Low-efficiency space development method and system based on live-action three-dimensional and digital stretching

Abstract

The invention discloses a low-efficiency space development method and system based on live-action three-dimensional and digital stretching, and particularly relates to the technical field of space development, and the method comprises the steps: obtaining a live-action picture set and point cloud data of a target building; creating a live-action three-dimensional model according to the live-action picture set and the point cloud data of the target building; obtaining a building evaluation coefficient of the live-action three-dimensional model, judging whether the target building is a low-efficiency space according to the building evaluation coefficient, and if yes, generating a space stretching instruction; receiving a space stretching instruction, obtaining a cloud point stretching formula, and performing digital stretching on the live-action three-dimensional model according to the cloud point stretching formula; according to the method, the deep neural network model is adopted to predict the point cloud coordinates, the accuracy of the stretched model is ensured, and the efficiency and adaptability of model calculation are improved in combination with optimization iteration of the cloud point stretching parameters. Through the fusion of the live-action image and the point cloud data and the use of the texture projection method, the generated three-dimensional model is strong in sense of reality, and the transformed space effect is visually displayed.

Description

Technical Field

[0001] The present invention relates to the technical field of space development. More specifically, the present invention relates to a method and system for inefficient space development based on real-scene three-dimensional and digital stretching. Background Art

[0002] With the acceleration of the urbanization process and the increasing shortage of land resources, the development and utilization of inefficient space have become an important direction for enhancing urban functions and optimizing resource allocation. However, inefficient space usually has complex building structures, and the space planning fails to fully consider functionality and accessibility, or is restricted by factors such as terrain, building density, and historical protection, making it difficult to use traditional development means.

[0003] In existing methods, for example, Chinese Patent No. CN118736139A discloses a method for constructing a three-dimensional model of building equipment and a digital twin system. CAD drawings are obtained and preprocessed to display the corresponding two-dimensional model in a 3D space; on the two-dimensional model, the model layer and model object of the three-dimensional model to be created are determined; the position information of each model object on the two-dimensional plane, the position information in the vertical direction, and its own height information are determined to determine the three-dimensional coordinates of the three-dimensional model in the 3D space; the type of the three-dimensional model is determined, and the corresponding prefabricated model is selected from a preset model resource library; an instance of a prefabricated model is created for each model object, and the position of each instance in the 3D space is determined according to the calculated three-dimensional coordinates to obtain the three-dimensional model. Although the above method can effectively improve the development efficiency, through research and application of the above method and the existing technology, it is found that the above method and the existing technology have at least the following partial defects:

[0004] Based on two-dimensional CAD drawings, only static three-dimensional models can be generated, and dynamic digital stretching operations cannot be directly performed on the models, and it is even more difficult to make real-time adjustments.

[0005] Therefore, the present invention provides a method and system for inefficient space development based on real-scene three-dimensional and digital stretching. Summary of the Invention

[0006] In order to overcome the above defects of the prior art, the present invention provides a method and system for inefficient space development based on real-scene three-dimensional and digital stretching to solve the problems raised in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides a method for inefficient space development based on real-scene three-dimensional and digital stretching, including:

[0009] Step 1: Obtain the collection of real - scene images and point - cloud data of the target building; create a real - scene 3D model based on the collection of real - scene images and point - cloud data of the target building;

[0010] Step 2: Obtain the building evaluation coefficient of the real - scene 3D model, and determine whether the target building is an inefficient space according to the building evaluation coefficient. If so, generate a space stretching instruction;

[0011] Step 3: Receive the space stretching instruction, obtain the cloud - point stretching formula, and complete the digital stretching of the real - scene 3D model according to the cloud - point stretching formula.

[0012] Furthermore, the method for obtaining the collection of real - scene images of the target building includes:

[0013] Step a11: Divide the target building according to the division rule to obtain N target sub - regions;

[0014] Step a12: Set the basic flight parameters of the drone, and take pictures of the nth target sub - region according to the preset flight path; n = 1, 2, ……, N;

[0015] The basic flight parameters include flight altitude, forward overlap rate, and side - overlap rate; the drone is equipped with a vertical camera and an oblique camera. The vertical camera is used to obtain orthophotos, and the oblique camera collects the side information of the target building from the front, back, left, and right angles;

[0016] After the shooting of the nth target sub - region is completed, let n=n + 1, and return to Step a11. When n = N, obtain the collection of real - scene images of the target building.

[0017] Furthermore, the method for obtaining the point - cloud data of the target building includes:

[0018] Step a21: Determine U stations of the target building, and there is a 30% - 50% overlapping area between each station;

[0019] Step a22: Calibrate the horizontal and vertical directions of the scanner, and adjust the scanning resolution. Scan the target building based on the uth station to assist in point - cloud alignment, and record the coordinates of each scanned point;

[0020] Step a23: Use point - cloud processing software to calibrate the point cloud of the uth station. After calibration, the error between every two point clouds ≤2mm. The point - cloud processing software includes CloudCompare and Trimble RealWorks;

[0021] Let u = u + 1, repeat Step a22 - Step a23 until u = U to end the loop, and count the scanned point - cloud sets of all U stations to obtain the point - cloud data of the target building;

[0022] The real scene map set and the point cloud data are in the same coordinate system.

[0023] Furthermore, the method for creating a real scene three-dimensional model based on the real scene map set and the point cloud data of the target building includes:

[0024] Step a31: Mesh the point cloud data of the target building using the Delaunay triangulation or Poisson surface reconstruction algorithm to obtain a point cloud mesh model;

[0025] Step a32: Perform similarity matching between the real scene map set and the point cloud data, and project the texture onto the mesh surface through the camera parameters to obtain an initial real scene model;

[0026] Step a33: Trim the overlapping areas in the initial real scene model to obtain a real scene three-dimensional model;

[0027] Among them, the method for projecting the texture onto the mesh surface through the camera parameters includes:

[0028] Step a321: Calculate the projection coordinates (μ, δ) of each point in the plane according to each point of the point cloud;

[0029] Step a322: Sample the texture from the real scene map according to the projection coordinates (μ, δ) and map the texture to the corresponding points on the mesh surface.

