Urban compensation modeling method and system combined with image recognition
By identifying and optimizing the overlapping areas of drone vision, performing texture map file retrieval and image resolution analysis, the problems of low accuracy and efficiency in drone city modeling are solved, and efficient urban compensation modeling is achieved.
Patent Information
- Application Number
- CN202411963131.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-10-17
- Estimated Expiration
- 2044-12-30
AI Technical Summary
In the existing technology, when using drones to collect urban images, the flying speed is too fast or the sampling interval is too large, resulting in low urban modeling accuracy, low compensation modeling efficiency, and a long model update cycle.
By obtaining the historical urban 3D model of the target city and the set of UAV acquisition paths, the overlapping areas of the UAV field of view are determined, the interactive adversarial sub-areas are divided, texture map file retrieval and image resolution analysis are performed, the fuzzy areas are identified, the UAV flight parameters are optimized, and image acquisition and 3D modeling are performed to compensate and update the model.
The recognition accuracy of fuzzy areas in urban three-dimensional models and the efficiency of compensation modeling are improved, and the model update cycle is shortened.
Smart Images
Figure CN119904360B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of urban compensation modeling, and particularly relates to a kind of urban compensation modeling method and system combined with image recognition. BACKGROUND
[0002] Through image recognition technology, the expansion of the city or the changes of the facilities such as buildings, roads and greenery within the region can be accurately monitored. However, when using drones to collect images of the city, the overlap between adjacent paths is insufficient due to the excessive speed of flight or the excessively large sampling interval, resulting in the lack of details in some urban areas, which leads to low accuracy in urban modeling. When compensating for areas with low accuracy in urban modeling, a long analysis time is required, further increasing the model update cycle. SUMMARY
[0003] The present application provides a kind of urban compensation modeling method and system combined with image recognition, to solve the technical problems of low accuracy and low efficiency of urban compensation modeling in the prior art.
[0004] In view of the above problems, the present application provides a kind of urban compensation modeling method and system combined with image recognition.
[0005] In a first aspect of the present application, a kind of urban compensation modeling method combined with image recognition is provided, the method comprises:
[0006] obtaining a historical urban three-dimensional model of a target city and a historical drone collection path set, wherein K historical drone collection paths in the historical drone collection path set are parallel to each other and the interval distance of any two adjacent historical drone collection paths is the same, K is an integer greater than or equal to 1;
[0007] based on the K environmental parameters of the K historical drone collection paths of the target city and the model parameters of the drone, the overlap area of the field of view of the drone between two adjacent historical drone collection paths during image collection is determined, and K-1 path interaction areas are obtained;
[0008] obtaining a preset sampling interval and a preset driving speed of the drone, combining the K-1 region lengths of the K-1 path interaction areas, dividing the K-1 path interaction areas, and obtaining a K-1 interaction sub-area set;
[0009] using the K-1 interaction sub-area set as an index, searching the texture mapping file of the historical urban three-dimensional model, and analyzing the image resolution and overlap rate of the search results to determine a K-1 fuzzy area set, wherein the K-1 fuzzy area set includes K-1 fuzzy degree sets;
[0010] determine the concentrated ambiguity to be compensated for;
[0011] Collect a set of historical unmanned aerial vehicle flight parameters for constructing the historical city three-dimensional model, optimize the set of historical unmanned aerial vehicle flight parameters according to the concentrated ambiguity to be compensated for, and determine an optimized set of unmanned aerial vehicle flight parameters;
[0012] Control the unmanned aerial vehicle based on the set of unmanned aerial vehicle flight parameters, collect images of the K-1 fuzzy region sets according to the set of historical unmanned aerial vehicle collection paths, and perform three-dimensional modeling according to the collection results, update the historical city three-dimensional model according to the modeling results, and obtain a target compensated city three-dimensional model.
[0013] In a second aspect of the present application, a city compensation modeling system combined with image recognition is provided, and the system comprises:
[0014] A collection path set acquisition module is configured to acquire a historical city three-dimensional model of a target city and a set of historical unmanned aerial vehicle collection paths, wherein K historical unmanned aerial vehicle collection paths in the set of historical unmanned aerial vehicle collection paths are parallel to each other, and the interval distance of any two adjacent historical unmanned aerial vehicle collection paths is the same, and K is an integer greater than or equal to 1;
[0015] A path interaction confrontation region acquisition module is configured to determine an overlap region of the unmanned aerial vehicle field of view of two adjacent historical unmanned aerial vehicle collection paths when image collection is performed based on K environmental parameters of the K historical unmanned aerial vehicle collection paths of the target city and model parameters of the unmanned aerial vehicle, and obtain K-1 path interaction confrontation regions;
[0016] An interaction confrontation sub-region set acquisition module is configured to acquire a preset sampling interval and a preset driving speed of the unmanned aerial vehicle, divide the K-1 path interaction confrontation regions based on K-1 region lengths of the K-1 path interaction confrontation regions, and obtain K-1 interaction confrontation sub-region sets;
[0017] A fuzzy region set determination module is configured to use the K-1 interaction confrontation sub-region sets as indexes to search a texture mapping file of the historical city three-dimensional model, analyze the search results in terms of image resolution and overlap rate, and determine K-1 fuzzy region sets, wherein the K-1 fuzzy region sets comprise K-1 ambiguity sets;
[0018] A concentrated ambiguity to be compensated for determination module is configured to perform ambiguity concentrated compensation analysis on the K-1 ambiguity sets, and determine the concentrated ambiguity to be compensated for;
[0019] The optimization unmanned aerial vehicle flight parameter set determination module is configured to collect a historical unmanned aerial vehicle flight parameter set for constructing the historical city three-dimensional model, and optimize the historical unmanned aerial vehicle flight parameter set according to the concentrated fuzzy degree to be compensated, to determine an optimized unmanned aerial vehicle flight parameter set;
[0020] The target compensated city three-dimensional model acquisition module is configured to control an unmanned aerial vehicle based on the unmanned aerial vehicle flight parameter set, perform image collection of the K-1 fuzzy region sets according to the historical unmanned aerial vehicle collection path set, perform three-dimensional modeling according to the collection result, perform compensated updating of the historical city three-dimensional model according to the modeling result, and acquire a target compensated city three-dimensional model.
