A binocular 3D scanning system and method with texture mapping function

Abstract

This invention relates to the field of 3D scanning technology and discloses a binocular 3D scanning system and method with texture mapping functionality. The system comprises five modules: depth data acquisition, depth data correction, point cloud topology construction, texture mapping engine, and texture optimization. The depth data acquisition module simultaneously acquires the left and right views of a target object using a binocular camera array, generating a raw depth image dataset. The depth data correction module performs outlier detection on the raw depth data, extracts discrete distributions through boundary aggregation analysis, and replaces outliers. The point cloud topology construction module generates a 3D point cloud spatial topology structure and associates point cloud coordinates with the pixel positions of the left view. The texture mapping engine module projects the RGB data of the left view onto the point cloud to generate an initial texture map. The texture optimization module verifies local color consistency using a generative adversarial network and outputs an optimized 3D texture model. This system improves the texture accuracy and realism of the 3D model.

Description

Technical Field

[0001] This invention relates to the field of 3D scanning technology, specifically to a binocular 3D scanning system and method with texture mapping function. Background Technology

[0002] With the widespread application of 3D modeling technology across various industries, the demand for 3D models that combine accurate shape and realistic texture is increasing. Binocular 3D scanning systems, with their non-contact measurement and cost-effectiveness, have become a common tool for acquiring 3D information about objects, and are widely used in reverse engineering, virtual display, and medical imaging. However, traditional binocular 3D scanning systems still face many technical bottlenecks in the texture mapping stage.

[0003] During the depth data acquisition phase, traditional binocular cameras often suffer from matching errors between the left and right views due to shooting angle deviations, ambient light intensity fluctuations, or differences in object surface materials. This results in a large number of anomalous data points mixed in with the generated raw depth image. If these outliers are used directly in subsequent processing, they will cause local bulges, depressions, or breaks in the 3D point cloud during reconstruction, destroying the original geometry of the object and creating potential problems for subsequent texture mapping.

[0004] In the point cloud topology construction stage, traditional methods often rely on simple coordinate transformation algorithms to associate the spatial position of the point cloud with the view pixels. This approach is susceptible to factors such as lens distortion and blurred object edges, resulting in mismatches between point cloud coordinates and corresponding pixels. When performing texture projection, RGB color data cannot be accurately attached to the point cloud surface, leading to texture stretching, overlapping, or missing textures, which severely affects the visual effect of the model.

[0005] In terms of texture mapping optimization, existing technologies generally employ traditional algorithms such as mean filtering and median filtering. These methods can only smooth the texture as a whole and are unable to handle local color differences in complex scenes. For example, in areas with shadows, highlights, or abrupt texture changes on the surface of an object, traditional optimization algorithms cannot accurately identify color boundaries, which can easily lead to harsh texture transitions and even obscure the original details of the object. The resulting 3D texture model is difficult to meet the requirements of high-precision application scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a binocular 3D scanning system and method with texture mapping function to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a binocular 3D scanning system with texture mapping function, the system comprising:

[0008] The depth data acquisition module is used to simultaneously acquire the left and right views of the target object through a binocular camera array and generate the original depth image dataset.

[0009] The depth data correction module is used to perform outlier detection on the original depth image dataset. It extracts the discrete distribution of depth data through aggregated boundary analysis and replaces the abnormal depth values ​​in the discrete distribution based on the confidence interval threshold.

[0010] The point cloud topology construction module is used to generate a 3D point cloud spatial topology structure based on the depth data output by the depth data correction module, and associate the spatial coordinates of each point cloud with the pixel position mapping relationship of the left view.

[0011] The texture mapping engine module is used to project the RGB color data of the left view onto the 3D point cloud spatial topology based on the pixel position mapping relationship to generate the initial texture map;

[0012] The texture optimization module is used to perform local color consistency verification on the initial texture map through an adversarial generative network, and output an optimized 3D texture model.

[0013] Preferably, the specific steps for the depth data correction module to perform abnormal depth value detection include:

[0014] The original depth image dataset is sorted in ascending order of depth value. Twenty-five percent of the position values ​​in the depth sequence are extracted as the lower quartiles of the depth distribution, and seventy-five percent of the position values ​​are extracted as the upper quartiles of the depth distribution.

[0015] The numerical difference between the upper and lower quartiles is calculated and defined as the depth aggregation interval. The depth confidence interval boundary is set based on the multiple relationship of the depth aggregation interval.

