Real scene three-dimensional surveying and mapping system based on big data

Abstract

The invention discloses a live-action three-dimensional surveying and mapping system based on big data, and relates to the technical field of three-dimensional surveying and mapping, and the system comprises a collection module which is used for a camera collection device to collect video data of a range region, and the video data comprises the northern view angle data of the range region, the above view angle data of the range region, and the west view angle data of the range region; through collaborative acquisition of multi-view video data of the north side, the upper side and the west side, complete space information of real objects in a scene can be captured, the problem of single-view shielding is solved, through processing of the big data identification model, rapid identification of various types of real objects in the scene and automatic extraction of three-dimensional feature parameters are realized, the identification precision is improved, and the real object identification efficiency is improved. Compared with the prior art, manual intervention is reduced, a dynamic three-dimensional model splicing and calibration method is adopted, the position, posture and size of the object model are adjusted in real time according to the actual recognition result, it is ensured that the generated three-dimensional map is accurate in structure and continuous in boundary, and the technical level of automatic three-dimensional mapping is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of three-dimensional surveying and mapping technology, in particular to a real scene three-dimensional surveying and mapping system based on big data. BACKGROUND

[0002] Surveying and mapping technology is a key science and engineering means for obtaining, processing and expressing spatial information of the earth's surface. Its core purpose is to provide high-precision spatial data support for land resource management, urban planning, engineering construction, environmental monitoring and many other fields. Traditional surveying and mapping technology mainly relies on manual ground measurement, total station, GPS positioning and other methods. Although it can obtain relatively accurate two-dimensional or a small amount of three-dimensional spatial data, the operation cycle is long and the coverage is limited. Real scene surveying and mapping technology refers to the use of multi-source sensors (such as cameras) to collect data on real world scenes, combined with computer vision, three-dimensional modeling and big data analysis, to realize the three-dimensional digital reconstruction and expression of scene spatial structure, feature attributes and spatial relationships. Compared with traditional surveying and mapping methods, real scene surveying and mapping technology can obtain more comprehensive, intuitive and high-precision spatial information, and is widely used in smart city construction, land space planning, digital twin, traffic management, cultural heritage protection and other fields. Existing surveying and mapping systems rely on a single perspective or fixed process to model collected data, lack cross comparison mechanism of multi-perspective physical object arrangement, and exist abnormal areas with incorrect category and arrangement order. The discovery has obvious lag and needs manual review for positioning and correction. In view of the above technical defects, a solution is proposed. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a real scene three-dimensional surveying and mapping system based on big data.

[0004] To achieve the above purpose, the following technical scheme is used: a real scene three-dimensional surveying and mapping system based on big data, comprising: A collection module is used to collect video data in the range area by a camera collection device. The video data includes range area north side view data, range area overhead view data and range area west side view data. An identification module uses a big data identification model to identify the scene physical object of the range area overhead view data. The big data identification model identifies the range area overhead view data and marks the type of the scene physical object. A three-dimensional object model is selected and combined into a range area map. A height identification module uses a big data identification model to identify the height value of the scene physical object of the range area north side view data. The three-dimensional object model in the range area map is modified according to the range area map. The verification module outputs a left-to-right object model arrangement sequence according to the range area map, uses a big data recognition model to identify a real object arrangement verification sequence of a scene real object in the left view angle data of the range area, verifies the object model arrangement sequence according to the real object arrangement verification sequence, and obtains a region to be reviewed; The region to be reviewed is subjected to secondary collection and secondary recognition.

[0005] The identification process of the big data recognition model on the scene real object in the view angle data above the range area comprises: S21, input the collected view angle data above the range area into the big data recognition model, the model adopts a deep convolutional neural network structure, performs image normalization, edge enhancement and convolution feature extraction on the input image, and obtains an image feature map; S22, generate and pixel-level divide the image by using a semantic segmentation algorithm on the image feature map, output a plurality of candidate boxes, and the candidate boxes correspond to potential scene object regions in the image; S23, apply a multi-type classification network to the candidate box region respectively, identify and output corresponding scene real object type labels, and the scene real object includes buildings, roads, plants and rivers.