[0021] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:
[0022] In the present application, a historical city three-dimensional model and a historical unmanned aerial vehicle collection path set of a target city are acquired, wherein K historical unmanned aerial vehicle collection paths in the historical unmanned aerial vehicle collection path set are parallel to each other and the interval distance of any two adjacent historical unmanned aerial vehicle collection paths is the same, K is an integer greater than or equal to 1, then based on K environmental parameters of the K historical unmanned aerial vehicle collection paths of the target city and model parameters of an unmanned aerial vehicle, an overlapping region of the unmanned aerial vehicle field of view of two adjacent historical unmanned aerial vehicle collection paths during image collection is determined, K-1 path interaction counter regions are acquired, a preset sampling interval and a preset driving speed of the unmanned aerial vehicle are acquired, the K-1 path interaction counter regions are divided based on K-1 region lengths of the K-1 path interaction counter regions, K-1 interaction counter sub-region sets are acquired, the K-1 interaction counter sub-region sets are used as indexes to search a texture mapping file of the historical city three-dimensional model, and the search result is analyzed in terms of image resolution and overlap rate to determine K-1 fuzzy region sets, wherein the K-1 fuzzy region sets include K-1 fuzzy degree sets, the K-1 fuzzy degree sets are subjected to fuzzy degree set compensation analysis, concentrated fuzzy degrees to be compensated are determined, a historical unmanned aerial vehicle flight parameter set for constructing the historical city three-dimensional model is collected, the historical unmanned aerial vehicle flight parameter set is optimized according to the concentrated fuzzy degrees to be compensated, an optimized unmanned aerial vehicle flight parameter set is determined, the unmanned aerial vehicle is controlled based on the unmanned aerial vehicle flight parameter set, image collection of the K-1 fuzzy region sets is performed according to the historical unmanned aerial vehicle collection path set, three-dimensional modeling is performed according to the collection result, the historical city three-dimensional model is subjected to compensated updating according to the modeling result, and a target compensated city three-dimensional model is acquired. The technical effect of efficiently identifying fuzzy regions and fuzzy degrees in a city three-dimensional model is achieved, and the accuracy and efficiency of compensated modeling are improved. BRIEF DESCRIPTION OF DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0024] Figure 1 A schematic diagram of a flow chart of a city compensation modeling method combined with image recognition provided in an embodiment of the present application;
[0025] Figure 2 A schematic diagram of a process for determining the concentrated ambiguity to be compensated in an urban compensation modeling method combined with image recognition provided in an embodiment of the present application;
[0026] Figure 3 A schematic diagram of the structure of an urban compensation modeling system combined with image recognition provided in an embodiment of the present application.
[0027] Explanation of the reference numerals: acquisition module for collection path set 11, acquisition module for path interactive confrontation area 12, acquisition module for interactive confrontation sub-area set 13, fuzzy area set determination module 14, centralized fuzziness to be compensated determination module 15, determination module for optimized UAV flight parameter set 16, target compensation city three-dimensional model acquisition module 17. DETAILED DESCRIPTION
[0028] This application provides a city compensation modeling method and system combined with image recognition, which is used to solve the technical problems of low accuracy and low efficiency of city compensation modeling in the existing technology.
[0029] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0030] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.
[0031] Example 1, as Figure 1 As shown, the present application provides a city compensation modeling method combined with image recognition, wherein the method includes:
[0032] S100: obtaining a historical city three-dimensional model of a target city and a historical unmanned aerial vehicle (UAV) collection path set, wherein K historical UAV collection paths in the historical UAV collection path set are parallel to each other and any two adjacent historical UAV collection paths have the same interval distance, and K is an integer greater than or equal to 1;
[0033] In one possible embodiment, the target city is any city that needs to be three-dimensionally modeled. The historical city three-dimensional model is obtained by scanning and modeling the target city according to the historical UAV collection path set using a UAV loaded with a laser radar or a camera at a historical time. Obtaining the historical city three-dimensional model of the target city provides basic reference data for subsequent compensation and updating.
[0034] Preferably, the historical UAV collection path set includes K historical UAV collection paths, wherein the K historical UAV collection paths are parallel to each other and any two adjacent historical UAV collection paths have the same interval distance, so as to ensure omission or repetition of image collection by the UAV and improve the accuracy and splicing efficiency when splicing the images obtained by scanning the target city. K is an integer greater than or equal to 1. Optionally,
[0035] S200: determining an overlapping area of a UAV field of view of two adjacent historical UAV collection paths when performing image collection based on K environmental parameters of the K historical UAV collection paths of the target city and a model parameter of the UAV, and obtaining K-1 path interaction areas;
[0036] In the embodiments of the present application, K environmental parameters of K historical UAV collection paths of a target city are first needed to be obtained. The environmental parameters of the historical UAV collection paths include but are not limited to weather conditions (such as wind speed and humidity), light conditions, terrain undulations (such as building height and density), and other factors that can affect the flight stability and field of view range of the UAV. The UAV model parameter is used to describe the type and performance of the UAV, including flight height, flight speed, field of view angle (lens field of view angle), image resolution, camera type, etc. These parameters are crucial for calculating the field of view range and image collection accuracy of the UAV.
[0037] For each two adjacent paths in the historical UAV collection paths, the image collection range (i.e., field of view) of the UAV on each path needs to be calculated based on the flight height, viewing angle and other parameters of the UAV. Generally, the field of view range of the UAV is calculated based on its camera field of view angle (FOV) and flight height, which can be determined by a person skilled in the art using geometric methods.
[0038] According to the spatial relationship between two adjacent historical unmanned aerial vehicle collection paths and the range of the unmanned aerial vehicle field of view, the field of view overlapping area of the two paths is calculated, and the field of view overlapping area is taken as the K-1 path interaction confrontation areas. The K-1 path interaction confrontation areas obtained lay the foundation for the model splicing after the three-dimensional modeling of the image obtained after the unmanned aerial vehicle collects the target city according to the K historical unmanned aerial vehicle collection paths.
[0039] Since there are K historical unmanned aerial vehicle collection paths, there are K-1 adjacent historical unmanned aerial vehicle collection path pairs. The unmanned aerial vehicle field of view overlapping area formed when the unmanned aerial vehicle collects images on the K-1 adjacent historical unmanned aerial vehicle collection path pairs is taken as the K-1 path interaction confrontation areas. Each path interaction confrontation area represents the field of view overlapping situation between two adjacent paths.
[0040] By obtaining the K-1 path interaction confrontation areas, the subsequent image analysis and three-dimensional modeling compensation are prepared, so that the image data fusion, reconstruction and compensation can be better performed, and the accuracy and efficiency of the compensation modeling are improved.
[0041] For example, assuming that there are 3 historical unmanned aerial vehicle collection paths in the target city (i.e. K=3), the overlapping area between two adjacent paths needs to be calculated: one interaction confrontation area is formed between the first path and the second path; a second interaction confrontation area is formed between the second path and the third path; the two interaction confrontation areas will be used as the basis for subsequent compensation modeling analysis, and finally 2 path interaction confrontation areas are obtained.
[0042] S300: Obtain the preset sampling interval and the preset driving speed of the unmanned aerial vehicle, divide the K-1 path interaction confrontation areas in combination with the K-1 region lengths of the K-1 path interaction confrontation areas, and obtain a K-1 interaction confrontation sub-region set;
[0043] Further, the preset sampling interval and the preset driving speed of the unmanned aerial vehicle are obtained, the K-1 path interaction confrontation areas are divided in combination with the K-1 region lengths of the K-1 path interaction confrontation areas, and a K-1 interaction confrontation sub-region set is obtained. The embodiment S300 of the application further comprises:
[0044] The preset sampling interval is multiplied by the preset driving speed to obtain a preset sampling length;
[0045] According to the preset sampling length and the K-1 region lengths, the ratio of the preset sampling length is obtained.
