each point included in each layer, and the height of the center point of each layer are counted.
  • the cylindrical neighborhood classification feature corresponding to the first sample point is calculated according to the number of points included in each layer, the elevation value of each point included in each layer, and the height of the center point of each layer.
  • the above radius R and height H can be set in advance.
  • the tower has the characteristics of continuous vertical direction.
  • the point cloud is layered according to a certain height from bottom to top, and the number of points in each layer and each layer included in each layer are counted.
  • the elevation value of each point and the height of the center point of each layer calculate the cylindrical neighborhood classification feature corresponding to the first sample point.
  • Cylindrical neighborhood classification features include maximum height deviation, average points, point deviation, elevation difference, elevation variance, normalized elevation values, and non-empty layer numbers. The following describes the calculation process of each cylindrical neighborhood classification feature in turn:
  • N is the total number of layers
  • N i is the number of points included in the i-th layer
  • H i is the height of the center point of the i-th layer
  • V 1 is the average height deviation.
  • V 2 max
  • ,i 1,2,...,N
  • V 2 is the maximum height deviation
  • H i is the height of the center point of the i-th layer
  • V 1 is the average height deviation
  • N ave is the average number of points
  • N is the total number of layers
  • N i is the number of points included in the i-th layer.
  • N dev max
  • ,i 2,3,...,N
  • N dev is the point deviation and N i is the number of points included in the i-th layer.
  • the elevation variance of all points in the cylindrical neighborhood is calculated from the elevation values of each point included in each layer within the cylindrical neighborhood.
  • the height of the center point in the cylindrical neighborhood relative to the ground point is calculated according to the center point height of each layer in the cylindrical neighborhood and the point cloud data corresponding to the ground point included in the sample point cloud data.
  • the total number of layers is 1, and the total number of layers of the statistics is The number of non-empty layers.
  • A4 Perform spherical neighborhood division on sample point cloud data and obtain spherical neighborhood classification features.
  • the first sample point is any point in the sample point cloud data.
  • a spherical neighborhood centered on the first sample point and having a radius r is divided.
  • the elevation value of each point in the spherical neighborhood is counted, and the elevation variance of all points in the spherical neighborhood is calculated according to the elevation value of each point, and the elevation variance is determined as the spherical neighborhood classification feature corresponding to the first sample point.
  • the above radius r can be set in advance.
  • Step 103 classify the classified lidar point cloud data by the point cloud classifier.
  • the lidar point cloud data to be classified is input into a point cloud classifier, and the point cloud classifier automatically classifies the classified lidar point cloud data to obtain a point cloud classification result, and the point cloud classification result includes a ground point, a power line, a pole and other Class object.
  • Other types of objects may be vegetation, billboards, bus stop signs, etc. along the transmission line.
  • the point cloud classification results are also subjected to speckle consolidation optimization, and the point cloud classification results are classified and optimized according to the tower position file and the preset optimization rules.
  • Speckle merging optimization refers to the point where there may be scattered within a preset distance range around an classified object. At this time, the scattered points are also classified as the points included in the classified object. For example, the scattered points existing in the preset distance range around the classified tower are classified as points included in the tower.
  • the above-mentioned tower position file includes coordinate information of each tower in the transmission line.
  • the coordinates of each point included in the classified tower are obtained.
  • For the coordinates of each point it is determined whether the coordinates of the point exist in the tower position file, and if so, it is determined that the point actually belongs to the tower. If it does not exist, it is determined that the point does not belong to the tower, and the point is removed from the classified tower. This can eliminate misclassified tower points.
  • the embodiment of the present invention specifies that the point within a certain distance from the ground point is not the power line point by the preset optimization rule. After the classification result is obtained, the distance between each point classified as the power line and the ground point is calculated, and the distance between the ground point and the ground point is less than a certain distance specified by the preset optimization rule, so that a part of the misclassified power line can be eliminated. point.
  • the preset optimization rule stipulates that if the number of points in the neighborhood around a certain point is less than the preset point threshold, the point is an isolated point. After the classification result is obtained, it is determined according to the neighborhood statistics manner whether the number of points in the neighborhood around each point classified is less than a preset threshold number specified by the preset optimization rule, and if so, the point is determined as an isolated point, and the point is eliminated. Isolated point, which removes some of the isolated noise points.
  • the embodiment of the present invention obtains the original lidar point cloud data from the laser radar device mounted on the flight platform, and selects sample point cloud data from the original lidar point cloud data, and the remaining point cloud data is used as Lidar point cloud data to be classified.
  • the feature point cloud data is extracted, and the point cloud classifier is trained according to the result of feature extraction.
  • the point cloud classifier then automatically classifies the classified lidar point cloud data to obtain the classified power lines, towers, ground points, vegetation, and other objects.
  • the point cloud classifier is trained by the sample point cloud data, and the lidar point cloud data to be classified by the point cloud classifier is automatically classified, which greatly improves the accuracy and efficiency of the automatic classification of the power line and the tower. After testing, the efficiency of automatic classification is tested. As follows: average 40 seconds / file (average about 3 million points / file). The automatic classification accuracy can reach 95% when the selected sample point cloud data is sufficiently representative.
  • the sample point cloud data and the lidar point cloud data to be classified are acquired; the point cloud classifier is established according to the sample point cloud data; and the classified lidar point cloud data is classified by the point cloud classifier.
  • the invention trains the point cloud classifier through the sample point cloud data, and automatically classifies the classified lidar point cloud data by the point cloud classifier, thereby greatly reducing the manual intervention component in the classification process, with high automation degree and low cost.
  • the sample point cloud data used to train the point cloud classifier covers various tower type tower data and various line type power line data, and the data is comprehensive.
  • the point cloud classifier trained by the sample point cloud data has high accuracy. , not easy to make mistakes.
  • the speckle combination optimization is performed, and the optimization is performed according to the tower position file and the preset optimization rule, thereby further improving the classification accuracy.
  • an embodiment of the present invention provides a laser radar point cloud data classification device, which is configured to perform the laser radar point cloud data classification method provided in Embodiment 1 above, and the device includes:
