A real-time model construction device based on cruise photography

Abstract

The application provides a real-time model construction device based on cruise photography, and belongs to the technical field of building modeling. The real-time model construction method based on cruise photography comprises an acquisition module, which is used for acquiring image information collected by cruise photography of a binocular camera and obtaining key construction point image data; a conversion module, which is used for converting the key construction point image data obtained by the acquisition module into a key construction point BIM model; a metadata updating unit, which is used for updating the key construction point BIM model in real time; and a model deconstruction module, which is used for receiving the key construction point BIM model obtained by the conversion module for deconstruction processing, deconstructing the key construction point BIM model, and obtaining model information. The application can construct a building model in real time according to the construction progress when a UAV cruises, and realize real-time control of the construction progress.

Description

Technical Field

[0001] This invention belongs to the field of architectural modeling technology, and more specifically, relates to a real-time model building device based on cruise photography. Background Technology

[0002] For construction projects, drones are used to take photos and then create architectural models, which allows for a convenient and intuitive understanding of the construction progress.

[0003] Chinese invention patent (application number: CN202110455680.6) with publication number CN113240809A discloses a method for constructing a three-dimensional building model. The method includes: dividing the current area into a grid region composed of multiple grids, and determining the outer expansion region of the grid region; dividing the grid region into a first grid and a second grid according to whether it is adjacent to the outer expansion region; sequentially performing aerial photography along each grid to acquire image data, and after the flight of each first grid is completed, continuing to perform aerial photography along the outer expansion region adjacent to the current first grid; and processing the image data.

[0004] The aforementioned invention can only build a building model once when conducting aerial photography. For buildings under construction, it cannot update the building model in real time according to the construction progress. It can only re-construct the building model after a new flight, and cannot keep track of the construction progress in real time. Summary of the Invention

[0005] In view of this, the present invention provides a real-time model building device based on cruise photography, which can build building models in real time according to the construction progress while the UAV is cruising, so as to realize real-time monitoring of the construction progress.

[0006] This invention is implemented as follows:

[0007] This invention provides a real-time model building device based on cruise photography, comprising:

[0008] The acquisition module is used to acquire image information collected by the binocular camera's cruise photography and obtain image data of key construction points;

[0009] A conversion module is used to convert the key construction point image data obtained by the acquisition module into a key construction point BIM model.

[0010] Metadata update unit, which is used to update the BIM model of the key construction points in real time;

[0011] The model deconstruction module is used to receive the key construction point BIM model obtained by the conversion module, deconstruct it, and obtain model information by deconstructing the key construction point BIM model.

[0012] The model deconstruction module includes:

[0013] The receiving unit is used to receive the key construction point BIM model obtained by the conversion module, detect the key construction point BIM model, and determine whether the BIM model belongs to the true key construction point BIM model.

[0014] The judgment unit is used to scan the true critical construction point BIM model and determine whether the true critical construction point BIM model has erroneous objects.

[0015] An error statistics unit is used to count the total number of errors of the error objects and determine the degree of error of the true critical construction point BIM model based on the total number of errors.

[0016] An error repair unit is used to repair the error object based on the error severity obtained by the error statistics unit;

[0017] The deconstruction unit is used to decompose the repaired BIM model of the true critical construction points to obtain the model information; wherein, the model information includes geometric model, material information, three-dimensional coordinate parameters and rendering information;

[0018] The model reconstruction module is used to build an architectural space model based on the model information.

[0019] Based on the above technical solution, the real-time model building device based on cruise photography of the present invention can be further improved as follows:

[0020] The metadata update unit is used to perform the following steps:

[0021] S01: Establish a blockchain network and key construction point data nodes along the route of the binocular camera. The blockchain network includes multiple key construction point blockchain nodes, and each key construction point has one key construction point data node and one key construction point blockchain node.

[0022] S02: Network the blockchain nodes at each of the key construction points using the P2P protocol to ensure that the blockchain nodes at each of the key construction points can communicate with each other.

[0023] S03: Establish a communication connection between each of the key construction point data nodes and at least one of the key construction point blockchain nodes, and register the ID information and stored content of the key construction point data nodes to the corresponding key construction point blockchain node.