[0046] The K-1 path interaction confrontation regions are sequentially divided according to a path collection direction based on the K sub-region division quantity and the preset sampling length, to obtain the K-1 sets of interaction confrontation sub-regions.
[0047] In one possible embodiment, the preset sampling interval is a time interval at which the UAV collects image data at a certain interval during flight, which is preset by a person skilled in the art. The person skilled in the art sets the preset sampling interval according to the required image resolution and image update frequency. A smaller sampling interval can improve the quality of image data, but may require higher computing power and storage capacity. The preset driving speed is the speed of the UAV along the collection path during flight, which is preset by a person skilled in the art. The driving speed affects the visual relationship between the UAV and the ground target, thereby affecting the definition of the image and the coverage of the collection. Generally, a lower flight speed can obtain a higher image resolution, but also results in a longer flight time.
[0048] Optionally, based on the preset sampling interval and the driving speed, the flight distance (sampling length) corresponding to each sampling can be calculated. The formula is: sampling length = preset sampling interval × preset driving speed. For example, assuming that the preset sampling interval of the UAV is 10 seconds and the preset driving speed is 5 meters / second, then the sampling length of each sampling is 50 meters. The sampling length indicates that the UAV collects image data once every 50 meters of flight.
[0049] Preferably, each path interaction confrontation region is an overlapping region of the field of view of the UAV when the UAV collects image data of the target city according to two adjacent historical UAV collection paths, and the UAV collects image data according to the sampling interval and the preset driving speed. Therefore, each path interaction confrontation region can be equally divided into a set of interaction confrontation sub-regions. The area of each interaction confrontation sub-region in each set of interaction confrontation sub-regions is the same. By dividing the K-1 path interaction confrontation regions, K-1 sets of interaction confrontation sub-regions are obtained. The number of interaction confrontation sub-regions in each set of interaction confrontation sub-regions is determined by the region length of the path interaction confrontation region. The longer the region length, the more the number of interaction confrontation sub-regions corresponding to the division.
[0050] In one possible embodiment, the length of each of the K-1 path interaction-antagonistic regions is counted, and a ratio of the length of each of the K-1 path interaction-antagonistic regions to the preset sampling length is calculated to obtain the K sub-region division quantity. The ratio represents how many times of image acquisition can be performed in the region according to the sampling interval and the flight speed. Preferably, the ratio is rounded. For example, if the length of a region is 100 meters and the sampling length is 50 meters, the ratio is 2, indicating that 2 times of image acquisition can be performed in the region.
[0051] The K-1 path interaction-antagonistic regions are divided according to the path acquisition direction based on the K sub-region division quantity and the preset sampling length to obtain the K-1 set of interaction-antagonistic sub-regions. This means that the length of each region is the same as the preset sampling length. Assuming that the length of a path interaction-antagonistic region is 100 meters and the preset sampling length is 50, the path interaction-antagonistic region is divided into two 50-meter-long interaction-antagonistic sub-regions. Each interaction-antagonistic sub-region will become a basic unit for subsequent image acquisition and texture mapping.
[0052] S400: The historical urban three-dimensional model is searched based on the K-1 set of interaction-antagonistic sub-regions as an index, and the search result is analyzed in terms of image resolution and overlap rate to determine the K-1 set of fuzzy regions, wherein the K-1 set of fuzzy regions includes K-1 sets of fuzziness.
[0053] In one embodiment of the present application, the historical urban three-dimensional model is also obtained by simulating the texture mapping file collected by the unmanned aerial vehicle, so that the historical urban three-dimensional model is searched based on the K-1 set of interaction-antagonistic sub-regions as an index to obtain the texture mapping file corresponding to the sub-regions. The texture mapping file is image data for describing the urban surface details (such as buildings, roads, trees, etc.) of the K-1 set of interaction-antagonistic sub-regions in the historical urban three-dimensional model.
[0054] Further, the image data in the search result is analyzed in terms of image resolution and overlap rate to determine the sub-regions in the K-1 set of interaction-antagonistic sub-regions that are fuzzy, and the sub-regions are added to the K-1 set of fuzzy regions. The K-1 set of fuzzy regions includes K-1 sets of fuzziness. The fuzziness is used to describe the image blurring degree of the interaction-antagonistic sub-region. The lower the image resolution and the overlap rate, the greater the fuzziness. In the case of a certain image resolution, the greater the overlap rate, the more images can be acquired by the interaction-antagonistic sub-region, and the smaller the fuzziness.
[0055] By searching and image quality analysis on the texture mapping file of the historical city three-dimensional model according to the K-1 interactive confrontation sub-region set, in combination with the resolution and overlap rate of the image, the fuzzy region is identified, the region needing compensation or optimization can be effectively determined, and the technical effects of improving the quality and accuracy of the city three-dimensional model are achieved.
[0056] Further, the texture mapping file of the historical city three-dimensional model is searched according to the K-1 interactive confrontation sub-region set as an index, and the image resolution and overlap rate analysis is performed on the search result to determine the K-1 fuzzy region set. The step S400 of the embodiment of the present application further includes:
[0057] The texture mapping file of the historical city three-dimensional model is searched according to the K-1 interactive confrontation sub-region set as an index, and K-1 sub-region unmanned aerial vehicle image group sets are obtained. Each sub-region unmanned aerial vehicle image group includes two region unmanned aerial vehicle images collected by the unmanned aerial vehicle for image collection of the corresponding interactive confrontation sub-region according to two adjacent historical unmanned aerial vehicle collection paths.
[0058] The pixel width and pixel height extraction is performed on the K-1 sub-region unmanned aerial vehicle image group set, and the K-1 sub-region image resolution group set is obtained according to the extraction result.
[0059] The in-group image overlap area identification is performed on the K-1 sub-region unmanned aerial vehicle image group set, and the K-1 overlap area identification result set is compared with the area of any one of the K-1 interactive confrontation sub-region set, and the K-1 sub-region image overlap rate set is obtained.
[0060] The blur degree identifier is called to perform blur degree identification on the K-1 sub-region image resolution group set and the K-1 sub-region image overlap rate set respectively, and the K-1 initial blur degree set is obtained.
[0061] It is judged whether the K-1 initial blur degree set is greater than or equal to the preset blur degree respectively. If yes, the corresponding interactive confrontation sub-region is taken as a fuzzy region, and the K-1 fuzzy region set is obtained.
[0062] Further, the step S400 of the embodiment of the present application further includes:
[0063] A plurality of sample sub-region image resolution groups, a plurality of sample sub-region image overlap rates and a plurality of sample initial blur degrees are collected as a training sample data set.
