Main inspection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration
By using convex hull adaptive partitioning and CPU-GPU heterogeneous acceleration, the low computational efficiency and massive data processing challenges of the triangulation method in marine mapping are solved, enabling efficient and accurate verification of seabed topographic data and improving processing capabilities and accuracy.
Patent Information
- Application Number
- CN202511062333.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2025-11-14
AI Technical Summary
Among existing marine surveying technologies, the triangulation method suffers from problems such as cumbersome manual processing, low computational efficiency, and inability to handle massive amounts of data, especially in complex terrain areas and large-scale data processing.
A master inspection comparison method based on convex hull adaptive partitioning and CPU-GPU heterogeneous acceleration is adopted. The inspection line coverage area is automatically delineated by the convex hull adaptive partitioning algorithm, and interpolation calculation and cross-difference calculation are performed by combining CPU-GPU heterogeneous parallel computing strategy, which replaces manual operation, optimizes memory usage and utilizes multi-CUDA stream asynchronous transmission.
It achieves intelligent and accurate segmentation of complex scenes, avoids human error, improves the speed of triangulation network construction by 4-10 times, breaks through the memory limitations of traditional software, stably processes more than 100 million deep point clouds, and significantly improves the efficiency of massive data processing.
Smart Images
Figure CN120950466A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine surveying and mapping technology, specifically involving a master inspection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration. Background Technology
[0002] In the field of marine surveying, acquiring high-precision seabed topographic data is crucial for applications such as resource exploration, navigation safety, and marine engineering. Multibeam bathymetry and airborne lidar bathymetry, as mainstream technologies, can efficiently acquire seabed topographic point cloud data. However, the accuracy of this data needs to be verified through a master-check comparison calculation, that is, by comparing the discrepancies at the intersection points of the master survey line and the check line data to assess data quality. The core of this process lies in constructing the topographic surface and calculating the discrepancies at the intersection points; its efficiency and accuracy directly affect the reliability of the surveying results.
[0003] Currently, the main methods for discrepancy calculations rely on two main technical approaches: the regular grid method and the triangular mesh method. The regular grid method, which constructs a regular grid surface for interpolation calculations, is widely used by mainstream international software (such as CARISHIPS / SIPS and Hedermaus). This method uses parameters such as grid spacing and search radius to mesh the bathymetry point cloud and calculate discrepancies at intersection points. However, the regular grid method has a significant drawback: parameter selection can introduce artificial smoothing effects, leading to terrain distortion. This distortion directly affects the accuracy of discrepancy calculations, especially in complex terrain areas (trenches or reef areas).
[0004] In contrast, the triangulation method constructs an irregular triangular network (TIN) based on the original sounding point cloud, performing interpolation calculations directly on the original points. This avoids the smoothing problem of regular grids, thus better preserving terrain integrity. This method has been adopted by the domestic software DepthPro and the international software Qimera, becoming the mainstream technology for main inspection and comparison. The triangulation method generates a triangular mesh through Delaunay triangulation, ensuring that each intersection point can be accurately interpolated within the triangle, resulting in more comprehensive calculation results. Although the triangulation method is superior to the regular grid method in terms of accuracy, it has revealed a series of shortcomings in practical applications:
[0005] Spatial constraints rely on manual intervention: software like DepthPro requires manual user interaction when defining the coverage area of check lines. For example, in non-convex set scenarios, manually defining boundaries is not only cumbersome but also prone to subjective errors, leading to inaccurate region division. This not only reduces efficiency but also affects the reliability of subsequent calculations.
[0006] Inadequate computing architecture performance: DepthPro software was developed in the 1990s and uses a pure CPU serial computing architecture. When processing large-scale data, this architecture cannot fully utilize modern hardware resources, resulting in low computing efficiency and failing to meet real-time or near-real-time processing requirements.
[0007] Poor adaptability to massive data: DepthPro software often crashes due to memory overflow when processing more than 3.65 million depth points. With the development of marine mapping technology, a single mission can collect hundreds of millions of point cloud data, which traditional software cannot handle.