[0006] The step of combining the scene real objects into a range area map comprises: S31, according to the scene real object type label, call the corresponding standard three-dimensional object model template in the three-dimensional model library, the three-dimensional model library is a collection of model templates; S32, extract the geographic plane coordinates (X, Y) from the view angle data above the range area; S33, according to the pixel size of the scene real object in the view angle data above the range area, scale the selected three-dimensional object model in proportion; and perform pose rotation on the model by referring to the direction in the view angle data above the range area; S34, superimpose the three-dimensional object model to the map model according to the geographic plane coordinates (X, Y), and process the boundaries between the three-dimensional object models by using an automatic splicing algorithm; S35, perform texture mapping on the map model to obtain a range area three-dimensional map.

[0007] Extracting the geographic plane coordinates (X, Y) from the view angle data above the range area comprises: extracting the two-dimensional projection center point coordinates (u, v) and the boundary box size information of the scene real object from the view angle data above the range area, and mapping the two-dimensional coordinates to the geographic plane coordinates (X, Y); Wherein, u0, v0 is the image origin, (X0, Y0) is the geographical coordinate origin, Sx, Sy is the resolution of pixel to plane distance, u, v is the two-dimensional projection center point coordinates.

[0008] The height value of the scene real object in the range area north side view angle data is identified by using the big data identification model. S41, image preprocessing is performed on the range area north side view angle data, and the image preprocessing includes distortion correction, edge enhancement and light balance processing. S42, input the range area north side view angle data into the big data identification model, and use the semantic segmentation and object detection algorithm to extract the target bounding box; S43, the height value of the target bounding box is calculated by applying the monocular image depth regression model set to the target bounding box. S44, match the target bounding box with the scene real object; S45, adjust the height of the corresponding three-dimensional object model according to the height value of the target bounding box.

[0009] The monocular image depth regression model set calculation includes: projecting the upper center pixel and the lower center pixel of the target bounding box into the camera coordinate system according to the camera intrinsic matrix, obtaining the vertical coordinates of the two points by calculation, and taking the absolute value of the difference between the vertical coordinates of the two points as the height value of the target bounding box.

[0010] According to the range area map, the object model arrangement sequence from left to right is output, including: S501, extract the two-dimensional projection center point coordinates of each three-dimensional object model in the range area three-dimensional map; S502, sort the center point coordinates according to the horizontal coordinate value from small to large to form the object model arrangement sequence L1; S503, input the range area left side view image into the big data identification model, and the big data identification model performs target detection on the image to output the real object arrangement verification sequence L2 under the left side view.

[0011] The real object arrangement verification sequence of the scene real object in the range area left side view angle data is identified, including: S511, corresponding comparison is performed on sequence L1 and sequence L2 to judge the position order consistency; S512, when the comparison exists type inconsistency or order inconsistency, mark the sequence L1 as an abnormal sequence; S513, define the abnormal sequence corresponding space area as a to-be-reviewed area.

[0012] The secondary collection and secondary identification of the region to be reviewed include: generating a secondary collection task according to the region to be reviewed, instructing a camera collection device to collect image data of the region to be reviewed, using a big data identification model to identify scene real objects of the range region overhead view data, selecting a three-dimensional object model, and updating the three-dimensional object model to the range region map; outputting a left-to-right object model arrangement sequence from the range region map, using a big data identification model to identify a real object arrangement verification sequence of scene real objects of the range region left side view data, and verifying the object model arrangement sequence according to the real object arrangement verification sequence.