[0064] The network framework of the blur recognizer based on the feedforward neural network is iteratively trained using the training sample dataset, and a one-to-two mapping relationship between the sub-region image resolution group, the sub-region image overlap rate, and the initial blur is learned until the convergence requirement is met, and the trained blur recognizer is obtained.
[0065] In one possible embodiment, the K-1 sets of interactive confrontation sub-regions are used as indexes to retrieve corresponding texture map files in the historical urban three-dimensional model, and the K-1 sets of sub-region UAV collected image groups are obtained. Each sub-region UAV collected image group includes two sub-region UAV collected images collected by the UAV for the corresponding interactive confrontation sub-region along two adjacent historical UAV collected paths. For example, each sub-region UAV collected image group includes two images: one is a collected image of the interactive confrontation sub-region along path A, and the other is a collected image of the interactive confrontation sub-region along path B. Path A is adjacent to path B.
[0066] The resolution of each sub-region UAV collected image group is analyzed, the pixel width (horizontal pixel number) and the pixel height (vertical pixel number) of each sub-region UAV collected image are extracted, and then the resolution of the image is calculated according to the pixel width and the pixel height, and the K-1 sets of image resolution groups of the K-1 sub-regions are obtained.
[0067] The overlap area of each K-1 set of sub-region UAV collected image groups is calculated, and the ratio of the overlap area to the area of any one of the corresponding interactive confrontation sub-regions is calculated, so as to obtain the overlap rate of the image, and the K-1 sets of image overlap rates of the K-1 sub-regions are obtained.
[0068] In one possible embodiment, the blur recognizer is a pre-trained functional model for intelligently analyzing the image resolution group and the image overlap rate to determine the blur. The blur recognizer combines the image resolution group and the image overlap rate to calculate the blur value of the image of each interactive confrontation sub-region, and obtains the initial blur set of the K-1 sub-regions.
[0069] Optionally, a plurality of sample sub-region image resolution groups, a plurality of sample sub-region image overlap rates, and a plurality of sample initial blurs are collected to form a training sample dataset. A blur recognizer is constructed using a feedforward neural network architecture. The network learns a one-to-two mapping relationship between the image resolution and the image overlap rate and the initial blur. The neural network is iteratively trained using the training sample dataset, and the network weights are continuously adjusted until the network converges and meets the preset accuracy requirement. A trained blur recognizer that can accurately identify blurred regions is obtained.
[0070] The trained ambiguity identifier is used to identify the ambiguity of the K-1 sets of sub-region image resolution groups and the K-1 sets of sub-region image overlap rates respectively, to determine and output all ambiguities, and to obtain the K-1 sets of initial ambiguities.
[0071] Further, according to the size of the initial ambiguities in the K-1 sets of initial ambiguities, it is determined which interactive adversarial sub-regions need to be marked as ambiguous regions. It is determined whether the K-1 sets of initial ambiguities are greater than or equal to a preset ambiguity. If so, the corresponding interactive adversarial sub-region is taken as an ambiguous region, and the K-1 sets of ambiguous regions are obtained.
[0072] By combining the resolution, overlap rate and ambiguity identification technology of the image, the ambiguous regions in the historical city three-dimensional model are effectively identified, and the accuracy of the identification is further improved by the ambiguity identifier. In the process of optimizing image acquisition and modeling, these ambiguous regions will be used as key regions for supplementary acquisition or optimization processing, thereby improving the quality and accuracy of the final city compensation three-dimensional model.
[0073] S500: Perform ambiguity set compensation analysis on the K-1 sets of ambiguities to determine the set of ambiguities to be compensated;
[0074] Further, as shown in Figure 2 The K-1 sets of ambiguities are analyzed for ambiguity set compensation, and the set of ambiguities to be compensated is determined. The embodiment S500 of the application further comprises:
[0075] Each ambiguity in the K-1 sets of ambiguities is taken as an analysis particle point, and a compensation analysis space is constructed, wherein the compensation analysis space includes a plurality of analysis particle points;
[0076] The mean value of the ambiguities of the plurality of analysis particle points is calculated to obtain a central analysis particle point;
[0077] The central analysis particle point is taken as a starting point, and a neighborhood is constructed in the compensation analysis space according to a preset iteration step size to determine a central analysis neighborhood, wherein the plurality of analysis particle points in the central analysis neighborhood are analysis particle points whose distance to the starting point is within the preset iteration step size;
[0078] The central analysis neighborhood is iteratively expanded according to the preset iteration step size to obtain a target analysis neighborhood;
[0079] The mean value of the ambiguities of the plurality of analysis particle points in the target analysis neighborhood is calculated to obtain the set of ambiguities to be compensated.
[0080] Further, the center analysis neighborhood is iteratively expanded according to the preset iteration step, to obtain a target analysis neighborhood, and the step S500 further includes:
[0081] The leftmost analysis particle point of the center analysis neighborhood is iterated to the left and the rightmost analysis particle point of the center analysis neighborhood is iterated to the right according to the preset iteration step respectively, to obtain a first left-side iteration analysis neighborhood and a first right-side iteration analysis neighborhood;
[0082] It is judged whether the concentration density of the first left-side iteration analysis neighborhood and the concentration density of the first right-side iteration analysis neighborhood are both greater than or equal to the concentration density of the center analysis neighborhood, and if yes, a union set of the first left-side iteration analysis neighborhood and the first right-side iteration analysis neighborhood is obtained, to obtain a first-stage iteration analysis neighborhood;
[0083] The leftmost analysis particle point of the first-stage iteration analysis neighborhood is iterated to the left and the rightmost analysis particle point of the first-stage iteration analysis neighborhood is iterated to the right again according to the preset iteration step respectively;
[0084] After multiple bilateral iterations, when the concentration density of the left-side stage iteration analysis neighborhood and the concentration density of the right-side stage iteration analysis neighborhood obtained by the current iteration are both less than the concentration density of the stage iteration analysis neighborhood obtained by the last iteration, the iteration is stopped, and the stage iteration analysis neighborhood obtained by the last iteration is taken as the target analysis neighborhood.
[0085] Further, the step S500 further includes:
[0086] When only the concentration density of the first left-side iteration analysis neighborhood or the concentration density of the first right-side iteration analysis neighborhood is greater than or equal to the concentration density of the center analysis neighborhood, the first left-side iteration analysis neighborhood or the first right-side iteration analysis neighborhood is taken as the first-stage iteration analysis neighborhood;
[0087] The leftmost analysis particle point of the first-stage iteration analysis neighborhood is iterated to the left or the rightmost analysis particle point of the first-stage iteration analysis neighborhood is iterated to the right according to the preset iteration step;
[0088] After multiple unilateral iterations, when the concentration density of the left-side stage iteration analysis neighborhood or the concentration density of the right-side stage iteration analysis neighborhood obtained by the current iteration is less than the concentration density of the stage iteration analysis neighborhood obtained by the last iteration, the iteration is stopped, and the stage iteration analysis neighborhood obtained by the last iteration is taken as the target analysis neighborhood.