[0016] Traverse the original depth image dataset, mark depth values ​​that exceed the depth confidence interval as invalid data, and fill in the invalid data with the weighted average of adjacent depth values.

[0017] Preferably, the operational logic of the point cloud topology construction module includes:

[0018] Establish three-dimensional spatial meshing rules, divide the surface of the target object into high curvature region mesh units and low curvature region mesh units, and assign high-density topological nodes to the high curvature region mesh units;

[0019] Based on the corrected depth data, a set of spatial coordinates of topological nodes is generated, and a spatial connection matrix of adjacent topological nodes is constructed.

[0020] The pixel coordinates of the left view are mapped to the spatial coordinate set of the topological nodes to form a pixel position mapping index table.

[0021] Preferably, the texture mapping engine module includes:

[0022] The color projection unit is used to project the RGB color data of the left view onto the three-dimensional point cloud spatial topology according to the topological node coordinates based on the pixel position mapping relationship index table.

[0023] The texture blending unit is used to perform bilinear interpolation calculations on the color data of adjacent topological nodes to eliminate color banding during the projection process.

[0024] The illumination compensation unit is used to analyze the ambient light intensity distribution of the left view and perform normalization compensation processing on the projected color data based on the light intensity distribution.

[0025] Preferably, the texture optimization module includes:

[0026] The generator unit is used to receive the 3D color distribution data of the initial texture map and generate adversarial synthetic texture samples;

[0027] The discriminator unit is used to compare the synthetic texture sample with the real color features of the left view and output the local texture distortion probability matrix.

[0028] An optimized feedback unit is used to adjust the color interpolation weight parameters of the generator unit in reverse according to the local texture distortion probability matrix, and iteratively update the 3D texture model.

[0029] Preferably, the system further includes:

[0030] The dynamic scanning parameter control module is used to calculate the point cloud stability coefficient based on the curvature distribution data of the three-dimensional point cloud spatial topology.

[0031] The scanning path planning module is used to retrieve the optimal binocular camera sampling step size from the scanning parameter configuration space, using the point cloud stability coefficient as an index.

[0032] The optimal sampling step size of the binocular camera is fed back to the depth data acquisition module, and the exposure time and focal length parameters of the binocular camera array are dynamically adjusted.

[0033] Preferably, the operating logic of the dynamic scanning parameter control module includes:

[0034] Construct a three-dimensional parametric space coordinate system, where the horizontal axis represents the variance of the point cloud curvature distribution, the vertical axis represents the depth data aggregation interval, and the vertical axis represents the standard deviation of ambient light intensity.

[0035] Establish a scanning parameter configuration space in a three-dimensional parameter space coordinate system, and label the corresponding historical camera sampling step size for each parameter coordinate point;

[0036] Based on the current point cloud stability coefficient, the target parameter subspace is located in the scanning parameter configuration space, and the optimal camera sampling step size in the target parameter subspace is retrieved by the neighborhood density clustering algorithm.

[0037] Preferably, the present invention also includes a binocular 3D scanning method with texture mapping function, applied to a binocular 3D scanning system with texture mapping function as described above, the method comprising the following steps:

[0038] Simultaneously acquire the left and right views from the stereo camera, generate the original depth image dataset, and perform depth data discrete distribution correction;

[0039] A 3D point cloud spatial topology is constructed based on the corrected depth data, and an index is established to map the topology node coordinates to the pixels in the left view.

[0040] Based on the mapping relationship index, the RGB color data of the left view is projected onto the three-dimensional point cloud spatial topology to generate an initial three-dimensional texture map;

[0041] By performing local color distortion detection on the initial 3D texture map using an adversarial generative network, an optimized 3D texture model is generated.

[0042] Preferably, the specific steps for generating the initial 3D texture map include:

[0043] Identify high-curvature region mesh cells in the spatial topology of a 3D point cloud and assign additional topology nodes to these high-curvature region mesh cells;

[0044] Spatial remapping of the left view pixel coordinates according to the additional topological node coordinates generates an enhanced pixel position mapping index;

[0045] RGB color data projection is performed based on the enhanced pixel location mapping relationship index, and Gaussian filtering color smoothing is performed between adjacent additional topology nodes.

