UAV flight path planning method for building image acquisition

Abstract

This application provides a drone flight path planning method for building image acquisition, comprising: acquiring a target area for building image acquisition, acquiring a first building image within a first sub-area according to a preset flight altitude and a first flight path; acquiring an implicit building spacing distribution and an implicit building height distribution within the target area based on analysis of the first building image, indexing the spacing distribution and height distribution of similar-family buildings, and processing the obtained building spacing distribution and building height distribution; performing flight altitude path planning within a second sub-area based on the building spacing and building height to obtain an optimal altitude flight path; acquiring building images within the second sub-area according to the optimal altitude flight path, and continuing to acquire building images until acquisition of the target area is complete. This method solves the technical problem of poor building image acquisition quality of drones in complex scenes with high building density and significant height differences in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of path planning, and in particular to a method for planning the flight path of an unmanned aerial vehicle (UAV) for acquiring building images. Background Art

[0002] Currently, when using drones to collect building images, in complex scenes with high building density and significant height differences (such as rural self-built residential areas), the occlusion of adjacent buildings will cause the side and top details of some buildings to be blocked, resulting in poor building image acquisition quality. Summary of the Invention

[0003] The present invention addresses the technical problem in the prior art that drones have poor building image acquisition quality in complex scenes with high building density and significant height differences, and provides a drone flight path planning method for building image acquisition.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] The present invention provides a UAV flight path planning method for building image acquisition, comprising:

[0006] Acquire a target area for building image acquisition, and control the drone to fly and acquire an image of a first building within a first sub-area according to a preset flight altitude and a first flight path;

[0007] Obtaining implicit building spacing distribution and implicit building height distribution of the target area based on the first building image analysis, combining regional feature information of the target area, indexing the spacing distribution and height distribution of buildings of the same family, and processing to obtain building spacing distribution and building height distribution;

[0008] Performing flight altitude path planning within the second sub-area based on the building spacing and building heights to obtain an optimal altitude flight path, wherein the planning includes analyzing an obstruction rate of oblique photography;

[0009] According to the optimal altitude flight path, the building image of the second sub-area is collected, and the building image collection is continued until the collection of the target area is completed.

[0010] The beneficial effects of the present invention are:

[0011] Compared with the prior art, the present application first obtains the target area for building image acquisition, and controls the drone to fly and acquire the first building image in the first sub-area according to the preset flight altitude and the first flight path, thereby providing the necessary basic information for subsequent path optimization. Secondly, the implicit building spacing distribution and the implicit building height distribution of the target area are acquired based on the analysis of the first building image. Combined with the regional feature information of the target area, the spacing distribution and the height distribution of the same family of buildings are indexed, and the building spacing distribution and the building height distribution are processed to obtain accurate spatial feature information of the target area, thereby providing more accurate data input for the subsequent drone path planning. Thirdly, based on the building spacing and building height, the flight altitude path planning is performed in the second sub-area to obtain the optimal altitude flight path, and the influence of the occlusion rate and the absolute deviation amplitude of the two previous and subsequent flight altitudes on the acquisition effect is comprehensively considered, and the optimal altitude flight path for the second sub-area is generated is output. Finally, according to the optimal altitude flight path, the building image of the second sub-area is acquired, and the building image acquisition is continued until the acquisition of the target area is completed, thereby obtaining accurate image information of the target area.

[0012] Through the above technical solution, this application fully considers the impact of the obstruction rate of oblique photography on the image acquisition quality. According to the highest building height, a preset flight altitude is configured to acquire the first building image in the first sub-area, and the first acquisition score is calculated based on the first flight altitude, building spacing distribution and building height distribution. Through iterative optimization, the flight altitude with the largest acquisition score is obtained as the optimal altitude flight path, and the building image of the second sub-area is acquired based on this. The building images collected in the two rounds are fused to obtain accurate image information of the target area, thereby improving the quality of image acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the flow of a UAV flight path planning method for building image acquisition provided by the present invention;

[0014] Figure 2 A schematic diagram of the flow of calculating the first acquisition occlusion rate in the UAV flight path planning method for building image acquisition provided by the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0017] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0018] Example 1, as Figure 1 As shown, an embodiment of the present invention provides a UAV flight path planning method for building image acquisition, comprising:

[0019] S10: Acquire a target area for building image acquisition, and control the drone to fly and acquire an image of a first building within a first sub-area according to a preset flight altitude and a first flight path;

[0020] When capturing architectural images using drones, poor image quality can occur due to occlusion in densely populated areas with uneven building heights, such as densely packed rural areas. Specifically, in densely populated areas, low-rise buildings are easily obscured by surrounding high-rise buildings. Furthermore, the staggered rooftop structures often create blind spots due to the limited viewing angle in traditional fixed-altitude flight modes. This significantly reduces the integrity and detail of captured architectural images.