  • the obtaining module 20 is configured to acquire sample point cloud data and lidar point cloud data to be classified;
  • the establishing module 21 is configured to establish a point cloud classifier according to the sample point cloud data
  • the classification module 22 is configured to classify the lidar point cloud data to be classified by the point cloud classifier.
  • the foregoing establishing module 21 includes:
  • the feature extraction unit 210 is configured to perform feature extraction on the sample point cloud data to obtain a classification feature.
  • the training unit 211 is configured to perform machine learning training on the classification features to obtain a point cloud classifier.
  • the above feature extraction unit includes:
  • the K-neighbor feature extraction sub-unit is configured to perform K-neighbor partitioning on the sample point cloud data, and obtain a K-neighbor classification feature;
  • the grid neighborhood feature extraction sub-unit is configured to perform grid neighborhood classification on the sample point cloud data, and obtain a grid neighborhood classification feature
  • the cylindrical neighborhood feature extraction sub-unit is configured to perform cylindrical neighborhood division on the sample point cloud data, and obtain a cylindrical neighborhood classification feature
  • the spherical neighborhood feature extraction sub-unit is configured to perform spherical neighborhood partitioning on the sample point cloud data, and obtain a spherical neighborhood classification feature.
  • the K neighborhood feature extraction subunit is configured to select K neighborhood points adjacent to the first sample point from the sample point cloud data, and the first sample point is any point in the sample point cloud data; A covariance matrix of a sample point and K neighborhood points; according to the covariance matrix, the K neighborhood classification feature corresponding to the first sample point is calculated.
  • the grid neighborhood feature extraction sub-unit is configured to divide the sample point cloud data into a plurality of preset size grids; obtain a maximum point cloud elevation value in the first grid and a number adjacent to the first grid The minimum point cloud elevation value in the second grid, the first grid is any grid divided; the difference between the maximum point cloud elevation value and the minimum point cloud elevation value is calculated, and the difference is determined as the first grid
  • the grid neighborhood classification feature corresponding to the network.
  • the cylindrical neighborhood feature extraction subunit is configured to, in the sample point cloud data, divide a cylindrical neighborhood centered on the first sample point and having a radius R height H, and the first sample point is a sample point Any point in the cloud data; performing point cloud layering on the cylindrical neighborhood corresponding to the first sample point according to the preset layering rule; counting the number of points included in each layer, the elevation value of each point included in each layer, and each The height of the center point of the layer; the cylindrical neighborhood classification feature corresponding to the first sample point is calculated according to the number of points included in each layer, the elevation value of each point included in each layer, and the height of the center point of each layer.
  • the spherical neighborhood feature extraction sub-unit is configured to, in the sample point cloud data, divide a spherical neighborhood centered on the first sample point and having a radius r, and the first sample point is any one of the sample point cloud data. One point; calculate the elevation variance of all points in the spherical neighborhood, and determine the elevation variance as the spherical neighborhood classification feature corresponding to the first sample point.
  • the classification module 22 is configured to input the lidar point cloud data to be classified into the point cloud classifier to obtain a point cloud classification result, where the point cloud classification result includes a ground point, a power line, and a pole tower.
  • the device further includes:
  • the optimization module 23 is configured to perform speckle merging optimization on the point cloud classification result, and classify and optimize the point cloud classification result according to the tower position file and the preset optimization rule.
  • the sample point cloud data and the lidar point cloud data to be classified are acquired; the point cloud classifier is established according to the sample point cloud data; and the classified lidar point cloud data is classified by the point cloud classifier.
  • the invention trains the point cloud classifier through the sample point cloud data, and automatically classifies the classified lidar point cloud data by the point cloud classifier, thereby greatly reducing the manual intervention component in the classification process, with high automation degree and low cost.
  • the sample point cloud data used to train the point cloud classifier covers various tower type tower data and various line type power line data, and the data is comprehensive.
  • the point cloud classifier trained by the sample point cloud data has high accuracy. , not easy to make mistakes.
  • the speckle combination optimization is performed, and the optimization is performed according to the tower position file and the preset optimization rule, thereby further improving the classification accuracy.
  • the lidar point cloud data classification device provided by the embodiment of the present invention may be specific hardware on the device or software or firmware installed on the device.
  • the implementation principle and the technical effects of the device provided by the embodiments of the present invention are the same as those of the foregoing method embodiments.
  • a person skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working processes of the foregoing system, the device and the unit can refer to the corresponding processes in the foregoing method embodiments, and details are not described herein again.
  • an embodiment of the present invention provides a laser radar point cloud data classification device, including a memory and a processor, where the memory is configured to store a program supporting the processor to perform any one of the foregoing laser radar point cloud data classification methods, and the processor is configured to Execute the program stored in the memory.
  • an embodiment of the present invention provides a computer storage medium configured to store computer software instructions for use in any of the foregoing lidar point cloud data classification methods.
  • the disclosed apparatus and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • multiple units or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some communication interface, device or unit, and may be electrical, mechanical or otherwise.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the functions may be stored in a computer readable storage medium if implemented in the form of a software functional unit and sold or used as a standalone product.
  • the technical solution of the present invention which is essential or contributes to the prior art, or a part of the technical solution, may be embodied in the form of a software product, which is stored in a storage medium, including
  • the instructions are used to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .
  • the trained point cloud classifier can automatically classify the lidar point cloud data to be classified, greatly reducing the manual intervention component in the classification process, high degree of automation, low cost, efficiency and accuracy. They are very high and not easy to make mistakes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Remote Sensing (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Mathematical Physics (AREA)
  • Electromagnetism (AREA)
  • Optical Radar Systems And Details Thereof (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种激光雷达点云数据分类方法、装置、设备及存储介质，该方法包括：获取样本点云数据及待分类的激光雷达点云数据（101）；根据样本点云数据建立点云分类器（102）；通过点云分类器对待分类的激光雷达点云数据进行分类（103）。该方法通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低。用于训练点云分类器的样本点云数据涵盖各种塔型的杆塔数据及各种线型的电力线数据，数据全面，通过该样本点云数据训练出的点云分类器的准确性很高，不易出错。而且自动分类得到分类结果后还进行散斑合并优化，以及根据杆塔位置文件及预设优化规则进行优化，进一步提高分类准确性。