[0046] Preferably, the operation flow of the adversarial generative network includes:

[0047] Extract the local color feature vectors of the 3D texture model and input them into the generator unit to generate synthetic texture data blocks;

[0048] Input the synthesized texture data block and the real color data block of the left view into the discriminator unit, and output the texture distortion probability distribution map;

[0049] The color interpolation kernel function parameters of the generator unit are corrected in reverse based on the texture distortion probability distribution map, and the texture output of the 3D texture model is updated.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] The depth data acquisition module uses a binocular camera array to simultaneously acquire left and right views, generating a raw depth image dataset that provides comprehensive foundational data for subsequent processing. The depth data correction module addresses outliers in the raw depth images by combining outlier detection with aggregated boundary analysis. It extracts the discrete distribution of the depth data and replaces outlier depth values ​​based on confidence interval thresholds, effectively eliminating erroneous data caused by environmental interference or equipment errors, making the depth data more closely reflect the actual shape of the object.

[0052] The point cloud topology construction module generates a 3D point cloud spatial topology structure based on the corrected depth data and precisely associates the spatial coordinates of each point cloud with the pixel positions in the left view. This precise association ensures the accuracy of subsequent color projection and avoids texture misalignment caused by position matching deviations. The texture mapping engine module projects the RGB color data of the left view onto the 3D point cloud based on the above mapping relationship. The generated initial texture map can initially present the color and texture features of the object's surface.

[0053] The texture optimization module introduces a generative adversarial network to perform local color consistency verification on the initial texture map. Through the learning and adjustment of the network model, it can effectively improve the problem of uneven color in local areas. Even in areas with complex textures or minor occlusions on the object's surface, it can make the color transition natural, improving the overall coordination of the texture map. Through the synergistic effect of this series of modules, the final output 3D texture model is improved in both morphological accuracy and texture realism, better meeting the needs of high-precision 3D models in fields such as industrial design, digital twins, and cultural relic digitization. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the working principle of the binocular 3D scanning system with texture mapping function described in this invention.

[0055] Figure 2 A flowchart for outlier detection in the depth data correction module;

[0056] Figure 3 A flowchart illustrating how the texture mapping engine module works;

[0057] Figure 4 A flowchart for the texture optimization module to generate anti-aliasing;

[0058] Figure 5 This is a flowchart of the dynamic scanning parameter control module. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 The present invention provides a binocular 3D scanning system with texture mapping function. The system includes: a depth data acquisition module, a depth data correction module, a point cloud topology construction module, a texture mapping engine module, and a texture optimization module.

[0061] The depth data acquisition module simultaneously acquires the left and right views of the target object using a binocular camera array, generating a raw depth image dataset. The depth data correction module performs outlier detection on the raw depth image dataset, extracts the discrete distribution of the depth data through aggregated boundary analysis, and replaces outlier depth values ​​in the discrete distribution based on confidence interval thresholds. The point cloud topology construction module generates a 3D point cloud spatial topology structure based on the corrected depth data and associates the spatial coordinates of each point cloud with the pixel positions in the left view. The texture mapping engine module projects the RGB color data of the left view onto the 3D point cloud spatial topology structure based on the pixel position mapping relationship, generating an initial texture map. The texture optimization module performs local color consistency verification on the initial texture map using a generative adversarial network, outputting an optimized 3D texture model.

[0062] Example 1: See Figure 2 The depth data correction module first sorts and organizes the acquired depth data during the processing of the raw depth image dataset. The raw depth data is acquired synchronously by a binocular camera array and contains depth information of the target object in the left and right views. This data is stored in matrix form, with each pixel corresponding to a depth value. To detect outliers, the depth data is sorted in ascending order of numerical value, forming an ordered depth sequence. In this sequence, the values ​​at the 25th percentile are extracted as the lower quartiles of the depth distribution, and the values ​​at the 75th percentile are extracted as the upper quartiles. These two quartiles are used to describe the central tendency and dispersion of the depth data.

[0063] The difference between the upper and lower quartiles is defined as the depth aggregation interval, which reflects the overall distribution range of the depth data. Based on the depth aggregation interval, the boundaries of the depth confidence interval are set, typically using a fixed multiple of this interval as the threshold. For example, if the depth aggregation interval is D, the confidence interval boundaries can be set as the lower quartile minus k times D, and the upper quartile plus k times D, where k is an adjustable coefficient. This coefficient can be adjusted according to actual application needs to adapt to the depth data distribution characteristics of different scenarios.

[0064] After setting the confidence interval boundaries, the system iterates through the original depth image dataset, checking each depth value one by one to see if it exceeds the confidence interval range. If a depth value is below the lower boundary or above the upper boundary, it is marked as invalid data. Invalid data is imputed using a weighted average of adjacent valid depth values, with weights calculated based on pixel distance or depth similarity. For example, for an invalid depth point, several surrounding valid depth points can be selected, assigned different weights based on spatial distance, and the final weighted average is calculated as the imputed value. This process effectively reduces noise and outliers in the depth data, improving the accuracy of subsequent 3D reconstruction.