[0021] To address the above problems, this application obtains the highest building height within the target area based on the target area, and configures a preset flight altitude based on the highest building height. The target area is then divided into multiple sub-areas, and the drone is controlled to fly according to the preset flight altitude to collect the image of the first building in the first sub-area.

[0022] Specifically, step S10 in the method includes:

[0023] Acquire a target area for building image acquisition and obtain the highest building height in the target area;

[0024] Configuring a preset flight altitude based on the maximum building height;

[0025] The target area is divided into multiple sub-areas, and the drone is controlled to fly according to the preset flight altitude to collect an image of a first building in a first sub-area.

[0026] In this embodiment, a target area for building image acquisition is first obtained, and the highest building height within the target area is determined. Specifically, the geographic boundaries of the target area and the highest building height within the area can be obtained by querying design drawings or third-party data sources (such as LiDAR point clouds or building databases). For example, a LiDAR device with an accuracy of ±0.5 meters (such as the Riegl VQ series) is used to scan the target area at a 240kHz pulse frequency and a ±30° scanning angle. Rooftop points are then filtered based on their relative height in the point cloud (Z coordinate minus ground height). The maximum value is taken as the highest building height within the target area (e.g., 15 meters).

[0027] Secondly, a preset flight altitude is configured based on the maximum building height. Specifically, the preset flight altitude is dynamically configured based on the maximum building height and a safety factor (typically 1.3-1.8 times, derived from extensive data summaries to ensure collision-free drone imagery). For example, if the highest building in an area is 15 meters, the preset altitude could be set to 15*1.5=22.5 meters, where * represents multiplication. This configuration ensures collision-free drone imagery and minimizes blind spots.

[0028] Finally, the target area is divided into multiple sub-areas, and the drone is controlled to fly at the preset altitude to capture a first building image within the first sub-area. Specifically, the target area is evenly divided into multiple grid-like sub-areas, and the drone is controlled to fly a parallel flight path at the preset altitude, capturing images of buildings within each of the multiple grid-like sub-areas as the first building image. The drone is equipped with a high-resolution camera (e.g., 8192×5460 pixels) and automatic exposure mode to ensure that the effective information coverage of a single sub-area image reaches over 95%.

[0029] In summary, compared to existing technologies, this application obtains the highest building height within a target area, configures a preset flight altitude based on the highest building height, then divides the target area into multiple sub-areas. At the preset flight altitude, the drone is controlled to capture an image of the first building within the first sub-area. In this way, the preset flight altitude is dynamically configured based on the highest building height, and a regional acquisition strategy is used to collect initial data, providing the necessary basic information for subsequent path optimization.

[0030] S20: Analyze the first building image to obtain implicit building spacing distribution and implicit building height distribution of the target area, combine regional feature information of the target area, index the spacing distribution and height distribution of buildings in the same family, and process to obtain building spacing distribution and building height distribution;

[0031] In scenarios with high building density and non-uniform heights (such as densely packed buildings in rural areas), there may be implicit building spacing and implicit building heights. These implicit building spacing and implicit building heights may be blind spots in the drone's field of view in images collected by the drone, thus affecting the quality of image acquisition.

[0032] To address the above problems, this application inputs the first building image collected by the drone into the pre-trained building spacing recognition network and building height recognition network, and obtains the implicit building spacing distribution and the implicit building height distribution through recognition output. Based on the regional feature information of the target area, multiple similar areas are indexed and obtained, and all the similar building spacings and all the similar building heights in the multiple similar areas are obtained as the similar building spacing distribution and the similar building height distribution. Finally, the implicit building spacing distribution, the implicit building height distribution, the similar building spacing distribution and the similar building height distribution are combined to obtain the building spacing distribution and the building height distribution.

[0033] Specifically, step S20 in the method includes:

[0034] transmitting the first building image to a cloud server;

[0035] Receive the implicit building spacing distribution and the implicit building height distribution obtained by identification, wherein the first building image is input into a building spacing recognition network and a building height recognition network, and the implicit building spacing distribution and the implicit building height distribution are obtained by identification output, wherein the implicit building spacing distribution includes multiple implicit building spacings, and the implicit building height distribution includes multiple implicit building heights. The building spacing recognition network and the building height recognition network use a sample building image set as input and use a sample building spacing distribution set and a sample building height distribution set as output for training respectively.

[0036] In this embodiment of the present application, the first building image is first transmitted to a cloud server. Specifically, the first building image (e.g., a 512×512 pixel RGB image) captured by the drone is transmitted to the cloud in real time via a 4G / 5G network and preprocessed (e.g., image normalization and resizing).

