Underground mine three-dimensional real scene modeling method fusing multi-source information
By employing a comprehensive approach that combines image enhancement, multi-source pose optimization, and 3D model optimization, the accuracy and efficiency issues of 3D reconstruction in underground mining environments are resolved, generating high-precision 3D reality models suitable for mining digitization and intelligent applications.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional 3D reconstruction methods suffer from insufficient accuracy, low efficiency, and poor adaptability in underground mining environments. They are particularly inadequate in image processing, pose estimation, and 3D modeling, and struggle to effectively handle low illumination, high noise, monotonous textures, and complex spatial structures.
A comprehensive approach combining image enhancement, multi-source pose optimization, and 3D model optimization is employed, including image enhancement with adaptive illumination correction and homomorphic filtering, pose estimation based on multi-source information fusion, and 3D Gaussian model optimization based on density clustering and energy minimization, to generate high-quality 3D reality models.
It improves the accuracy and efficiency of 3D reality modeling of underground mines, effectively copes with complex mining environments, generates high-precision 3D models, and provides technical support for the digital and intelligent application of mines.
Smart Images
Figure CN121661271A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of computer vision, computer graphics, and mine digitization technology, specifically a method for three-dimensional real-scene modeling of underground mines that integrates multi-source information. Background Technology
[0002] Underground mining environments are extremely complex, presenting numerous challenges such as the lack of RTK signals, poor lighting, abundant dust, monotonous texture features, and repetitive spatial structures. These challenges pose significant difficulties for 3D reality modeling. Traditional 3D reconstruction methods often exhibit shortcomings in such environments, including insufficient accuracy, low efficiency, and poor adaptability.
[0003] In the field of image processing, images acquired in mining environments generally suffer from significant problems such as low illumination, high noise, and blurred details. Although existing techniques such as histogram equalization and contrast stretching have been widely used, in special scenarios with low light and strong dust, these techniques often lead to improper global pixel distribution adjustment, resulting in overexposure in bright areas and amplification of noise in dark areas, causing insufficient adaptability and distortion of details.
[0004] In pose estimation, traditional methods mainly rely on a single data source, failing to fully utilize the advantages of multi-source information fusion. Visual feature methods are prone to failure when the texture is simple; IMU methods suffer from cumulative errors; and control point methods require manual placement and have uneven distribution.
[0005] In the field of 3D modeling, although 3D Gaussian Splatting technology can achieve high-quality rendering effects, it still needs to be optimized to improve its redundancy and applicability after being applied to 3D real-world modeling of complex mining environments.
[0006] Therefore, there is an urgent need for a comprehensive 3D modeling technology solution that can integrate mine image enhancement, multi-source pose estimation, and 3D modeling optimization to improve the accuracy, efficiency, and practicality of 3D reality modeling of underground mines. Summary of the Invention
[0007] The main objective of this invention is to address the problems existing in the prior art and provide a method for three-dimensional real-scene modeling of underground mines that integrates multi-source information. This method achieves high-quality and high-precision three-dimensional real-scene modeling of mines by integrating image enhancement, multi-source pose optimization, and model optimization in three-dimensional reconstruction.
[0008] The technical solution of the present invention is as follows: a method for three-dimensional real-scene modeling of underground mines that integrates multi-source information, comprising the following steps: S1, data acquisition and preprocessing: synchronously acquiring image data, IMU data and control point information of the mine environment. The process involves several steps: S1) Preliminary processing of the acquired raw image data, including removing obviously blurred or overexposed images, performing corrections, and format conversion; S2) Image enhancement processing: using an image enhancement method based on adaptive illumination correction and homomorphic filtering to improve image quality; S3) Camera pose optimization: using a pose optimization method based on multi-source information fusion to accurately estimate the camera pose; S4) 3D Gaussian model initialization: initializing a 3D Gaussian distribution using the optimized pose and enhanced image; S5) 3D Gaussian model optimization: optimizing the 3D Gaussian model based on the principles of density clustering and energy minimization to accurately adapt to the spatial density differences in the mine, focusing on key structures, suppressing redundant information, and effectively balancing the accuracy and computational efficiency of the 3D model; S6) Model construction and fusion: combining the optimized 3D Gaussian model with other mine data to construct a comprehensive 3D reality model; simultaneously, fusing the 3D Gaussian model with ore body models, equipment models, etc.; S7) Model update and application: updating the model based on real-time data and applying it to scenarios such as mine safety monitoring and resource management.