Description

激光雷达点云数据分类方法、装置、设备及存储介质
相关申请的交叉引用
本申请要求于2017年11月29日提交中国专利局的申请号为201711222953.2，名称为“激光雷达点云数据分类方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及电力巡检技术领域，具体而言，涉及一种激光雷达点云数据分类方法、装置、设备及存储介质。
背景技术
输电线路是电网的重要组成部分，快速高效地巡检电力线附近的植被和其它地物的状况，对电力部门实现对电力系统的实时监测、快速评估和科学预测具有深远意义。
随着激光雷达技术及无人机技术的发展，当前通常采用无人机搭载激光雷达扫描设备进行输电线路巡检，获得输电线路对应的激光雷达点云数据。通过激光雷达点云数据排查输电线路的故障及隐患之前，首先需要对激光雷达点云数据进行分类。当前通常采用人工手动分类的方式，从激光雷达点云数据中划分出电力线、杆塔、地面点及植被等地物类型。
由于激光雷达点云数据的数据量很大，依赖人工分类，工作量非常大，成本高，效率低。且人工分类自动化程度低，易出错，分类准确性也很低。
发明内容
有鉴于此，本发明实施例的目的在于提供一种激光雷达点云数据分类方法、装置、设备及存储介质，通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低，效率和准确性都很高，不易出错。
第一方面，本发明实施例提供了一种激光雷达点云数据分类方法，所述方法包括：
获取样本点云数据及待分类的激光雷达点云数据；
根据所述样本点云数据建立点云分类器；
通过所述点云分类器对所述待分类的激光雷达点云数据进行分类。
结合第一方面，本发明实施例提供了上述第一方面的第一种可能的实现方式，其中，所述根据所述样本点云数据建立点云分类器，包括：
对所述样本点云数据进行特征提取，获得分类特征；
对所述分类特征进行机器学习训练，得到点云分类器。
结合第一方面的第一种可能的实现方式，本发明实施例提供了上述第一方面的第二种可能的实现方式，其中，所述对所述样本点云数据进行特征提取，获得分类特征，包括：
对所述样本点云数据进行K邻域划分，并获取K邻域分类特征；
对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征；
对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
结合第一方面的第二种可能的实现方式，本发明实施例提供了上述第一方面的第三种可能的实现方式，其中，所述对所述样本点云数据进行K邻域划分，并获取K邻域分类特征，包括：
从所述样本点云数据中选取与第一样本点相邻的K个邻域点，所述第一 样本点为所述样本点云数据中的任一点；
构建所述第一样本点与所述K个邻域点的协方差矩阵；
根据所述协方差矩阵，计算所述第一样本点对应的K邻域分类特征。
结合第一方面的第二种可能的实现方式，本发明实施例提供了上述第一方面的第四种可能的实现方式，其中，所述对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征，包括：
将所述样本点云数据划分为多个预设尺寸的格网；
获取第一格网中的最大点云高程值及与所述第一格网相邻的第二格网中的最小点云高程值，所述第一格网为划分出的任一格网；
计算所述最大点云高程值与所述最小点云高程值之间的差值，将所述差值确定为所述第一格网对应的格网邻域分类特征。
结合第一方面的第二种可能的实现方式，本发明实施例提供了上述第一方面的第五种可能的实现方式，其中，所述对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征，包括：
在所述样本点云数据中，划分出以第一样本点为中心且半径为R高度为H的圆柱形邻域，所述第一样本点为所述样本点云数据中的任一点；
按照预设分层规则对所述第一样本点对应的圆柱形邻域进行点云分层；
统计每层包括的点数、每层包括的每个点的高程值及每层的中心点高度；
根据所述每层包括的点数、所述每层包括的每个点的高程值及所述每层的中心点高度，计算所述第一样本点对应的圆柱形邻域分类特征。
结合第一方面的第二种可能的实现方式，本发明实施例提供了上述第一方面的第六种可能的实现方式，其中，所述对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征，包括：
在所述样本点云数据中，划分出以第一样本点为中心且半径为r的球形 邻域，所述第一样本点为所述样本点云数据中的任一点；
计算所述球形邻域内所有点的高程方差，将所述高程方差确定为所述第一样本点对应的球形邻域分类特征。
结合第一方面至第一方面的第六种可能的实现方式之一，本发明实施例提供了上述第一方面的第七种可能的实现方式，其中，所述通过所述点云分类器对所述待分类的激光雷达点云数据进行分类，包括：
将所述待分类的激光雷达点云数据输入所述点云分类器，获得点云分类结果，所述点云分类结果包括地面点、电力线和杆塔。
结合第一方面的第七种可能的实现方式，本发明实施例提供了上述第一方面的第八种可能的实现方式，其中，所述通过所述点云分类器对所述待分类的激光雷达点云数据进行分类之后，还包括：
对所述点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对所述点云分类结果进行分类优化。
结合第一方面至第一方面的第八种可能的实现方式之一，本发明实施例提供了上述第一方面的第九种可能的实现方式，其中，所述样本点云数据包括经选择得到的杆塔点云数据、电力线点云数据和地面点点云数据。
第二方面，本发明实施例提供了一种激光雷达点云数据分类装置，所述装置包括：
获取模块，用于获取样本点云数据及待分类的激光雷达点云数据；
建立模块，用于根据所述样本点云数据建立点云分类器；
分类模块，用于通过所述点云分类器对所述待分类的激光雷达点云数据进行分类。
结合第二方面，本发明实施例提供了上述第二方面的第一种可能的实现方式，其中，所述建立模块包括：
特征提取单元，配置成对所述样本点云数据进行特征提取，获得分类特 征；
训练单元，配置成对所述分类特征进行机器学习训练，得到点云分类器。
结合第二方面的第一种可能的实现方式，本发明实施例提供了上述第二方面的第二种可能的实现方式，其中，所述特征提取单元包括：
K邻域特征提取子单元，配置成对所述样本点云数据进行K邻域划分，并获取K邻域分类特征；
格网邻域特征提取子单元，配置成对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征；
圆柱形邻域特征提取子单元，配置成对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
球形邻域特征提取子单元，配置成对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
结合第二方面至第二方面的第二种可能的实现方式之一，本发明实施例提供了上述第二方面的第三种可能的实现方式，其中，所述分类模块配置成：将所述待分类的激光雷达点云数据输入所述点云分类器，获得点云分类结果，所述点云分类结果包括地面点、电力线和杆塔。
结合第二方面的第三种可能的实现方式，本发明实施例提供了上述第二方面的第四种可能的实现方式，所述装置还包括：优化模块，配置成对所述点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对所述点云分类结果进行分类优化。
第三方面，本发明实施例提供了一种激光雷达点云数据分类设备，包括存储器以及处理器，所述存储器配置成存储支持处理器执行第一方面任一项所述方法的程序，所述处理器配置成执行所述存储器中存储的程序。
第四方面，本发明实施例提供了一种计算机存储介质，配置成储存为第一方面任一项所述方法所用的计算机软件指令。
在本发明实施例提供的方法、装置、设备及存储介质中，获取样本点云数据及待分类的激光雷达点云数据；根据样本点云数据建立点云分类器；通过点云分类器对待分类的激光雷达点云数据进行分类。本发明通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低。用于训练点云分类器的样本点云数据涵盖各种塔型的杆塔数据及各种线型的电力线数据，数据全面，通过该样本点云数据训练出的点云分类器的准确性很高，不易出错。而且自动分类得到分类结果后还进行散斑合并优化，以及根据杆塔位置文件及预设优化规则进行优化，进一步提高分类准确性。