[0037] Next, the preprocessed first building image is fed into a pretrained building spacing recognition network and a building height recognition network. The recognition outputs are the implicit building spacing distribution (including multiple implicit building spacings) and the implicit building height distribution (including multiple implicit building heights). Furthermore, the building spacing recognition network is based on a ResNet50 architecture, with a spatial pyramid pooling layer added after the convolutional layer. This captures building spacing features at different scales (such as the distance between adjacent building outlines) and outputs a probability distribution with 15 bins (e.g., a 25% probability for a spacing of 5-10 meters). The building height recognition network uses a U-Net architecture, extracting semantic features of the building tops and bottoms (e.g., roof slope and base position) through an encoder-decoder, generating a probability distribution of building heights (e.g., a 30% probability for a height of 12-13 meters). Furthermore, both the building spacing recognition network and the building height recognition network are trained using over 5,000 images annotated with multiple implicit building spacings and multiple implicit building heights. The data annotations are calibrated using LiDAR point clouds and total station measurements to ensure the accuracy of the implicit feature recognition.

[0038] For example, a preprocessed first building image is fed into a building spacing recognition network and a building height recognition network, which output an implicit building spacing distribution (e.g., 2.3m, 3.1m) and an implicit building height distribution (e.g., 0.8m, 1.4m). In this way, combined with artificial intelligence, multiple implicit building spacing and height distributions in the target area are quickly and accurately derived based on the first building image, providing a reliable data foundation for subsequent planning.

[0039] Furthermore, the step of “combining the regional characteristic information of the target area, indexing the spacing distribution and height distribution of buildings of the same family, and processing to obtain the spacing distribution and height distribution of buildings” includes:

[0040] Collecting regional characteristic information of the target area;

[0041] According to the regional feature information, multiple same-family regions for collecting building images are indexed and obtained, and all the same-family building spacings and all the same-family building heights in the multiple same-family regions are obtained to obtain the same-family building spacing distribution and the same-family building height distribution;

[0042] The implicit building spacing distribution, the implicit building height distribution, the same-family building spacing distribution, and the same-family building height distribution are combined to obtain a building spacing distribution and a building height distribution.

[0043] In the embodiment of the present application, the regional characteristic information of the target area is first collected. Specifically, the multi-dimensional characteristic information of the target area (such as building density, maximum building spacing, maximum building height, etc.) is first collected.

[0044] Secondly, using the multidimensional feature information of the target area, multiple similar areas are indexed from the historical database and obtained as homologous areas. The total distances between buildings in the same area and the total heights of buildings in the same area are then obtained as the homologous building distance distribution and homologous building height distribution. The homologous building distance distribution and homologous building height distribution are obtained from the historical database based on the similarity between the target area and the historical data. This effectively addresses the problem of missing features due to occlusion during drone image acquisition and more accurately reflects the distribution of building distances and building heights in the target area. The higher the similarity, the more accurate the reflection. Furthermore, the indexing process can retrieve multiple homologous areas from the historical database based on the cosine similarity algorithm (e.g., similar building density and similar maximum building distance), extract the total distances between buildings and total building heights in these areas, and use them as the homologous building distance distribution and homologous building height distribution.

[0045] For example, a cosine similarity algorithm is used to retrieve multiple areas with a similarity greater than 0.7 from a historical database as homologous areas, and then all building spacings (such as 3.3m, 4.2m) and all building heights (such as 2.6m, 3.8m) of the multiple homologous areas are obtained respectively as the homologous building spacing distribution and the homologous building height distribution.

[0046] Finally, the implicit building spacing distribution, implicit building height distribution, same-family building spacing distribution, and same-family building height distribution are combined to obtain the building spacing distribution and building height distribution. For example, by combining the implicit building spacing distribution (e.g., 2.3m, 3.1m), the implicit building height distribution (e.g., 7.8m, 8.4m, 6.5m, 9.3m), the same-family building spacing distribution (e.g., 3.3m, 4.2m), and the same-family building height distribution (e.g., 5.6m, 3.8m, 4.7m, 8.0m), the building spacing distribution (2.3m, 3.1m, 3.3m, 4.2m) and the building height distribution (7.8m, 8.4m, 6.5m, 9.3m, 5.6m, 3.8m, 4.7m, 8.0m) are obtained. In this way, based on historical similar data, combined with the building spacing distribution and implicit building height distribution obtained by artificial intelligence output and the spacing distribution and height distribution of buildings of the same family obtained based on historical data, some features missing due to occlusion in the drone image acquisition process are supplemented, providing more accurate data input for subsequent drone path planning.