[0024] S04: After the image information collected by the binocular camera during patrol photography is saved to one of the key construction point data nodes, the key construction point data node will send the metadata information of the image information to the key construction point blockchain node linked to it for on-chain processing; the metadata information includes data ID, source data node ID, data size, data type and data update time;

[0025] S05: The metadata information in step S04 is encapsulated into a message and broadcast to the blockchain network. The message is eventually recorded in the acquisition module.

[0026] S06: The acquisition module links to one of the key construction point data nodes and informs the key construction point data node it links to the latest metadata corresponding to the data ID located on the blockchain network that it needs to acquire;

[0027] S07: The key construction point data node linked to the acquisition unit synchronously obtains the latest metadata corresponding to the data ID on the blockchain network from the key construction point blockchain node; updates the metadata to the acquisition unit, and communicates with the source key construction point data node ID of the metadata record, and synchronizes the content of the entire blockchain network through the P2P protocol. If the key construction point data node finds that the metadata has not changed and there is already corresponding data, it directly provides data services.

[0028] S08: When the metadata of the key construction point data node changes, a message about the metadata change will be sent through the linked key construction point blockchain node, including a message about the source key construction point data node ID changing.

[0029] S09: After receiving the message that the metadata has changed, the blockchain node at the key construction point will forward it to the key construction point data node linked to it. If the key construction point data node linked to it does not contain this data, the next step will be performed. If the key construction point data node linked to it contains this data, steps S06 and S07 will be executed.

[0030] S10: Ignore this retweet.

[0031] Furthermore, the key construction point data nodes are connected to one or more cloud networks, which are used to enable data communication and cloud storage between the key construction points.

[0032] The conversion module is used to perform the following steps:

[0033] Step 1: Periodically collect image data of the key construction points from the acquisition module;

[0034] Step 2: If the current time is not the start time of the current calculation cycle, input the key construction point image data into the trained first target image detection neural network model to filter and obtain valid target detection image data;

[0035] Step 3: If the current time is the start time of the current calculation cycle, input the key construction point image data into the trained second target image detection neural network model to filter and obtain valid target detection image data;

[0036] Step 4: Obtain the current status data of key construction points based on the effective target detection image data;

[0037] Step 5: Build a BIM model using the key construction point status data.

[0038] Furthermore, the training of the first target image detection neural network model includes the following steps:

[0039] Step 1: Establish a neural network model for detecting the first target image;

[0040] Step 2: Collect construction image training data and preprocess the construction image training data;

[0041] Step 3: Use the construction image training data to train the neural network model for detecting the first target image.

[0042] Furthermore, the specific operation method of step two is as follows:

[0043] The first target image detection neural network model is trained using the construction image training data as training input data and whether it is target detection image data as training output data.

[0044] Furthermore, when using the construction image training data as training input data, the construction image training data is occluded and superimposed with randomly selected preset occlusion block image data to generate a new first type of secondary generated construction image training data. The pixels of the two construction image training data are linearly superimposed to obtain a second type of secondary generated construction image training data. The set of the construction image training data, the first type of secondary generated construction image training data, and the second type of secondary generated construction image training data is used as the training set.

[0045] Furthermore, the target detection image data has multiple target tracking boxes, each containing a tracking target.

[0046] The training of the second target image detection neural network model includes the following steps:

[0047] Step 1: Establish a neural network model for detecting the second target image;

[0048] Step 2: Collect the construction image training data and the magnified construction image training data;

[0049] Step 3: Train the second target image detection neural network model using the construction image training data and the magnified image training data of the construction image.

[0050] Furthermore, the specific operation method of step two is as follows:

[0051] The second target image detection neural network model is trained using the construction image training data and the magnified image training data of the construction image as training input data, and whether it is a target detection image data as training output data.

[0052] Furthermore, the method for filtering to obtain effective target detection image data includes the following steps:

[0053] Step 1: Input the image data of the key construction point collected at the current moment and the magnified image data of the key construction point into the trained second target image detection neural network model to obtain the corresponding two target detection image data;

[0054] Step 2: If the number of target tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image is greater than the number of tracking targets in the target detection image data corresponding to the key construction point image data, then proceed to Step 3; if the number of target tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image is less than the number of tracking targets in the target detection image data corresponding to the key construction point image data, then proceed to Step 4.