为使本发明的上述目的、特征和优点能更明显易懂，下文特举较佳实施例，并配合所附附图，作详细说明如下。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，应当理解，以下附图仅示出了本发明的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他相关的附图。
图1示出了本发明实施例1所提供的一种激光雷达点云数据分类方法的流程图；
图2示出了本发明实施例1所提供的激光雷达点云数据分类示意图；
图3示出了本发明实施例2所提供的一种激光雷达点云数据分类装置的结构示意图；
图4示出了本发明实施例2所提供的另一种激光雷达点云数据分类装置的结构示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。通常在此处附图中描述和示出的本发明实施例的组件可以以各种不同的配置来布置和设计。因此，以下对在附图中提供的本发明的实施例的详细描述并非旨在限制要求保护的本发明的范围，而是仅仅表示本发明的选定实施例。基于本发明的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本发明保护的范围。
考虑到现有技术中采用人工手动分类的方式，从激光雷达点云数据中划分出电力线、杆塔、地面点及植被等地物类型。由于激光雷达点云数据的数据量很大，依赖人工分类，工作量非常大，成本高，效率低。且人工分类自动化程度低，易出错，分类准确性也很低。基于此，本发明实施例提供了一种激光雷达点云数据分类方法、装置、设备及存储介质，下面通过实施例进行描述。
实施例1
参见图1，本发明实施例提供了一种激光雷达点云数据分类方法，该方法具体包括以下步骤：
步骤101：获取样本点云数据及待分类的激光雷达点云数据。
本发明实施例中，在直升机或无人机等飞行平台上搭载激光雷达(Light Detection and Ranging,LiDAR)设备，然后通过直升机或无人机等飞行平台对输电线路进行巡检，在巡检过程中飞行平台上搭载的激光雷达设备对输电线路进行数据采集，得到输电线路对应的激光雷达点云数据。
上述样本点云数据即为从输电线路对应的原始的激光雷达点云数据中人工选择出的训练样本。输电线路沿线包含电力线、杆塔、植被等，因此激光雷达设备采集的激光雷达点云数据中包括电力线、杆塔、植被等事物对应的 点云数据。为了保证分类的准确性，人工选择训练样本时应考虑不同杆塔的塔型，如猫头塔、酒杯塔、干字型塔、门型塔等，以及应综合考虑不同电力线的线型，如单导线、分裂导线等，并人工分出电力线和杆塔的类别，而且通过滤波算法分出地面点。
样本点云数据包括经选择得到的杆塔点云数据、电力线点云数据和地面点点云数据。也即，将人工选择出的不同塔型的杆塔的点云数据、不同线型的电力线的点云数据及地面点对应的点云数据等组成上述样本点云数据。待分类的激光雷达点云数据即为激光雷达设备采集的除样本点云数据以外的其他点云数据。
步骤102：根据样本点云数据建立点云分类器。
对样本点云数据进行特征提取，获得分类特征；对分类特征进行机器学习训练，得到点云分类器。本发明实施例通过如下步骤A1-A4的操作来获取分类特征，具体包括：
A1：对样本点云数据进行K邻域划分，并获取K邻域分类特征。
为了便于描述，本发明实施例中将样本点云数据中的任一点称为第一样本点。从样本点云数据中选取与第一样本点相邻的K个邻域点。构建第一样本点与K个邻域点的协方差矩阵。根据该协方差矩阵，计算第一样本点对应的K邻域分类特征。
首先根据该协方差矩阵计算特征值λ 1、λ 2、λ 3。其中，λ 1≥λ 2≥λ 3≥0，基于特征值λ 1、λ 2、λ 3计算第一样本点对应的K邻域分类特征：
Sum＝λ 1+λ 2+λ 3
Figure PCTCN2017117607-appb-000001
Figure PCTCN2017117607-appb-000002
Figure PCTCN2017117607-appb-000003
Figure PCTCN2017117607-appb-000004
Figure PCTCN2017117607-appb-000005
其中，Sum为特征值的和，Omnivariance为特征值全方差，Eigenentropy为特征熵，Anisotropy为各向异性，Planarity为平面度，Linearity为线性度。
对于样本点云数据中的其他每个样本点，都同第一样本点，按照上述方式划分出其他每个样本点对应的K邻域，并计算出其他每个样本点对应的K邻域分类特征。
A2：对样本点云数据进行格网邻域划分，并获取格网邻域分类特征。
将样本点云数据划分为多个预设尺寸的格网。为了便于描述，本发明实施例中将划分出的任一格网称为第一格网。获取第一格网中的最大点云高程值及与第一格网相邻的第二格网中的最小点云高程值。计算最大点云高程值与最小点云高程值之间的差值，将该差值确定为第一格网对应的格网邻域分类特征。
对于划分出的其他每个格网，都同第一格网，按照上述方式计算出其他每个格网对应的格网邻域分类特征。
A3：对样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
同样地，第一样本点为样本点云数据中的任一点。在样本点云数据中，划分出以第一样本点为中心且半径为R高度为H的圆柱形邻域。按照预设分层规则对第一样本点对应的圆柱形邻域进行点云分层。统计每层包括的点数、每层包括的每个点的高程值及每层的中心点高度。根据每层包括的点数、每层包括的每个点的高程值及每层的中心点高度，计算第一样本点对应的圆柱形邻域分类特征。上述半径R及高度H可以预先设置。
杆塔具有垂直方向连续的特点，对第一样本点及其圆柱形邻域范围内的 点，自下往上按照一定高度进行点云分层，统计每一层的点数、每层包括的每个点的高程值和每一层的中心点高度，根据统计结果计算第一样本点对应的圆柱形邻域分类特征。圆柱形邻域分类特征包括最大高度偏差、平均点数、点数偏差、高程差、高程方差、归一化的高程值及非空层数。下面依次说明各圆柱形邻域分类特征的计算过程：
(1)、最大高度偏差
根据每一层的点数和每一层的中心点高度，按照如下公式计算平均高度偏差：
Figure PCTCN2017117607-appb-000006
在上述公式中，N为总层数，N i为第i层包括的点数，H i为第i层的中心点高度，V 1为平均高度偏差。
然后通过如下公式计算最大高度偏差：
V 2＝max|H i-V 1|,i＝1,2,...,N
其中，V 2为最大高度偏差，H i为第i层的中心点高度，V 1为平均高度偏差。
(2)、平均点数
通过如下公式计算平均点数：
Figure PCTCN2017117607-appb-000007
其中，N ave为平均点数，N为总层数，N i为第i层包括的点数。
(3)、点数偏差
通过如下公式计算点数偏差：
N dev＝max|N i-N dev|,i＝2,3,...,N
其中，N dev为点数偏差，N i为第i层包括的点数。
(4)、高程差
从圆柱形邻域内每层包括的每个点的高程值中，确定出最大高程值和最小高程值，计算最大高程值和最小高程值的差，得到高程差。
(5)、高程方差
根据圆柱形邻域内每层包括的每个点的高程值，计算圆柱形邻域内所有点的高程方差。
(6)、归一化的高程值
根据圆柱形邻域内每一层的中心点高度及样本点云数据中包括的地面点对应的点云数据，计算圆柱形邻域内的中心点相对于地面点的高度。
(7)非空层数
在统计每一层的点数过程中，如果第i层包括的点数大于0，则记为1，反之，则记为0，最后统计值为1的总层数，该统计的总层数即为非空层数。
A4：对样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
同样地，第一样本点为样本点云数据中的任一点。在样本点云数据中，划分出以第一样本点为中心且半径为r的球形邻域。统计该球形邻域内每个点的高程值，根据每个点的高程值，计算球形邻域内所有点的高程方差，将该高程方差确定为第一样本点对应的球形邻域分类特征。上述半径r可以预先设置。
通过上述A1-A4的操作获得分类特征后，对分类特征进行机器学习训练，即可得到点云分类器。
步骤103：通过点云分类器对待分类的激光雷达点云数据进行分类。
将待分类的激光雷达点云数据输入点云分类器，点云分类器对待分类的激光雷达点云数据进行自动分类，获得点云分类结果，点云分类结果包括地面点、电力线、杆塔和其他类物体。其他类物体可以为输电线路沿线的植被、广告牌、公交站牌等。