[0008] Functional limitations: Existing software such as Qimera has relatively simple output results, only including in-depth statistical tables, lacking in-depth analysis of discrepancies, which limits the comprehensiveness of data quality assessment.
[0009] To address the above issues, we propose a master check comparison method based on adaptive convex hull partitioning and heterogeneous acceleration. Summary of the Invention
[0010] The purpose of this invention is to address the shortcomings of existing technologies by providing a master inspection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration, which solves the problems of cumbersome manual processing, low computational efficiency, and inability to handle massive amounts of data in traditional methods.
[0011] Currently, when performing master-level comparison calculations on seabed topographic bathymetry data, the triangular mesh method for accuracy verification suffers from cumbersome manual processing, low computational efficiency, and inability to handle massive amounts of data. To address these issues, we propose a master-level comparison method based on convex hull adaptive partitioning and heterogeneous acceleration. In implementing this method, the data in the master survey line folder and the inspection line folder are first converted to a pre-defined data structure; the data is then processed using the convex hull adaptive partitioning algorithm; interpolation calculations and cross-difference calculations are performed using a CPU-GPU heterogeneous parallel computing strategy; finally, statistical analysis and kernel density map visualization are output from the cross-difference files. In this embodiment of the invention, an adaptive convex hull partitioning algorithm is used to replace the manual delineation of regions, achieving intelligent and accurate partitioning of the inspection line coverage area. This solves the problems of complex scenarios such as ring non-convex sets and avoids human error. At the same time, a CPU-GPU heterogeneous parallel computing strategy is used to achieve multi-threaded scheduling and block-based streaming data management, optimizing memory usage and using asynchronous transmission of multiple CUDA streams to hide latency. The triangulation construction speed is 4-10 times faster than DepthPro. Thus, it breaks through the memory limitations of traditional software, can stably process point clouds with depths exceeding 100 million, and significantly improves the processing efficiency of massive amounts of data.
[0012] This invention is implemented as follows: a master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration, the method comprising:
[0013] S10, export the main survey line data and check line data after various corrections to the corresponding main survey line folder and check line folder respectively, and convert the data in the main survey line folder and check line folder into a preset data structure in a unified format.
[0014] S20: Obtain the data in the main test line folder and the inspection line folder after format conversion, and process the data based on the convex hull adaptive subdivision algorithm;
[0015] S30 constructs a triangular mesh model based on the main survey line data after clipping. It adopts a block strategy to divide the overall data of the triangular mesh model into multiple data blocks. It performs interpolation calculation and cross-difference calculation based on the CPU-GPU heterogeneous parallel computing strategy. The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to the storage file.
[0016] S40 performs statistical analysis on cross-difference files and outputs kernel density map visualizations.
[0017] Preferably, the method for processing data based on the convex hull adaptive partitioning algorithm includes:
[0018] S101, load the data in the main test line folder and the inspection line folder after format conversion, and identify the main test line data and the inspection line data;
[0019] S102, Construct a convex hull boundary model of the inspection line based on the inspection line data;
[0020] S103, adaptively trim the main test line data according to the convex hull boundary of the inspection line to obtain trimmed main test line data that matches the inspection line region.
[0021] Preferably, when adaptively clipping the main test line data based on the convex hull boundary of the inspection line, the cv2.convexHull function of OpenCV is used to calculate its convex hull polygon boundary, and then the convex hull polygon boundaries of all inspection lines are integrated as the spatial constraint for clipping the main test line data.
[0022] Preferably, the method for interpolation calculation and cross-difference calculation based on CPU-GPU heterogeneous parallel computing strategy includes:
[0023] S201 constructs a triangulation model based on the clipped main survey line data, and uses a block-based strategy to divide the overall data of the triangulation model into multiple data blocks, and transmits the block-based triangulation model data to the GPU.
