CN117589167A

CN117589167A - Unmanned aerial vehicle routing inspection route planning method based on three-dimensional point cloud model

Info

Publication number: CN117589167A
Application number: CN202311463729.8A
Authority: CN
Inventors: 宋海彬; 杨海峰; 江海; 谢仁杰; 姚传涛; 孙浩; 王健; 邓柱锋; 杨宇; 蒋益; 郭运宏; 董军
Original assignee: Tianshengqiao Bureau of Extra High Voltage Power Transmission Co
Current assignee: Tianshengqiao Bureau of Extra High Voltage Power Transmission Co
Priority date: 2023-11-06
Filing date: 2023-11-06
Publication date: 2024-02-23

Abstract

The invention relates to the technical field of unmanned aerial vehicle route planning, in particular to an unmanned aerial vehicle routing inspection route planning method based on a three-dimensional point cloud model. According to the invention, the inspection area is thoroughly scanned by a laser radar scanning technology, high precision and integrity of the obtained original three-dimensional point cloud data are ensured, high-quality processing and feature extraction of the point cloud data are ensured by application of a RANSAC filtering and DBSCAN clustering algorithm, accurate identification and association of the area are realized by combining a feature matching method with a random forest classifier, and an unmanned aerial vehicle can adaptively plan a route according to topography and obstacles by application of an A-type search algorithm and a topography constraint factor in path planning, so that operability of the unmanned aerial vehicle in a complex environment is enhanced, real-time acquisition and processing of the data are ensured by introduction of an edge computing technology, and timeliness and accuracy of an inspection report are ensured.

Description

Unmanned aerial vehicle routing inspection route planning method based on three-dimensional point cloud model

Technical Field

The invention relates to the technical field of unmanned aerial vehicle route planning, in particular to an unmanned aerial vehicle routing inspection route planning method based on a three-dimensional point cloud model.

Background

Unmanned aerial vehicle route planning technical field has covered the design and has implemented unmanned aerial vehicle's route to ensure that unmanned aerial vehicle can carry out various tasks effectively, such as inspection, monitoring, reconnaissance and search etc.. This field combines knowledge in the fields of aeronautics engineering, geographic Information Systems (GIS), computer science and autonomous flight control.

The unmanned aerial vehicle routing planning method is a route design method aiming at specific tasks, and aims to improve efficiency, comprehensively cover a target area, ensure flight safety and realize unmanned aerial vehicle autonomy. The aim is achieved by utilizing map data, a target recognition technology, a path planning algorithm and a real-time adjustment mechanism, so that the unmanned aerial vehicle can efficiently and safely autonomously plan and fly a route when the unmanned aerial vehicle executes a task. The technology has wide application, and covers the knowledge of various fields such as aviation engineering, geographic information systems, computer science, autonomous flight control and the like.

In the unmanned aerial vehicle routing planning method, most of the existing methods adopt the traditional method on point cloud data processing, and are easily affected by noise, so that the data processing result is unstable. In addition, many existing schemes do not fully utilize advanced algorithms to segment data and extract features, so that accuracy and efficiency of object identification are difficult to guarantee. In the aspect of route planning, the traditional method often depends on manual setting or a simple algorithm, lacks self-adaptive capability to complex environments, and is easy to generate conflict or detour in complex terrains, so that the route efficiency is low. Finally, many existing methods do not combine with edge computing technology to process real-time data, so that the feedback of the inspection result is delayed and can not respond to possible abnormal conditions in time, and the risk and uncertainty of inspection are increased.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides an unmanned aerial vehicle routing planning method based on a three-dimensional point cloud model.

In order to achieve the above purpose, the present invention adopts the following technical scheme: an unmanned aerial vehicle routing planning method based on a three-dimensional point cloud model comprises the following steps:

s1: scanning the inspection area or equipment based on a laser radar scanning method to generate original three-dimensional point cloud data;

s2: based on the original three-dimensional point cloud data, adopting a point cloud processing tool specifically for RANSAC filtering to carry out data filtering and downsampling, and generating processed three-dimensional point cloud data;

s3: based on the processed three-dimensional point cloud data, a DBSCAN clustering algorithm is adopted to segment the data and extract features, and a segmented point cloud area and feature description are generated;

s4: based on the segmented point cloud region, performing object identification by using a random forest classifier, and associating equipment or a region by combining a feature matching method to generate an object identification tag and an association relationship;

s5: according to the object identification tag and the feature description, a search algorithm A and a terrain constraint factor are adopted to conduct path planning, and a preliminary inspection path is generated;

S6: based on the preliminary inspection path, performing path optimization by using a simulated annealing algorithm and a path efficiency evaluation function, and generating an optimized inspection path;

s7: based on the optimized routing inspection path, adopting a route generating tool and a Geographic Information System (GIS) to determine a route, and generating an unmanned aerial vehicle route;

s8: defining flight parameters including height, speed and attitude based on the unmanned aerial vehicle route and combining with a flight dynamics model, and generating flight rule parameters;

s9: based on the flight rule parameters, performing unmanned aerial vehicle inspection by using an edge computing technology, collecting and processing data in real time, and generating an inspection report, wherein the inspection report comprises equipment state evaluation and abnormal condition analysis.

As a further scheme of the invention, based on a laser radar scanning method, scanning a patrol area or equipment, and generating original three-dimensional point cloud data comprises the following steps:

s101: selecting a full-field scanning algorithm based on the target area, configuring the equipment position, and generating an equipment configuration scheme;

s102: based on the equipment configuration scheme, setting radar scanning parameters by adopting a parameter initialization method, and generating initialization setting parameters;

S103: starting a continuous scanning method based on the initialization setting parameters, acquiring radar data, and generating original scanning data;

s104: based on the original scan data, performing data preprocessing by adopting a background denoising algorithm to generate optimized scan data;

s105: based on the optimized scanning data, converting the scanning data into a point cloud format by adopting a three-dimensional conversion method, and generating original three-dimensional point cloud data.