[0047] In summary, compared to the existing technology, this application inputs the first building image collected by the drone into the pre-trained building spacing recognition network and building height recognition network, and the recognition output obtains the implicit building spacing distribution and the implicit building height distribution. Based on the regional feature information of the target area, multiple homologous areas are indexed and obtained, and all homologous building spacings and all homologous building heights in multiple homologous areas are obtained as homologous building spacing distribution and homologous building height distribution. Finally, the implicit building spacing distribution, implicit building height distribution, homologous building spacing distribution and homologous building height distribution are combined to obtain the building spacing distribution and building height distribution. In this way, feature information is extracted based on the data collected by the drone, and necessary feature information is supplemented through historical data to obtain accurate spatial feature information of the target area, providing more accurate data input for subsequent drone path planning.

[0048] S30: performing flight altitude path planning within the second sub-area based on the building spacing and building heights to obtain an optimal altitude flight path, wherein the planning includes analyzing an obstruction rate of oblique photography;

[0049] Oblique photography is used to capture images of buildings. Oblique photography is a photography method that can capture images of the sides and tops of target buildings to construct three-dimensional images. However, in scenarios where the building density is high and the heights are not uniform, the acquisition quality is poor due to occlusion. In the aforementioned steps, a preset flight altitude is configured based on the highest building height, and the image of the first building in the first sub-area is captured. In order to improve the image acquisition quality, the flight altitude of the drone can be optimized for two rounds of image acquisition. Furthermore, if the second flight altitude varies too much from the preset flight altitude, although the obstruction rate of the oblique photography can be kept at a low level, it may also cause torque distortion when the fused image is finally constructed.

[0050] To address the above problems, the present application obtains the flight altitude range for drones to collect building images, randomly generates a first flight altitude within the flight altitude range, and calculates the first acquisition occlusion rate of drone flight to collect building images in the second sub-area based on the first flight altitude, building spacing distribution, and building height distribution. In combination with the preset flight altitude, the first height change rate is calculated, and the first acquisition score is calculated. The flight altitude is continued to be randomly generated, and iterative optimization is performed to obtain the flight altitude with the largest acquisition score as the optimal flight altitude, and the optimal altitude flight path is generated in combination with the second sub-area.

[0051] Specifically, step S30 in the method includes:

[0052] Obtain the flight altitude range of the drone for building image acquisition;

[0053] randomly generating a first flight altitude within the flight altitude interval;

[0054] Based on the first flight altitude, building spacing distribution, and building height distribution, performing an obstruction rate analysis of the oblique photography of building images collected by the drone in the second sub-area to obtain a first acquisition obstruction rate, and calculating the height change rate in combination with the preset flight altitude to obtain a first height change rate, thereby obtaining a first acquisition score;

[0055] Continue to randomly generate flight heights, perform iterative optimization, and obtain the flight height with the largest collection score as the optimal flight height. Combined with the second sub-area, generate the optimal height flight path.

[0056] In this embodiment, the drone's flight altitude range for capturing building images is first determined. Specifically, based on the target area's preset flight altitude and safety rules, the flight altitude range is determined (typically 80%-120% of the preset altitude; for example, if the preset altitude is 15 meters, the flight altitude range can be set to 12-18 meters) to ensure that the drone covers the target area as much as possible while avoiding collisions.

[0057] Secondly, a first flight altitude is randomly generated within the flight altitude interval. For example, a first flight altitude (such as 14 meters) is randomly generated within the flight altitude interval (such as 12-18 meters).

[0058] Next, based on the first flight altitude, building spacing distribution, and building height distribution, the occlusion rate analysis network predicts the obstruction rate for oblique photography at the first flight altitude to obtain the first acquisition occlusion rate. The altitude change rate is then calculated to obtain the first altitude change rate. The obstruction rate for oblique photography reflects the occlusion between adjacent buildings. The first altitude change rate = |first flight altitude - preset flight altitude| / preset flight altitude. For example, for a first flight altitude (e.g., 14 meters) and a preset flight altitude (e.g., 15 meters), the first altitude change rate = |14 - 15| / 15 = 6.67%. Finally, based on the first acquisition occlusion rate and the first altitude change rate, a first quality score and a first continuous score are calculated. These scores are then added together to obtain a first acquisition score. The first acquisition score is a rating of the current UAV flight parameters. A higher first acquisition score indicates higher image acquisition quality under these flight parameters.

[0059] Finally, the flight altitude is randomly generated and iteratively optimized to determine the altitude with the highest acquisition score. This altitude is then used to generate the optimal flight path for the second sub-area. For example, after 50 rounds of randomly generated flight altitudes and iterative optimization, the altitude with the highest score (e.g., 15.5 meters) is selected as the optimal flight altitude. This is then used to generate the optimal flight path for the second sub-area. This ensures that each building's facade and roof are not significantly obstructed during oblique photography, while also minimizing unnecessary altitude fluctuations.