[0024] S202, the GPU solves for the coefficients of each triangle through parallel computing;
[0025] S203 streams check line data and uses CUDA streaming technology to achieve asynchronous parallel transmission of check line data.
[0026] S204 uses the KD tree algorithm to spatially index the data points of the inspection line, matches them to the corresponding triangular regions, and performs interpolation calculations and cross-difference calculations.
[0027] S205: The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to a storage file.
[0028] Preferably, when the GPU solves for each triangle coefficient through parallel computing, it loads the triangle index block by block and constructs the coefficient matrix, and asynchronously transmits data such as water depth point coordinates and triangle coefficients to the GPU block by block, reads the water depth point data of the inspection line in a block-stream manner, and starts the kernel function to calculate the interpolated water depth point value and the cross-point discrepancy value.
[0029] Preferably, when performing statistical analysis and kernel density plot visualization on the cross-difference files, the output includes cross-difference intervals, arithmetic mean, absolute mean, mean error, standard deviation, segmented error limit table, cross-difference distribution histogram, absolute value interval statistical table, and kernel density plot.
[0030] Compared with the prior art, the embodiments of this application have the following main advantages:
[0031] In this embodiment of the invention, an adaptive convex hull partitioning algorithm is used to replace the manual delineation of regions, achieving intelligent and accurate partitioning of the inspection line coverage area. This solves the problems of complex scenarios such as ring non-convex sets and avoids human error. At the same time, a CPU-GPU heterogeneous parallel computing strategy is used to achieve multi-threaded scheduling and block-based streaming data management, optimizing memory usage and using asynchronous transmission of multiple CUDA streams to hide latency. The triangulation construction speed is 4-10 times faster than DepthPro. Thus, it breaks through the memory limitations of traditional software, can stably process point clouds with depths exceeding 100 million, and significantly improves the processing efficiency of massive amounts of data.
[0032] In this embodiment of the invention, by constructing a minimum convex hull boundary model for the inspection line data, the algorithm can automatically fit the coverage area of inspection lines of arbitrary shapes, completely replacing the high-cost operation of manually delineating areas. Especially for complex boundary scenarios that are difficult to handle with traditional methods, and by achieving global adaptive clipping through geometric calculations, it solves problems such as easy omissions and blurred boundaries in manual annotation, ensuring the spatial matching accuracy between the main survey line data and the inspection line coverage area, and avoiding subjective errors from manual area delineation from the source. Attached Figure Description
[0033] Figure 1 This is a schematic diagram illustrating the implementation process of the master inspection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration provided by the present invention.
[0034] Figure 2 A schematic diagram of the region obtained by using only the convex hull algorithm for cropping is shown.
[0035] Figure 3 A schematic diagram of the region obtained by cropping using the convex hull adaptive subdivision algorithm is shown.
[0036] Figure 4 The figure shows a comparison of the main detection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration in an embodiment of the present invention with the mesh construction time of DepthPro software.
[0037] Figure 5 The figure shows a comparison between the master detection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration in an embodiment of the present invention and the time taken by DepthPro software to calculate the discrepancy value. Detailed Implementation
[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.
[0039] Currently, when performing master-level comparison calculations on seabed topographic bathymetry data, the triangular mesh method for accuracy verification suffers from cumbersome manual processing, low computational efficiency, and inability to handle massive amounts of data. To address these issues, we propose a master-level comparison method based on convex hull adaptive partitioning and heterogeneous acceleration. In implementing this method, the data in the master survey line folder and the inspection line folder are first converted to a pre-defined data structure; the data is then processed using the convex hull adaptive partitioning algorithm; interpolation calculations and cross-difference calculations are performed using a CPU-GPU heterogeneous parallel computing strategy; finally, statistical analysis and kernel density map visualization are output from the cross-difference files. In this embodiment of the invention, an adaptive convex hull partitioning algorithm is used to replace the manual delineation of regions, achieving intelligent and accurate partitioning of the inspection line coverage area. This solves the problems of complex scenarios such as ring non-convex sets and avoids human error. At the same time, a CPU-GPU heterogeneous parallel computing strategy is used to achieve multi-threaded scheduling and block-based streaming data management, optimizing memory usage and using asynchronous transmission of multiple CUDA streams to hide latency. The triangulation construction speed is 4-10 times faster than DepthPro. Thus, it breaks through the memory limitations of traditional software, can stably process point clouds with depths exceeding 100 million, and significantly improves the processing efficiency of massive amounts of data.