[0030] Furthermore, the method for trimming the overlapping areas in the initial real scene model includes:

[0031] Step a331: Calculate the camera view weight of the r-th real scene map based on the mesh surface;

[0032] Step a332: Calculate the distance weight of the r-th real scene map;

[0033] Step a333: Normalize the camera view weight and the distance weight to obtain the total weight of the r-th real scene map;

[0034] Step a334: Analyze the overlapping areas and calculate the texture color values;

[0035] Step a335: Let r = r + 1, and repeat steps a331 - a334 until r = R to end the loop, and obtain a real scene three-dimensional model with seamless texture.

[0036] Furthermore, the method for obtaining the building evaluation coefficient of the real scene three-dimensional model includes:

[0037] Obtain the building utilization data of the target area, where the building utilization data includes building density, floor area ratio, and building function type score;

[0038] Analyze the building utilization data to obtain the building evaluation coefficient;

[0039] The method for determining whether the target building is an inefficient space based on the building evaluation coefficient includes:

[0040] Preset the evaluation coefficient threshold, where the evaluation coefficient threshold includes a first evaluation threshold and a second evaluation threshold, and the first evaluation threshold is less than the second evaluation threshold; compare the building evaluation coefficient with the preset evaluation coefficient threshold;

[0041] If the building evaluation coefficient is less than the first evaluation threshold, or the building evaluation coefficient is greater than the second evaluation threshold, then determine that the target building is an inefficient space and generate a space stretching instruction;

[0042] If the building evaluation coefficient is greater than or equal to the first evaluation threshold and less than or equal to the second evaluation threshold, then do not determine that the target building is an inefficient space and do not generate a space stretching instruction.

[0043] Furthermore, the method for obtaining the cloud point stretching formula includes:

[0044] Step b1: According to the generated space stretching instruction, subtract the actual point cloud coordinates from the target point cloud coordinates to obtain the coordinate point difference, and use the coordinate point difference as the point cloud deviation; the point cloud coordinates are the X coordinate value, the Y coordinate value, or the Z coordinate value;

[0045] Step b2: According to the coordinate point difference, obtain the cloud point stretching parameters, where the cloud point stretching parameters include the cloud point proportional gain, the cloud point integral gain, and the cloud point differential gain;

[0046] Step b3: Calculate the cloud point stretching formula according to the obtained cloud point stretching parameters and the point cloud deviation.

[0047] Furthermore, the method for obtaining the cloud point stretching parameters includes:

[0048] Step c1: Preset the initial point cloud coordinate YD min , the maximum point cloud coordinate YD max , the stretching coefficient θ, and the maximum number of iterations and let the current point cloud coordinate YD = YD min ;

[0049] Step c2: Randomly set a feasible solution K, where the feasible solution K is the cloud point stretching parameter, obtain the cloud point stretching parameter range, where the cloud point stretching parameter range includes the cloud point proportional gain range, the cloud point integral gain range, and the cloud point differential gain range, and the cloud point stretching parameter range is the range of the feasible solution K;

[0050] Step c3: Determine the fitness function;

[0051] Wherein, YD' represents the target point cloud coordinates, and YD" represents the predicted point cloud coordinates; g represents the coordinates of the g-th cloud point, and G represents the total number of cloud point coordinates;

[0052] Step c4: Calculate the fitness f corresponding to the feasible solution K; using the feasible solution K as the current point, perform random perturbation within the neighborhood of the current point to obtain a new feasible solution K', and calculate the fitness f' corresponding to the new feasible solution K';

[0053] Step c5: Calculate the fitness difference f", and the expression of the fitness difference f" is f" = f' - f;

[0054] If the fitness difference f" > 0, then let K = K', that is, assign the value of the new feasible solution K' to the feasible solution K; if the fitness difference f" ≤ 0, then calculate the probability P', and according to the probability P', let K = K';

[0055] Step c6: Loop steps c4 to c5 until the number of loops reaches the maximum number of iterations ZD, then the loop ends and enter step c7;

[0056] Step c7: Let the current point cloud coordinate YD = YD × θ, that is, stretch the current point cloud coordinate in step c1, and assign the stretched value to the current point cloud coordinate; let the maximum number of iterations ZD = ZD × θ, that is, assign the stretched value of the maximum number of iterations to the maximum number of iterations; if the stretched maximum number of iterations is not an integer, then round up the stretched maximum number of iterations;

[0057] Step c8: Loop steps c4 to c7 until the current point cloud coordinate YD > YD max When the loop ends, obtain the cloud point stretching parameter corresponding to the feasible solution K.

[0058] Furthermore, the method for obtaining the predicted point cloud coordinates includes:

[0059] Taking the cloud point stretching parameter and the point cloud deviation corresponding to the feasible solution K as cloud point feature data; inputting the cloud point feature data into the trained point cloud coordinate prediction model to predict the corresponding predicted point cloud coordinates;

[0060] The specific training process of the point cloud coordinate prediction model includes:

[0061] Pre-collect the predicted point cloud coordinates corresponding to H groups of cloud point feature data, and convert the cloud point feature data and the corresponding predicted point cloud coordinates into a corresponding set of feature vectors;

[0062] Use each set of feature vectors as the input of the point cloud coordinate prediction model. The point cloud coordinate prediction model outputs a set of predicted point cloud coordinates corresponding to each set of cloud point feature data, and uses the actual predicted point cloud coordinates corresponding to each set of cloud point feature data as the prediction target. The actual predicted point cloud coordinates are the predicted point cloud coordinates corresponding to the cloud point feature data collected in advance as described above. Use minimizing the sum of the prediction errors of all cloud point feature data as the training target. Train the point cloud coordinate prediction model until the sum of the prediction errors converges and then stop training. The point cloud coordinate prediction model is a deep neural network model.