As a further scheme of the present invention, based on the original three-dimensional point cloud data, a point cloud processing tool, specifically, RANSAC filtering is adopted to perform data filtering and downsampling, and the steps of generating the processed three-dimensional point cloud data specifically include:

s201: based on the original three-dimensional point cloud data, adopting a RANSAC filtering algorithm to perform preliminary filtering to generate point cloud data after the RANSAC filtering;

s202: based on the RANSAC filtered point cloud data, adopting an outlier removal method to clean the data to obtain denoised point cloud data;

s203: based on the denoised point cloud data, performing data dimension reduction by adopting a downsampling algorithm to generate downsampled point cloud data;

s204: and based on the down-sampled point cloud data, adopting a quality inspection method to confirm the data integrity and establish the processed three-dimensional point cloud data.

As a further scheme of the invention, based on the processed three-dimensional point cloud data, a DBSCAN clustering algorithm is adopted to segment and extract features of the data, and the steps of generating the segmented point cloud region and feature description are specifically as follows:

s301: setting parameters for an algorithm based on the processed three-dimensional point cloud data to obtain DBSCAN parameter configuration;

s302: based on the DBSCAN parameter configuration, a DBSCAN clustering algorithm is adopted to perform data segmentation, and a preliminary clustering result is generated;

s303: based on the preliminary clustering result, removing outlier areas by adopting an area optimization method, and obtaining an optimized clustering result;

s304: describing by adopting a feature extraction method based on the optimized clustering result to obtain a point cloud feature description;

s305: based on the point cloud feature description, classifying by adopting a region labeling method, and generating a segmented point cloud region and feature description.

As a further scheme of the invention, based on the segmented point cloud region, the method uses a random forest classifier to perform object recognition, and combines a feature matching method to associate equipment or regions, and the step of generating an object recognition tag and an association relation is specifically as follows:

S401: based on the segmented point cloud region, carrying out feature description on the point cloud by adopting a feature extraction algorithm to generate a point cloud feature set;

s402: classifying the features by adopting a random forest classifier based on the point cloud feature set to generate a preliminary object identification tag;

s403: based on the preliminary object identification tag, adopting a feature matching algorithm to match with known equipment or areas to generate a detailed object association tag;

s404: and based on the detailed object association tag, determining the interrelationship among the objects by adopting an association analysis algorithm, and generating an object identification tag and an association relation.

As a further scheme of the present invention, according to the object identification tag and the feature description, a search algorithm and a terrain constraint factor are adopted to perform path planning, and the step of generating a preliminary inspection path specifically includes:

s501: based on scene data, extracting topographic features by adopting a topographic analysis method to generate topographic data;

s502: based on the topographic data and the object identification tag, generating a preliminary path plan by adopting an A search algorithm;

s503: based on the preliminary path planning, carrying out path adjustment by combining with a terrain constraint factor, and generating an adjusted path planning;

S504: and performing path detail improvement based on the adjusted path planning to generate a preliminary inspection path.

As a further scheme of the invention, based on the preliminary inspection path, path optimization is performed by using a simulated annealing algorithm and a path efficiency evaluation function, and the step of generating an optimized inspection path specifically comprises the following steps:

s601: based on the preliminary inspection path, performing path evaluation by adopting a path efficiency evaluation function, and generating a path efficiency evaluation result;

s602: based on the path efficiency evaluation result, adopting a simulated annealing algorithm to perform optimization search on the path to generate an optimized routing inspection path;

s603: based on the optimized routing inspection path, carrying out path feasibility verification and generating a verified routing inspection path;

s604: and carrying out detail optimization based on the verified inspection path to generate an optimized inspection path.

As a further scheme of the invention, based on the optimized routing inspection path, a route generating tool and a geographic information system GIS are adopted to determine a route, and the steps of generating the unmanned aerial vehicle route are specifically as follows:

s701: based on the optimized routing inspection path, performing path conversion by adopting a Dijkstra algorithm to generate preliminary route data;

S702: based on the preliminary route data, screening an optimal flight area by using a spatial analysis method of a Geographic Information System (GIS) to obtain screened route data;

s703: based on the screened route data, determining the waypoints of the unmanned aerial vehicle by adopting a waypoint optimization algorithm, and acquiring the route details of the unmanned aerial vehicle;

s704: and integrating the unmanned aerial vehicle route details, and generating the unmanned aerial vehicle route by adopting a linear interpolation method.

As a further scheme of the invention, based on the unmanned aerial vehicle route and in combination with a flight dynamics model, the steps of defining flight parameters including height, speed and attitude and generating flight rule parameters are specifically as follows:

s801: based on the unmanned aerial vehicle route, a flight dynamics model is adopted to predict the flight demand, and a flight demand analysis report is obtained;

s802: setting a flight altitude parameter by using a PID control algorithm based on the flight demand analysis report;

s803: referring to the flight demand analysis report, formulating a flight speed parameter and a flight attitude parameter by utilizing an attitude control algorithm;

s804: and generating flight rule parameters by adopting a rule formulation algorithm based on the flight altitude parameters, the flight speed parameters and the flight attitude parameters.

As a further scheme of the invention, based on the flight rule parameters, the unmanned aerial vehicle inspection is executed by utilizing an edge computing technology, data are collected and processed in real time, and an inspection report is generated, wherein the steps of equipment state evaluation and abnormal condition analysis comprise:

s901: according to the flight rule parameters, a real-time data stream processing technology is adopted to execute unmanned aerial vehicle inspection tasks and obtain real-time inspection data;

s902: processing and storing the real-time inspection data by using an edge computing technology and a quick storage algorithm, and storing the processed inspection data;

s903: based on the processed inspection data, performing equipment state evaluation and abnormal condition analysis by using an abnormal detection algorithm to obtain equipment state evaluation and abnormal condition analysis;

s904: and based on the equipment state evaluation and the abnormal condition analysis, making a patrol report by using a report generation algorithm.