[0040] This invention provides a master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration, such as... Figure 1As shown, the master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration specifically includes:
[0041] S10, export the main survey line data and check line data after various corrections to the corresponding main survey line folder and check line folder respectively, and convert the data in the main survey line folder and check line folder into a preset data structure in a unified format.
[0042] S20: Obtain the data in the main test line folder and the inspection line folder after format conversion, and process the data based on the convex hull adaptive subdivision algorithm;
[0043] It should be noted that the method for processing data based on the convex hull adaptive partitioning algorithm includes:
[0044] S101, load the data in the main test line folder and the inspection line folder after format conversion, and identify the main test line data and the inspection line data;
[0045] S102, Construct a convex hull boundary model of the inspection line based on the inspection line data;
[0046] S103, adaptively clip the main test line data according to the convex hull boundary of the inspection line to obtain clip-main-data of the clipped main test line that matches the inspection line region.
[0047] In this embodiment of the invention, when adaptively clipping the main test line data based on the convex hull boundary of the inspection line, the cv2.convexHull function of OpenCV is used to calculate its convex hull polygon boundary, and then the convex hull polygon boundaries of all inspection lines are integrated as the spatial constraint for clipping the main test line data.
[0048] It should be noted that in the convex hull adaptive partitioning algorithm, the convex hull is usually represented as the sequence of vertices of a minimal convex polygon of a planar point set. Its mathematical definition is: given a planar point set... The convex hull conv(S) is a minimal convex set that satisfies the following conditions:
[0049]
[0050] In computational geometry, a convex hull is typically represented as a vertex sequence H = (h1, h2, ..., h...). m ), where h i ∈S and the polygon boundary satisfies the convexity condition (counterclockwise direction, non-negative cross product):
[0051]
[0052] As defined above, the convex hull algorithm only searches for polygons within convex sets. However, in practical applications, when evaluating the accuracy of inner conformity, the check line may be a non-convex set. For example, if the result surface of the check line is a ring, the generated convex hull will become a circle covering the ring, failing to accurately identify the inner and outer rings. Regardless of the type of ring, it must be composed of multiple convex point sets or can be composed of multiple check lines. Therefore, when constructing a convex hull for a set of check line depth points, constructing the convex hull for each check line separately and then integrating them can avoid the occurrence of non-convex sets.
[0053] Based on the above analysis, a convex hull adaptive subdivision algorithm is proposed: For each inspection line, the cv2.convexHull function of OpenCV (where OpenCV is an open-source computer vision software library, and the cv2.convexHull function is the core function in the OpenCV library used to calculate the convex hull of a point set) is used to calculate its convex hull polygon boundary. Then, the convex hull polygon boundaries of all inspection lines are integrated as the spatial constraints for clip-main-data.
[0054] In this embodiment of the invention, by constructing a minimum convex hull boundary model for the inspection line data, the algorithm can automatically fit the coverage area of inspection lines of arbitrary shapes, completely replacing the high-cost operation of manually delineating areas. Especially for complex boundary scenarios that are difficult to handle with traditional methods, and by achieving global adaptive clipping through geometric calculations, it solves problems such as easy omissions and blurred boundaries in manual annotation, ensuring the spatial matching accuracy between the main survey line data and the inspection line coverage area, and avoiding subjective errors from manual area delineation from the source.