[0065] The point cloud topology construction module generates a 3D point cloud spatial topology structure based on the corrected depth data. This process first establishes 3D spatial mesh generation rules, dividing the target object's surface into several mesh units. The mesh unit division is based on the target object's geometric features, particularly its curvature variations. High-curvature regions, such as edges and corners, have significant curvature variations and require denser meshes to improve detail reproduction. Low-curvature regions, such as flat surfaces, have smaller curvature variations and can use sparser meshes to reduce computational complexity.

[0066] Each mesh cell corresponds to a set of topological nodes, and the node density is dynamically adjusted according to the type of mesh cell. Mesh cells in high-curvature regions are assigned high-density topological nodes to ensure that the geometric features of the details can be accurately represented. Mesh cells in low-curvature regions are assigned lower-density topological nodes to reduce redundant calculations. The spatial coordinates of the topological nodes are directly generated from the corrected depth data, and each node corresponds to a three-dimensional spatial point.

[0067] After generating the set of spatial coordinates of the topological nodes, the system further constructs a spatial connectivity matrix between adjacent nodes. This matrix describes the adjacency relationships between nodes and is used for subsequent point cloud surface reconstruction and texture mapping. For example, if two nodes are less than a set threshold apart in 3D space, they are marked as adjacent nodes in this matrix. This relationship matrix helps optimize the structure of the point cloud data, making it more consistent with the surface continuity of real objects.

[0068] The system maps the pixel coordinates of the left view to the set of spatial coordinates of the topological nodes, forming a pixel position mapping index table. This index table records the left view pixel coordinates corresponding to each topological node, enabling subsequent texture mapping to accurately project the color information of the 2D image onto the 3D point cloud structure. The mapping process is based on camera calibration parameters and geometric transformation relationships to ensure accurate correspondence between 2D pixels and 3D spatial points.

[0069] Example 2: See Figure 3 The texture mapping engine module is responsible for accurately mapping color information from a 2D image onto a 3D point cloud structure. This process involves three key steps: color projection, texture blending, and lighting compensation. Through sophisticated algorithm design, this module ensures that the final generated texture map accurately reflects the visual features of the object's surface.

[0070] The color projection unit operates based on a pixel position mapping index table. This index table, generated by the point cloud topology construction module, records in detail the coordinate correspondence between each 3D topology node and its corresponding pixel in the left view. During projection, the system precisely maps the RGB color data of the left view according to the spatial coordinates of the topology nodes based on this correspondence. Due to the viewing angle differences during binocular camera acquisition, direct projection may result in missing or misaligned color information in some areas. To address this issue, the system employs a multi-view color compensation mechanism. When a topology node is occluded in the left view, it automatically references the corresponding area in the right view for color compensation.

[0071] The primary task of the texture blending unit is to eliminate potential discontinuities during color projection. When adjacent topological nodes have significant color differences, direct mapping can lead to noticeable seams and discontinuous areas. The system employs a bilinear interpolation algorithm to smooth these transition areas. This algorithm first identifies combinations of adjacent nodes with significant color changes and then inserts transition color values ​​between these nodes. The calculation of these transition values ​​comprehensively considers the color distribution characteristics of multiple surrounding nodes, resulting in a natural gradient effect in the final texture. For high-curvature regions, due to the high density of topological nodes, the system appropriately increases the number of interpolation sampling points to ensure color continuity across complex geometric structures.

[0072] The illumination compensation unit aims to eliminate the impact of uneven ambient lighting on texture quality. Because the lighting conditions in the acquisition environment may vary locally, directly projected color data often contains unnecessary variations in brightness. The system establishes an illumination influence model by analyzing the illumination intensity distribution characteristics of the left view. This model first decomposes the image into basic color components and illumination components, and then normalizes the illumination components. During projection, the system dynamically adjusts the color value of each node based on the spatial orientation information of the 3D point cloud, ensuring that the final texture exhibits consistent brightness across different viewing angles. For highly reflective or shadowed areas, the system combines data from multiple frames for comprehensive compensation to prevent illumination anomalies in a single frame from affecting the overall texture quality.