得到分类结果后，还对点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对点云分类结果进行分类优化。
散斑合并优化是指在分类出的某物体周围预设距离范围内可能存在散落的点，此时将这些散落的点也归为分类出的该物体所包括的点。例如，将分类出的杆塔周围预设距离范围存在的散落的点归为杆塔包括的点。
上述杆塔位置文件中包括输电线路中每个杆塔的坐标信息。获取分类出的杆塔中包括的每个点的坐标，对于每个点的坐标，确定杆塔位置文件中是否存在该点的坐标，若存在，则确定该点确实属于杆塔。若不存在，则确定该点不属于杆塔，从分类出的杆塔中剔除该点。如此能够剔除误分类的杆塔点。
由于树木的树干有时会被误分为电力线，因此本发明实施例通过上述预设优化规则规定距离地面点以上一定距离内的点不是电力线点。得到分类结果后，计算分类为电力线的每个点与地面点之间的距离，将与地面点之间的距离小于预设优化规则规定的一定距离的点剔除，从而可以剔除一部分误分类的电力线点。
由于激光雷达设备采集数据时不可避免的会产生一些噪声点，因此上述预设优化规则还规定若某个点周围邻域内的点数若小于预设点数阈值，则该点为孤立点。得到分类结果后，根据邻域统计方式，判断分类出的每个点周围邻域内的点数是否小于预设优化规则规定的预设点数阈值，如果是，则将该点确定为孤立点，剔除该孤立点，如此可以去除一部分孤立的噪声点。
如图2所示，本发明实施例从飞行平台上搭载的激光雷达设备获取原始的激光雷达点云数据，从原始的激光雷达点云数据中选取样本点云数据，剩下的点云数据作为待分类的激光雷达点云数据。对样本点云数据进行特征提取，根据特征提取的结果训练点云分类器。然后通过点云分类器对待分类的激光雷达点云数据进行自动分类，得到分类出的电力线、杆塔、地面点、植被和其他类物体。通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大提高了电力线和杆塔自动分 类的准确性和效率，经过测试，自动分类的效率如下：平均40秒/档(平均约300万点/档)。在选取的样本点云数据具备足够代表性的情况下，自动分类精度可达到95％。
在本发明实施例中，获取样本点云数据及待分类的激光雷达点云数据；根据样本点云数据建立点云分类器；通过点云分类器对待分类的激光雷达点云数据进行分类。本发明通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低。用于训练点云分类器的样本点云数据涵盖各种塔型的杆塔数据及各种线型的电力线数据，数据全面，通过该样本点云数据训练出的点云分类器的准确性很高，不易出错。而且自动分类得到分类结果后还进行散斑合并优化，以及根据杆塔位置文件及预设优化规则进行优化，进一步提高分类准确性。
实施例2
如图3所示，本发明实施例提供了一种激光雷达点云数据分类装置，该装置配置成执行上述实施例1所提供的激光雷达点云数据分类方法，该装置包括：
获取模块20，配置成获取样本点云数据及待分类的激光雷达点云数据；
建立模块21，配置成根据样本点云数据建立点云分类器；
分类模块22，配置成通过点云分类器对待分类的激光雷达点云数据进行分类。
如图4所示，上述建立模块21包括：
特征提取单元210，配置成对样本点云数据进行特征提取，获得分类特征；
训练单元211，配置成对分类特征进行机器学习训练，得到点云分类器。
上述特征提取单元包括：
K邻域特征提取子单元，配置成对样本点云数据进行K邻域划分，并获取 K邻域分类特征；
格网邻域特征提取子单元，配置成对样本点云数据进行格网邻域划分，并获取格网邻域分类特征；
圆柱形邻域特征提取子单元，配置成对样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
球形邻域特征提取子单元，配置成对样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
上述K邻域特征提取子单元，配置成从样本点云数据中选取与第一样本点相邻的K个邻域点，第一样本点为样本点云数据中的任一点；构建第一样本点与K个邻域点的协方差矩阵；根据协方差矩阵，计算第一样本点对应的K邻域分类特征。
上述格网邻域特征提取子单元，配置成将样本点云数据划分为多个预设尺寸的格网；获取第一格网中的最大点云高程值及与第一格网相邻的第二格网中的最小点云高程值，第一格网为划分出的任一格网；计算最大点云高程值与最小点云高程值之间的差值，将差值确定为第一格网对应的格网邻域分类特征。
上述圆柱形邻域特征提取子单元，配置成在样本点云数据中，划分出以第一样本点为中心且半径为R高度为H的圆柱形邻域，第一样本点为样本点云数据中的任一点；按照预设分层规则对第一样本点对应的圆柱形邻域进行点云分层；统计每层包括的点数、每层包括的每个点的高程值及每层的中心点高度；根据每层包括的点数、每层包括的每个点的高程值及每层的中心点高度，计算第一样本点对应的圆柱形邻域分类特征。
上述球形邻域特征提取子单元，配置成在样本点云数据中，划分出以第一样本点为中心且半径为r的球形邻域，第一样本点为样本点云数据中的任一点；计算球形邻域内所有点的高程方差，将高程方差确定为第一样本点对应的球形邻域分类特征。
上述分类模块22，配置成将待分类的激光雷达点云数据输入点云分类器，获得点云分类结果，点云分类结果包括地面点、电力线和杆塔。
如图4所示，该装置还包括：
优化模块23，配置成对点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对点云分类结果进行分类优化。
在本发明实施例中，获取样本点云数据及待分类的激光雷达点云数据；根据样本点云数据建立点云分类器；通过点云分类器对待分类的激光雷达点云数据进行分类。本发明通过样本点云数据训练出点云分类器，通过点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低。用于训练点云分类器的样本点云数据涵盖各种塔型的杆塔数据及各种线型的电力线数据，数据全面，通过该样本点云数据训练出的点云分类器的准确性很高，不易出错。而且自动分类得到分类结果后还进行散斑合并优化，以及根据杆塔位置文件及预设优化规则进行优化，进一步提高分类准确性。
本发明实施例所提供的激光雷达点云数据分类装置可以为设备上的特定硬件或者安装于设备上的软件或固件等。本发明实施例所提供的装置，其实现原理及产生的技术效果和前述方法实施例相同，为简要描述，装置实施例部分未提及之处，可参考前述方法实施例中相应内容。所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，前述描述的系统、装置和单元的具体工作过程，均可以参考上述方法实施例中的对应过程，在此不再赘述。
进一步，本发明实施例提供了一种激光雷达点云数据分类设备，包括存储器以及处理器，存储器配置成存储支持处理器执行前述任一项激光雷达点云数据分类方法的程序，处理器配置成执行存储器中存储的程序。
进一步，本发明实施例提供了一种计算机存储介质，配置成储存为前述任一项激光雷达点云数据分类方法所用的计算机软件指令。
在本发明所提供的实施例中，应该理解到，所揭露装置和方法，可以通过其它的方式实现。以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，又例如，多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些通信接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本发明提供的实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步定义和解释，此外，术语“第一”、“第二”、“第三”等仅用于区分描述，而不能理解为 指示或暗示相对重要性。
最后应说明的是：以上所述实施例，仅为本发明的具体实施方式，用以说明本发明的技术方案，而非对其限制，本发明的保护范围并不局限于此，尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本发明实施例技术方案的精神和范围。都应涵盖在本发明的保护范围之内。因此，本发明的保护范围应以所述权利要求的保护范围为准。
工业实用性：
通过应用本申请的技术方案，能够利用训练出的点云分类器对待分类的激光雷达点云数据进行自动分类，大大减少分类过程中的人工干预成份，自动化程度高，成本低，效率和准确性都很高，不易出错。