[0055] S30 constructs a triangular mesh model based on the main survey line data after clipping. It adopts a block strategy to divide the overall data of the triangular mesh model into multiple data blocks. It performs interpolation calculation and cross-difference calculation based on the CPU-GPU heterogeneous parallel computing strategy. The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to the storage file.
[0056] S40 performs statistical analysis and kernel density plot visualization on the cross-difference files. The output includes cross-difference intervals, arithmetic mean, absolute mean, mean error, standard deviation, segmented error limit table, cross-difference distribution histogram, absolute value interval statistical table, and kernel density plot.
[0057] In this embodiment of the invention, an adaptive convex hull partitioning algorithm is used to replace the manual delineation of regions, achieving intelligent and accurate partitioning of the inspection line coverage area. This solves the problems of complex scenarios such as ring non-convex sets and avoids human error. At the same time, a CPU-GPU heterogeneous parallel computing strategy is used to achieve multi-threaded scheduling and block-based streaming data management, optimizing memory usage and using asynchronous transmission of multiple CUDA streams to hide latency. The triangulation construction speed is 4-10 times faster than DepthPro. Thus, it breaks through the memory limitations of traditional software, can stably process point clouds with depths exceeding 100 million, and significantly improves the processing efficiency of massive amounts of data.
[0058] In this embodiment of the invention, the method for interpolation calculation and cross-difference calculation based on the CPU-GPU heterogeneous parallel computing strategy specifically includes:
[0059] S201 constructs a triangulation model based on the clipped main survey line data, and uses a block-based strategy to divide the overall data of the triangulation model into multiple data blocks, and transmits the block-based triangulation model data to the GPU.
[0060] It should be noted that this embodiment embeds a CPU-GPU heterogeneous parallel computing strategy into a CPU-GPU heterogeneous collaborative architecture. Within this architecture, all system tasks are categorized into logical operation tasks and numerical operation tasks, with the CPU and GPU undertaking different types of processing tasks, thus forming a collaborative model of "CPU management and scheduling + GPU accelerated computing." Furthermore, to improve the robustness of the method, a KD-tree matching mechanism is introduced into the heterogeneous computing framework, and a block-based streaming processing mechanism is designed. It also optimizes asynchronous data transmission and computational overlap techniques.
[0061] Considering that loading all depth point data onto the GPU during the current process of checking line depth point matching triangle interpolation can easily lead to excessive single-batch memory consumption and process crashes, this embodiment adopts a block-based strategy to divide the overall data into multiple data blocks. The amount of data loaded in each block is dynamically set by the system after obtaining GPU parameters. Furthermore, the KD-tree is used to accelerate the process of matching check line depth points with triangles, significantly improving the processing efficiency of massive amounts of data.
[0062] S202, the GPU solves for the coefficients of each triangle through parallel computing;
[0063] S203 streams check line data and uses CUDA streaming technology to achieve asynchronous parallel transmission of check line data.
[0064] S204 uses the KD tree algorithm to spatially index the data points of the inspection line, matches them to the corresponding triangular regions, and performs interpolation calculations and cross-difference calculations.
[0065] S205: The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to a storage file.
[0066] In this embodiment, when the GPU solves for the coefficients of each triangle through parallel computation, it loads the triangle index block by block and constructs the coefficient matrix. It then asynchronously transmits data such as depth point coordinates and triangle coefficients to the GPU in blocks. Simultaneously, it streams the inspection line depth point data in blocks and starts the kernel function to calculate the interpolated depth point values and the crosspoint discrepancies. By executing multiple CUDA streams alternately, the data communication latency between the CPU and GPU is hidden, achieving time overlap between computation and transmission, thus improving the overall pipeline efficiency.
[0067] When verifying the effectiveness of the master-detector alignment method based on convex hull adaptive subdivision and heterogeneous acceleration provided by this invention, since the master-detector alignment discrepancy calculation module in the domestic software DepthPro has more complete calculation functions and more comprehensive statistical analysis results, this embodiment selects DepthPro and this method for comparative verification. The specific data used is measured multibeam data from a certain sea area to verify the correctness of the calculation results of this method and the efficiency improvement in various aspects compared with DepthPro software. The hardware used in this experiment is completely identical.