[0063] In a second aspect, the present invention provides an inefficient space development system based on real scene three-dimensional and digital stretching, which is used to implement the above-mentioned inefficient space development method based on real scene three-dimensional and digital stretching, and includes:

[0064] A model creation module, which is used to obtain a set of real scene images and point cloud data of the target building; create a real scene three-dimensional model according to the set of real scene images and point cloud data of the target building;

[0065] A judgment module, which is used to obtain the building evaluation coefficient of the real scene three-dimensional model, and judge whether the target building is an inefficient space according to the building evaluation coefficient. If so, generate a space stretching instruction;

[0066] A stretching module, which is used to receive the space stretching instruction, obtain a cloud point stretching formula, and complete digital stretching of the real scene three-dimensional model according to the cloud point stretching formula.

[0067] The technical effects and advantages of the present invention:

[0068] 1. The present invention uses a deep neural network model to predict point cloud coordinates to ensure the accuracy of the stretched model. At the same time, combined with the optimization iteration of cloud point stretching parameters, it improves the efficiency and adaptability of model calculation. The fusion of real scene images and point cloud data and the use of texture projection methods make the generated three-dimensional model have a strong sense of reality and can more intuitively display the transformed space effect. The present invention avoids the time-consuming and laborious manual intervention in traditional methods, improves the convenience of the development process, and provides an innovative technical path for building planning and renovation by optimizing the allocation of space resources.

[0069] 2. The present invention realizes the accurate identification and efficient development of inefficient spaces by combining real - scene three - dimensional modeling and digital stretching technology. Real - scene images of target buildings are collected using drones and multi - angle lenses, and at the same time, through high - precision point cloud data scanning, the integrity and consistency of model data are ensured. On this basis, by calculating the building evaluation coefficient, it is possible to scientifically judge whether a building belongs to an inefficient space, avoiding the subjectivity and inaccuracy existing in traditional development methods. The digital stretching process uses stretching formulas and optimization algorithms to ensure the accuracy and continuity of the transformation of the point cloud model, significantly improving the space utilization rate and development efficiency, and providing a scientific basis and technical support for the transformation of urban inefficient spaces. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flowchart of the development method for inefficient spaces based on real - scene three - dimensional and digital stretching in Embodiment 1;

[0071] Figure 2 It is a flowchart of the method for obtaining the set of real - scene images of the target building in Embodiment 1;

[0072] Figure 3 It is a flowchart of the method for obtaining the point cloud data of the target building in Embodiment 1;

[0073] Figure 4 It is a schematic structural diagram of the development system for inefficient spaces based on real - scene three - dimensional and digital stretching in Embodiment 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0075] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although terms such as "first" and "second" may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0077] Embodiment 1

[0078] Please refer to Figure 1 As shown, this embodiment discloses and provides an inefficient space development method based on real-scene three-dimensional and digital stretching. The method includes:

[0079] Step 1: Obtain a set of real-scene images and point cloud data of the target building; create a real-scene three-dimensional model according to the set of real-scene images and point cloud data of the target building.

[0080] It should be noted that: the target building includes an old industrial park, an abandoned commercial area, or a village in the city, etc.

[0081] Please refer to Figure 2 As shown, in implementation, the method for obtaining the set of real-scene images of the target building includes:

[0082] Step a11: Divide the target building according to the division rule to obtain N target sub-regions;

[0083] It should be noted that: the division rule is to divide by shape size or area, which is specifically determined by those skilled in the art and will not be overly limited herein.

[0084] Step a12: Set the basic flight parameters of the drone and take pictures of the nth target sub-region according to the preset flight route; n = 1, 2,..., N;

[0085] It should be noted that: the drone is equipped with a vertical lens and an oblique lens. The vertical lens is used to obtain orthophotos, and the oblique lens collects side information of the target building from multiple angles such as front, back, left, and right to construct a three-dimensional model. The basic flight parameters include flight altitude, forward overlap rate, and side overlap rate. Among them, the flight altitude is determined by the mapping scale and the ground resolution requirement. The larger the scale and the higher the resolution requirement, the lower the flight altitude. The forward overlap rate is set between 60% and 80%. A higher overlap rate is beneficial for feature matching and three-dimensional reconstruction; the side overlap rate is maintained at 30% - 40% to ensure sufficient overlap between the images taken by adjacent flight routes to prevent data gaps.

[0086] Step a13: After the shooting of the nth target sub-region is completed, let n = n + 1, and return to step a11. Until n = N, a set of real-scene images of the target building is obtained.

[0087] Please refer to Figure 3 As shown, in the implementation, the method for obtaining the point cloud data of the target building includes:

[0088] Step a21: Determine U stations of the target building, and there is a 30% - 50% overlapping area between each station;

[0089] Step a22: Calibrate the horizontal and vertical directions of the scanner, and adjust the scanning resolution. Based on the u-th station, scan the target building to assist in point cloud alignment, and record the coordinates of each scanning point;

[0090] It should be noted that: The auxiliary point cloud is a point cloud data set generated during the point cloud alignment process by calibrating the scanner and adjusting the scanning parameters, and is used to improve the alignment accuracy.