Claims (17)

  1. 一种激光雷达点云数据分类方法，其特征在于，所述方法包括：
    获取样本点云数据及待分类的激光雷达点云数据；
    根据所述样本点云数据建立点云分类器；
    通过所述点云分类器对所述待分类的激光雷达点云数据进行分类。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述样本点云数据建立点云分类器，包括：
    对所述样本点云数据进行特征提取，获得分类特征；
    对所述分类特征进行机器学习训练，得到点云分类器。
  3. 根据权利要求2所述的方法，其特征在于，所述对所述样本点云数据进行特征提取，获得分类特征，包括：
    对所述样本点云数据进行K邻域划分，并获取K邻域分类特征；
    对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征；
    对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
    对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
  4. 根据权利要求3所述的方法，其特征在于，所述对所述样本点云数据进行K邻域划分，并获取K邻域分类特征，包括：
    从所述样本点云数据中选取与第一样本点相邻的K个邻域点，所述第一样本点为所述样本点云数据中的任一点；
    构建所述第一样本点与所述K个邻域点的协方差矩阵；
    根据所述协方差矩阵，计算所述第一样本点对应的K邻域分类特征。
  5. 根据权利要求3所述的方法，其特征在于，所述对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征，包括：
    将所述样本点云数据划分为多个预设尺寸的格网；
    获取第一格网中的最大点云高程值及与所述第一格网相邻的第二格网中的最小点云高程值，所述第一格网为划分出的任一格网；
    计算所述最大点云高程值与所述最小点云高程值之间的差值，将所述差值确定为所述第一格网对应的格网邻域分类特征。
  6. 根据权利要求3所述的方法，其特征在于，所述对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征，包括：
    在所述样本点云数据中，划分出以第一样本点为中心且半径为R高度为H的圆柱形邻域，所述第一样本点为所述样本点云数据中的任一点；
    按照预设分层规则对所述第一样本点对应的圆柱形邻域进行点云分层；
    统计每层包括的点数、每层包括的每个点的高程值及每层的中心点高度；
    根据所述每层包括的点数、所述每层包括的每个点的高程值及所述每层的中心点高度，计算所述第一样本点对应的圆柱形邻域分类特征。
  7. 根据权利要求3所述的方法，其特征在于，所述对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征，包括：
    在所述样本点云数据中，划分出以第一样本点为中心且半径为r的球形邻域，所述第一样本点为所述样本点云数据中的任一点；
    计算所述球形邻域内所有点的高程方差，将所述高程方差确定为所述第一样本点对应的球形邻域分类特征。
  8. 根据权利要求1至7任一项所述的方法，其特征在于，所述通过所述点云分类器对所述待分类的激光雷达点云数据进行分类，包括：
    将所述待分类的激光雷达点云数据输入所述点云分类器，获得点云分类结果，所述点云分类结果包括地面点、电力线和杆塔。
  9. 根据权利要求8所述的方法，其特征在于，所述通过所述点云分类器 对所述待分类的激光雷达点云数据进行分类之后，还包括：
    对所述点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对所述点云分类结果进行分类优化。
  10. 根据权利要求1至9任一项所述的方法，其特征在于，所述样本点云数据包括经选择得到的杆塔点云数据、电力线点云数据和地面点点云数据。
  11. 一种激光雷达点云数据分类装置，其特征在于，所述装置包括：
    获取模块，配置成获取样本点云数据及待分类的激光雷达点云数据；
    建立模块，配置成根据所述样本点云数据建立点云分类器；
    分类模块，配置成通过所述点云分类器对所述待分类的激光雷达点云数据进行分类。
  12. 根据权利要求11所述的装置，其特征在于，所述建立模块包括：
    特征提取单元，配置成对所述样本点云数据进行特征提取，获得分类特征；
    训练单元，配置成对所述分类特征进行机器学习训练，得到点云分类器。
  13. 根据权利要求12所述的装置，其特征在于，所述特征提取单元包括：
    K邻域特征提取子单元，配置成对所述样本点云数据进行K邻域划分，并获取K邻域分类特征；
    格网邻域特征提取子单元，配置成对所述样本点云数据进行格网邻域划分，并获取格网邻域分类特征；
    圆柱形邻域特征提取子单元，配置成对所述样本点云数据进行圆柱形邻域划分，并获取圆柱形邻域分类特征；
    球形邻域特征提取子单元，配置成对所述样本点云数据进行球形邻域划分，并获取球形邻域分类特征。
  14. 根据权利要求11至13任一项所述的装置，其特征在于，所述分类模块配置成：
    将所述待分类的激光雷达点云数据输入所述点云分类器，获得点云分类结果，所述点云分类结果包括地面点、电力线和杆塔。
  15. 根据权利要求14所述的装置，其特征在于，所述装置还包括：
    优化模块，配置成对所述点云分类结果进行散斑合并优化，以及根据杆塔位置文件及预设优化规则对所述点云分类结果进行分类优化。
  16. 一种激光雷达点云数据分类设备，其特征在于，包括存储器以及处理器，所述存储器配置成存储支持处理器执行权利要求1至10任一项所述方法的程序，所述处理器配置成执行所述存储器中存储的程序。
  17. 一种计算机存储介质，其特征在于，配置成储存为权利要求1至10任一项所述方法所用的计算机软件指令。
PCT/CN2017/117607 2017-11-29 2017-12-21 激光雷达点云数据分类方法、装置、设备及存储介质 WO2019104780A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/304,673 US11636289B2 (en) 2017-11-29 2017-12-21 Method, apparatus, and device for classifying LiDAR point cloud data, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711222953.2 2017-11-29
CN201711222953.2A CN108133227A (zh) 2017-11-29 2017-11-29 激光雷达点云数据分类方法及装置