[0068] First, we verify the clipping effect of this method on non-convex regions, i.e., annular regions. Figure 2 This represents the region obtained by cropping using only the convex hull algorithm. Figure 3 For the region obtained by cropping using the convex hull adaptive subdivision algorithm, from Figure 2 and Figure 3 It can be seen that the convex hull adaptive subdivision algorithm can completely extract the boundary of the circular inspection line and cut the main survey line according to this boundary to obtain the required area.
[0069] To verify the consistency of the calculation results, DepthPro software and the proposed method were used to calculate the two sets of survey area data respectively. Result1 and Result2 represent the analysis results obtained by the proposed method, while DepthPro1 and DepthPro2 represent the analysis results obtained by DepthPro software. Table 1 shows the data analysis results, and as shown in Table 1, the data analysis results are completely identical.
[0070] Table 1. Comparison of Consistency Verification Experiment Results for Calculation Results (Unit: m)
[0071] data Number of data points Mutual difference range Arithmetic Mean of Difference Absolute value of difference average Standard deviation Mean error result1 102466 -3.60～1.90 0.02 0.18 0.20 0.20 DepthPro1 102464 -3.60～1.90 0.02 0.18 0.20 0.20 result2 298961 -1.75～2.58 0.12 0.19 0.16 0.18 DepthPro2 298959 -1.75～2.57 0.12 0.19 0.16 0.18
[0072] Then, data from six different survey areas were selected and calculated using DepthPro software and this method respectively. The processing time of the two methods was recorded, and a line graph of the processing time was plotted based on these records for intuitive comparison and analysis. The comparative analysis results are as follows: Figure 4 and Figure 5 As shown, where, Figure 4 This figure shows a comparison of the main detection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration in an embodiment of the present invention with the mesh construction time of DepthPro software. Figure 5 The figure shows a comparison between the master detection comparison method based on convex hull adaptive subdivision and heterogeneous acceleration in an embodiment of the present invention and the time taken by DepthPro software to calculate the discrepancy value.
[0073] Finally, discrepancy calculations were performed on massive datasets. The results showed that DepthPro software experienced memory crashes when processing more than 3.65 million depth points on the main survey lines. To verify the processing capability of this method with massive datasets, all the previously collected test data were used for performance testing. This batch of data included 65 main survey lines and 11 check lines. Before trimming, the total number of depth points on the main survey lines was 124,766,352; after trimming, the total number of depth points on the main survey lines was 32,903,786, and the total number of depth points on the check lines was 17,557,301. The experimental results are shown in Table 2, indicating that this method still has stable processing capabilities for datasets exceeding 3.65 million depth points, reaching the scale of 100 million depth points.
[0074] Table 2. Time taken by this method to process massive amounts of data (unit: seconds)
[0075] Processing content time consuming Processing main survey line data 95.367 Calculate the convex hull of the inspection line 15.871 Cut the main test line 25.103 Constructing a triangular network 506.513 GPU preprocessing 20.872 Load check line data into the GPU 68.022 GPU interpolates matching points and calculates discrepancies. 30.092 Total time 761.840
[0076] This invention addresses the current challenge in marine surveying and mapping information processing: the large volume of data makes it difficult to quickly and efficiently analyze and evaluate the quality of massive surveying results. It proposes and implements a master-detector comparison method based on adaptive convex hull triangulation and heterogeneous CPU-GPU systems, integrating these two methods. Table 2 shows that experimental results demonstrate that, under the same hardware environment, this invention can stably process over 100 million water depth points. Under the same data volume, the accuracy of the calculation results obtained by this method is completely consistent with DepthPro, but the calculation speed for triangulation and intersection discrepancies is 4-10 times faster than DepthPro, significantly improving the processing efficiency of massive data.