[0073] In the specific implementation of texture mapping, the system adopts a layered processing strategy. First, coarse-grained color projection is performed on the overall point cloud structure to quickly establish a basic texture framework. Then, fine-grained processing is performed on detailed areas, especially high-frequency feature areas such as edges and corners, which undergo multiple iterations for optimization. This layered approach ensures both processing efficiency and texture quality in key areas. The system also maintains a texture quality evaluation mechanism to monitor the mapping effect of each region in real time, automatically triggering a reprojection process for areas that do not meet the quality standards.

[0074] For texture mapping of dynamic objects, the system introduces a temporal consistency maintenance mechanism. When processing consecutive frames of 3D scan data, the system records the variation patterns of texture features between adjacent frames and makes predictive adjustments during the mapping process of new frames, referencing these patterns. This method effectively reduces inter-frame texture flickering, making the texture representation of dynamic objects more stable and natural. For fast-moving regions, the system automatically increases the sampling frequency and uses motion compensation algorithms to correct texture distortion caused by object displacement.

[0075] In terms of hardware acceleration, the texture mapping engine fully leverages the parallel computing capabilities of modern GPUs. It organizes the topological data, color projection matrix, and lighting compensation parameters of the 3D point cloud into a data structure that facilitates efficient GPU processing. Particularly for computationally intensive operations such as bilinear interpolation, the system employs a dedicated shader program for optimization, significantly improving the texture mapping speed for large-scale point clouds. Simultaneously, the system supports dynamic load balancing, automatically allocating computing resources based on the point cloud density distribution to ensure that high-density areas receive more processing power.

[0076] The texture mapping engine also includes an adaptive parameter adjustment subsystem. This system monitors various quality metrics during texture generation in real time, including color consistency, edge sharpness, and lighting uniformity. Based on the monitoring results, it dynamically adjusts key parameters such as projection parameters, interpolation coefficients, and compensation intensity. This adaptive mechanism enables the system to adapt to the texture mapping needs of different materials and lighting conditions, maintaining stable output quality. For special materials such as transparent or highly reflective surfaces, the system activates a dedicated material processing mode, employing physically based rendering techniques to enhance the realism of the texture.

[0077] In the final texture output stage, the system performs a comprehensive quality check. This check includes the integrity of color channels, the correctness of texture coordinates, and the accuracy of matching with the geometric model. Areas with defects are automatically repaired using methods such as neighbor interpolation or multi-view information fusion. After all processing is complete, the system outputs an optimized 3D texture model that retains the visual details of the original image while perfectly fitting the 3D geometric structure.

[0078] Example 3: See Figure 4 The core function of the texture optimization module is to perform local color consistency verification on the initial texture map using a generative adversarial network (GAN) to eliminate texture distortion or color deviation that may occur during projection. This module consists of a generator unit, a discriminator unit, and an optimization feedback unit, and gradually improves the texture quality of the 3D texture model through an iterative optimization mechanism.

[0079] The generator unit receives the 3D color distribution data of the initial texture map, with its input being the set of RGB values ​​associated with topological nodes. This unit employs a convolutional neural network (CNN) with an encoder-decoder structure to extract multi-scale color features. The encoder consists of four convolutional layers, each with a 3×3 kernel size and a stride of 2, using the LeakyReLU activation function. The encoding process progressively reduces the spatial resolution of the feature map while increasing the number of channels, ultimately generating a 128-dimensional latent spatial feature vector. The decoder progressively restores the spatial resolution through transposed convolutional layers, outputting a synthetic texture sample with the same size as the input. The generator's loss function includes two terms: content loss and adversarial loss. The content loss uses the L1 norm to measure the difference between the synthetic sample and the real color data, while the adversarial loss is calculated using the feedback signal from the discriminator.

[0080] The discriminator unit consists of a five-layer convolutional network. The input is a 64×64 pixel local texture patch, and the output is a texture distortion probability matrix. The discriminator first downsamples the input patch, followed by batch normalization and LeakyReLU activation after each convolutional layer. The final convolutional layer outputs a single-channel feature map, which is mapped to a probability value between 0 and 1 using the sigmoid function, representing the likelihood of texture distortion in the corresponding region. The training objective of the discriminator is to minimize the misclassification rate of real texture patches while maximizing the recognition accuracy of synthetic texture patches.

[0081] The optimization feedback unit dynamically adjusts the generator parameters based on the distortion probability matrix output by the discriminator. Each element in the probability matrix... Indicates coordinates as The probability of texture block distortion is calculated as follows:

[0082] ;

[0083] in, This represents the output value of the discriminator. The input is a texture patch. The feedback unit calculates the gradient of the generator parameters using the backpropagation algorithm, and the gradient update amount... It is determined by the following rules:

[0084] During the iterative optimization process, the generator and discriminator are updated alternately. After each iteration, the optimization feedback unit resamples the local region of the 3D texture model and calculates a new distortion probability matrix. When the mean change of the probability matrix over three consecutive iterations is less than a preset tolerance... When the model converges, the optimization is terminated.