[0009] Furthermore, the data acquisition in step S1 includes: acquiring image data using a color panoramic camera, acquiring inertial measurement data using an IMU, acquiring control point data using a total station, and recording the timestamps of each data source for synchronization.
[0010] Further, in step S2, the image enhancement processing includes: converting the image from the BGR color space to the HSV color space and separating the luminance channel V; performing adaptive gamma correction on the luminance channel V; performing homomorphic filtering on the corrected luminance channel; using guided filtering to enhance details and suppress noise on the homomorphically filtered luminance channel; adaptively adjusting the saturation channel S in the HSV color space according to the degree of luminance correction; merging the processed H, S, and V channels and converting them back to the BGR color space to obtain the enhanced image.
[0011] Furthermore, in step S3, the construction and solution of the optimization model includes: using camera pose as the optimization variable, constructing an objective function using visual reprojection error, IMU measurement error, and control point projection error; dynamically adjusting the weight parameters of the corresponding error terms in the objective function according to the confidence level of each data source; and solving the nonlinear least squares problem corresponding to the objective function to obtain the optimized camera pose.
[0012] Furthermore, step S4 includes: generating an initial point cloud using the optimized camera pose and enhanced image through a motion reconstruction method; initializing a 3D Gaussian distribution based on the initial point cloud, where each Gaussian distribution contains parameters such as position, rotation, scaling, opacity, and color.
[0013] Furthermore, in step S5, the optimization of the 3D Gaussian model includes: dividing the mine scene into densely populated areas and open areas using a density clustering algorithm; constructing an energy function containing data fitting terms and redundancy suppression terms; iteratively adjusting the position, scaling, opacity, and color parameters of each Gaussian element in the 3D Gaussian model based on the gradient of the energy function; periodically pruning redundant Gaussian elements whose contribution to the model is below a preset threshold during the iteration process; and stopping the optimization when the energy function converges or reaches a preset number of iterations, thereby obtaining the optimized 3D Gaussian model.
[0014] Furthermore, the model construction and fusion in step S6 includes: combining the optimized 3D Gaussian model with other mine geological data to construct a comprehensive three-dimensional real-scene model; at the same time, fusing the 3D Gaussian model with ore body models, equipment models, etc.
[0015] Furthermore, the model update and application in step S7 includes: regularly collecting new mine geological data, real-time sensor data, model data, etc., and updating the 3D model; applying the model to scenarios such as mine safety monitoring, resource management, and mining planning; and supporting virtual reality and augmented reality visualization.
[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for three-dimensional real-scene modeling of underground mines that integrates multi-source information as described above.
[0017] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for three-dimensional real-scene modeling of underground mines that integrates multi-source information as described above.
[0018] Compared with the prior art, the present invention has the following advantages: 1. It provides a complete technical process, which systematically integrates image enhancement, robust multi-source pose optimization and adaptive 3D model optimization technology for the mining environment, and solves the bottleneck problem of single technical links.
[0019] 2. An image enhancement method combining adaptive gamma correction and homomorphic filtering effectively improves the visual quality of mine images with low illumination and uneven lighting, suppresses noise, enhances details, and provides high-quality input for subsequent processing.
[0020] 3. A pose optimization method based on multi-source information fusion and dynamic weight adjustment is proposed, which comprehensively utilizes the advantages of vision, IMU and control points, makes up for the shortcomings of single sensor, and significantly improves the accuracy of camera pose estimation and system robustness.
[0021] 4. Innovatively, density clustering and energy minimization are introduced into the optimization process of the 3D Gaussian splash model. It can adaptively optimize the distribution of Gaussian elements according to the density differences of the spatial structure of the mine, effectively suppressing redundancy while preserving key structural details, and achieving a good balance between model accuracy and computational efficiency.
[0022] 5. The method has strong adaptability and practicality, and can effectively cope with the complex and harsh working environment of underground mines. The generated model has high accuracy and reasonable data volume, providing a reliable technical foundation for the digital and intelligent application of mines. Attached Figure Description
[0023] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation
[0024] like Figure 1 As shown, this embodiment provides a method for 3D real-scene modeling of underground mines that integrates multi-source information, specifically including the following steps: S1, Data Acquisition and Preprocessing. Image data, IMU data, and control point information of the mine environment are acquired simultaneously. Image data is acquired using a color camera, inertial measurement data is acquired using an IMU, and control point data is acquired using a total station. All data sources are synchronized and calibrated in time to ensure data consistency. The acquired raw image data undergoes preliminary processing. First, based on image sharpness evaluation functions (such as Laplacian variance) and brightness histograms, obviously blurred or overexposed / underexposed images are automatically removed. Second, image distortion correction is performed using camera calibration parameters. Finally, the images are uniformly converted to a preset format and resolution.