[0077] In summary, this invention provides a master inspection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration. In the embodiments of this invention, the convex hull adaptive partitioning algorithm replaces the manual delineation of regions, achieving intelligent and accurate division of the inspection line coverage area, solving the problems of complex scenarios such as ring non-convex sets, and avoiding human error. At the same time, the CPU-GPU heterogeneous parallel computing strategy realizes multi-threaded scheduling and block-based streaming data management, optimizes memory usage, and utilizes asynchronous transmission of multiple CUDA streams to hide latency. The triangulation construction speed is 4-10 times faster than DepthPro. Thus, it breaks through the memory limitations of traditional software, can stably process point clouds with depths exceeding 100 million, and significantly improves the processing efficiency of massive data.
[0078] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.
[0079] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.
Claims
1. A master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration, characterized in that, The method includes: S10, export the main survey line data and check line data after various corrections to the corresponding main survey line folder and check line folder respectively, and convert the data in the main survey line folder and check line folder into a preset data structure in a unified format; S20: Obtain the data in the main test line folder and the inspection line folder after format conversion, and process the data based on the convex hull adaptive subdivision algorithm; S30 constructs a triangular mesh model based on the main survey line data after clipping. It adopts a block strategy to divide the overall data of the triangular mesh model into multiple data blocks. It performs interpolation calculation and cross-difference calculation based on the CPU-GPU heterogeneous parallel computing strategy. The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to the storage file. S40 performs statistical analysis on cross-difference files and outputs kernel density map visualizations.
2. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in claim 1, characterized in that: The method for processing data based on the convex hull adaptive partitioning algorithm includes: S101, load the data in the main test line folder and the inspection line folder after format conversion, and identify the main test line data and the inspection line data; S102, Construct a convex hull boundary model of the inspection line based on the inspection line data; S103, adaptively trim the main test line data according to the convex hull boundary of the inspection line to obtain trimmed main test line data that matches the inspection line region.
3. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in claim 2, characterized in that: When adaptively clipping the main test line data based on the convex hull boundary of the inspection line, the cv2.convexHull function of OpenCV is used to calculate its convex hull polygon boundary, and then the convex hull polygon boundaries of all inspection lines are integrated as the spatial constraint for clipping the main test line data.
4. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in claim 1, characterized in that: Methods for interpolation calculation and cross-difference calculation based on CPU-GPU heterogeneous parallel computing strategy include: S201 constructs a triangulation model based on the clipped main survey line data, and uses a block-based strategy to divide the overall data of the triangulation model into multiple data blocks, and transmits the block-based triangulation model data to the GPU. S202, the GPU solves for the coefficients of each triangle through parallel computing; S203 streams check line data and uses CUDA streaming technology to achieve asynchronous parallel transmission of check line data.
5. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in claim 4, characterized in that: Methods for interpolation calculation and cross-difference calculation based on CPU-GPU heterogeneous parallel computing strategies also include: S204 uses the KD tree algorithm to spatially index the data points of the inspection line, matches them to the corresponding triangular regions, and performs interpolation calculations and cross-difference calculations. S205: The CPU synchronously receives the cross-difference results asynchronously returned by the GPU and writes them to a storage file.
6. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in claim 5, characterized in that: When the GPU solves for the coefficients of each triangle through parallel computing, it loads the triangle index block by block and constructs the coefficient matrix. It then asynchronously transmits data such as the water depth point coordinates and triangle coefficients to the GPU block by block, reads the water depth point data of the inspection line in a streaming manner, and simultaneously starts the kernel function to calculate the interpolated water depth point value and the cross-point discrepancy value.
7. The master detection comparison method based on convex hull adaptive partitioning and heterogeneous acceleration as described in any one of claims 2-6, characterized in that: When performing statistical analysis and kernel density plot visualization on cross-difference files, the output includes cross-difference intervals, arithmetic mean, absolute mean, mean error, standard deviation, segmented error limit table, cross-difference distribution histogram, absolute value interval statistical table, and kernel density plot.