[0091] Step a23: Use point cloud processing software to calibrate the point cloud of the u-th station. After calibration, the error between every two point clouds ≤ 2mm. The point cloud processing software includes CloudCompare and Trimble RealWorks;

[0092] It should be noted that: The point cloud is a set composed of multiple scanning points.

[0093] Step a24: Let u = u + 1, repeat steps a22 - a23 until u = U to end the loop, and count the scanned point cloud sets of all U stations to obtain the point cloud data of the target building.

[0094] It is worth noting that: The set of real-scene images and the point cloud data are in the same coordinate system to ensure the accuracy of the positional relationship between the two.

[0095] In the implementation, according to the set of real-scene images and the point cloud data of the target building, the method for creating a real-scene three-dimensional model includes:

[0096] Step a31: Use the Delaunay triangulation or Poisson surface reconstruction algorithm to grid the point cloud data of the target building to obtain a point cloud grid model;

[0097] The Delaunay triangulation or Poisson surface reconstruction algorithm is a prior art and will not be repeated here.

[0098] Step a32: Perform similarity matching between the set of real-scene images and the point cloud data, and project the texture onto the grid surface through the camera parameters to obtain an initial real-scene model;

[0099] It should be noted that the similarity algorithms include, but are not limited to, cosine similarity algorithm or Euclidean distance algorithm, etc.; before the similarity matching between the set of real-scene images and the point cloud data, the set of real-scene images needs to be preprocessed, and the preprocessing includes, but is not limited to, image enhancement, image denoising, and image segmentation, etc.

[0100] Step a33: Trim the overlapping regions in the initial real-scene model to obtain a real-scene 3D model.

[0101] Among them, the method of projecting texture onto the mesh surface through camera parameters includes:

[0102] Step a321: Calculate the projection coordinates (μ, δ) of each point in the point cloud on the plane, and its calculation formula is:

[0103]

[0104] In the formula, fx and fy are the focal lengths, cx and cy are the principal point coordinates, and (X, Y, Z) represents the three-dimensional coordinates in the point cloud.

[0105] It should be noted that: the principal point coordinates refer to the coordinates of the intersection point of the optical axis (the principal ray passing through the center of the lens) and the imaging plane (usually the photosensitive element or image sensor of the camera) during the camera imaging process. In the process of 3D reconstruction and image processing, the principal point coordinates are usually expressed as (cx, cy) in pixel units.

[0106] Step a322: Sample the texture from the real-scene image according to the projection coordinates (μ, δ) and map the texture to the corresponding points on the mesh surface.

[0107] Among them, the method of trimming the overlapping regions in the initial real-scene model includes:

[0108] Step a331: Based on the mesh surface, calculate the camera view angle weight of the r-th real-scene image, and its calculation formula is:

[0109] SJ r =cos(ω)

[0110] In the formula, SJ r represents the camera view angle weight, and ω represents the angle between the line of sight of the r-th real-scene image and the mesh normal vector.

[0111] It should be noted that: when the shooting direction of the real-scene image is directly facing the mesh surface, the weight is larger, and when the angle deviates, the weight is lower.

[0112] Step a332: Calculate the distance weight of the r-th real-scene image, and its calculation formula is:

[0113]

[0114] where LD r represents the distance weight of the r-th real scene image, d represents the distance from the camera to the center of the grid surface, and ∈ represents a constant greater than zero to avoid a zero denominator.

[0115] It should be noted that: the closer the distance, the greater the camera view weight, ensuring that the details of the real scene image at close range are prioritized.

[0116] Step a333: Normalize the camera view weight and the distance weight to obtain the total weight of the r-th real scene image;

[0117]

[0118] where W r represents the total weight of the r-th real scene image.

[0119] Step a334: Analyze the overlapping area and calculate the texture color value, and its calculation formula is:

[0120]

[0121] where Cyc r represents the texture color value of the r-th real scene image, C r represents the color value of the pixel corresponding to the r-th real scene image; W r represents the total weight of the r-th real scene image.

[0122] Step a335: Let r = r + 1, and repeat steps a331 - a334 until r = R to end the loop, and obtain a seamless texture real scene 3D model.

[0123] Step 2: Obtain the building evaluation coefficient of the real scene 3D model, and judge whether the target building is an inefficient space according to the building evaluation coefficient. If so, generate a space stretching instruction;

[0124] In implementation, the method for obtaining the building evaluation coefficient of the real scene 3D model includes:

[0125] Obtain the building utilization data of the target area, and the building utilization data includes building density, floor area ratio, and building function type score;

[0126] Among them, the formula for obtaining the building density is:

[0127] where MD represents the building density, YJM represents the research area, and JZM represents the building floor area.

[0128] The method for obtaining the floor area ratio includes:

[0129] Wherein, RJ represents the floor area ratio, ZM represents the total building area, and YJM represents the area of the research region;

[0130] The total building area is obtained by using LiDAR point cloud data or building height data to calculate the total area of each building (building footprint × number of floors).

[0131] Analyze the building utilization data to obtain the building evaluation coefficient, and its calculation formula is:

[0132] PG = MD × β 1 + RJ × β 2 + PF

[0133] Wherein, PG represents the building evaluation coefficient, MD represents the building density, RJ represents the floor area ratio, PF represents the building function type score, and β 1 and β 2 are weight factors, and the weight factors are all set by those skilled in the art according to experience.

[0134] It should be noted that: the building function type score is preset, and different values are set according to the building function type. Exemplarily, the building function type score corresponding to the building function type of a commercial building is set to 3, the building function type score corresponding to the building function type of a residence is set to 2, and the building function type score corresponding to the building function type of a factory is set to 1.