[0060] Specifically, “performing an obstruction rate analysis of oblique photography of building images collected by a drone in the second sub-area based on the first flight altitude, building spacing distribution, and building height distribution to obtain a first acquisition obstruction rate” includes:

[0061] Randomly selecting a building spacing and two building heights from the building spacing distribution and the building height distribution as a building feature group, and obtaining K building feature groups, where K is an integer greater than 1;

[0062] According to the first flight altitude and K building feature groups, an oblique photography occlusion rate analysis is performed to obtain K oblique photography occlusion rates, and the average is calculated to obtain a first acquisition occlusion rate.

[0063] In the embodiment of the present application, a building spacing and two building heights are first randomly selected from the building spacing distribution and building height distribution as building feature groups, and K building feature groups are selected, where K is an integer greater than 1. Specifically, two buildings are randomly selected, and their building spacing and building heights are determined as building feature groups. This step is repeated until a total of K building feature groups are randomly selected. In this way, the obstruction rate of oblique photography at the first flight altitude is analyzed using the randomly collected building feature groups. For example, from the building spacing distribution (e.g., 2.3m, 3.1m, 3.3m, 4.2m) and height distribution (e.g., 7.8m, 8.4m, 6.5m, 9.3m, 5.6m, 3.8m, 4.7m, 8.0m), two building feature groups are randomly selected (e.g., building spacing of 2.3m, building heights of 7.8m and 8.4m; building spacing of 3.3m, building heights of 5.6m and 3.8m), each of which represents a building adjacency relationship.

[0064] Secondly, based on the first flight altitude and K building feature groups, an oblique photography occlusion rate analysis is performed to obtain K oblique photography occlusion rates, and the average is calculated to obtain the first acquisition occlusion rate. Specifically, the first flight altitude and K building feature groups are sequentially input into the pre-trained occlusion rate analysis network, and the K oblique photography occlusion rates for the current flight altitude and building feature group are output. Further, the average of the K oblique photography occlusion rates is calculated as the first acquisition occlusion rate. For example, the first flight altitude (e.g., 14 meters) and two building feature groups (e.g., building spacing of 2.3 meters, building heights of 7.8 meters and 8.4 meters; building spacing of 3.3 meters, building heights of 5.6 meters and 3.8 meters) are input into the pre-trained occlusion rate analysis network, and the output is the first acquisition occlusion rates of 30% and 15%. In this case, the first acquisition occlusion rate = (30% + 15%) / 2 = 22.5%. In this way, by taking the average of the occlusion rates of K feature groups, the accidental interference of a single feature group can be effectively reduced, and a more reliable first-collection occlusion rate can be obtained. Through a large number of random sampling, the occlusion rate prediction error can be controlled within ±5%, providing a reliable basis for subsequent scoring calculations.

[0065] Furthermore, if Figure 2 As shown, the “performing an oblique photography occlusion rate analysis based on the first flight altitude and K building feature groups to obtain K oblique photography occlusion rates” includes:

[0066] Based on the historical data of building images collected by drone oblique photography, a set of sample flight altitudes and a set of sample building feature groups are collected. The occlusion rates of building images collected by oblique photography at different sample flight altitudes and sample building feature groups are collected, and the sample oblique photography occlusion rate sets are obtained by annotation.

[0067] Use machine learning to build an occlusion rate analysis network;

[0068] Using the sample flight altitude set, the sample building feature group set, and the sample oblique photography obstruction rate set, the obstruction rate analysis network is trained until convergence;

[0069] The first flight altitude is combined with the K building feature groups respectively, input into the occlusion rate analysis network, and K oblique photography occlusion rates are outputted, and the average is calculated to obtain the first acquisition occlusion rate.

[0070] In an embodiment of the present application, first, historical data of building images are collected based on oblique photography of drones (oblique photography is a photography method that can capture images of the sides and tops of buildings to construct a three-dimensional image of the target building), a sample flight altitude set, a sample building feature group set, and the occlusion rates of the oblique photography of building images at different sample flight altitudes and sample building feature groups are collected, and the sample oblique photography occlusion rate set is obtained by annotation. For example, 2000+ groups of sample data are collected, each group of data contains the flight altitude of the drone, the building feature group (including the building spacing and the height of two buildings) and the actual occlusion rate, and each group of data is labeled accordingly by manual annotation or LiDAR verification. For example, when the flight altitude is 10 meters, the spacing is 2.5 meters, and the heights of the two buildings are 7.8 meters and 5.6 meters respectively, the occlusion rate is 28%. In this way, the sample oblique photography occlusion rate set is obtained.