Compared with the prior art, the invention has the advantages and positive effects that:

according to the invention, the inspection area is thoroughly scanned by the laser radar scanning technology, so that the high precision and the integrity of the obtained original three-dimensional point cloud data are ensured. The application of RANSAC filtering and DBSCAN clustering algorithms further ensures high-quality processing and feature extraction of point cloud data, and provides a solid foundation for subsequent object recognition. The random forest classifier combines with the feature matching method to realize the precise identification and association of the areas. In addition, the application of the A-search algorithm and the terrain constraint factor in path planning enables an unmanned aerial vehicle to adaptively plan a route according to terrain and obstacles, and operability of the unmanned aerial vehicle in a complex environment is enhanced. The introduction of the edge computing technology ensures the real-time acquisition and processing of data and ensures the timeliness and accuracy of the inspection report.

Drawings

FIG. 1 is a schematic diagram of the main steps of the present invention;

FIG. 2 is a detailed schematic of the S1 of the present invention;

FIG. 3 is a schematic diagram of an S2 refinement of the present invention;

FIG. 4 is a schematic diagram of an S3 refinement of the present invention;

FIG. 5 is a schematic diagram of an S4 refinement of the present invention;

FIG. 6 is a schematic diagram of an S5 refinement of the present invention;

FIG. 7 is a schematic diagram of an S6 refinement of the present invention;

FIG. 8 is a schematic diagram of an S7 refinement of the present invention;

FIG. 9 is a schematic diagram of an S8 refinement of the present invention;

fig. 10 is a schematic diagram of the S9 refinement of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Example 1

Referring to fig. 1, the present invention provides a technical solution: an unmanned aerial vehicle routing planning method based on a three-dimensional point cloud model comprises the following steps:

s3: based on the processed three-dimensional point cloud data, segmenting the data and extracting features by adopting a DBSCAN clustering algorithm to generate segmented point cloud areas and feature descriptions;

s4: based on the segmented point cloud region, performing object identification by using a random forest classifier, and associating equipment or a region by combining a feature matching method to generate an object identification tag and an association relation;

S7: based on the optimized routing inspection path, adopting a route generating tool and a Geographic Information System (GIS) to carry out route determination, and generating an unmanned aerial vehicle route;

s9: based on flight rule parameters, unmanned aerial vehicle inspection is executed by utilizing an edge computing technology, data are collected and processed in real time, and inspection reports are generated, wherein the inspection reports comprise equipment state evaluation and abnormal condition analysis.

First, from the viewpoint of data acquisition, a laser radar scanning technology is adopted to ensure the careful and complete scanning of the inspection area. The high-precision scanning mode not only greatly improves the accuracy of data, but also reduces errors caused by data missing, and provides a solid foundation for subsequent processing and application.

Secondly, the RANSAC filtering and the DBSCAN clustering are introduced, so that remarkable improvement is realized in data processing and feature extraction. The method can effectively remove noise, ensure data quality, accurately extract key features from mass data and greatly enhance the identification rate of the unmanned aerial vehicle to the target.

Furthermore, the fusion of the random forest classifier and the feature matching ensures the high efficiency and accuracy of object recognition. The unmanned aerial vehicle provides strong support for the unmanned aerial vehicle in complex actual environments, such as scenes of coexistence of multiple devices, relatively close distance between the devices and the like, and ensures that the unmanned aerial vehicle can accurately and rapidly determine the inspection target.

In addition, the combination of the A-search algorithm and the terrain constraint factor enables routing inspection path planning to be more strategic and intelligent. The method can plan a safe and efficient route for the unmanned aerial vehicle no matter in mountainous regions, building groups or other complex terrains. And by combining with a simulated annealing algorithm, the optimization of the route is further ensured, so that the unmanned aerial vehicle can save energy and improve the working efficiency when executing tasks.

The GIS application not only enables the route planning to have more geographic consciousness, but also can ensure that the unmanned aerial vehicle avoids various geographic barriers in actual operation, and increases the inspection safety. In addition, the introduction of the flight dynamics model provides more scientific reference for the flight of the unmanned aerial vehicle, and ensures the stability and reliability of the flight.

Finally, the adoption of the edge computing technology ensures the real-time performance and the processing speed of the data. The unmanned aerial vehicle can rapidly capture abnormal conditions in the inspection process, generate a timely inspection report, greatly strengthen the early warning capability of potential risks of equipment and provide powerful support for related decisions.

Referring to fig. 2, based on a laser radar scanning method, scanning a patrol area or equipment to generate original three-dimensional point cloud data specifically includes:

s104: based on the original scanning data, adopting a background denoising algorithm to perform data preprocessing, and generating optimized scanning data;

In S101, device position configuration and full field scanning algorithm selection are performed

And in the target area, determining the installation position and angle of the laser radar, and ensuring complete coverage of the inspection area. The full-field scanning algorithm is selected, so that the equipment can scan the surrounding environment 360 degrees without dead angles. Configuration equipment comprises parameters such as horizontal and vertical angles, scanning resolution and the like of the laser radar.

Example code #

device_position = set_device_position(area_coordinates)

full_view_scan_algorithm = select_full_view_algorithm()

device_configuration = configure_device(device_position, full_view_scan_algorithm)

In S102, the initialization of radar scan parameters is performed

# import necessary libraries

import numpy as np

# initializing scan parameters

scan_frequency=10# sweep frequency (Hz)

angular_resolution=0.5# angular resolution (degree/dot)

max_range=50# maximum detection distance (meters)

def initialize_scan_parameters(frequency, resolution, max_range):

scan_parameters = {

"frequency": frequency,

"resolution": resolution,

"max_range": max_range

}

return scan_parameters

In S103, continuous scanning is started to acquire radar data

In actual operation, it is necessary to scan and acquire raw scan data using an API or a driver of the lidar device.