[0089] In a possible embodiment, the K-1 sets of ambiguity degree reflect the ambiguity degrees of K-1 sets of ambiguous regions in the K-1 sets of interactive confrontation sub-regions in the historical urban three-dimensional model. By performing a set compensation analysis on the K-1 sets of ambiguity degree, the general ambiguity degree of the K-1 sets of ambiguous regions is determined, and the set of ambiguity degrees to be compensated is obtained. The technical effect of providing data support for subsequent urban compensation modeling is achieved.
[0090] By taking each ambiguity degree in the K-1 sets of ambiguity degree as an analysis particle point, a compensation analysis space is constructed, wherein the compensation analysis space includes a plurality of analysis particle points. Further, the average ambiguity degree of the plurality of analysis particle points is calculated to determine the average ambiguity degree of the compensation analysis space when considering analysis particle points at edge positions, thereby obtaining a central analysis particle point.
[0091] Further, a central analysis neighborhood is constructed in the compensation analysis space according to a preset iteration step (a difference value of ambiguity degree in a single iteration preset by a person skilled in the art) from the central analysis particle point as a starting point, wherein the plurality of analysis particle points in the central analysis neighborhood are analysis particle points having a distance from the starting point within the preset iteration step.
[0092] Further, the central analysis neighborhood is iteratively expanded according to the preset iteration step until the edge of a region with a relatively dense distribution of analysis particle points is reached, and a target analysis neighborhood is obtained. By calculating the average ambiguity degree of the plurality of analysis particle points in the target analysis neighborhood, the general ambiguity degree of the K-1 sets of ambiguity degree is determined without considering ambiguity degrees that are relatively biased and have a low frequency of occurrence, and the calculation result is taken as the set of ambiguity degrees to be compensated.
[0093] Preferably, the leftmost analysis particle point of the central analysis neighborhood is iterated to the left and the rightmost analysis particle point of the central analysis neighborhood is iterated to the right according to the preset iteration step, respectively, to obtain a first left iteration analysis neighborhood and a first right iteration analysis neighborhood. The distance of an analysis particle point in the first left iteration analysis neighborhood from the leftmost analysis particle point of the central analysis neighborhood is less than or equal to the preset iteration step. The distance of an analysis particle point in the first right iteration analysis neighborhood from the rightmost analysis particle point of the central analysis neighborhood is less than or equal to the preset iteration step.
[0094] In one embodiment, the center neighborhood blur difference of the leftmost analysis particle and the rightmost analysis particle of the center analysis neighborhood is calculated, and the number of analysis particle points in the center analysis neighborhood is counted. The ratio of the counted number of analysis particle points to the center neighborhood blur difference is taken as the concentration density of the center analysis neighborhood. The concentration density of the center analysis neighborhood reflects the degree of concentration of the analysis particle points in the center analysis neighborhood. Based on the same principle, the concentration density of the first left iterative analysis neighborhood and the concentration density of the first right iterative analysis neighborhood are calculated.
[0095] If the concentration density of the first left iterative analysis neighborhood and the concentration density of the first right iterative analysis neighborhood are both greater than or equal to the concentration density of the center analysis neighborhood, it indicates that the distribution of the analysis particle points in the first left iterative analysis neighborhood and the first right iterative analysis neighborhood is relatively dense. In this case, the union of the first left iterative analysis neighborhood and the first right iterative analysis neighborhood is calculated to obtain the first-stage iterative analysis neighborhood.
[0096] Further, the leftmost analysis particle point of the first-stage iterative analysis neighborhood is iterated to the left and the rightmost analysis particle point of the first-stage iterative analysis neighborhood is iterated to the right according to the preset iterative step size, respectively. After multiple iterations, when the concentration density of the left-stage iterative analysis neighborhood and the concentration density of the right-stage iterative analysis neighborhood obtained in the current iteration are both less than the concentration density of the stage iterative analysis neighborhood obtained in the last iteration, the iteration is stopped. The stage iterative analysis neighborhood obtained in the last iteration is taken as the target analysis neighborhood.
[0097] In one embodiment, when only the concentration density of the first left iterative analysis neighborhood or the concentration density of the first right iterative analysis neighborhood is greater than or equal to the concentration density of the center analysis neighborhood, it indicates that the analysis particle point concentration degree of the first left iterative analysis neighborhood or the analysis particle point concentration degree of the first right iterative analysis neighborhood is greater than that of the center analysis neighborhood. In this case, the first left iterative analysis neighborhood or the first right iterative analysis neighborhood is taken as the first-stage iterative analysis neighborhood. The leftmost analysis particle point of the first-stage iterative analysis neighborhood is iterated to the left or the rightmost analysis particle point of the first-stage iterative analysis neighborhood is iterated to the right according to the preset iterative step size.
[0098] After multiple unilateral iterations, when the concentration density of the left-stage iterative analysis neighborhood or the concentration density of the right-stage iterative analysis neighborhood obtained in the current iteration is less than the concentration density of the stage iterative analysis neighborhood obtained in the last iteration, the iteration is stopped. The stage iterative analysis neighborhood obtained in the last iteration is taken as the target analysis neighborhood.
[0099] The determination of the concentrated blur degree to be compensated provides data support for subsequent city compensation modeling, and a technical effect of improving compensation modeling accuracy is achieved.
[0100] S600: Collect a set of historical unmanned aerial vehicle flight parameters for constructing the historical city three-dimensional model, optimize the set of historical unmanned aerial vehicle flight parameters according to the concentrated blur degree to be compensated, and determine a set of optimized unmanned aerial vehicle flight parameters;
[0101] S700: Control the unmanned aerial vehicle based on the set of unmanned aerial vehicle flight parameters, collect images of the K-1 sets of blur regions according to the set of historical unmanned aerial vehicle collection paths, and perform three-dimensional modeling according to the collection results, perform compensation update on the historical city three-dimensional model according to the modeling results, and obtain a target compensated city three-dimensional model.
[0102] In one embodiment, a set of historical unmanned aerial vehicle flight parameters during model construction is extracted from the historical city three-dimensional model. The set of historical unmanned aerial vehicle flight parameters includes flight path, flight height, speed, attitude, and the like. Based on the concentrated analysis result of the blur degree, the flight parameters (such as flight height, flight attitude, etc.) are adjusted to reduce the error caused by the blur region.