[0085] The adversarial training mechanism adaptively locates texture distortion regions without requiring a predefined mathematical model for color deviation; the generator's multi-scale feature extraction capability can simultaneously handle high-frequency details and low-frequency color transitions; and the dynamic weight adjustment strategy of the optimized feedback unit balances the needs of texture restoration and feature preservation. The final output 3D texture model maintains geometric accuracy while exhibiting higher visual realism.

[0086] Example 4: The core function of the dynamic scanning parameter control module is to dynamically adjust the acquisition parameters of the binocular camera based on the spatial characteristics of the 3D point cloud to optimize the quality of depth data and scanning efficiency. This module analyzes the point cloud stability coefficient, retrieves the optimal camera sampling step size from a predefined parameter configuration space, and feeds it back to the depth data acquisition module to achieve adaptive scanning control.

[0087] The point cloud stability coefficient is calculated based on the curvature distribution and the depth data aggregation interval. The curvature distribution reflects the geometric complexity of the object's surface. Local curvature values ​​are obtained by calculating the mean principal curvature at each topological node, and then the curvature variance of the entire point cloud is statistically analyzed. The depth data aggregation interval characterizes the dispersion of the depth data and is determined by the difference between the upper and lower quartiles. The stability coefficient is defined as the normalized product of the curvature variance and the depth aggregation interval. A larger value indicates that the point cloud data in the current scanned area is more unstable, requiring adjustment of camera parameters to improve data quality. The scan parameter configuration space stores historical optimal parameter combinations under different scan conditions, and its structure is shown in the table below.

[0088] Table 1: Historical optimal parameter combination structure under different scanning conditions.

[0089]

[0090] During the scanning process, the dynamic scanning parameter control module monitors the curvature variance, depth aggregation spacing, and ambient light intensity standard deviation of the current point cloud in real time, and matches the optimal parameters in the configuration space. The matching process uses a multi-dimensional interval query algorithm. First, it locates the parameter subspace based on the curvature variance and depth aggregation spacing, and then combines the lighting conditions to filter the final optimal parameter set. For example, when the curvature variance is detected to be 0.35, the depth aggregation spacing to be 3.2mm, and the ambient light intensity standard deviation to be 80lux, the module will match the second row of parameter sets, set the sampling step size to 1.5mm, adjust the exposure time to 15ms, and increase the focal length by 5%.

[0091] The scan path planning module generates the camera's motion trajectory based on the optimal sampling step size. For highly stable regions (such as flat surfaces), a larger sampling step size is used to improve scanning efficiency; for low-stability regions (such as areas with high curvature or complex textures), a smaller sampling step size is used to ensure data accuracy. The trajectory planning algorithm is based on B-spline curve interpolation, ensuring smooth camera movement and coverage of all critical areas.

[0092] Dynamic compensation for ambient light intensity is achieved by analyzing the brightness histogram of the left view. The module calculates the mean and standard deviation of the brightness of the current frame. If a sudden change in illumination (such as shadows or reflections) is detected, fine-tuning of the exposure time and gain parameters is triggered. The compensation strategy prioritizes ensuring exposure uniformity in high-curvature areas to avoid abrupt changes in depth data due to illumination variations.

[0093] The historical parameter configuration space enables data-driven adaptive adjustment, avoiding oversampling or undersampling problems caused by relying on fixed parameters; the multi-dimensional parameter matching mechanism can simultaneously take into account the effects of geometric complexity, data dispersion, and ambient lighting; dynamic exposure compensation and focus adjustment work together to ensure data consistency under different scanning conditions.

[0094] Example 5: See Figure 5 The dynamic scanning parameter adjustment module achieves fine-tuning of camera sampling parameters by constructing a three-dimensional parameter space coordinate system. This coordinate system uses the variance of point cloud curvature distribution as the horizontal axis, the depth data aggregation interval as the vertical axis, and the standard deviation of ambient light intensity as the vertical axis, forming a three-dimensional parameter mapping structure. Each discrete point in the coordinate system is associated with a set of historical scanning parameters and a corresponding point cloud quality score, which is calculated based on a combination of geometric consistency and texture sharpness.