Example frame #

def perform_continuous_scan(scan_parameters):

# initializing laser radar apparatus

laser_device = initialize_laser_device()

# continuous scanning and acquisition of data

raw_scan_data = []

for _ in range(scan_parameters["duration"]):

scan_data = laser_device.scan()

raw_scan_data.append(scan_data)

return raw_scan_data

In S104, data preprocessing and background denoising are performed

Background denoising algorithms include simple moving average filtering. The following is one example code to perform moving average filtering:

def apply_background_noise_removal(raw_scan_data, window_size=5):

filtered_scan_data = []

for i in range(len(raw_scan_data)):

if i<window_size:

# does not perform filtering, directly adds raw data to the output

filtered_scan_data.append(raw_scan_data[i])

else:

# calculate the mean within the sliding window, add the result to the output

smoothed_value = np.mean(raw_scan_data[i - window_size:i])

filtered_scan_data.append(smoothed_value)

return filtered_scan_data

In S105, three-dimensional conversion is performed to generate original three-dimensional point cloud data

The preprocessed polar coordinate data is converted into three-dimensional point cloud data, using the following code example:

def convert_to_3D_point_cloud(preprocessed_scan_data):

point_cloud_data = []

for i, scan_value in enumerate(preprocessed_scan_data):

converting polar coordinates to rectangular coordinates using lidar parameters #

x = scan_value * np.cos(i * angular_resolution)

y = scan_value * np.sin(i * angular_resolution)

z=0# assuming on the ground, can be modified according to the actual situation

point_cloud_data.append((x, y, z))

return point_cloud_data

Referring to fig. 3, based on original three-dimensional point cloud data, a point cloud processing tool, specifically RANSAC filtering, is adopted to perform data filtering and downsampling, and the steps of generating the processed three-dimensional point cloud data are specifically as follows:

s202: based on the point cloud data after RANSAC filtering, adopting an outlier removing method to clean the data to obtain denoised point cloud data;

s203: based on the denoised point cloud data, adopting a downsampling algorithm to perform data dimension reduction to generate downsampled point cloud data;

In S201, lansa filtering is performed

RANSAC (Random Sample Consensus) is an algorithm for estimating model parameters and removing outliers. In this step RANSAC will be used to filter out points that do not belong to the main plane.

# import into PCL library

import pcl

# creation Point cloud object

cloud = pcl.PointCloud()

From_list (point_from_data) # uses previously generated point cloud data

# create RANSAC object

ransac = cloud.make_segmenter()

ransac.set_model_type(pcl.SACMODEL_PLANE)

ransac.set_method_type(pcl.SAC_RANSAC)

The range_distance_threshold (0.01) # threshold needs to be adjusted according to data

# perform RANSAC

inliers, coefficients = ransac.segment()

Point cloud after# extraction and RANSAC filtering

filtered_cloud = cloud.extract(inliers, negative=False)

In S202, outlier removal is performed

Outlier removal is to further remove residual outliers that do not belong to the principal plane.

# create a filtering object

outlier_filter = filtered_cloud.make_statistical_outlier_filter()

Number of points in the neighborhood outlier_filter.set_mean_k (50) #

The outlier_filter.set_std_dev_mul_thresh (1.0) # threshold needs to be adjusted according to data

# perform outlier removal

cleaned_cloud = outlier_filter.filter()

In S203, downsampling is performed

The downsampling is to reduce the density of the point cloud data, reduce the computational complexity and improve the efficiency of subsequent processing.

# create a downsampled filter object

voxel_filter = cleaned_cloud.make_voxel_grid_filter()

leaf_size=0.01# downsampled grid size, which needs to be adjusted according to the data

voxel_filter.set_leaf_size(leaf_size, leaf_size, leaf_size)

# perform downsampling

downsampled_cloud = voxel_filter.filter()

In S204, quality inspection is performed

The quality check is to confirm the integrity and validity of the data.

Size of # checkpoint cloud to ensure data integrity

if downsampled_cloud.size<1000:

print ('incomplete point cloud data, please re-collect')

else:

print ('Point cloud processing completed, data available')

Referring to fig. 4, based on the processed three-dimensional point cloud data, the data is segmented and feature extracted by adopting a DBSCAN clustering algorithm, and the steps for generating the segmented point cloud region and feature description specifically include:

s302: based on DBSCAN parameter configuration, a DBSCAN clustering algorithm is adopted to conduct data segmentation, and a preliminary clustering result is generated;

s303: based on the preliminary clustering result, removing the outlier region by adopting a region optimization method, and obtaining an optimized clustering result;

s305: based on the point cloud feature description, classifying by adopting a region labeling method to generate a segmented point cloud region and feature description.

In S301, setting DBSCAN parameters is performed

First, parameters of the DBSCAN algorithm, including eps (neighborhood radius) and min_samples (minimum number of samples in the neighborhood), need to be set. The values of these parameters need to be adjusted depending on the data and application.

from sklearn.cluster import DBSCAN

# setting DBSCAN parameters

eps = 0.1 # neighborhood radius

min_samples=minimum number of samples in 10# neighborhood

dbscan = DBSCAN(eps=eps, min_samples=min_samples)

In S302, DBSCAN clustering is performed

Clustering the point cloud data by using a DBSCAN algorithm. This will generate a preliminary clustering result.

# use DBSCAN clustering

cluster_labels = dbscan.fit_predict(processed_point_cloud)

In S303, region optimization is performed

In the clustering result, some outliers or unwanted areas are included. The outlier region may be removed using a region optimization method, such as based on cluster size or other attributes.

import numpy as np

Removal of outlier region #

unique_labels = np.unique(cluster_labels)

optimal_clusters = []

for label in unique_labels:

if label == -1:

continuous # skip noise point

cluster_points = processed_point_cloud[cluster_labels == label]

if len(cluster_points)>= min_cluster_size:

optimal_clusters.append(cluster_points)

In S304, feature extraction is performed

For each optimized cluster region, they can be described using various feature extraction methods. The following is an example, using the mean and standard deviation of point cloud data as features.