[0103] Optionally, a plurality of sample sets of unmanned aerial vehicle flight parameters, a plurality of sample concentrated blur degrees to be compensated, and a plurality of sample sets of optimized unmanned aerial vehicle flight parameters are obtained as sample data, the network framework constructed based on the feedforward neural network is supervised and trained using the sample data until convergence is achieved, and a trained flight parameter optimizer is obtained. The concentrated blur degree to be compensated and the set of historical unmanned aerial vehicle flight parameters are input into the flight parameter optimizer for intelligent analysis, and the set of optimized unmanned aerial vehicle flight parameters is obtained. It is ensured that when the unmanned aerial vehicle collects images according to the set of optimized unmanned aerial vehicle flight parameters, the influence of the blur region in the historical model can be effectively compensated.
[0104] The set of optimized unmanned aerial vehicle flight parameters obtained in the S600 step is used to control the unmanned aerial vehicle to fly according to the historical flight path. According to the set of historical unmanned aerial vehicle collection paths, the unmanned aerial vehicle is organized to collect images of the K-1 sets of blur regions in the S500 step, to ensure that the image data covers the blur region and to improve the image quality as much as possible. The image data collected by the unmanned aerial vehicle is used to perform three-dimensional reconstruction using an image processing algorithm to obtain a high-precision three-dimensional model of the target region. The historical city three-dimensional model is compensated and updated in combination with the collected three-dimensional data. By comparing the differences between the updated three-dimensional model and the original model, error correction and detail supplementation are performed, and finally a target compensated city three-dimensional model is obtained.
[0105] By optimizing the flight parameters and precise image acquisition, the influence of the fuzzy area is compensated, the accuracy of three-dimensional modeling is improved, and the effective compensation update of the historical city three-dimensional model is realized.
[0106] In summary, the embodiments of the present application have at least the following technical effects:
[0107] 1. The present application can determine the overlapping area between adjacent paths by combining the historical unmanned aerial vehicle collection path and environmental parameters, thereby effectively filling in the missing areas during image acquisition. The analysis and compensation of the fuzzy area ensure that the details of the historical city three-dimensional model are more accurately reconstructed, avoiding model distortion caused by insufficient angle of view or resolution, thereby significantly improving the accuracy of the compensation model.
[0108] 2. By analyzing historical unmanned aerial vehicle flight parameters and sampling intervals, flight speed and other factors, the flight path and parameters of the unmanned aerial vehicle can be accurately optimized to ensure the overlap rate and image resolution of the images obtained during image acquisition of the fuzzy area, correct the fuzzy area in the historical city three-dimensional model, and achieve the technical effects of improving modeling efficiency and compensation modeling quality.
[0109] Embodiment two, based on the same inventive concept as the city compensation modeling method combined with image recognition in the foregoing embodiments, as shown in Figure 3 The present application provides a city compensation modeling system combined with image recognition, and the system and method embodiments in the present application are based on the same inventive concept. The system comprises:
[0110] The collection path set acquisition module 11 is used to acquire the historical city three-dimensional model of the target city and the historical unmanned aerial vehicle collection path set, wherein the K historical unmanned aerial vehicle collection paths in the historical unmanned aerial vehicle collection path set are parallel to each other and the interval distance of any two adjacent historical unmanned aerial vehicle collection paths is the same, and K is an integer greater than or equal to 1;
[0111] The path interaction area obtaining module 12 is used to determine the overlapping area of the unmanned aerial vehicle field of view of two adjacent historical unmanned aerial vehicle collection paths during image acquisition based on the K environmental parameters of the K historical unmanned aerial vehicle collection paths of the target city and the model parameters of the unmanned aerial vehicle, and obtain K-1 path interaction areas;
[0112] The interaction sub-area set obtaining module 13 is used to acquire the preset sampling interval and the preset driving speed of the unmanned aerial vehicle, combine the K-1 region lengths of the K-1 path interaction areas, divide the K-1 path interaction areas, and obtain K-1 interaction sub-area sets;
[0113] The fuzzy region set determination module 14 is used for indexing the K-1 interactive confrontation sub-region sets to retrieve the texture mapping files of the historical urban three-dimensional model, and performing image resolution and overlap rate analysis on the retrieval results to determine K-1 fuzzy region sets, wherein the K-1 fuzzy region sets include K-1 fuzzy degree sets;
[0114] The concentrated fuzzy degree to be compensated determination module 15 is used for performing fuzzy degree concentrated compensation analysis on the K-1 fuzzy degree sets to determine the concentrated fuzzy degree to be compensated;
[0115] The optimized unmanned aerial vehicle flight parameter set determination module 16 is used for collecting a historical unmanned aerial vehicle flight parameter set used for constructing the historical urban three-dimensional model, and optimizing the historical unmanned aerial vehicle flight parameter set according to the concentrated fuzzy degree to be compensated to determine an optimized unmanned aerial vehicle flight parameter set;
[0116] The target compensated urban three-dimensional model obtaining module 17 is used for controlling an unmanned aerial vehicle based on the unmanned aerial vehicle flight parameter set, collecting images of the K-1 fuzzy region sets according to the historical unmanned aerial vehicle collection path set, and performing three-dimensional modeling according to the collection results, and performing compensation update on the historical urban three-dimensional model according to the modeling results to obtain a target compensated urban three-dimensional model.
[0117] Further, the fuzzy region set determination module 14 is used for performing the following steps:
[0118] Indexing the K-1 interactive confrontation sub-region sets to retrieve the texture mapping files of the historical urban three-dimensional model to obtain K-1 sub-region unmanned aerial vehicle image group sets, and each sub-region unmanned aerial vehicle image group includes two regional unmanned aerial vehicle images collected by an unmanned aerial vehicle according to two adjacent historical unmanned aerial vehicle collection paths for a corresponding interactive confrontation sub-region;
[0119] Iterating the K-1 sub-region unmanned aerial vehicle image group sets to extract pixel width and pixel height, and obtaining K-1 sub-region image resolution group sets according to the extraction results;
[0120] Iterating the K-1 sub-region unmanned aerial vehicle image group sets to identify the image overlap area within the group, and comparing K-1 overlap area identification result sets with the area of any one of the K-1 interactive confrontation sub-region sets to obtain K-1 sub-region image overlap rate sets;
[0121] Calling a fuzzy degree identifier to perform fuzzy degree identification on the K-1 sub-region image resolution group sets and the K-1 sub-region image overlap rate sets respectively to obtain K-1 initial fuzzy degree sets;
[0122] respectively, if yes, the corresponding interactive counter-sub-region is taken as a blur region, and K-1 blur region sets are obtained.
[0123] Further, the blur region set determination module 14 is configured to perform the following steps:
[0124] A plurality of sample sub-region image resolution groups, a plurality of sample sub-region image overlap rates, and a plurality of sample initial blurs are collected as a training sample data set.
[0125] The network framework of the blur identifier based on the feedforward neural network is iteratively trained using the training sample data set, and a one-to-two mapping relationship between the sub-region image resolution group and the sub-region image overlap rate and the initial blur is learned until a convergence requirement is met, and the trained blur identifier is obtained.