[0025] S2. Image Enhancement Processing. An image enhancement method based on adaptive illumination correction and homomorphic filtering is used to process the mine image. Specifically, this includes: converting the image from the BGR color space to the HSV color space and separating the luminance channel (V channel); performing adaptive gamma correction on the luminance channel, dynamically adjusting the gamma value according to the image's luminance distribution characteristics; performing homomorphic filtering on the corrected luminance channel to suppress low-frequency illumination unevenness and enhance high-frequency details; using guided filtering for detail enhancement and noise suppression; adaptively adjusting the saturation channel according to the degree of luminance correction; and merging the processed channels and converting them back to the BGR color space to obtain the enhanced image.
[0026] S3. Camera Pose Optimization. A pose optimization method based on multi-source information fusion is used to accurately estimate the camera pose. A graph optimization model is constructed, with camera pose as the optimization variable and visual reprojection error, IMU measurement error, and control point projection error as constraints. The optimization objective function is:
[0027]
[0028] Where xi is the coordinate of the image feature point, Pi is the camera projection function of the i-th frame, X is the coordinate of the point in three-dimensional space, qi is the pose calculated by the IMU, Qi is the pose transformation function from IMU to camera, cj is the projected coordinate of the control point in the image, c'j is the actual coordinate of the control point, and λ1, λ2, and λ3 are adaptive weight parameters that are dynamically adjusted according to the following rules: the visual weight λ1 is positively correlated with the number of feature points successfully matched in each frame (e.g., λ1=1.0 when the number of feature points > 500; λ1=0.7 when 100-500; λ1=0.3 when <100); the IMU weight λ2 decreases when the variance of the accelerometer and gyroscope readings is greater than a threshold (e.g., 0.1) (e.g., λ2=0.5), and increases when it is less than a threshold (e.g., λ2=1.0); the control point weight λ3 is positively correlated with the number of identified control points and their uniformity of distribution in the image (e.g., calculated by entropy). The nonlinear least squares problem is solved using the Levenberg-Marquardt algorithm or the Gauss-Newton algorithm to obtain the optimized camera pose.
[0029] S4. 3D Gaussian Model Initialization. Using the optimized camera pose and enhanced images, an initial point cloud is generated via the Structure for Motion Reconstruction (SfM) method. Based on this point cloud, a 3D Gaussian splash model is initialized. Each 3D point location is initialized with a Gaussian primitive, its initial position being the point's coordinates. The rotation parameter is initialized to the identity matrix, the scaling parameter is initialized to a small value estimated based on the local density of the point cloud, the opacity is initialized to 0.5, and the color is initialized to the average color of that point across all visible images.
[0030] S5, 3D Gaussian model optimization. An optimization method based on density clustering and energy minimization is employed. First, the mine scene is divided into densely populated and open areas using a density clustering algorithm. Then, an energy function containing data fitting terms and redundancy suppression terms is constructed. Based on the gradient of the energy function, the position, scaling, opacity, and color of Gaussian elements are iteratively adjusted. Redundant Gaussian elements that contribute minimally to the model are periodically pruned until the energy function converges or the required number of iterations is reached, completing the 3D Gaussian model optimization.
[0031] S6. Model Construction and Fusion. Integrate the optimized 3D Gaussian model with other mine data to construct a complete 3D reality model. Fusion the 3D model with geological data, equipment data, etc., to form a complete 3D reality model of the underground mine.
[0032] S7. Model Updates and Applications. Regularly collect new mine geological data, real-time sensor data, and model data to update the 3D model, maintaining its timeliness and accuracy. Apply the model to scenarios such as mine safety monitoring, resource management, and mining planning. Support virtual reality and augmented reality visualization, providing an immersive mine environment experience.
[0033] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above method embodiments.
[0034] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method embodiments.
[0035] Compared with traditional methods, the three-dimensional reconstruction results of this invention in underground tunnel environments show significant improvements in geometric consistency, detail preservation, and visual effects.