Publications (1)

Publication Number Publication Date
WO2019104780A1 true WO2019104780A1 (zh) 2019-06-06

Family

ID=62389766

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/117607 WO2019104780A1 (zh) 2017-11-29 2017-12-21 激光雷达点云数据分类方法、装置、设备及存储介质

Country Status (3)

Country Link
US (1) US11636289B2 (zh)
CN (1) CN108133227A (zh)
WO (1) WO2019104780A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110910407A (zh) * 2019-11-25 2020-03-24 南京林业大学 基于移动激光扫描点云数据的行道树树干提取方法
CN111258331A (zh) * 2020-01-20 2020-06-09 北京拓维思科技有限公司 无人机电力线路运维检修系统及方法
CN113761238A (zh) * 2021-08-27 2021-12-07 广州文远知行科技有限公司 点云存储方法、装置、设备及存储介质
CN113759947A (zh) * 2021-09-10 2021-12-07 中航空管系统装备有限公司 基于激光雷达的机载飞行避障辅助方法、装置及系统
CN114332848A (zh) * 2022-03-10 2022-04-12 北京云圣智能科技有限责任公司 点云标注方法、装置及服务器
CN114563796A (zh) * 2022-03-03 2022-05-31 北京华宜信科技有限公司 基于无人机激光雷达探测数据的电力线识别方法和装置
CN115685222A (zh) * 2022-11-14 2023-02-03 国网湖北省电力有限公司超高压公司 一种基于激光点云数据的电力线塔架自动检测方法

Families Citing this family (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108959173B (zh) * 2018-07-06 2022-02-11 北京数字绿土科技有限公司 一种激光雷达点云数据的解算方法及装置
CN109063789A (zh) * 2018-09-07 2018-12-21 北京数字绿土科技有限公司 杆塔和电力线的分类方法、装置和系统
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
CN110097077B (zh) * 2019-03-26 2021-12-10 深圳市速腾聚创科技有限公司 点云数据分类方法、装置、计算机设备和存储介质
US11618502B2 (en) * 2019-03-28 2023-04-04 Volkswagen Aktiengesellschaft On-road localization methodologies and equipment utilizing road surface characteristics
CN112053365B (zh) * 2019-06-06 2022-06-14 武汉星源云意科技有限公司 一种基于线性特征的输电线路导线激光点云数据的提取方法
CN110794424B (zh) * 2019-11-13 2022-04-05 湖北大学 基于特征选择的全波形机载激光雷达地物分类方法及系统
CN111145159B (zh) * 2019-12-27 2023-05-23 北京数字绿土科技股份有限公司 提取巡检关键部件点的方法及装置
CN112488215A (zh) * 2020-12-04 2021-03-12 广东电网有限责任公司 一种电缆沟点云自动分类方法
CN112446343B (zh) * 2020-12-07 2024-03-15 园测信息科技股份有限公司 一种融合多尺度特征的车载点云道路杆状物机器学习自动提取方法
CN113076870A (zh) * 2021-03-31 2021-07-06 广东电网有限责任公司 一种基于激光点云的杆塔识别方法及系统
CN114639024A (zh) * 2022-03-03 2022-06-17 江苏方天电力技术有限公司 一种输电线路激光点云自动分类方法
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
CN115049938B (zh) * 2022-08-16 2022-11-29 广州地铁设计研究院股份有限公司 一种点云图像处理方法及系统
WO2024050594A1 (en) * 2022-09-05 2024-03-14 Baraja Pty Ltd Spatial profiling systems and methods
CN115546544B (zh) * 2022-09-30 2023-11-17 深圳市规划和自然资源数据管理中心 基于图神经网络的LiDAR点云与OSM标注信息流耦合分类方法
CN115422981B (zh) * 2022-11-04 2023-04-07 自然资源部第一海洋研究所 面向单频机载激光测深数据的水陆分类方法、系统及应用
CN116129144A (zh) * 2023-02-23 2023-05-16 北京数字绿土科技股份有限公司 一种基于激光点云的光伏面板单体的自动提取方法和系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103324945A (zh) * 2013-07-08 2013-09-25 南京大学 一种基于模式识别的森林点云分类方法
CN106022381A (zh) * 2016-05-25 2016-10-12 厦门大学 基于车载激光扫描点云的路灯杆自动提取技术
CN106650809A (zh) * 2016-12-20 2017-05-10 福州大学 一种车载激光点云目标分类方法和系统
CN107292335A (zh) * 2017-06-06 2017-10-24 云南电网有限责任公司信息中心 一种基于随机森林模型的输电线路点云数据自动分类方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9396545B2 (en) * 2010-06-10 2016-07-19 Autodesk, Inc. Segmentation of ground-based laser scanning points from urban environment
CN105373814B (zh) * 2014-08-26 2019-04-02 南京林业大学 一种基于激光点云数据的真实阔叶树器官分类识别方法
CN104282038A (zh) * 2014-09-26 2015-01-14 北京航空航天大学 基于局部子邻域划分的点云特征点提取方法
US10032267B2 (en) * 2016-06-09 2018-07-24 Lockheed Martin Corporation Automating the assessment of damage to infrastructure assets