[0071] Secondly, machine learning was used to construct an occlusion rate analysis network. Specifically, the occlusion rate analysis network consists of a three-layer fully connected neural network: the input layer is a 4-dimensional vector (flight altitude, distance, building height 1, building height 2), the hidden layer uses the ReLU activation function to extract nonlinear features, and the output layer uses the Sigmoid function to normalize the occlusion rate (0-1).

[0072] Next, the occlusion rate analysis network is trained using the sample flight altitude set, sample building feature group set, and sample oblique photography occlusion rate set until convergence. Specifically, the sample flight altitude set, sample building feature group set, and sample oblique photography occlusion rate set are divided into training set, validation set, and test set at a ratio of 7:1.5:1.5. The network is trained using the Adam optimizer, with the mean squared error (MSE) as the loss function. Training is stopped when the model loss on the validation set decreases by < 0.001 for 10 consecutive rounds. When the prediction accuracy on the test set reaches 92% (MSE=0.025), it is considered converged and training is stopped, resulting in the completed occlusion rate analysis network.

[0073] Finally, the first flight altitude is combined with the K building feature groups and input into the occlusion rate analysis network, which outputs K oblique photography occlusion rates, and the average is calculated to obtain the first acquisition occlusion rate. Specifically, the current flight altitude and K random feature groups are input into the trained occlusion rate analysis network, which outputs K oblique photography occlusion rates, and the K oblique photography occlusion rates are averaged to obtain the first acquisition occlusion rate. Taking the average can improve data accuracy. For example, the first flight altitude (e.g., 14 meters) and two building feature groups (e.g., building spacing of 2.3 meters, building heights of 7.8 meters and 8.4 meters; building spacing of 3.3 meters, building heights of 5.6 meters and 3.8 meters) are input into the pre-trained occlusion rate analysis network, and the output obtains first acquisition occlusion rates of 30% and 15%. The first acquisition occlusion rate at this time = (30% + 15%) / 2 = 22.5%.

[0074] Furthermore, the “calculating the altitude change rate in combination with the preset flight altitude to obtain a first altitude change rate, and calculating a first acquisition score” includes:

[0075] calculating an absolute deviation of the first flight altitude from the preset flight altitude to obtain a first altitude change rate;

[0076] Calculating a first quality score based on the first acquisition occlusion rate, and calculating a first continuous score based on the first height change rate;

[0077] A first acquisition score is calculated based on the first quality score and the first continuous score.

[0078] In this embodiment of the present application, the absolute deviation of the first flight altitude from the preset flight altitude is first calculated to obtain a first altitude change rate, where the first altitude change rate = |first flight altitude - preset flight altitude| / preset flight altitude. The first altitude change rate reflects the altitude adjustment amplitude of the drone between the first and second sub-areas. If the altitude adjustment amplitude is too large, while the obstruction rate of oblique photography can be kept low, it may also cause torque distortion when the fused image is finally constructed. For example, if the first flight altitude is (e.g., 14 meters) and the preset flight altitude is (e.g., 15 meters), the first altitude change rate at this time = |14 - 15| / 15 = 6.67%.

[0079] Secondly, based on the first acquisition occlusion rate, a first quality score is calculated, and based on the first altitude change rate, a first continuous score is calculated. The first quality score = 1-first acquisition occlusion rate, which reflects the degree of influence of the occlusion on the acquisition effect. The smaller the first acquisition occlusion rate, the larger the first quality score. The first continuous score = 1-first altitude change rate, which reflects the degree of influence of the absolute deviation of the altitude between the two previous and subsequent flights on the acquisition effect. The smaller the first altitude change rate, the higher the first continuous score. For example, if the first acquisition occlusion rate is 22.5% and the first altitude change rate is 6.67%, then the first quality score = 1-22.5% = 77.5%, and the first continuous score = 1-6.67% = 93.33%.

[0080] Finally, the first acquisition score is calculated based on the first quality score and the first continuous score: (First Quality Score + First Continuous Score) * 0.01. For example, if the first quality score is 77.5% and the first continuous score is 93.33%, then the first acquisition score = (77.5% + 93.33%) * 0.01 = 1.71. This first acquisition score combines the impact of the occlusion rate and the absolute deviation in altitude between the two previous and subsequent flight levels on the acquisition quality. While ensuring that the occlusion rate remains within a reasonable range, it ensures that torque distortion is not generated when constructing the fused image.