[0135] In implementation, the method for judging whether a target building is an inefficient space according to the building evaluation coefficient includes:

[0136] Preset an evaluation coefficient threshold, and the evaluation coefficient threshold includes a first evaluation threshold and a second evaluation threshold, and the first evaluation threshold is less than the second evaluation threshold; compare the building evaluation coefficient with the preset evaluation coefficient threshold;

[0137] If the building evaluation coefficient is less than the first evaluation threshold, or the building evaluation coefficient is greater than the second evaluation threshold, then judge that the target building is an inefficient space and generate a space stretching instruction;

[0138] If the building evaluation coefficient is greater than or equal to the first evaluation threshold and less than or equal to the second evaluation threshold, then do not judge that the target building is an inefficient space, do not generate a space stretching instruction, and store the real scene 3D model of the target building in the database.

[0139] It should be noted that: the preset evaluation coefficient threshold is that those skilled in the art respectively obtain the building utilization data multiple times during the historical normal working stage, calculate the building evaluation coefficients of multiple groups of building utilization data, and take the average value of the multiple building evaluation coefficients as the preset evaluation coefficient threshold.

[0140] This step analyzes the spatial utilization efficiency by measuring the building density and plot ratio, and determines the inefficiently utilized spaces such as idle houses or abandoned factories to be optimized.

[0141] Step 3: Receive the spatial stretching instruction, obtain the cloud point stretching formula, and perform digital stretching on the real-scene 3D model according to the cloud point stretching formula.

[0142] In implementation, the method for obtaining the cloud point stretching formula includes:

[0143] Step b1: According to the generated spatial stretching instruction, subtract the actual point cloud coordinates from the target point cloud coordinates to obtain the coordinate point difference, and use the coordinate point difference as the point cloud deviation; the point cloud coordinates are the X coordinate value, Y coordinate value, or Z coordinate value. In this embodiment, the Z coordinate value is taken as an example for illustration.

[0144] It should be noted that: the target point cloud coordinates represent the expected point cloud position after stretching, and the actual point cloud coordinates represent the original coordinates obtained by scanning.

[0145] Step b2: According to the coordinate point difference, obtain the cloud point stretching parameters, and the cloud point stretching parameters include the cloud point proportional gain, cloud point integral gain, and cloud point derivative gain;

[0146] It should be noted that: the cloud point proportional gain directly gives the control amount according to the deviation between the current stretching state and the target stretching state. The greater the deviation, the stronger the adjustment force output based on the proportional gain, driving the cloud point to quickly change towards the expected stretching form to ensure fast response; the cloud point integral gain continuously accumulates all the stretching deviations from the start time of stretching to the present. When relying solely on the proportional gain cannot completely eliminate the slight difference between the final stretching position of the cloud point and the target point cloud coordinates, the integral gain will continuously accumulate over time to help achieve precise positioning and eliminate the steady-state error; the cloud point derivative gain focuses on the change rate of the cloud point stretching deviation. If the cloud point stretching speed is too fast and exceeds the expected target, the derivative gain can sensitively capture the situation of the deviation change accelerating and output a reverse control amount to avoid overstretching and maintain the smoothness and controllability of the stretching process.

[0147] The method for obtaining the cloud point stretching parameters includes:

[0148] Step c1: Preset the initial point cloud coordinate YD min , the maximum point cloud coordinate YD max , the stretching coefficient θ, and the maximum number of iterations and let the current point cloud coordinate YD = YD min ;

[0149] Step c2: Randomly set a feasible solution K, where the feasible solution K is the cloud point stretching parameter. Obtain the range of the cloud point stretching parameter, which includes the cloud point proportional gain range, the cloud point integral gain range, and the cloud point derivative gain range. The range of the cloud point stretching parameter is obtained according to the technical parameters of the PID controller, and the range of the cloud point stretching parameter is the range of the feasible solution K.

[0150] Step c3: Determine the fitness function.

[0151] The expression of the fitness function is:

[0152] In the formula, YD′ g represents the target point cloud coordinates, and YD" g represents the predicted point cloud coordinates; g represents the g-th cloud point coordinate, and G represents the total number of cloud point coordinates.

[0153] The method for obtaining the predicted point cloud coordinates includes:

[0154] Take the cloud point stretching parameter corresponding to the feasible solution K and the point cloud deviation as the cloud point feature data; input the cloud point feature data into the trained point cloud coordinate prediction model to predict the corresponding predicted point cloud coordinates.

[0155] The specific training process of the point cloud coordinate prediction model includes:

[0156] Pre-collect the predicted point cloud coordinates corresponding to H groups of cloud point feature data, and convert the cloud point feature data and the corresponding predicted point cloud coordinates into a corresponding set of feature vectors.

[0157] Take each set of feature vectors as the input of the point cloud coordinate prediction model. The point cloud coordinate prediction model takes a set of predicted point cloud coordinates corresponding to each group of cloud point feature data as the output, takes the actual predicted point cloud coordinates corresponding to each group of cloud point feature data as the prediction target, and the actual predicted point cloud coordinates are the above-mentioned pre-collected predicted point cloud coordinates corresponding to the cloud point feature data; take minimizing the sum of the prediction errors of all cloud point feature data as the training target; among them, the calculation formula of the prediction error: where YWC is the prediction error, E is the number of groups of feature vectors corresponding to the cloud point feature data, A e is the predicted point cloud coordinate corresponding to the e-th group of cloud point feature data, and O e is the actual predicted point cloud coordinate corresponding to the e-th group of cloud point feature data; train the point cloud coordinate prediction model until the sum of the prediction errors reaches convergence and then stop training; the point cloud coordinate prediction model is a deep neural network model.