# feature extraction example: mean and standard deviation

cluster_features = []

for cluster in optimal_clusters:

cluster_mean = np.mean(cluster, axis=0)

cluster_std = np.std(cluster, axis=0)

cluster_features.append(np.concatenate((cluster_mean, cluster_std)))

In S305, region labeling and classification are performed

Finally, the feature descriptions may be used to classify the clustered regions. May be a machine learning classification task, for example using a Support Vector Machine (SVM) or a deep learning model.

from sklearn.svm import SVC

# create SVM classifier and train

svm_classifier = SVC()

svm_classifier.fit(features, labels)

Sorting new data #)

predicted_labels = svm_classifier.predict(new_features)

Referring to fig. 5, based on the segmented point cloud area, the method uses a random forest classifier to perform object recognition, and combines a feature matching method to associate equipment or an area, so as to generate an object recognition tag and an association relation specifically as follows:

S404: based on the detailed object association tag, an association analysis algorithm is adopted to determine the interrelationship among the objects, and an object identification tag and an association relation are generated.

In S401, feature extraction is performed

First, feature extraction needs to be performed for each point cloud region for object recognition. The following is an example, using the geometric and statistical features of the point cloud.

# feature extraction example: geometric and statistical features

features = []

for region in segmented_regions:

Extraction of geometric features, e.g. volume, surface area, etc. of regions

geometric_features = extract_geometric_features(region)

# extract statistical features, e.g. mean, standard deviation, etc. of regions

statistical_features = extract_statistical_features(region)

region_features = np.concatenate((geometric_features, statistical_features))

features.append(region_features)

In S402, random forest classification is performed

The extracted features are classified using a random forest classifier, thereby generating a preliminary object identification tag.

from sklearn.ensemble import RandomForestClassifier

Random forest classifier for# creation and training

rf_classifier = RandomForestClassifier(n_estimators=100, random_state=0)

rf_classifier.fit(features, initial_labels)

# predictive object identification tag

predicted_labels = rf_classifier.predict(features)

In S403, feature matching is performed

At this step, the preliminary object identification tags need to be matched against known devices or areas to generate detailed object association tags. To using feature matching algorithms such as nearest neighbor searches or feature-based matching.

from sklearn.neighbors import NearestNeighbors

# creation of nearest neighbor model

k=1# can be adjusted as desired

nn_model = NearestNeighbors(n_neighbors=k)

nn_model.fit(known_features)

Matching each region #)

detailed_labels = []

for feature in features:

Finding nearest neighbor known device or region #

distances, indices = nn_model.kneighbors([feature])

Taking the label closest to the label as a detailed object association label

detailed_labels.append(known_labels[indices[0][0]])

In S404, association analysis is performed

Finally, based on the detailed object association tags, association analysis algorithms may be used to determine interrelationships between objects, such as association rule mining or network analysis.

# performing correlation analysis

# includes determining connections, dependencies, or other relationships between objects

association_rules = perform_association_analysis(detailed_labels)

Referring to fig. 6, according to the object identification tag and the feature description, a search algorithm and a terrain constraint factor are adopted to perform path planning, and the step of generating a preliminary inspection path specifically includes:

s502: based on the topographic data and the object identification label, generating a preliminary path plan by adopting an A-search algorithm;

s503: based on the preliminary path planning, combining with the terrain constraint factor, carrying out path adjustment to generate an adjusted path planning;

In S501, topographic data extraction is performed

First, scene data needs to be acquired and topographical features extracted therefrom for path planning. The terrain data may include elevation, grade, obstacle location, etc. Such data may be obtained from remote sensing images, laser scan data, or geographic information systems.

# terrain data extraction example: acquiring elevation information from an elevation map

elevation_map = load_elevation_map("elevation_data.GIF")

# can use various methods to parse the terrain data, such as GDAL library or other GIS tools

In S502, a path planning is performed

Using an a-search algorithm, a preliminary path plan is generated based on the terrain data and the object identification tag.

Path planning example #A #)

from queue import PriorityQueue

def astar(start, goal, terrain_map, object_labels):

open_list = PriorityQueue()

open_list.put(start)

ca_from= { } # is used to track paths

g_score = {point: float('inf') for point in terrain_map}

g_score[start] = 0

while not open_list.empty():

current = open_list.get()

if current == goal:

return reconstruct_path(came_from, current)

for neighbor in get_neighbors(current):

tentative_g_score = g_score[current] + terrain_cost(current, neighbor, terrain_map)

if tentative_g_score<g_score[neighbor]:

came_from[neighbor] = current

g_score[neighbor] = tentative_g_score

f_score = tentative_g_score + heuristic(neighbor, goal)

open_list.put((f_score, neighbor))

return None # does not find a path

Detailed implementation requires adjustment according to terrain and object identification information

In S503, path adjustment is performed

At this step, the preliminary path may be adjusted in conjunction with terrain constraint factors to ensure that the path takes into account terrain features, including avoiding cliffs, rivers or other obstacles.

# Path adjustment example: avoiding steep slopes

adjusted_path = adjust_path(initial_path, elevation_map)

In S504, execution path detail is perfected

And finally, performing detail refinement of the path, including considering the position and the size of the object, so as to generate a final inspection path.

# Path detail perfection example: avoiding objects

final_path = refine_path(adjusted_path, object_labels)

Referring to fig. 7, based on the preliminary inspection path, path optimization is performed by using a simulated annealing algorithm and a path efficiency evaluation function, and the steps of generating the optimized inspection path are specifically as follows:

s603: based on the optimized routing inspection path, carrying out feasibility verification of the path, and generating a verified routing inspection path;

s604: and carrying out detail optimization based on the verified inspection path, and generating an optimized inspection path.

In S601, path evaluation is performed

First, the quality of the preliminary patrol path is evaluated using a path efficiency evaluation function. The path efficiency evaluation function may consider path length, ability to avoid obstacles, and other path-related metrics.