[0013] The application provides a real scene three-dimensional surveying and mapping system based on big data. The application can capture complete spatial information of real objects in a scene through the cooperative collection of north side, overhead and west side multi-view video data, overcome the problem of single view occlusion, realize the rapid identification of various types of real objects in the scene and the automatic extraction of three-dimensional feature parameters through the processing of a big data identification model, improve the identification accuracy, reduce manual intervention, adopt a dynamic three-dimensional model splicing and calibration method, adjust the position, posture and size of the object model in real time according to the actual identification result, ensure that the generated three-dimensional map is accurate in structure and continuous in boundary, and improve the technical level of automatic three-dimensional mapping. The application intelligently compares the initially generated object model arrangement sequence with the real object arrangement verification sequence collected from different views, detects category inconsistency and sequence abnormality, finds potential abnormal areas and locates them, issues a secondary collection instruction, performs more detailed multi-view supplementary collection on the abnormal areas, reanalyzes and corrects the abnormal areas by combining a secondary identification algorithm, improves the timeliness of problem discovery and processing, effectively reduces false positives and missed detections, and significantly improves the overall reliability and application accuracy of three-dimensional surveying and mapping results. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 It is a structural framework schematic diagram of the system of the application. Figure 2 It is a flowchart schematic diagram of the system of the application. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0016] Embodiment 1: Please refer toFigure 1 The application provides a real three-dimensional surveying and mapping system based on big data, comprising: A collection module is configured to collect video data in a range area by a camera collection device, wherein the video data comprises north side view data, overhead view data and west side view data of the range area. An identification module is configured to identify a scene real object in the overhead view data of the range area by using a big data identification model, mark the type of the scene real object by identifying the overhead view data of the range area by the big data identification model, select a three-dimensional object model, and combine the three-dimensional object model into a range area map. A height identification module is configured to identify the height value of the scene real object in the north side view data of the range area by using the big data identification model, and modify the three-dimensional object model in the range area map according to the range area map. A verification module is configured to output an object model arrangement sequence from left to right according to the range area map, identify a real object arrangement verification sequence of the scene real object in the left side view data of the range area by using the big data identification model, verify the object model arrangement sequence according to the real object arrangement verification sequence, and obtain a region to be reviewed. The region to be reviewed is subjected to secondary collection and secondary identification.

[0017] The identification process of the big data identification model on the scene real object in the overhead view data of the range area comprises: S21. The overhead view data of the range area collected is input into the big data identification model, the model adopts a deep convolutional neural network structure, performs image normalization, edge enhancement and convolution feature extraction on the input image, and obtains an image feature map. S22. A semantic segmentation algorithm is used to generate a candidate region and perform pixel-level division on the image based on the image feature map, and a plurality of candidate boxes are output, wherein the candidate boxes correspond to potential scene object regions in the image. S23. A multi-type classification network is applied to the candidate box region, and a corresponding scene real object type label is identified and output, wherein the scene real object comprises a building, a road, a plant and a river.

[0018] It is worth mentioning that the big data recognition model adopts ResNet50 as the basic network structure. Specifically, the ResNet50 model is pre-trained on large-scale image datasets such as ImageNet to obtain good parameter initialization. The system integrates mainstream deep learning frameworks (including but not limited to PyTorch, TensorFlow, Keras), and can directly call the official ResNet50 pre-training weight. In the application process, users do not need to train the model from scratch, but only need to load the trained ResNet50 model parameters to efficiently extract features or classify categories for newly collected scene pictures, greatly reducing the training cost and required time, and improving the recognition efficiency and practicality of the overall system.

[0019] The step of combining the scene real objects into a range area map includes: S31, according to the scene real object type label, calling the corresponding standard three-dimensional object model template in the three-dimensional model library, which is a collection of model templates; S32, extracting geographic plane coordinates (X, Y) from the range area overhead view data; S33, scaling the selected three-dimensional object model according to the pixel size of the scene real object in the range area overhead view data; and performing pose rotation on the model by referring to the direction in the range area overhead view data; S34, superimposing the three-dimensional object model into the map model according to the geographic plane coordinates (X, Y), and processing the boundaries between the three-dimensional object models through an automatic splicing algorithm; S35, performing texture mapping on the map model to obtain a range area three-dimensional map.

[0020] The output type label (such as "building", "road", "vegetation", "river") automatically retrieves and calls the standard three-dimensional object model template corresponding to the type label in the three-dimensional model library. The three-dimensional model library usually contains rich parametric model data, and each template not only contains basic geometric structure, but also contains metadata suitable for various scaling and rotation transformations.

[0021] The system uses image measurement, sensor positioning or spatial reference points and other means to automatically extract the two-dimensional projection center point coordinates (u, v) of each scene real object from the overhead view image, combines camera parameters (such as intrinsic matrix, distortion coefficient) and known geographic reference, and accurately converts the two-dimensional pixel coordinates into actual geographic plane coordinates (X, Y) through a mapping function (such as affine transformation or homography matrix).

[0022] After the structure splicing of the map model is completed, the system performs texture mapping processing on the surface of each three-dimensional object model. Specifically, the original image or high-resolution texture photo collected is projected onto the model surface, so that the appearance details conform to the real scene. This process usually uses UV coordinate mapping or multi-view splicing technology to achieve detail restoration and visual realism improvement.