[0095] The parameter configuration space is constructed using an incremental update strategy. The initial space is filled with scan data from standard test scenarios, containing parameter combinations under different curvature, depth dispersion, and lighting conditions. As actual scanning tasks are performed, the system continuously records the optimal parameters for new scenarios and dynamically expands the configuration space. New data points undergo outlier detection to exclude abnormal records caused by equipment malfunctions or environmental interference. The distribution density of parameter points within the space reflects common operating conditions in real-world applications, with high-density areas corresponding to frequently occurring scanning conditions.

[0096] The calculation of point cloud stability coefficients integrates three factors: curvature distribution, depth discretization, and illumination fluctuations. The curvature distribution variance is calculated using a moving window method to statistically determine the rate of change of local surfaces, with the window size adaptively adjusted according to the object size. The depth aggregation spacing employs a dynamic interquartile range algorithm to avoid misjudging complex geometries using a fixed threshold. The illumination intensity standard deviation is calculated based on the brightness channel of the left view, with the sampling area strictly aligned with the depth data acquisition field of view. The stability coefficient is ultimately normalized to a scalar value between 0 and 1; a higher value indicates more complex current scanning conditions.

[0097] The target parameter subspace is located using a hierarchical search strategy. First, the main search level is determined based on stability coefficients: coefficients between 0 and 0.3 correspond to simple conditions, 0.3 to 0.7 to medium-complexity conditions, and coefficients above 0.7 to high-difficulty conditions. Each main level is further divided into secondary subspaces according to the weighted proportions of curvature, depth, and illumination. An octree spatial index structure is used to accelerate the search process, and bounding box collision detection quickly eliminates irrelevant parameter regions.

[0098] The neighborhood density clustering algorithm operates within the target subspace, and the clustering features include parameter variation trends and operational continuity. The algorithm first calculates the Mahalanobis distance between each parameter point, excluding isolated points with excessively large statistical differences. The core clustering process employs the adaptive radius DBSCAN method, with the density threshold dynamically adjusted based on the sparsity of the points within the subspace. The center point of each cluster represents a typical parameter configuration for that operational condition, and the optimal parameters are ultimately determined through a weighted voting mechanism. The voting weights are determined by both historical scores and recent usage frequency, with parameters that have been successfully applied recently receiving higher weights.

[0099] The parameter feedback mechanism employs a gradual adjustment strategy. When a change in scanning conditions is detected, the system does not immediately switch to a completely new parameter set, but adjusts it in stages according to the magnitude of the change. For gradual changes in ambient lighting, the exposure parameters transition gradually using linear interpolation; for sudden changes in geometric complexity, the sampling step size is adjusted using an exponential decay method. This mechanism avoids discontinuous data acquisition caused by parameter jumps, while giving the algorithm sufficient time to converge to the optimal state.

[0100] The environmental adaptability enhancement module continuously monitors the effect of parameter execution. After each parameter adjustment, the system collects verification frames and calculates the actual point cloud quality indicators. If the indicators are lower than the expected threshold, the parameter backtracking mechanism is triggered, automatically restoring the parameter set to the previous stable state, and marking the current operating condition as an area requiring special attention.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A binocular 3D scanning system with texture mapping function, characterized in that, include: The depth data acquisition module is used to simultaneously acquire the left and right views of the target object through a binocular camera array and generate the original depth image dataset. The depth data correction module is used to perform outlier detection on the original depth image dataset. It extracts the discrete distribution of depth data through aggregated boundary analysis and replaces the abnormal depth values ​​in the discrete distribution based on the confidence interval threshold. The point cloud topology construction module is used to generate a 3D point cloud spatial topology structure based on the depth data output by the depth data correction module, and associate the spatial coordinates of each point cloud with the pixel position mapping relationship of the left view. The texture mapping engine module is used to project the RGB color data of the left view onto the 3D point cloud spatial topology based on the pixel position mapping relationship to generate the initial texture map; The texture optimization module is used to perform local color consistency verification on the initial texture map through an adversarial generative network and output an optimized 3D texture model. The specific steps for the depth data correction module to perform abnormal depth value detection include: The original depth image dataset is sorted in ascending order of depth value. Twenty-five percent of the position values ​​in the depth sequence are extracted as the lower quartiles of the depth distribution, and seventy-five percent of the position values ​​are extracted as the upper quartiles of the depth distribution. The numerical difference between the upper and lower quartiles is calculated and defined as the depth aggregation interval. The depth confidence interval boundary is set based on the multiple relationship of the depth aggregation interval. Traverse the original depth image dataset, mark depth values ​​that exceed the depth confidence interval as invalid data, and fill in the invalid data with the weighted average of adjacent depth values; The texture mapping engine module includes: The color projection unit is used to project the RGB color data of the left view onto the three-dimensional point cloud spatial topology according to the topological node coordinates based on the pixel position mapping relationship index table. The texture blending unit is used to perform bilinear interpolation calculations on the color data of adjacent topological nodes to eliminate color banding during the projection process. The illumination compensation unit is used to analyze the ambient light intensity distribution of the left view and perform normalization compensation processing on the projected color data based on the light intensity distribution. The texture optimization module includes: The generator unit is used to receive the 3D color distribution data of the initial texture map and generate adversarial synthetic texture samples; The discriminator unit is used to compare the synthetic texture sample with the real color features of the left view and output the local texture distortion probability matrix. The optimized feedback unit is used to adjust the color interpolation weight parameters of the generator unit in reverse according to the local texture distortion probability matrix, and iteratively update the 3D texture model. The system also includes: The dynamic scanning parameter control module is used to calculate the point cloud stability coefficient based on the curvature distribution data of the three-dimensional point cloud spatial topology. The scanning path planning module is used to retrieve the optimal binocular camera sampling step size from the scanning parameter configuration space, using the point cloud stability coefficient as an index. The optimal sampling step size of the binocular camera is fed back to the depth data acquisition module, and the exposure time and focal length parameters of the binocular camera array are dynamically adjusted.