[0126] Further, the centralized blur to be compensated determination module 15 is configured to perform the following steps:
[0127] Each blur in the K-1 blur set is taken as an analysis particle point, and a compensation analysis space is constructed, wherein the compensation analysis space includes a plurality of analysis particle points.
[0128] The average blur of the plurality of analysis particle points is calculated to obtain a central analysis particle point.
[0129] The central analysis neighborhood is determined by taking the central analysis particle point as a starting point and constructing a neighborhood in the compensation analysis space according to a preset iteration step length, wherein the plurality of analysis particle points in the central analysis neighborhood are the analysis particle points whose distances to the starting point are within the preset iteration step length.
[0130] The central analysis neighborhood is iteratively expanded according to the preset iteration step length to obtain a target analysis neighborhood.
[0131] The average blur of the plurality of analysis particle points in the target analysis neighborhood is calculated to obtain the centralized blur to be compensated.
[0132] Further, the centralized blur to be compensated determination module 15 is configured to perform the following steps:
[0133] The leftmost analysis particle point of the central analysis neighborhood is iteratively moved to the left and the rightmost analysis particle point of the central analysis neighborhood is iteratively moved to the right according to the preset iteration step length, respectively, to obtain a first left iterative analysis neighborhood and a first right iterative analysis neighborhood.
[0134] determining whether the concentration density of the first left iterative analysis neighborhood and the concentration density of the first right iterative analysis neighborhood are both greater than or equal to the concentration density of the center analysis neighborhood, if yes, performing a set union on the first left iterative analysis neighborhood and the first right iterative analysis neighborhood to obtain a first-stage iterative analysis neighborhood;
[0135] again performing left iteration on the leftmost analysis particle point of the first-stage iterative analysis neighborhood and right iteration on the rightmost analysis particle point of the first-stage iterative analysis neighborhood according to the preset iteration step size, and stopping iteration when the concentration density of the left-stage iterative analysis neighborhood and the concentration density of the right-stage iterative analysis neighborhood obtained in the current iteration are both less than the concentration density of the stage iterative analysis neighborhood obtained in the last iteration, and taking the stage iterative analysis neighborhood obtained in the last iteration as the target analysis neighborhood.
[0136] Further, the concentration ambiguity determination module 15 is configured to perform the following steps:
[0137] when the concentration density of the first left iterative analysis neighborhood or the concentration density of the first right iterative analysis neighborhood is greater than or equal to the concentration density of the center analysis neighborhood, taking the first left iterative analysis neighborhood or the first right iterative analysis neighborhood as a first-stage iterative analysis neighborhood;
[0138] performing left iteration on the leftmost analysis particle point of the first-stage iterative analysis neighborhood or right iteration on the rightmost analysis particle point of the first-stage iterative analysis neighborhood according to the preset iteration step size, and stopping iteration when the concentration density of the left-stage iterative analysis neighborhood or the concentration density of the right-stage iterative analysis neighborhood obtained in the current iteration is less than the concentration density of the stage iterative analysis neighborhood obtained in the last iteration, and taking the stage iterative analysis neighborhood obtained in the last iteration as the target analysis neighborhood.
[0139] Further, the interactive antagonistic sub-region set obtaining module 13 is configured to perform the following steps:
[0140] multiplying the preset sampling interval by the preset driving speed to obtain a preset sampling length;
[0141] calculating the ratio of the K-1 region length ratios to the preset sampling length to obtain K sub-region division numbers;
[0142] dividing the K-1 path interactive antagonistic regions according to the path collection direction based on the K sub-region division numbers and the preset sampling length to obtain the K-1 interactive antagonistic sub-region sets.
[0143] It should be noted that the above-mentioned embodiment sequences of the present application are merely for description only, but not for representing the advantages and disadvantages of the embodiments. And the above-mentioned embodiments of the present specification have been described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.
[0144] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
[0145] The specification and drawings are merely exemplary of the present application, and any and all modifications, variations, combinations or equivalents that are within the scope of the present application should be considered covered by the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is intended to include these modifications and variations.
Claims
1. A city compensation modeling method combined with image recognition, characterized in that: The method comprises: Obtain a historical 3D city model and a set of historical drone collection paths for the target city, wherein K historical drone collection paths in the set are parallel to each other and the distance between any two adjacent historical drone collection paths is the same, and K is an integer greater than or equal to 1; Based on K environmental parameters of K historical drone collection paths and drone model parameters of the target city, the overlapping area of the drone field of view of two adjacent historical drone collection paths during image collection is determined to obtain K-1 path interaction confrontation areas; Obtaining a preset sampling interval and a preset driving speed of the UAV, and dividing the K-1 path interaction confrontation areas into K-1 interaction confrontation sub-area sets based on the K-1 area lengths of the K-1 path interaction confrontation areas; Using the K-1 interactive adversarial sub-region sets as indexes, searching for texture mapping files of the historical city three-dimensional model, and performing image resolution and overlap rate analysis on the search results to determine K-1 fuzzy region sets, wherein the K-1 fuzzy region sets include K-1 fuzziness degree sets; performing centralized ambiguity compensation analysis on the K-1 ambiguity sets to determine centralized ambiguities to be compensated; Collecting a set of historical UAV flight parameters for constructing the historical city three-dimensional model, optimizing the set of historical UAV flight parameters according to the centralized ambiguity to be compensated, and determining an optimized set of UAV flight parameters; The drone is controlled based on the drone flight parameter set, images of the K-1 fuzzy area sets are acquired according to the historical drone acquisition path set, three-dimensional modeling is performed based on the acquisition results, and the historical three-dimensional city model is compensated and updated based on the modeling results to obtain a target compensated three-dimensional city model; The step of performing centralized ambiguity compensation analysis on the K-1 ambiguity sets to determine centralized ambiguities to be compensated includes: Taking each ambiguity in the K-1 ambiguity sets as an analysis particle point, constructing a compensation analysis space, wherein the compensation analysis space includes a plurality of analysis particle points; Calculating the fuzziness mean of the multiple analysis particle points to obtain a central analysis particle point; Taking the central analysis particle point as a starting point, constructing a neighborhood in the compensation analysis space according to a preset iteration step length to determine a central analysis neighborhood, wherein a plurality of analysis particle points in the central analysis neighborhood are analysis particle points whose distance from the starting point is within the preset iteration step length; Iteratively expanding the central analysis neighborhood according to the preset iteration step size to obtain a target analysis neighborhood; The ambiguity mean of multiple analysis particle points in the target analysis neighborhood is calculated to obtain the concentrated ambiguity to be compensated.