[0055] Step 3: Define the magnified image data of the key construction point images within the current calculation cycle as valid target detection image input data, and define the target detection image corresponding to the valid target detection image input data as valid target detection image data;

[0056] Step 4: Define the key construction point image data within the current calculation cycle as the valid target detection image input data, and define the target detection image data corresponding to the valid target detection image input data as the valid target detection image data.

[0057] Compared with existing technologies, the beneficial effects of the real-time model construction method based on cruise photography provided by this invention are as follows: By establishing a blockchain network and key construction point data nodes, key construction point image data is uploaded in real time during drone cruise, synchronizing the content of the entire blockchain network, converting the key construction point image data into a key construction point BIM model, and establishing a building space model. This allows for real-time and rapid monitoring of construction progress, high efficiency in building space model component updates, and fast updates. Each key construction point data node exchanges data based on the information of the key construction point blockchain node, separating distributed computing from storage and transmission, thus improving the utilization rate of computing and storage. Both the key construction point blockchain node and the key construction point data node adopt a decentralized peer-to-peer network, avoiding the probability of single-point failure caused by a single node and ensuring availability. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 The present invention provides a flowchart of a real-time model construction method based on cruise photography. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0062] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0063] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0064] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0065] like Figure 1 The image shown is a first embodiment of a real-time model construction method based on cruise photography provided by the present invention. In this embodiment, it includes:

[0066] The acquisition module is used to acquire image information collected by the binocular camera's cruise photography and obtain image data of key construction points;

[0067] The conversion module is used to convert the key construction point image data obtained by the acquisition module into a key construction point BIM model.

[0068] Metadata update unit, which is used to update the BIM model of key construction points in real time;

[0069] The model deconstruction module is used to receive the key construction point BIM model obtained from the conversion module, deconstruct it, and obtain model information.

[0070] The model deconstruction module includes:

[0071] The receiving unit is used to receive the key construction point BIM model obtained by the conversion module, detect the key construction point BIM model, and determine whether the BIM model belongs to the true key construction point BIM model.

[0072] The judgment unit is used to scan the BIM model of the true critical construction points and determine whether the BIM model of the true critical construction points has erroneous objects.

[0073] The error statistics unit is used to count the total number of errors for erroneous objects and to determine the degree of error in the BIM model of critical construction points based on the total number of errors.

[0074] The error repair unit is used to repair the erroneous object based on the error severity obtained from the error statistics unit.

[0075] The deconstruction unit is used to decompose the repaired true critical construction point BIM model to obtain model information; the model information includes geometric model, material information, three-dimensional coordinate parameters and rendering information.

[0076] The model reconstruction module is used to create an architectural space model based on the model information.

[0077] Rendering information includes color gamut, color resolution, and rendering resolution.

[0078] The binocular camera is mounted on the drone.

[0079] In the above technical solution, the metadata update unit is used to perform the following steps:

[0080] S01: Establish a blockchain network and key construction point data nodes along the route of the binocular camera cruise. The blockchain network includes multiple key construction point blockchain nodes, and each key construction point has a key construction point data node and a key construction point blockchain node.

[0081] S02: Network the blockchain nodes at each key construction site using the P2P protocol to ensure that the blockchain nodes at each key construction site can communicate with each other.

[0082] S03: Establish a communication connection between each key construction point data node and at least one of the key construction point blockchain nodes, and register the ID information and stored content of the key construction point data node to the corresponding key construction point blockchain node.

[0083] S04: When image information collected by a binocular camera during patrol photography is saved to one of the key construction point data nodes, the key construction point data node will send the metadata information of the image information to the key construction point blockchain node linked to it for on-chain processing; the metadata information includes data ID, source data node ID, data size, data type and data update time;

[0084] S05: The metadata information in step S04 is encapsulated into a message and broadcast to the blockchain network. The message is eventually recorded in the acquisition module.

[0085] S06: Obtain the module's connection to one of the key construction point data nodes, and inform the key construction point data node connected to it of the latest metadata corresponding to the data ID located on the blockchain network that needs to be obtained;

[0086] S07: The key construction point data node linked to the acquisition unit synchronously obtains the latest metadata corresponding to the data ID on the blockchain network from the key construction point blockchain node; updates the metadata to the acquisition unit, and communicates with the source key construction point data node ID of the metadata record, and synchronizes the content of the entire blockchain network through the P2P protocol. If the key construction point data node finds that the metadata has not changed and there is already corresponding data, it directly provides data services.