Example of Path efficiency evaluation function

def path_efficiency(path, terrain_map, object_labels):

# evaluation according to factors such as path Length and obstacle avoidance

efficiency_score = calculate_efficiency(path, terrain_map, object_labels)

return efficiency_score

In S602, performing simulated annealing algorithm optimization

And (5) performing optimized search on the preliminary inspection path by using a simulated annealing algorithm. The simulated annealing algorithm is a heuristic algorithm and can be used for global optimization. In the algorithm, an energy function needs to be defined, and energy is calculated according to the efficiency evaluation result of the path.

Example # simulated annealing algorithm

import random

import math

def simulated_annealing(initial_path, temperature, cooling_rate):

current_path = initial_path

best_path = initial_path

current_energy = path_efficiency(current_path)

best_energy = current_energy

while temperature>1:

new_path = generate_neighbor_path(current_path)

new_energy = path_efficiency(new_path)

energy_diff = new_energy - current_energy

if energy_diff<0 or random.random()<math.exp(-energy_diff / temperature):

current_path = new_path

current_energy = new_energy

if current_energy<best_energy:

best_path = current_path

best_energy = current_energy

temperature *= 1 - cooling_rate

return best_path

In S603, path verification is performed

And verifying the feasibility of the optimized routing inspection path. The path is ensured not to pass through the obstacle, and the requirements of the robot on the movement capability and the like are met. If a problem is found, a path correction is performed.

# Path verification example

def validate_path(path, object_labels, terrain_map):

# check if the path collides with an obstacle, if it is within the terrain allowance

is_valid = is_path_valid(path, object_labels, terrain_map)

if not is_valid:

path = correct_invalid_path(path, object_labels, terrain_map)

return path

In S604, detail optimization is performed

Finally, the verified inspection path is optimized in detail to ensure that the inspection path meets specific inspection task requirements, such as specific field coverage, time limitation and the like.

# detail optimization example

def optimize_path_details(validated_path, task_requirements):

optimized_path = fine_tune_path(validated_path, task_requirements)

return optimized_path

Referring to fig. 8, based on the optimized routing path, the route determination is performed by adopting a route generation tool and a geographic information system GIS, and the steps of generating the unmanned aerial vehicle route are specifically as follows:

s701: based on the optimized routing inspection path, performing path conversion by adopting Dijkstra algorithm to generate preliminary route data;

In S701, path conversion is performed

Based on the optimized routing inspection path, the path is converted into route data by using Dijkstra algorithm. The Dijkstra algorithm is used to find the shortest path between nodes in the graph.

Example Dijkstra algorithm

def dijkstra(graph, start, end):

shortest_distance = {}

predecessor = {}

unseen_nodes = graph

infinity = float('inf')

path = []

for node in unseen_nodes:

shortest_distance[node] = infinity

shortest_distance[start] = 0

while unseen_nodes:

min_node = None

for node in unseen_nodes:

if min_node is None:

min_node = node

elif shortest_distance[node]<shortest_distance[min_node]:

min_node = node

for child_node, weight in graph[min_node].items():

if weight + shortest_distance[min_node]<shortest_distance[child_node]:

shortest_distance[child_node] = weight + shortest_distance[min_node]

predecessor[child_node] = min_node

unseen_nodes.pop(min_node)

current_node = end

while current_node != start:

try:

path.insert(0, current_node)

current_node = predecessor[current_node]

except KeyError:

print("Path not reachable")

break

path.insert(0, start)

if shortest_distance[end] != infinity:

return path

In S702, route screening is performed

And (3) performing spatial screening on the preliminary route data by using a spatial analysis method in a Geographic Information System (GIS) to ensure that the route cannot collide with a restricted area, a no-fly area or other spatial restrictions.

# GIS spatial analysis method example

def spatial_analysis(initial_route, spatial_data):

filtered_route = apply_spatial_filter(initial_route, spatial_data)

return filtered_route

In S703, performing waypoint determination

And determining the waypoints of the unmanned aerial vehicle by using the waypoint optimization algorithm, and ensuring the stability and flight safety of the route. The waypoint optimization algorithm can consider factors such as wind speed, wind direction, topography and the like.

Example # waypoint optimization algorithm

def optimize_waypoints(filtered_route, weather_data, terrain_data):

optimized_waypoints = calculate_optimal_waypoints(filtered_route, weather_data, terrain_data)

return optimized_waypoints

In S704, route generation is performed

And finally, generating a route of the unmanned aerial vehicle by using a linear interpolation method according to the determined waypoints, and ensuring the smoothness and the high efficiency of the flight path.

Example of linear interpolation

def linear_interpolation(waypoints):

drone_flight_path = interpolate_waypoints(waypoints)

return drone_flight_path

Referring to fig. 9, based on the unmanned aerial vehicle route and in combination with the flight dynamics model, defining the flight parameters including altitude, speed and attitude, the steps of generating the flight rule parameters are specifically as follows:

s803: referring to a flight demand analysis report, formulating a flight speed parameter and a flight attitude parameter by utilizing an attitude control algorithm;

s804: based on the flight altitude parameter, the flight speed parameter and the flight attitude parameter, a rule formulation algorithm is adopted to generate flight rule parameters.

In S801, flight demand prediction is performed

Flight dynamics models are used to estimate flight parameters. Comprising the following codes:

example of code #

def predict_flight_requirements(route, aircraft_model):

flight_requirements = {}

for waypoint in route:

altitude = aircraft_model.estimate_altitude(waypoint)

speed = aircraft_model.estimate_speed(waypoint)

attitude = aircraft_model.estimate_attitude(waypoint)

flight_requirements[waypoint] = {

'altitude': altitude,

'speed': speed,

'attitude': attitude

}

return flight_requirements

S802, setting fly height parameters is performed

A PID control algorithm is used to set the fly-height parameter. The following are code examples:

example of code #

def set_altitude_parameters(current_altitude, target_altitude):

kp=0.1# proportional gain

ki=0.01# integral gain

kd=0.1# differential gain

error = target_altitude - current_altitude

integral = 0

derivative = 0

previous_error = 0

while True:

error = target_altitude - current_altitude

integral += error

derivative = error - previous_error

control_output = kp * error + ki * integral + kd * derivative

previous_error = error

# adjust fly height

if abs(error)<0.01:

break

return control_output

In S803, the method for making the flight speed and attitude parameters is performed