2. The urban compensation modeling method combined with image recognition according to claim 1, characterized in that: Using the K-1 interactive adversarial sub-region sets as indexes, the texture map files of the historical city three-dimensional model are retrieved, and the image resolution and overlap rate of the retrieval results are analyzed to determine the K-1 fuzzy region sets, including: Using the K-1 interactive adversarial sub-region sets as indexes, a texture map file of the historical city 3D model is retrieved to obtain a set of K-1 sub-region drone-collected image groups, each sub-region drone-collected image group including two region drone-collected images of the corresponding interactive adversarial sub-region acquired by a drone along two adjacent historical drone acquisition paths; Traversing the K-1 sub-region drone-collected image group sets to extract pixel width and pixel height, and obtaining the K-1 sub-region image resolution group sets according to the extraction results; Traversing the K-1 sub-region drone-collected image groups to identify the overlapping areas of the images within the groups, and comparing the K-1 overlapping area identification result sets with the area of any interactive adversarial sub-region in the K-1 interactive adversarial sub-region sets to obtain the K-1 sub-region image overlapping rate set; Calling a fuzziness identifier to perform fuzziness identification on the K-1 sub-region image resolution group sets and the K-1 sub-region image overlap rate sets respectively, to obtain K-1 initial fuzziness sets; It is determined whether the K-1 initial fuzzy degree sets are greater than or equal to the preset fuzzy degree. If so, the corresponding interactive confrontation sub-regions are used as fuzzy regions to obtain the K-1 fuzzy region sets.
3. The urban compensation modeling method combined with image recognition according to claim 2, characterized in that: include: Collecting multiple sample sub-region image resolution groups, multiple sample sub-region image overlap rates, and multiple sample initial blurs as a training sample data set; The training sample data set is used to iteratively train the network framework of the ambiguity identifier constructed based on the feedforward neural network, and the two-to-one mapping relationship between the sub-region image resolution group and the sub-region image overlap rate and the initial ambiguity is learned until the convergence requirement is met, thereby obtaining the trained ambiguity identifier.
4. The urban compensation modeling method combined with image recognition according to claim 1, characterized in that: Iteratively expanding the central analysis neighborhood according to the preset iteration step size to obtain a target analysis neighborhood includes: Iterating the leftmost analysis particle point of the central analysis neighborhood to the left and iterating the rightmost analysis particle point of the central analysis neighborhood to the right according to the preset iteration step size, respectively, to obtain a first left iterative analysis neighborhood and a first right iterative analysis neighborhood; Determine whether the concentration density of the first left iterative analysis neighborhood and the concentration density of the first right iterative analysis neighborhood are both greater than or equal to the concentration density of the center analysis neighborhood; if so, perform a union of the first left iterative analysis neighborhood and the first right iterative analysis neighborhood to obtain a first-stage iterative analysis neighborhood; Again iterating leftward on the leftmost analysis particle point of the iterative analysis neighborhood of the first stage and iterating rightward on the rightmost analysis particle point of the iterative analysis neighborhood of the first stage according to the preset iteration step size; After multiple bilateral iterations, when the concentration density of the left stage iterative analysis neighborhood and the concentration density of the right stage iterative analysis neighborhood obtained in the current iteration are both less than the concentration density of the stage iterative analysis neighborhood obtained in the previous iteration, the iteration is stopped and the stage iterative analysis neighborhood obtained in the previous iteration is used as the target analysis neighborhood.
5. The urban compensation modeling method combined with image recognition according to claim 4, characterized in that: include: When only the concentration density of the first left iterative analysis neighborhood or the concentration density of the first right iterative analysis neighborhood is greater than or equal to the concentration density of the central analysis neighborhood, the first left iterative analysis neighborhood or the first right iterative analysis neighborhood is used as the first-stage iterative analysis neighborhood; Iterate the leftmost analysis particle point of the iterative analysis neighborhood of the first stage to the left or iterate the rightmost analysis particle point of the iterative analysis neighborhood of the first stage to the right according to the preset iteration step size; After multiple unilateral iterations, when the concentration density of the left stage iterative analysis neighborhood or the concentration density of the right stage iterative analysis neighborhood obtained in the current iteration is less than the concentration density of the stage iterative analysis neighborhood obtained in the previous iteration, the iteration is stopped and the stage iterative analysis neighborhood obtained in the previous iteration is used as the target analysis neighborhood.
6. The urban compensation modeling method combined with image recognition according to claim 1, characterized in that: Obtain a preset sampling interval and a preset driving speed of the UAV, and divide the K-1 path interaction confrontation areas into K-1 interaction confrontation sub-area sets based on the K-1 area lengths of the K-1 path interaction confrontation areas, including: Multiplying the preset sampling interval by the preset driving speed to obtain a preset sampling length; Calculate the ratio of the length of the K-1 regions to the preset sampling length to obtain the number of K sub-region divisions; Based on the K sub-area division numbers and the preset sampling length, the K-1 path interaction confrontation areas are divided in sequence according to the path collection direction to obtain the K-1 interaction confrontation sub-area sets.
7. An urban compensation modeling system combined with image recognition, characterized in that: The system is used to implement the urban compensation modeling method combined with image recognition according to any one of claims 1 to 6, and the system includes: A collection path set acquisition module is used to obtain a historical three-dimensional city model of a target city and a set of historical drone collection paths, wherein K historical drone collection paths in the set are parallel to each other and the distance between any two adjacent historical drone collection paths is the same, and K is an integer greater than or equal to 1; A path interaction confrontation area acquisition module is used to determine the overlapping area of the drone field of view of two adjacent historical drone collection paths during image collection based on K environmental parameters of K historical drone collection paths and drone model parameters of the target city, and obtain K-1 path interaction confrontation areas; An interactive confrontation sub-area set acquisition module is used to obtain a preset sampling interval and a preset driving speed of the UAV, and divide the K-1 path interactive confrontation areas into K-1 area lengths to obtain a K-1 interactive confrontation sub-area set; a fuzzy region set determination module, configured to retrieve a texture map file of the historical city three-dimensional model using the K-1 interactive adversarial sub-region sets as indexes, and perform image resolution and overlap rate analysis on the retrieval results to determine K-1 fuzzy region sets, wherein the K-1 fuzzy region sets include K-1 fuzziness degree sets; a centralized ambiguity determination module for performing centralized ambiguity compensation analysis on the K-1 ambiguity sets to determine centralized ambiguities for compensation; an optimized UAV flight parameter set determination module, configured to collect a historical UAV flight parameter set for constructing the historical city three-dimensional model, optimize the historical UAV flight parameter set according to the centralized ambiguity to be compensated, and determine an optimized UAV flight parameter set; The target compensated urban three-dimensional model acquisition module is used to control the drone based on the drone flight parameter set, perform image acquisition of the K-1 fuzzy area set according to the historical drone acquisition path set, perform three-dimensional modeling based on the acquisition results, and compensate and update the historical urban three-dimensional model based on the modeling results to obtain the target compensated urban three-dimensional model.
Citation Information
Patent Citations
Unmanned aerial vehicle path determination method and apparatus for reconstructing three-dimensional model
CN106296816A
Digital control method and system for project supervision
CN118196351A