[0087] S08: When the metadata of a key construction point data node changes, a message about the metadata change will be sent through the linked key construction point blockchain nodes, including a message about the change of the source key construction point data node ID;

[0088] S09: After receiving the message that the metadata has changed, the blockchain node at the key construction point will forward it to the key construction point data node connected to it. If the key construction point data node connected to it does not contain this data, proceed to the next step. If the key construction point data node connected to it contains this data, proceed to steps S06 and S07.

[0089] S10: Ignore this retweet.

[0090] In step S07, when the metadata of the key construction point data node changes, a new file is written to the corresponding key construction point data node.

[0091] Furthermore, in the above technical solution, there are one or more cloud network connections between the data nodes of the key construction points. The cloud network connections are used to realize data communication and cloud storage between the key construction points.

[0092] The key construction point data nodes that establish cloud network links can modify other key construction point data nodes in the same cloud network. After obtaining modification authorization, during the file writing or editing update process, a disk path cache needs to be created to cache the modified data. At this time, there is no need to process the metadata on the blockchain. After the operation is completed, the cached data file is stored in the storage layer, the cache is deleted, and then the metadata information of the newly stored file in the storage layer is obtained and updated to the key construction point blockchain node.

[0093] In the above technical solution, the conversion module is used to perform the following steps:

[0094] Step 1: Periodically collect image data of key construction points from the acquisition module;

[0095] Step 2: If the current time is not the start time of the current calculation cycle, input the key construction point image data into the trained first target image detection neural network model to filter and obtain valid target detection image data;

[0096] Step 3: If the current time is the start time of the current calculation cycle, input the key construction point image data into the trained second target image detection neural network model, and filter to obtain valid target detection image data;

[0097] Step 4: Obtain the current status data of key construction points based on the effective target detection image data;

[0098] Step 5: Build a BIM model using the status data of key construction points.

[0099] The first target image detection neural network model and the second target image detection neural network model are deep learning models based on the YOLOv3 neural network and the Deepsort algorithm. When the deep learning model is running, it first upsamples the feature map output by the YOLOv3 framework by 2 times, and then concatenates the upsampled feature map with the downsampled feature map output by the YOLOv3 framework by 4 times to output a fusion information layer downsampled by 4 times. Then, it upsamples the fusion information layer downsampled by 2 times and concatenates its output with the downsampled feature map output by the YOLOv3 framework by 2 times.

[0100] Furthermore, in the above technical solution, the training of the first target image detection neural network model includes the following steps:

[0101] Step 1: Establish a neural network model for detecting the first target image;

[0102] Step 2: Collect construction image training data and preprocess the construction image training data;

[0103] Step 3: Use the construction image training data to train the neural network model for detecting the first target image.

[0104] Preprocessing of construction image training data includes converting construction image information to the same size and labeling the construction image information.

[0105] Furthermore, in the above technical solution, the specific operation method of step two is as follows:

[0106] The first target image detection neural network model is trained using construction image training data as training input data and whether the image is a target detection image as training output data.

[0107] Furthermore, in the above technical solution, when using construction image training data as training input data, the construction graphic training data is occluded and superimposed with randomly selected preset occlusion block image data to generate a new first type of secondary generated construction image training data. The pixels of the two construction graphic training data are linearly superimposed to obtain a second type of secondary generated construction graphic training data. The set of construction image training data, the first type of secondary generated construction image training data, and the second type of secondary generated construction graphic training data is used as the training set.

[0108] The steps for obtaining the second type of secondary generation construction graphic training data by linearly superimposing points include:

[0109] By permutation and combination, two different construction images are used as a group to obtain multiple combinations of construction image training data;

[0110] For each combination of construction image training data, a new second type of secondary generated construction image training data is obtained. The pixel value of each pixel in the second type of secondary generated construction image training data is calculated according to the following formula:

[0111] ;

[0112] Where n(i) is the pixel value of the i-th pixel in the training data of the second type of secondary generated construction graphics. Let be the pixel value of the i-th pixel in the first construction image training data set. α is the pixel value of the i-th pixel in the training data of the second construction image in the UAV image composite; α is the scaling factor.