[0158] It should be noted that the predicted point cloud coordinates corresponding to the cloud point feature data are collected by those skilled in the art during the historical inefficient spatial stretching process. When adjusting the point cloud coordinates, multiple groups of different cloud point feature data are collected. Under the conditions of each group of cloud point feature data, the corresponding cloud point stretching formula is calculated. According to multiple cloud point stretching formulas, the 3D software is controlled to adjust the point cloud coordinates in sequence, and the adjusted point cloud coordinates are recorded in sequence. The adjusted point cloud coordinates are the predicted point cloud coordinates corresponding to the cloud point feature data;

[0159] Step c4: Calculate the fitness f corresponding to the feasible solution K; taking the feasible solution K as the current point, perform a random perturbation within the neighborhood of the current point to obtain a new feasible solution K', and calculate the fitness f' corresponding to the new feasible solution K';

[0160] Step c5: Calculate the fitness difference f", and the expression of the fitness difference f" is f" = f' - f;

[0161] If the fitness difference f" > 0, then let K = K', that is, assign the value of the new feasible solution K' to the feasible solution K; if the fitness difference f" ≤ 0, then calculate the probability P', and according to the probability P', let K = K'; the expression of the probability P' is:

[0162] Step c6: Loop steps c4 to c5 until the number of loops reaches the maximum iteration number ZD, then the loop ends and enter step c7;

[0163] Step c7: Let the current point cloud coordinate YD = YD × θ, that is, stretch the current point cloud coordinate in step c1, and assign the stretched value to the current point cloud coordinate; let the maximum iteration number ZD = ZD × θ, that is, assign the stretched value of the maximum iteration number to the maximum iteration number; if the stretched maximum iteration number is not an integer, then round up the stretched maximum iteration number to make the stretched maximum iteration number an integer;

[0164] Step c8: Loop steps c4 to c7 until the current point cloud coordinate YD > YD max When the loop ends, obtain the cloud point stretching parameters corresponding to the feasible solution K.

[0165] It should be noted that the initialized point cloud coordinate YD min The maximum point cloud coordinate YD max, the stretching coefficient θ and the maximum number of iterations ZD are used as preset parameters. These preset parameters are determined by those skilled in the art. During the historical inefficient spatial stretching process, when generating a spatial stretching instruction, for the same point cloud deviation, multiple different groups of preset parameters are sequentially set in advance. The point cloud stretching parameters are obtained in sequence. According to the obtained multiple groups of point cloud stretching parameters, the corresponding point cloud stretching formulas are calculated. According to the multiple point cloud stretching formulas, the 3D software is sequentially controlled to adjust the point cloud coordinates. The preset parameter corresponding to the adjusted point cloud coordinates that is closest to the target point cloud coordinates is used as the preset parameter corresponding to this point cloud deviation. By analogy, the preset parameters corresponding to multiple point cloud deviations are obtained. The average value of the multiple preset parameters (i.e., the average value of the initial point cloud coordinates, the maximum point cloud coordinates, the stretching coefficient average value, and the maximum number of iterations average value) is used as the initial point cloud coordinates YD preset in step c1 min , the maximum point cloud coordinates YD max , the stretching coefficient θ and the maximum number of iterations ZD;

[0166] Step b3: Calculate the point cloud stretching formula (stretching in the Z-axis direction) according to the obtained point cloud stretching parameters and the point cloud deviation;

[0167]

[0168] In the formula: ZZ 1 represents the stretched Z coordinate, ZZ 0 represents the actual Z coordinate, K P represents the point cloud proportional gain, K I represents the point cloud integral gain, K D represents the point cloud derivative gain, and ΔZ represents the point cloud deviation in the Z-axis.

[0169] It should be noted that: K P ×ΔZ represents directly giving a control quantity according to the current deviation (i.e., ΔZ) to push the point cloud closer to the target position; ∫ΔZ dt represents correcting the long-term deviation by accumulating historical deviations (i.e., the deviations at each moment). Especially when the proportional gain is not sufficient to eliminate subtle errors, the integral gain will play a key role; represents generating a reverse control through the rate of change of the deviation (i.e., ) to avoid overstretching, which helps to smooth the control process and avoid rapid overshoot.

[0170] Similarly, calculate the point cloud stretching formula (stretching in the X-axis direction) according to the obtained point cloud stretching parameters and the point cloud deviation;

[0171]

[0172] In the formula: XX 1 represents the stretched X coordinate, XX 0 represents the actual X coordinate, KP Represents the cloud point ratio gain, K I Represents the cloud point integral gain, K D Represents the cloud point differential gain, and ΔX represents the point cloud deviation of the X-axis.

[0173] Similarly, according to the obtained cloud point stretching parameters and point cloud deviation, calculate the cloud point stretching formula (stretching in the Y-axis direction);

[0174]

[0175] In the formula: YY 1 Represents the stretched Y coordinate, YY 0 Represents the actual Y coordinate, K P Represents the cloud point ratio gain, K I Represents the cloud point integral gain, K D Represents the cloud point differential gain, and ΔY represents the point cloud deviation of the Y-axis.

[0176] It should be noted that: the cloud point stretching parameters of the X-axis, Y-axis, and Z-axis are different because they are independently set according to the cloud point error characteristics and adjustment requirements of each axis. The steps of the setting are the same as the steps of obtaining the cloud point stretching parameters above and will not be repeated here.

[0177] This embodiment uses a deep neural network model to predict the point cloud coordinates to ensure the accuracy of the stretched model. At the same time, combined with the optimization iteration of the cloud point stretching parameters, the efficiency and adaptability of the model calculation are improved. The fusion of real-scene images and point cloud data and the use of texture projection methods make the generated 3D model have a strong sense of reality and can more intuitively display the renovated space effect. The present invention avoids the time-consuming and laborious manual intervention in the traditional method, improves the convenience of the development process, and provides an innovative technical path for building planning and renovation by optimizing the spatial resource allocation.