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103324945A (zh) * 2013-07-08 2013-09-25 南京大学 一种基于模式识别的森林点云分类方法
CN106022381A (zh) * 2016-05-25 2016-10-12 厦门大学 基于车载激光扫描点云的路灯杆自动提取技术
CN106650809A (zh) * 2016-12-20 2017-05-10 福州大学 一种车载激光点云目标分类方法和系统
CN107292335A (zh) * 2017-06-06 2017-10-24 云南电网有限责任公司信息中心 一种基于随机森林模型的输电线路点云数据自动分类方法

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110910407A (zh) * 2019-11-25 2020-03-24 南京林业大学 基于移动激光扫描点云数据的行道树树干提取方法
CN110910407B (zh) * 2019-11-25 2023-09-15 南京林业大学 基于移动激光扫描点云数据的行道树树干提取方法
CN111258331A (zh) * 2020-01-20 2020-06-09 北京拓维思科技有限公司 无人机电力线路运维检修系统及方法
CN113761238A (zh) * 2021-08-27 2021-12-07 广州文远知行科技有限公司 点云存储方法、装置、设备及存储介质
CN113761238B (zh) * 2021-08-27 2022-08-23 广州文远知行科技有限公司 点云存储方法、装置、设备及存储介质
CN113759947A (zh) * 2021-09-10 2021-12-07 中航空管系统装备有限公司 基于激光雷达的机载飞行避障辅助方法、装置及系统
CN113759947B (zh) * 2021-09-10 2023-08-08 中航空管系统装备有限公司 基于激光雷达的机载飞行避障辅助方法、装置及系统
CN114563796A (zh) * 2022-03-03 2022-05-31 北京华宜信科技有限公司 基于无人机激光雷达探测数据的电力线识别方法和装置
CN114563796B (zh) * 2022-03-03 2022-11-04 北京华宜信科技有限公司 基于无人机激光雷达探测数据的电力线识别方法和装置
CN114332848A (zh) * 2022-03-10 2022-04-12 北京云圣智能科技有限责任公司 点云标注方法、装置及服务器
CN115685222A (zh) * 2022-11-14 2023-02-03 国网湖北省电力有限公司超高压公司 一种基于激光点云数据的电力线塔架自动检测方法
CN115685222B (zh) * 2022-11-14 2023-06-23 国网湖北省电力有限公司超高压公司 一种基于激光点云数据的电力线塔架自动检测方法

Also Published As

Publication number Publication date
US20210224613A1 (en) 2021-07-22
US11636289B2 (en) 2023-04-25
CN108133227A (zh) 2018-06-08

Similar Documents

Publication Publication Date Title
WO2019104780A1 (zh) 激光雷达点云数据分类方法、装置、设备及存储介质
WO2019104781A1 (zh) 一种点云数据处理方法、装置、电子设备及计算机可读存储介质
CN111784685B (zh) 一种基于云边协同检测的输电线路缺陷图像识别方法
CN109118479B (zh) 基于胶囊网络的绝缘子缺陷识别定位装置及方法
CN110148130B (zh) 用于检测零件缺陷的方法和装置
CN111340787B (zh) 输电线路的导线缺陷检测识别方法、装置及计算机设备
WO2018028103A1 (zh) 一种基于人眼视觉特性的电力线路无人机巡检方法
JP6197659B2 (ja) 検出制御装置、プログラム及び検出システム
CN113095438B (zh) 晶圆缺陷分类方法及其装置、系统、电子设备和存储介质
US20150071541A1 (en) Automated method for measuring, classifying, and matching the dynamics and information passing of single objects within one or more images
CN109935080B (zh) 一种交通线路上的车流量实时计算的监测系统及方法
CN111144325A (zh) 变电站电力设备的故障识别定位方法、装置及设备
CN112613569B (zh) 图像识别方法、图像分类模型的训练方法及装置
CN110133443B (zh) 基于平行视觉的输电线路部件检测方法、系统、装置
CN113240623B (zh) 一种路面病害检测方法及装置
CN112036384B (zh) 精子头部形态识别方法、装置及设备
CN111105398A (zh) 一种基于可见光图像数据的输电线路部件裂纹检测方法
Xin et al. Three‐dimensional reconstruction of Vitis vinifera (L.) cvs Pinot Noir and Merlot grape bunch frameworks using a restricted reconstruction grammar based on the stochastic L‐system
Huang et al. Structural defect detection technology of transmission line damper based on UAV image
CN111597939B (zh) 一种基于深度学习的高铁线路鸟窝缺陷检测方法
CN109657540A (zh) 枯死树定位方法及系统
CN112529836A (zh) 高压线路缺陷检测方法、装置、存储介质及电子设备
CN112446231A (zh) 一种人行横道检测方法、装置、计算机设备及存储介质
CN115239947A (zh) 基于无监督学习的小麦条锈病严重度评估方法及装置
CN114782822A (zh) 电力设备的状态检测方法、装置、电子设备及存储介质

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: 17933284

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17933284

Country of ref document: EP

Kind code of ref document: A1