[0081] In summary, compared to the existing technology, this application obtains the flight altitude range for drones to collect building images, randomly generates a first flight altitude within the flight altitude range, calculates the first acquisition occlusion rate of drone flight images of buildings in the second sub-area based on the first flight altitude, building spacing distribution, and building height distribution, and calculates the first altitude change rate and the first acquisition score in combination with the preset flight altitude. The application then continues to randomly generate flight altitudes, performs iterative optimization, and obtains the flight altitude with the largest acquisition score as the optimal flight altitude. The optimal altitude flight path is generated in combination with the second sub-area. In this way, the effects of the occlusion rate and the absolute deviation amplitude of the two flight altitudes on the acquisition effect are comprehensively considered, and the optimal altitude flight path generated in the second sub-area is output.

[0082] S40: Acquire building images of the second sub-area according to the optimal altitude flight path, and continue acquiring building images until acquisition of the target area is completed.

[0083] In the embodiment of the present application, the drone is controlled to fly along the optimal altitude flight path to capture building images of the second sub-area, and this step is repeated until the target area is completely captured. In this way, accurate image information of the target area is obtained.

[0084] In summary, the embodiments of the present application have at least the following technical effects:

[0085] Compared to existing technologies, this application first obtains the highest building height within the target area and configures a preset flight altitude based on the highest building height. The target area is then divided into multiple sub-areas. At the preset flight altitude, the drone is controlled to capture an image of the first building within the first sub-area. In this way, the preset flight altitude is dynamically configured based on the highest building height, and a regional acquisition strategy is used to collect initial data, providing the necessary basic information for subsequent path optimization.

[0086] Secondly, the present application inputs the first building image collected by the drone into the pre-trained building spacing recognition network and building height recognition network, and obtains the implicit building spacing distribution and the implicit building height distribution through the recognition output. Based on the regional feature information of the target area, multiple homologous areas are indexed and all the homologous building spacings and all the homologous building heights in the multiple homologous areas are obtained as the homologous building spacing distribution and homologous building height distribution. Finally, the implicit building spacing distribution, implicit building height distribution, homologous building spacing distribution and homologous building height distribution are combined to obtain the building spacing distribution and building height distribution. In this way, feature information is extracted based on the data collected by the drone, and necessary feature information is supplemented through historical data to obtain accurate spatial feature information of the target area, providing more accurate data input for subsequent drone path planning.

[0087] Again, this application obtains the flight altitude interval for drones to collect building images, randomly generates a first flight altitude within the flight altitude interval, calculates the first acquisition occlusion rate of drone flight images of buildings in the second sub-area based on the first flight altitude, building spacing distribution, and building height distribution, and calculates the first altitude change rate and the first acquisition score in combination with the preset flight altitude. The application continues to randomly generate flight altitudes, performs iterative optimization, and obtains the flight altitude with the largest acquisition score as the optimal flight altitude. The optimal altitude flight path is generated in combination with the second sub-area. In this way, the effects of the occlusion rate and the absolute deviation amplitude of the two flight altitudes on the acquisition effect are comprehensively considered, and the optimal altitude flight path generated in the second sub-area is output.

[0088] Finally, the present application follows the optimal altitude flight path, and then controls the UAV to fly according to the optimal altitude flight path to collect building images of the second sub-area, and repeats this step until the target area is collected and accurate image information of the target area is obtained.

[0089] Through the above technical solution, this application fully considers the impact of the obstruction rate of oblique photography on the image acquisition quality. According to the highest building height, a preset flight altitude is configured to acquire the first building image in the first sub-area, and the first acquisition score is calculated based on the first flight altitude, building spacing distribution and building height distribution. Through iterative optimization, the flight altitude with the largest acquisition score is obtained as the optimal altitude flight path, and the building image of the second sub-area is acquired based on this. The building images collected in the two rounds are fused to obtain accurate image information of the target area, thereby improving the quality of image acquisition.

[0090] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0091] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0093] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0095] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0096] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A UAV flight path planning method for building image acquisition, characterized in that: The method comprises: Acquire a target area for building image acquisition, and control the drone to fly and acquire an image of a first building within a first sub-area according to a preset flight altitude and a first flight path; Obtaining implicit building spacing distribution and implicit building height distribution of the target area based on the first building image analysis, combining regional feature information of the target area, indexing the spacing distribution and height distribution of buildings of the same family, and processing to obtain building spacing distribution and building height distribution; Performing flight altitude path planning within the second sub-area based on the building spacing and building heights to obtain an optimal altitude flight path, wherein the planning includes analyzing an obstruction rate of oblique photography; According to the optimal altitude flight path, the building image of the second sub-area is collected, and the building image collection is continued until the collection of the target area is completed.