[0178] This embodiment realizes the accurate identification and efficient development of inefficient spaces by combining real-scene 3D modeling and digital stretching technology. Use drones and multi-angle cameras to collect real-scene images of the target building, and at the same time, through high-precision point cloud data scanning, ensure the integrity and consistency of the model data. On this basis, through the calculation of the building evaluation coefficient, it is possible to scientifically judge whether a building belongs to an inefficient space, avoiding the subjectivity and inaccuracy existing in the traditional development method. The digital stretching process uses stretching formulas and optimization algorithms to ensure the accuracy and continuity of the point cloud model transformation, significantly improving the space utilization rate and development efficiency, and providing a scientific basis and technical support for the renovation of urban inefficient spaces.

[0179] Embodiment 2

[0180] Please refer to Figure 4As shown in the figure, this embodiment provides an inefficient space development system based on real-scene three-dimensional and digital stretching. The system includes: a model creation module, a judgment module, and a stretching module; each module is connected by wired and / or wireless means to achieve data transmission between modules;

[0181] The model creation module is used to obtain the real-scene map set and point cloud data of the target building; according to the real-scene map set and point cloud data of the target building, create a real-scene three-dimensional model;

[0182] The judgment module is used to obtain the building evaluation coefficient of the real-scene three-dimensional model, and judge whether the target building is an inefficient space according to the building evaluation coefficient. If so, generate a space stretching instruction;

[0183] The stretching module is used to receive the space stretching instruction, obtain the cloud point stretching formula, and complete digital stretching of the real-scene three-dimensional model according to the cloud point stretching formula.

[0184] The formulas involved above are all calculated by removing the dimension and taking their numerical values. It is a formula obtained by software simulation of a large amount of collected data to be closest to the real situation. The weight factors in the formula and each preset threshold in the analysis process are set by those skilled in the art according to the actual situation or obtained by simulation of a large amount of data; the size of the weight factor is a specific value obtained by quantifying each parameter for subsequent comparison. Regarding the size of the weight factor, it depends on the amount of sample data and the processing coefficients initially set by those skilled in the art for each group of sample data; as long as it does not affect the proportional relationship between the parameters and the quantified values.

[0185] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application should not easily think of changes or substitutions, and should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0186] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An inefficient space development method based on real-life 3D and digital stretching, characterized in that: include: Step 1: Obtain the real-life image set and point cloud data of the target building; Create a real-life 3D model based on the real-life image collection and point cloud data of the target building; Step 2: Obtain the building evaluation coefficient of the real-scene 3D model, and determine whether the target building is an inefficient space based on the building evaluation coefficient. If so, generate a space stretching instruction; Step 3: Receive the space stretching instruction, obtain the cloud point stretching formula, and complete the digital stretching of the real scene 3D model according to the cloud point stretching formula.

2. The inefficient space development method based on real scene three-dimensional and digital stretching according to claim 1 is characterized in that: The method for obtaining the real scene picture set of the target building includes: Step a11: Divide the target building according to the division rules to obtain N target sub-areas; Step a12: setting the basic flight parameters of the drone, and photographing the nth target sub-area according to the preset route; n=1, 2, ..., N; The basic flight parameters include flight altitude, heading overlap rate and lateral repetition rate; the UAV is equipped with a vertical lens and an oblique lens, the vertical lens is used to obtain orthographic images, and the oblique lens collects side information of the target building from front, back, left and right angles; Step a13: After the nth target sub-area is photographed, let n=n+1 and return to step a11 until n=N, a set of real scene images of the target building is obtained.

3. The inefficient space development method based on real scene 3D and digital stretching according to claim 2 is characterized in that: Methods for acquiring point cloud data of target buildings include: Step a21: Determine U sites of the target building, and there is a 30% to 50% overlap area between each site; Step a22: calibrate the scanner in the horizontal and vertical directions, and adjust the scanning resolution, scan the target building based on the u-th site to assist in point cloud alignment, and record the coordinates of each scanning point; Step a23: Use point cloud processing software to calibrate the point cloud of the u-th station. After calibration, the error between every two point clouds is ≤2mm; Step a24: let u=u+1, repeat steps a22-a23 until u=U, end the loop, count the scan point cloud sets of all U sites, and obtain the point cloud data of the target building; The real scene image set and the point cloud data are in the same coordinate system.

4. The inefficient space development method based on real scene 3D and digital stretching according to claim 3 is characterized in that: The method of creating a realistic 3D model based on the realistic image set and point cloud data of the target building includes: Step a31: meshing the point cloud data of the target building using Delaunay triangulation or Poisson surface reconstruction algorithm to obtain a point cloud mesh model; Step a32: perform similarity matching between the real scene image set and the point cloud data, project the texture onto the mesh surface through camera parameters, and obtain an initial real scene model; Step a33: trimming the overlapping areas in the initial real scene model to obtain a real scene three-dimensional model; Among them, the method of projecting texture onto the mesh surface through camera parameters includes: Step a321: Calculate the projection coordinates (μ, δ) of each point on the plane according to each point of the point cloud; Step a322: Sample the texture from the real scene image according to the projection coordinates (μ, δ) and map the texture to the corresponding point on the mesh surface.

5. The inefficient space development method based on real scene three-dimensional and digital stretching according to claim 4 is characterized in that: Methods for trimming overlapping areas in the initial reality model include: Step a331: Calculate the camera perspective weight of the rth real scene image based on the grid surface; Step a332: Calculate the distance weight of the rth real scene image; Step a333: normalize the camera view weight and the distance weight to obtain the total weight of the rth real scene image; Step a334: Analyze the overlapping area and calculate the texture color value; Step a335: Let r=r+1, repeat steps a331-a334, and end the loop until r=R to obtain a real-life 3D model with seamless texture.