[0023] extracting geographical plane coordinates (X, Y) from the range area overhead perspective data, including: extracting two-dimensional projection center point coordinates (u, v) and boundary box size information of the scene real object from the range area overhead perspective data, and mapping the two-dimensional coordinates into geographical plane coordinates (X, Y); wherein u0, v0 are image origins, (X0, Y0) is a geographical coordinate origin, Sx, Sy are resolutions of pixel-to-plane distances, and u, v are two-dimensional projection center point coordinates.

[0024] The use of a big data recognition model to identify the height value of the scene real object in the range area north side perspective data includes: S41, performing image preprocessing on the range area north side perspective data, the image preprocessing including distortion correction, edge enhancement, and light balance processing; S42, inputting the range area north side perspective data into a big data recognition model, and extracting a target bounding box using a semantic segmentation and object detection algorithm; S43, applying a monocular image depth regression model set to the target bounding box to calculate the height value of the target bounding box; S44, matching the target bounding box with the scene real object; S45, adjusting the height of the corresponding three-dimensional object model according to the height value of the target bounding box.

[0025] It is worth noting that the preprocessed image is input into the big data recognition model. Inside the model, a semantic segmentation algorithm (such as DeepLabV3+ etc.) is used to divide the image at the pixel level, and the scene real objects in different regions are preliminarily identified; then a target detection algorithm (such as YOLOv5, Faster R-CNN, etc.) is used to accurately extract the bounding box (Bounding Box) of each target object, and record the pixel coordinates and size of each bounding box.

[0026] The system automatically matches the spatial position, type label, and other information of each bounding box with the original scene real object list, ensuring that the height information of each object is accurately associated with the corresponding three-dimensional object model. The matching strategy uses a spatial overlap degree determination method.

[0027] The monocular image depth regression model set is applied to calculation, including: projecting the upper edge center pixel and the lower edge center pixel on the target bounding box into the camera coordinate system according to the camera intrinsic matrix, obtaining the vertical coordinates of the two points by calculation, and taking the absolute value of the difference between the vertical coordinates of the two points as the height value of the target bounding box.

[0028] It is worth noting that the system obtains the intrinsic matrix of the camera used for shooting images, which is used to realize the transformation between pixel coordinates and three-dimensional coordinates of the camera. For each target bounding box to be measured, the system automatically determines the positions of the upper edge center pixel and the lower edge center pixel, and then projects the two pixels into the camera coordinate system by using the camera intrinsic matrix. Through the depth regression model, the system can obtain the vertical coordinate value of the projection point in the camera coordinate system, and further calculate the absolute value of the difference between the vertical coordinates of the upper edge center point and the lower edge center point, which is used as the actual height estimation of the real object corresponding to the target bounding box. In order to ensure the accuracy of the height estimation, the system can also combine the standard objects with known heights in the scene for scale correction and error correction.

[0029] Beneficial effects: The present application can capture the complete spatial information of the real objects in the scene through the cooperative collection of north side, upper side and west side multi-view video data, overcome the problem of single view occlusion, realize the rapid identification of various types of real objects in the scene and the automatic extraction of three-dimensional feature parameters through the processing of the big data recognition model, improve the identification accuracy, reduce the manual intervention, adopt the dynamic three-dimensional model splicing and calibration method, adjust the position, posture and size of the object model in real time according to the actual identification result, ensure that the generated three-dimensional map structure is accurate and the boundary is continuous, and improve the technical level of automatic three-dimensional mapping.

[0030] Embodiment 2: The range area map outputs a left-to-right object model arrangement sequence, including: S501, extracting the two-dimensional projection center point coordinates of each three-dimensional object model in the range area three-dimensional map; S502, sorting the center point coordinates in the horizontal coordinate value from small to large to form an object model arrangement sequence L1; S503, inputting the left side view image of the range area into a big data recognition model, and the big data recognition model performs target detection on the image to output a real object arrangement verification sequence L2 under the left side view.

[0031] The real object arrangement verification sequence of the scene real object identified from the left side view data of the range area includes: S511, performing corresponding comparison on the sequence L1 and the sequence L2 to judge the position order consistency; S512, when the comparison exists type inconsistency or order inconsistency, marking the sequence L1 as an abnormal sequence; S513, define the abnormal sequence corresponding to the spatial region as the area to be reviewed.

[0032] It is worth noting that the geometric center or the centroid of the bottom surface of each three-dimensional object model in the range area three-dimensional map is calculated, and it is accurately projected to the two-dimensional image plane using the camera extrinsic parameters and the projection matrix to obtain the pixel-level center point (u, v). For high-precision areas, the complete geometric center projection is used; for long-range or low-precision areas, the bounding box centroid is simplified, so as to balance the running efficiency and positioning accuracy.