2. The binocular 3D scanning system with texture mapping function according to claim 1, characterized in that, The operational logic of the point cloud topology construction module includes: Establish three-dimensional spatial meshing rules, divide the surface of the target object into high curvature region mesh units and low curvature region mesh units, and assign high-density topological nodes to the high curvature region mesh units; Based on the corrected depth data, a set of spatial coordinates of topological nodes is generated, and a spatial connection matrix of adjacent topological nodes is constructed. The pixel coordinates of the left view are mapped to the spatial coordinate set of the topological nodes to form a pixel position mapping index table.

3. A binocular 3D scanning system with texture mapping function according to claim 1, characterized in that, The operating logic of the dynamic scanning parameter control module includes: Construct a three-dimensional parametric space coordinate system, where the horizontal axis represents the variance of the point cloud curvature distribution, the vertical axis represents the depth data aggregation interval, and the vertical axis represents the standard deviation of ambient light intensity. Establish a scanning parameter configuration space in a three-dimensional parameter space coordinate system, and label the corresponding historical camera sampling step size for each parameter coordinate point; Based on the current point cloud stability coefficient, the target parameter subspace is located in the scanning parameter configuration space, and the optimal camera sampling step size in the target parameter subspace is retrieved by the neighborhood density clustering algorithm.

4. A binocular 3D scanning method with texture mapping function, applied to a binocular 3D scanning system with texture mapping function as described in any one of claims 1 to 3, characterized in that, Includes the following steps: Simultaneously acquire the left and right views from the stereo camera, generate the original depth image dataset, and perform depth data discrete distribution correction; A 3D point cloud spatial topology is constructed based on the corrected depth data, and an index is established to map the topology node coordinates to the pixels in the left view. Based on the mapping relationship index, the RGB color data of the left view is projected onto the three-dimensional point cloud spatial topology to generate an initial three-dimensional texture map; By performing local color distortion detection on the initial 3D texture map using an adversarial generative network, an optimized 3D texture model is generated.

5. A binocular 3D scanning method with texture mapping function according to claim 4, characterized in that, The specific steps for generating the initial 3D texture map include: Identify high-curvature region mesh cells in the spatial topology of a 3D point cloud and assign additional topology nodes to these high-curvature region mesh cells; Spatial remapping of the left view pixel coordinates according to the additional topological node coordinates generates an enhanced pixel position mapping index; RGB color data projection is performed based on the enhanced pixel location mapping relationship index, and Gaussian filtering color smoothing is performed between adjacent additional topology nodes.

6. A binocular 3D scanning method with texture mapping function according to claim 5, characterized in that, The operation process of the adversarial generative network includes: Extract the local color feature vectors of the 3D texture model and input them into the generator unit to generate synthetic texture data blocks; Input the synthesized texture data block and the real color data block of the left view into the discriminator unit, and output the texture distortion probability distribution map; The color interpolation kernel function parameters of the generator unit are corrected in reverse based on the texture distortion probability distribution map, and the texture output of the 3D texture model is updated.

Images