[0113] Furthermore, in the above technical solution, the target detection image data has multiple target tracking boxes, and each target tracking box contains a tracking target.

[0114] The training of the second target image detection neural network model includes the following steps:

[0115] Step 1: Establish a neural network model for detecting the second target image;

[0116] Step 2: Collect construction image training data and magnified construction image training data;

[0117] Step 3: Train the neural network model for detecting the second target image using training data of construction images and training data of magnified construction images.

[0118] Furthermore, in the above technical solution, the specific operation method of step two is as follows:

[0119] The neural network model for detecting second target images is trained using construction image training data and magnified construction image training data as training input data, and whether the image is a target detection image as training output data.

[0120] Furthermore, in the above technical solution, the method for selecting effective target detection image data includes the following steps:

[0121] Step 1: Input the key construction point image data collected at the current moment and the magnified image data of the key construction point image into the trained second target image detection neural network model to obtain the corresponding two target detection image data;

[0122] Step 2: If the number of target tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image is greater than the number of tracking targets in the target detection image data corresponding to the key construction point image, proceed to Step 3; if the number of target tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image is less than the number of tracking targets in the target detection image data corresponding to the key construction point image, proceed to Step 4.

[0123] Step 3: Define the magnified image data of the key construction point images within the current calculation cycle as the valid target detection image input data, and define the target detection image corresponding to the valid target detection image input data as the valid target detection image data;

[0124] Step 4: Define the key construction point image data within the current calculation cycle as the valid target detection image input data, and define the target detection image data corresponding to the valid target detection image input data as the valid target detection image data.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time model building device based on cruise photography, characterized in that, include: The acquisition module is used to acquire image information collected by the binocular camera's cruise photography and obtain image data of key construction points; A conversion module is used to convert the key construction point image data obtained by the acquisition module into a key construction point BIM model. Metadata update unit, which is used to update the BIM model of the key construction points in real time; The model deconstruction module is used to receive the key construction point BIM model obtained by the conversion module, deconstruct it, and obtain model information by deconstructing the key construction point BIM model. The model deconstruction module includes: The receiving unit is used to receive the key construction point BIM model obtained by the conversion module, detect the key construction point BIM model, and determine whether the BIM model belongs to the true key construction point BIM model. The judgment unit is used to scan the true critical construction point BIM model and determine whether the true critical construction point BIM model has erroneous objects. An error statistics unit is used to count the total number of errors of the error objects and determine the degree of error of the true critical construction point BIM model based on the total number of errors. An error repair unit is used to repair the error object based on the error severity obtained by the error statistics unit; The deconstruction unit is used to decompose the repaired true critical construction point BIM model to obtain model information; wherein, the model information includes geometric model, material information, three-dimensional coordinate parameters and rendering information; The model reconstruction module is used to build an architectural space model based on the model information. The conversion module is used to perform the following steps: Step 1: Periodically collect image data of the key construction points from the acquisition module; Step 2: If the current time is not the start time of the current calculation cycle, input the key construction point image data into the trained first target image detection neural network model to filter and obtain valid target detection image data; Step 3: If the current time is the start time of the current calculation cycle, input the key construction point image data into the trained second target image detection neural network model, and filter to obtain valid target detection image data; Step 4: Obtain the current status data of key construction points based on the effective target detection image data; Step 5: Build a BIM model using the key construction point status data.