2. The UAV flight path planning method for building image acquisition according to claim 1, characterized in that: Acquiring a target area for building image acquisition, and controlling the unmanned aerial vehicle to fly and acquire a first building image within a first sub-area according to a preset flight altitude and a first flight path, including: Acquire a target area for building image acquisition and obtain the highest building height in the target area; Configuring a preset flight altitude based on the maximum building height; The target area is divided into multiple sub-areas, and the drone is controlled to fly according to the preset flight altitude to collect an image of a first building in a first sub-area.

3. The method for UAV flight path planning for building image acquisition according to claim 1, characterized in that: Obtaining implicit building spacing distribution and implicit building height distribution in the target area according to the first building image analysis includes: transmitting the first building image to a cloud server; Receive the implicit building spacing distribution and the implicit building height distribution obtained by identification, wherein the first building image is input into a building spacing recognition network and a building height recognition network, and the implicit building spacing distribution and the implicit building height distribution are obtained by identification output, wherein the implicit building spacing distribution includes multiple implicit building spacings, and the implicit building height distribution includes multiple implicit building heights. The building spacing recognition network and the building height recognition network use a sample building image set as input and use a sample building spacing distribution set and a sample building height distribution set as output for training respectively.

4. The method for UAV flight path planning for building image acquisition according to claim 1, characterized in that: In combination with the regional characteristic information of the target area, indexing the spacing distribution and height distribution of buildings of the same family, and processing to obtain the building spacing distribution and building height distribution, including: Collecting regional characteristic information of the target area; According to the regional feature information, multiple same-family regions for collecting building images are indexed and obtained, and all the same-family building spacings and all the same-family building heights in the multiple same-family regions are obtained to obtain the same-family building spacing distribution and the same-family building height distribution; The implicit building spacing distribution, the implicit building height distribution, the same-family building spacing distribution, and the same-family building height distribution are combined to obtain a building spacing distribution and a building height distribution.

5. The method for UAV flight path planning for building image acquisition according to claim 1, characterized in that: Performing flight altitude path planning in the second sub-area based on the building spacing and building heights to obtain an optimal altitude flight path includes: Obtain the flight altitude range of the drone for building image acquisition; randomly generating a first flight altitude within the flight altitude interval; Based on the first flight altitude, building spacing distribution, and building height distribution, performing an obstruction rate analysis of the oblique photography of building images collected by the drone in the second sub-area to obtain a first acquisition obstruction rate, and calculating the height change rate in combination with the preset flight altitude to obtain a first height change rate, thereby obtaining a first acquisition score; Continue to randomly generate flight heights, perform iterative optimization, and obtain the flight height with the largest collection score as the optimal flight height. Combined with the second sub-area, generate the optimal height flight path.

6. The method for UAV flight path planning for building image acquisition according to claim 5, characterized in that: Based on the first flight altitude, building spacing distribution, and building height distribution, an obstruction rate analysis of oblique photography of building images collected by the UAV in the second sub-area is performed to obtain a first acquisition obstruction rate, including: Randomly selecting a building spacing and two building heights from the building spacing distribution and the building height distribution as a building feature group, and obtaining K building feature groups, where K is an integer greater than 1; According to the first flight altitude and K building feature groups, an oblique photography occlusion rate analysis is performed to obtain K oblique photography occlusion rates, and the average is calculated to obtain a first acquisition occlusion rate.

7. The method for UAV flight path planning for building image acquisition according to claim 6, characterized in that: According to the first flight altitude and the K building feature groups, an obstruction rate analysis of obstruction photography is performed to obtain K obstruction rates of obstruction photography, including: Based on the historical data of building images collected by drone oblique photography, a set of sample flight altitudes and a set of sample building feature groups are collected. The occlusion rates of building images collected by oblique photography at different sample flight altitudes and sample building feature groups are collected, and the sample oblique photography occlusion rate sets are obtained by annotation. Use machine learning to build an occlusion rate analysis network; Using the sample flight altitude set, the sample building feature group set, and the sample oblique photography obstruction rate set, the obstruction rate analysis network is trained until convergence; The first flight altitude is combined with the K building feature groups respectively, input into the occlusion rate analysis network, and K oblique photography occlusion rates are outputted, and the average is calculated to obtain the first acquisition occlusion rate.

8. The method for UAV flight path planning for building image acquisition according to claim 5, characterized in that: Calculating the altitude change rate in combination with the preset flight altitude to obtain a first altitude change rate, and calculating a first acquisition score, including: calculating an absolute deviation of the first flight altitude from the preset flight altitude to obtain a first altitude change rate; Calculating a first quality score based on the first acquisition occlusion rate, and calculating a first continuous score based on the first height change rate; A first acquisition score is calculated based on the first quality score and the first continuous score.

Images