Example of code #

def set_speed_and_attitude_parameters(route, aircraft_model):

speed_and_attitude_parameters = {}

for waypoint in route:

speed = aircraft_model.estimate_speed(waypoint)

attitude = aircraft_model.estimate_attitude(waypoint)

speed_and_attitude_parameters[waypoint] = {

'speed': speed,

'attitude': attitude

}

return speed_and_attitude_parameters

In S804, generating flight rule parameters is performed

Based on the altitude parameter, the speed parameter, and the attitude parameter, a flight rule parameter is generated. The following are code examples:

Example of code #

def generate_flight_rules(altitude_parameters, speed_attitude_parameters):

flight_rules = {}

for waypoint in altitude_parameters:

altitude_rule = determine_altitude_rule(altitude_parameters[waypoint])

speed_attitude_rule = determine_speed_attitude_rule(speed_attitude_parameters[waypoint])

flight_rules[waypoint] = {

'altitude_rule': altitude_rule,

'speed_attitude_rule': speed_attitude_rule

}

return flight_rules

Referring to fig. 10, based on flight rule parameters, using an edge computing technology, unmanned aerial vehicle inspection is performed, data is collected and processed in real time, and inspection reports are generated, including the steps of equipment state evaluation and abnormal condition analysis:

s901: according to flight rule parameters, a real-time data stream processing technology is adopted to execute unmanned aerial vehicle inspection tasks and obtain real-time inspection data;

s902: processing and storing the real-time inspection data by utilizing an edge computing technology and a quick storage algorithm, and storing the processed inspection data;

s904: based on the equipment state evaluation and the abnormal situation analysis, a report generation algorithm is utilized to formulate a patrol report.

In S901, an unmanned aerial vehicle inspection task is executed to obtain real-time inspection data

The unmanned aerial vehicle executes the inspection task according to the flight rule parameter, gathers real-time inspection data, includes:

the flight path of the drone is controlled using altitude, speed, and attitude information in the flight rules parameters. Unmanned aerial vehicles are equipped with sensors, such as cameras, infrared cameras, lidar, etc., for gathering environmental data, such as images, temperature, humidity, etc. And transmitting the data acquired in real time to the edge computing equipment through the communication module.

Example of code #

def perform_inspection(flight_rules):

for waypoint, rule in flight_rules.items():

control_clone (flight_wheels) # controls flight

collect_data () # acquisition data

Transmit_data () # transmit data

In S902, processing and storing real-time inspection data are performed

The edge computing device processes and stores the real-time inspection data using a fast storage algorithm to ensure efficient management and analysis of the data. The following are examples:

data preprocessing: and carrying out preprocessing steps such as noise filtering, data cleaning, correction and the like on the collected original data.

Example of code #

def preprocess_data(raw_data):

cleaned_data = data_cleanup(raw_data)

calibrated_data = calibrate(cleaned_data)

return calibrated_data

And (3) data storage: and storing the processed data in a local storage device or a cloud storage for subsequent analysis and report generation.

Example of code #

def store_data(processed_data):

local_storage.save(processed_data)

In S903, device state evaluation and abnormal situation analysis are performed

And analyzing the processed data by using an anomaly detection algorithm, evaluating the state of the equipment and detecting the anomaly. The method comprises the following steps:

device status assessment: based on the rules and the model, it is evaluated whether the device is in a normal operating state.

Example of code #

def assess_device_status(processed_data):

status = device_status_assessment(processed_data)

return status

Abnormality detection: statistical methods, machine learning models, or deep learning models are used to detect anomalies.

Example of code #

def detect_anomalies(processed_data):

anomalies = anomaly_detection(processed_data)

return anomalies

In S904, the preparation of the inspection report is performed

And (3) utilizing a report generation algorithm to formulate a patrol report based on the results of equipment state evaluation and abnormal situation analysis.

Example of code #

def generate_inspection_report(device_status, anomalies):

report = create_report(device_status, anomalies)

return report。

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims

1. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model is characterized by comprising the following steps of:

scanning the inspection area or equipment based on a laser radar scanning method to generate original three-dimensional point cloud data;

based on the original three-dimensional point cloud data, adopting a point cloud processing tool specifically for RANSAC filtering to carry out data filtering and downsampling, and generating processed three-dimensional point cloud data;

based on the processed three-dimensional point cloud data, a DBSCAN clustering algorithm is adopted to segment the data and extract features, and a segmented point cloud area and feature description are generated;

Based on the segmented point cloud region, performing object identification by using a random forest classifier, and associating equipment or a region by combining a feature matching method to generate an object identification tag and an association relationship;

according to the object identification tag and the feature description, a search algorithm A and a terrain constraint factor are adopted to conduct path planning, and a preliminary inspection path is generated;

based on the preliminary inspection path, performing path optimization by using a simulated annealing algorithm and a path efficiency evaluation function, and generating an optimized inspection path;

based on the optimized routing inspection path, adopting a route generating tool and a Geographic Information System (GIS) to determine a route, and generating an unmanned aerial vehicle route;

defining flight parameters including height, speed and attitude based on the unmanned aerial vehicle route and combining with a flight dynamics model, and generating flight rule parameters;

based on the flight rule parameters, performing unmanned aerial vehicle inspection by using an edge computing technology, collecting and processing data in real time, and generating an inspection report, wherein the inspection report comprises equipment state evaluation and abnormal condition analysis.

2. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein the step of scanning a routing area or equipment based on a laser radar scanning method to generate original three-dimensional point cloud data is specifically as follows:

Selecting a full-field scanning algorithm based on the target area, configuring the equipment position, and generating an equipment configuration scheme;

based on the equipment configuration scheme, setting radar scanning parameters by adopting a parameter initialization method, and generating initialization setting parameters;

starting a continuous scanning method based on the initialization setting parameters, acquiring radar data, and generating original scanning data;

based on the original scan data, performing data preprocessing by adopting a background denoising algorithm to generate optimized scan data;

based on the optimized scanning data, converting the scanning data into a point cloud format by adopting a three-dimensional conversion method, and generating original three-dimensional point cloud data.

3. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein based on the original three-dimensional point cloud data, a point cloud processing tool, specifically a RANSAC filter, is adopted to perform data filtering and downsampling, and the step of generating the processed three-dimensional point cloud data is specifically as follows:

based on the original three-dimensional point cloud data, adopting a RANSAC filtering algorithm to perform preliminary filtering to generate point cloud data after the RANSAC filtering;

based on the RANSAC filtered point cloud data, adopting an outlier removal method to clean the data to obtain denoised point cloud data;

Based on the denoised point cloud data, performing data dimension reduction by adopting a downsampling algorithm to generate downsampled point cloud data;

and based on the down-sampled point cloud data, adopting a quality inspection method to confirm the data integrity and establish the processed three-dimensional point cloud data.

4. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein the steps of dividing and extracting features of data by adopting a DBSCAN clustering algorithm based on the processed three-dimensional point cloud data to generate divided point cloud areas and feature descriptions are specifically as follows:

setting parameters for an algorithm based on the processed three-dimensional point cloud data to obtain DBSCAN parameter configuration;

based on the DBSCAN parameter configuration, a DBSCAN clustering algorithm is adopted to perform data segmentation, and a preliminary clustering result is generated;

based on the preliminary clustering result, removing outlier areas by adopting an area optimization method, and obtaining an optimized clustering result;

describing by adopting a feature extraction method based on the optimized clustering result to obtain a point cloud feature description;

based on the point cloud feature description, classifying by adopting a region labeling method, and generating a segmented point cloud region and feature description.

5. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein the steps of performing object recognition by using a random forest classifier based on the segmented point cloud region, associating equipment or regions by combining a feature matching method, and generating an object recognition tag and an association relation are specifically as follows:

based on the segmented point cloud region, carrying out feature description on the point cloud by adopting a feature extraction algorithm to generate a point cloud feature set;

classifying the features by adopting a random forest classifier based on the point cloud feature set to generate a preliminary object identification tag;

based on the preliminary object identification tag, adopting a feature matching algorithm to match with known equipment or areas to generate a detailed object association tag;

and based on the detailed object association tag, determining the interrelationship among the objects by adopting an association analysis algorithm, and generating an object identification tag and an association relation.

6. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein the steps of adopting an a-search algorithm and a terrain constraint factor to carry out path planning according to the object identification tag and the feature description and generating a preliminary routing path are specifically as follows:

Based on scene data, extracting topographic features by adopting a topographic analysis method to generate topographic data;

based on the topographic data and the object identification tag, generating a preliminary path plan by adopting an A search algorithm;

based on the preliminary path planning, carrying out path adjustment by combining with a terrain constraint factor, and generating an adjusted path planning;

and performing path detail improvement based on the adjusted path planning to generate a preliminary inspection path.

7. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein based on the preliminary routing path, path optimization is performed by using a simulated annealing algorithm and a path efficiency evaluation function, and the step of generating an optimized routing path is specifically as follows:

based on the preliminary inspection path, performing path evaluation by adopting a path efficiency evaluation function, and generating a path efficiency evaluation result;

based on the path efficiency evaluation result, adopting a simulated annealing algorithm to perform optimization search on the path to generate an optimized routing inspection path;

based on the optimized routing inspection path, carrying out path feasibility verification and generating a verified routing inspection path;

And carrying out detail optimization based on the verified inspection path to generate an optimized inspection path.

8. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein based on the optimized routing path, a route generation tool and a Geographic Information System (GIS) are adopted to carry out route determination, and the step of generating the unmanned aerial vehicle route is specifically as follows:

based on the optimized routing inspection path, performing path conversion by adopting a Dijkstra algorithm to generate preliminary route data;

based on the preliminary route data, screening an optimal flight area by using a spatial analysis method of a Geographic Information System (GIS) to obtain screened route data;

based on the screened route data, determining the waypoints of the unmanned aerial vehicle by adopting a waypoint optimization algorithm, and acquiring the route details of the unmanned aerial vehicle;

and integrating the unmanned aerial vehicle route details, and generating the unmanned aerial vehicle route by adopting a linear interpolation method.

9. The unmanned aerial vehicle routing planning method based on the three-dimensional point cloud model according to claim 1, wherein the steps of defining flight parameters including altitude, speed and attitude based on the unmanned aerial vehicle route and combining with a flight dynamics model and generating flight rule parameters are specifically as follows:

Based on the unmanned aerial vehicle route, a flight dynamics model is adopted to predict the flight demand, and a flight demand analysis report is obtained;

setting a flight altitude parameter by using a PID control algorithm based on the flight demand analysis report;

referring to the flight demand analysis report, formulating a flight speed parameter and a flight attitude parameter by utilizing an attitude control algorithm;

and generating flight rule parameters by adopting a rule formulation algorithm based on the flight altitude parameters, the flight speed parameters and the flight attitude parameters.

10. The unmanned aerial vehicle inspection route planning method based on the three-dimensional point cloud model according to claim 1, wherein based on the flight rule parameters, the unmanned aerial vehicle inspection is executed by utilizing an edge computing technology, data are collected and processed in real time, an inspection report is generated, and the steps including equipment state evaluation and abnormal condition analysis are specifically as follows:

according to the flight rule parameters, a real-time data stream processing technology is adopted to execute unmanned aerial vehicle inspection tasks and obtain real-time inspection data;

processing and storing the real-time inspection data by using an edge computing technology and a quick storage algorithm, and storing the processed inspection data;

Based on the processed inspection data, performing equipment state evaluation and abnormal condition analysis by using an abnormal detection algorithm to obtain equipment state evaluation and abnormal condition analysis;

and based on the equipment state evaluation and the abnormal condition analysis, making a patrol report by using a report generation algorithm.