2. The real-time model building device based on cruise photography according to claim 1, characterized in that, The metadata update unit is used to perform the following steps: S01: Establish a blockchain network and key construction point data nodes along the route of the binocular camera. The blockchain network includes multiple key construction point blockchain nodes, and each key construction point has one key construction point data node and one key construction point blockchain node. S02: Network the blockchain nodes at each key construction point using the P2P protocol to ensure that the blockchain nodes at each key construction point can communicate with each other. S03: Establish a communication connection between each of the key construction point data nodes and at least one of the key construction point blockchain nodes, and register the ID information and stored content of the key construction point data nodes to the corresponding key construction point blockchain node. S04: After the image information collected by the binocular camera during patrol photography is saved to one of the key construction point data nodes, the key construction point data node will send the metadata information of the image information to the key construction point blockchain node linked to it for on-chain processing; the metadata information includes data ID, source data node ID, data size, data type and data update time; S05: The metadata information in step S04 is encapsulated into a message and broadcast to the blockchain network. The message is eventually recorded in the acquisition module. S06: The acquisition module connects to one of the key construction point data nodes and informs the key construction point data node connected to it of the latest metadata information corresponding to the data ID located on the blockchain network that it needs to acquire; S07: The key construction point data node linked to the acquisition module synchronously acquires the latest metadata information corresponding to the data ID located on the blockchain network from the key construction point blockchain node; The metadata information is updated to the acquisition module, and at the same time, it communicates with the source of the metadata information record, the key construction point data node ID, and synchronizes the content of the entire blockchain network through the P2P protocol. If the key construction point data node finds that the metadata information has not changed and there is already corresponding data, it directly provides data services. S08: When the metadata information of the key construction point data node changes, a message about the metadata information change will be sent through the linked key construction point blockchain node, including a message about the source key construction point data node ID changing; S09: After receiving the message that the metadata information has changed, the blockchain node at the key construction point will forward it to the key construction point data node linked to it. If the key construction point data node linked to it does not contain this data, the next step will be performed. If the key construction point data node linked to it contains this data, steps S06 and S07 will be executed. S10: Ignore this retweet.

3. The real-time model building device based on cruise photography according to claim 2, characterized in that, The key construction site data nodes are connected to one or more cloud networks, which are used to enable data communication and cloud storage between the key construction sites.

4. The real-time model building device based on cruise photography according to claim 3, characterized in that, The training of the first target image detection neural network model includes the following steps: Step 1: Establish a neural network model for detecting the first target image; Step 2: Collect construction image training data and preprocess the construction image training data; Step 3: Use the construction image training data to train the neural network model for detecting the first target image.

5. The real-time model building device based on cruise photography according to claim 4, characterized in that, The specific operation method for step two is as follows: The first target image detection neural network model is trained using the construction image training data as training input data and whether it is target detection image data as training output data.

6. The real-time model building device based on cruise photography according to claim 5, characterized in that, When the construction image training data is used as training input data, the construction image training data is occluded and superimposed using randomly selected preset occlusion block image data to generate a new first type of secondary generated construction image training data. The pixels of the two construction image training data are linearly superimposed to obtain a second type of secondary generated construction image training data. The set of the construction image training data, the first type of secondary generated construction image training data and the second type of secondary generated construction image training data is used as the training set.

7. The real-time model building device based on cruise photography according to claim 1, characterized in that, The target detection image data has multiple target tracking boxes, each containing a target to be tracked. The training of the second target image detection neural network model includes the following steps: Step 1: Establish a neural network model for detecting the second target image; Step 2: Collect construction image training data and magnified construction image training data; Step 3: Train the second target image detection neural network model using the construction image training data and the construction image magnified image training data.

8. The real-time model building device based on cruise photography according to claim 7, characterized in that, The specific operation method for step two is as follows: The second target image detection neural network model is trained using the construction image training data and the construction image magnified image training data as training input data, and whether it is a target detection image data as training output data.

9. The real-time model building device based on cruise photography according to claim 8, characterized in that, The method for obtaining effective target detection image data includes the following steps: Step 1: Input the key construction point image data collected at the current moment and the magnified image data of the key construction point image data into the trained second target image detection neural network model to obtain the corresponding two target detection image data; Step 2: If the number of tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image data is greater than the number of tracking targets in the target detection image data corresponding to the key construction point image data, then proceed to Step 3; if the number of tracking targets in the target detection image data corresponding to the magnified image data of the key construction point image is less than the number of tracking targets in the target detection image data corresponding to the key construction point image data, then proceed to Step 4. Step 3: Define the magnified image data of the key construction point image data within the current calculation cycle as the valid target detection image input data, and define the target detection image corresponding to the valid target detection image input data as the valid target detection image data; Step 4: Define the key construction point image data within the current calculation cycle as the valid target detection image input data, and define the target detection image data corresponding to the valid target detection image input data as the valid target detection image data.

Images