6. The inefficient space development method based on real scene 3D and digital stretching according to claim 5 is characterized in that: The methods for obtaining the building assessment coefficient of the real-life 3D model include: Acquire building utilization data of the target area, wherein the building utilization data includes building density, volume ratio, and building function type score; Analyze building utilization data to obtain building assessment coefficients; Methods for determining whether a target building is an inefficient space based on the building assessment coefficient include: Preset an evaluation coefficient threshold, the evaluation coefficient threshold includes a first evaluation threshold and a second evaluation threshold, the first evaluation threshold is less than the second evaluation threshold; compare the building evaluation coefficient with the preset evaluation coefficient threshold; If the building assessment coefficient is less than the first assessment threshold, or the building assessment coefficient is greater than the second assessment threshold, the target building is judged to be an inefficient space, and a space stretching instruction is generated; If the building assessment coefficient is greater than or equal to the first assessment threshold, and the building assessment coefficient is less than or equal to the second assessment threshold, the target building is not judged as an inefficient space, and no space stretching instruction is generated.

7. The inefficient space development method based on real scene 3D and digital stretching according to claim 6 is characterized in that: The methods for obtaining the cloud point stretching formula include: Step b1: according to the generated space stretching instruction, subtract the actual point cloud coordinates from the target point cloud coordinates to obtain a coordinate point difference value, and use the coordinate point difference value as the point cloud deviation; the point cloud coordinates are X coordinate values, Y coordinate values ​​or Z coordinate values; Step b2: according to the coordinate point difference value, obtaining cloud point stretching parameters, the cloud point stretching parameters include cloud point proportional gain, cloud point integral gain and cloud point differential gain; Step b3: Calculate the cloud point stretching formula based on the obtained cloud point stretching parameters and point cloud deviation.

8. The inefficient space development method based on real scene three-dimensional and digital stretching according to claim 7 is characterized in that: The methods for obtaining cloud point stretching parameters include: Step c1: Preset the initial point cloud coordinates YD min , Maximum point cloud coordinate YD max , stretch factor θ and maximum number of iterations And let the current point cloud coordinates YD = YD min ; Step c2: randomly set a feasible solution K, the feasible solution K is the cloud point stretching parameter, and obtain the cloud point stretching parameter range, the cloud point stretching parameter range includes the cloud point proportional gain range, the cloud point integral gain range and the cloud point differential gain range, and the cloud point stretching parameter range is the range of the feasible solution K; Step c3: determine the fitness function; Step c4: Calculate the fitness f corresponding to the feasible solution K; take the feasible solution K as the current point, perform random perturbations in the neighborhood of the current point, obtain a new feasible solution K′, and calculate the fitness f′ corresponding to the new feasible solution K′; Step c5: Calculate the fitness difference f"; If the fitness difference f">0, let K=K′, that is, assign the value of the new feasible solution K′ to the feasible solution K; if the fitness difference f"≤0, calculate the probability P′, and according to the probability P′, let K=K′; Step c6: loop steps c4 to c5 until the number of loops reaches the maximum number of iterations ZD, the loop ends, and the process goes to step c7; Step c7: Let the current point cloud coordinates YD = YD × θ, that is, stretch the current point cloud coordinates in step c1, and assign the stretched value to the current point cloud coordinates; let the maximum number of iterations ZD = ZD × θ, that is, assign the value of the maximum number of iterations after stretching to the maximum number of iterations; if the maximum number of iterations after stretching is not an integer, round up the maximum number of iterations after stretching; Step c8: Loop steps c4 to c7 until the current point cloud coordinate YD>YD max When , the loop ends and the cloud point stretching parameters corresponding to the feasible solution K are obtained.

9. The inefficient space development method based on real scene three-dimensional and digital stretching according to claim 8 is characterized in that: The methods for obtaining the predicted point cloud coordinates include: The cloud point stretching parameters and point cloud deviations corresponding to the feasible solution K are used as cloud point feature data; the cloud point feature data are input into the trained point cloud coordinate prediction model to predict the corresponding predicted point cloud coordinates; The specific training process of the point cloud coordinate prediction model includes: Preliminarily collect the predicted point cloud coordinates corresponding to H groups of cloud point feature data, and convert the cloud point feature data and the corresponding predicted point cloud coordinates into a corresponding set of feature vectors; Each group of feature vectors is used as the input of the point cloud coordinate prediction model. The point cloud coordinate prediction model takes a group of predicted point cloud coordinates corresponding to each group of cloud point feature data as output, and takes the actual predicted point cloud coordinates corresponding to each group of cloud point feature data as the prediction target, and the actual predicted point cloud coordinates are the predicted point cloud coordinates corresponding to the cloud point feature data collected in advance as mentioned above; minimizing the sum of prediction errors of all cloud point feature data is used as the training goal; the point cloud coordinate prediction model is trained until the sum of prediction errors reaches convergence and the training is stopped; the point cloud coordinate prediction model is specifically a deep neural network model.

10. An inefficient space development system based on real scene 3D and digital stretching, used to implement the inefficient space development method based on real scene 3D and digital stretching as claimed in any one of claims 1 to 9, characterized in that: include: A model creation module is used to obtain a set of real-life images and point cloud data of a target building; Create a real-life 3D model based on the real-life image collection and point cloud data of the target building; A judgment module is used to obtain a building evaluation coefficient of the real-scene three-dimensional model, and judge whether the target building is an inefficient space according to the building evaluation coefficient. If so, a space stretching instruction is generated; The stretching module is used to receive the space stretching instruction, obtain the cloud point stretching formula, and complete the digital stretching of the real scene three-dimensional model according to the cloud point stretching formula.

Images