[0036] This invention provides a complete 3D reality modeling solution for underground mines by integrating image enhancement, pose optimization and 3D reconstruction technologies. It effectively improves modeling accuracy, efficiency and practicality, and provides technical support for the digitalization and intelligentization of mines.
Claims
1. A method for three-dimensional real-scene modeling of underground mines that integrates multi-source information, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Simultaneously acquire image data, IMU data, and control point information of the mine environment. Perform preliminary processing on the acquired raw image data, including removing obviously blurry or overexposed images, performing corrections, and format conversion; S2. Image Enhancement Processing: An image enhancement method based on adaptive illumination correction and homomorphic filtering is adopted to improve image quality. S3. Camera Pose Optimization: A pose optimization method based on multi-source information fusion is used to accurately estimate the camera pose. S4. 3D Gaussian Model Initialization: The 3D Gaussian distribution is initialized using the optimized pose and enhanced image. S5. 3D Gaussian Model Optimization: The 3D Gaussian model is optimized based on the principles of density clustering and energy minimization to accurately adapt to the spatial density differences in the mine, focusing on key structures, suppressing redundant information, and effectively balancing the accuracy and computational efficiency of the 3D model. S6. Model Construction and Fusion: The optimized 3D Gaussian model is combined with other mine data to construct a comprehensive 3D reality model. Simultaneously, the 3D Gaussian model is fused with ore body models, equipment models, etc. S7. Model Update and Application: The model is updated based on real-time data and applied to scenarios such as mine safety monitoring and resource management.
2. The method for three-dimensional real-scene modeling of underground mines that integrates multi-source information according to claim 1, characterized in that, The data acquisition in step S1 includes: acquiring image data using a color panoramic camera, acquiring inertial measurement data using an IMU, acquiring control point data using a total station, and recording the timestamps of each data source for synchronization.
3. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, In step S2, the image enhancement process includes: converting the image from the BGR color space to the HSV color space and separating the luminance channel V; performing adaptive gamma correction on the luminance channel V; performing homomorphic filtering on the corrected luminance channel; using guided filtering to enhance details and suppress noise on the homomorphically filtered luminance channel; adaptively adjusting the saturation channel S in the HSV color space according to the degree of luminance correction; merging the processed H, S, and V channels and converting them back to the BGR color space to obtain the enhanced image.
4. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, In step S3, the construction and solution of the optimization model includes: using camera pose as the optimization variable, constructing an objective function using visual reprojection error, IMU measurement error and control point projection error; dynamically adjusting the weight parameters of the corresponding error terms in the objective function according to the confidence of each data source; solving the nonlinear least squares problem corresponding to the objective function to obtain the optimized camera pose.
5. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, Step S4 includes: using the optimized camera pose and enhanced image, generating an initial point cloud through the motion reconstruction method; initializing a 3D Gaussian distribution based on the initial point cloud, where each Gaussian distribution contains parameters such as position, rotation, scaling, opacity, and color.
6. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, In step S5, the optimization of the 3D Gaussian model includes: dividing the mine scene into densely populated areas and open areas using a density clustering algorithm; constructing an energy function that includes data fitting terms and redundancy suppression terms; iteratively adjusting the position, scaling, opacity, and color parameters of each Gaussian element in the 3D Gaussian model based on the gradient of the energy function; periodically pruning redundant Gaussian elements whose contribution to the model is below a preset threshold during the iteration process; and stopping the optimization when the energy function converges or reaches a preset number of iterations, thereby obtaining the optimized 3D Gaussian model.
7. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, The model construction and fusion in step S6 includes: combining the optimized 3D Gaussian model with other mine geological data to construct a comprehensive three-dimensional real-scene model; and simultaneously, fusing the 3D Gaussian model with ore body models, equipment models, etc.
8. The method for three-dimensional real-scene modeling of underground mines integrating multi-source information according to claim 1, characterized in that, The model update and application in step S7 includes: periodically collecting new mine geological data, real-time sensor data, model data, etc., and updating the 3D model; applying the model to scenarios such as mine safety monitoring, resource management, and mining planning; and supporting virtual reality and augmented reality visualization.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for three-dimensional real-scene modeling of underground mines that integrates multi-source information as described in any one of claims 1-8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for three-dimensional real-scene modeling of underground mines that integrates multi-source information as described in any one of claims 1-8.