[0033] All center points are sorted in ascending order according to the horizontal pixel coordinate u to form a left-to-right model sequence L1. A pixel tolerance δ_x (such as ±3px) is introduced during the sorting process to filter out minor measurement errors; if the u difference between two models is less than the tolerance, it is considered as the same horizontal position, and a secondary sorting is performed according to the depth or category priority to ensure the order is stable and reliable.

[0034] The left side view original image is sent to the big data recognition model, first the target bounding box is detected by YOLOv8 and the category is assigned, then the accurate center point is obtained by SAM fine-tuning mask. The system discards the targets with detection confidence less than 0.25, and performs non-maximum suppression on the overlapping boxes (IoU>0.6), only keeping the one with the highest confidence; then the center points are sorted according to the horizontal coordinates to get the left side view verification sequence L2.

[0035] When the category is inconsistent or the arrangement is crossed, the system will mark the corresponding L1 segment as an abnormal sequence, and divide it into type abnormal, order abnormal or composite abnormal according to the abnormality; if m (≥2) consecutive models are abnormal, they are automatically merged into one segment to prevent fragmentation review and improve processing efficiency. According to the three-dimensional bounding box of the first and last models of the abnormal sequence, a buffer (such as 1.2 times) is generated according to the proportion to generate the area to be reviewed, and it is recorded to the task queue.

[0036] The two-level collection and secondary identification of the area to be reviewed include: generating a two-level collection task according to the area to be reviewed, instructing the camera collection device to collect image data of the area to be reviewed, using a big data recognition model to identify the scene real object of the range area overhead view data, selecting a three-dimensional object model, and updating the three-dimensional object model to the range area map; outputting a left-to-right object model arrangement sequence according to the range area map, using a big data recognition model to identify the real object arrangement verification sequence of the scene real object of the left side view data of the range area, and verifying the object model arrangement sequence according to the real object arrangement verification sequence.

[0037] When the system determines the area to be reviewed, first generate a secondary acquisition task, indicate the polygon range of the area, the priority acquisition angle (east or south), the target resolution and overlap rate requirements, select the optimal unmanned aerial vehicle or vehicle gimbal and flight path; The acquisition platform completes multi-angle image acquisition according to the planned path, and qualified images are then transmitted to the identification module for re-identification.

[0038] Beneficial effects: The application arranges the primary generated object model sequence and the real object arrangement verification sequence collected from different angles for intelligent comparison, detects category inconsistency and sequence abnormalities, locates potential abnormal areas, issues secondary acquisition instructions, and performs more detailed multi-angle supplementary acquisition on the abnormal areas, reanalyzes and corrects the abnormal areas by combining secondary identification algorithms, improves the timeliness of problem discovery and processing, effectively reduces misjudgment and missed detection, and significantly improves the overall reliability and application accuracy of three-dimensional mapping results.

[0039] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the flow or function described in the embodiments of the application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through a wired network or wireless network. The computer readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD) or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0040] Some data in the above formula are dimensionless for numerical calculation, and the contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0041] The above embodiments are only used to illustrate the technical method of the application and are not limited. Although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the application.

Claims

1. A real-scene 3D mapping system based on big data, characterized in that: include: An acquisition module is used for collecting video data of the range area by a camera acquisition device, wherein the video data includes the north side viewing angle data of the range area, the upper side viewing angle data of the range area, and the west side viewing angle data of the range area; A recognition module, using a big data recognition model to identify real objects in the scene from the perspective data above the range area, wherein the big data recognition model recognizes the perspective data from the range area and labels the types of the real objects in the scene; Selecting three-dimensional object models and combining the three-dimensional object models into a range area map; A height recognition module, using a big data recognition model, identifies the height values ​​of real objects in the scene from the north side perspective data of the range area, and modifies the three-dimensional object model in the range area map according to the range area map; a verification module, outputting a left-to-right object model arrangement sequence based on the range area map, using a big data recognition model to identify a real object arrangement verification sequence of real objects in the scene of the left side perspective data of the range area, and verifying the object model arrangement sequence based on the real object arrangement verification sequence to obtain an area to be reviewed; Conduct secondary collection and secondary identification on the area to be reviewed.

2. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The process of the big data recognition model identifying the scene objects in the perspective data above the range area includes: S21. Input the collected view angle data of the range area and the range area above into a big data recognition model. The model uses a deep convolutional neural network structure to perform image normalization, edge enhancement, and convolution feature extraction on the input image to obtain an image feature map. S22, using a semantic segmentation algorithm to generate candidate regions and perform pixel-level segmentation on the image feature map, and output a plurality of candidate boxes, each of which corresponds to a potential scene object region in the image; S23. Apply a multi-type classification network to the candidate frame areas respectively to identify and output corresponding scene object type labels, where the scene objects include buildings, lanes, plants, and rivers.

3. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The steps for combining scene objects into a range area map include: S31. Calling a corresponding standard 3D object model template in a 3D model library according to the scene object type label, wherein the 3D model library is a collection of model templates; S32. Extracting geographic plane coordinates (X, Y) from the perspective data above the range area; S33, scaling the selected three-dimensional object model in proportion to the pixel size of the scene object in the viewing angle data above the range area; and performing posture rotation on the model by referring to the direction in the viewing angle data above the range area; S34, superimposing the three-dimensional object model onto the map model according to the geographic plane coordinates (X, Y), and processing the boundaries between the three-dimensional object models using an automatic splicing algorithm; S35. Perform texture mapping on the map model to obtain a three-dimensional map of the range area.

4. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: Extracting geographic plane coordinates (X, Y) from the perspective data above the range area, including: extracting the two-dimensional projection center point coordinates (u, v) and bounding box size information of the scene object from the perspective data above the range area, and mapping the two-dimensional coordinates to geographic plane coordinates (X, Y); Among them, u0, v0 are the image origin, (X0, Y0) is the geographic coordinate origin, Sx, Sy are the resolutions of the pixel-to-plane distance, and u, v are the coordinates of the two-dimensional projection center point.

5. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The use of the big data recognition model to identify the height values ​​of real objects in the scene of the north side perspective data of the range area includes: S41, performing image preprocessing on the north side perspective data of the range area, wherein the image preprocessing includes distortion correction, edge enhancement and illumination equalization processing; S42, inputting the north side perspective data of the range area into the big data recognition model, and extracting the target bounding box using semantic segmentation and object detection algorithms; S43: Applying a monocular image depth regression model set to the target bounding box to calculate a height value of the target bounding box; S44, matching the target bounding box with the real object in the scene; S45 . Adjust the height of the corresponding three-dimensional object model according to the height value of the target bounding box.

6. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The calculation is performed using a monocular image depth regression model set, including: projecting the upper center pixel and the lower center pixel of the target bounding box to the camera coordinate system according to the camera intrinsic parameter matrix, obtaining the vertical coordinates of the two points by calculation, and using the absolute value of the difference between the vertical coordinates of the two points as the height value of the target bounding box.

7. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: Outputting a sequence of object models arranged from left to right according to the range area map includes: S501, extracting the coordinates of the two-dimensional projection center point of each three-dimensional object model in the three-dimensional map of the range area; S502, sorting the center point coordinates in ascending order of horizontal coordinate values ​​to form an object model arrangement sequence L1; S503 : Input the left-side perspective image of the range area into a big data recognition model. The big data recognition model performs target detection on the image and outputs a real object arrangement verification sequence L2 under the left-side perspective.

8. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The real object arrangement verification sequence for identifying the scene real objects of the left side viewing angle data of the range area includes: S511, aligning sequence L1 with sequence L2 to determine positional order consistency; S512. When there is a type discrepancy or order discrepancy in the comparison, mark the sequence L1 as an abnormal sequence; S513: Define the spatial region corresponding to the abnormal sequence as a region to be reviewed.

9. The real-scene 3D mapping system based on big data according to claim 1, characterized in that: The secondary collection and secondary identification of the area to be reviewed include: generating a secondary collection task based on the area to be reviewed, instructing a camera collection device to collect image data of the area to be reviewed, using a big data recognition model to identify real objects in the scene of the perspective data above the range area, selecting a three-dimensional object model, and updating the three-dimensional object model to the range area map; outputting an object model arrangement sequence from left to right according to the range area map, using a big data recognition model to identify a real object arrangement verification sequence of real objects in the scene of the perspective data on the left side of the range area, and verifying the object model arrangement sequence according to the real object arrangement verification sequence.

Images