Combined mapping method based on coal pile height field matching and space-time global optimization

Abstract

The invention discloses a joint mapping method based on coal pile height field matching and space-time global optimization, relates to the technical field of cabin cleaning machine positioning, and aims to solve the problem that point clouds acquired during high-altitude scanning of a ship unloader are sparse and are difficult to directly match with data of a cabin cleaning machine. The method comprises the following steps: S1, utilizing a multi-scale height field correlation matching algorithm of coal pile point clouds of a cabin cleaning machine and a ship unloader, utilizing a coal pile as a common characteristic body, and searching a common region by rasterizing and matching an overall height distribution structure of the coal pile; s2, point cloud registration is conducted on the common area, and the initial transformation relation between the cabin cleaning machine and the ship unloader is determined; and S3, constructing a global constraint factor graph optimization model, inputting the key frame point cloud and map data of the cabin cleaning machine and the point cloud data of the ship unloader, and outputting a joint map of the cabin cleaning machine and the ship unloader. According to the method, the initial registration robustness can be remarkably improved, cross-space-time error compensation is realized, and the global consistency of the map is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cargo hold cleaning machine positioning, and in particular to a joint mapping method based on coal pile height field matching and spatiotemporal global optimization. Background Art

[0002] During the bulk coal transfer process at ports, a tank cleaning machine first gathers the bulk coal in the center of the ship's hold, where it is then picked up by a ship unloader and placed on a conveyor belt for transport. Currently, tank cleaning and ship unloading operations rely primarily on manual labor, resulting in low efficiency and a harsh working environment. Automating tank cleaning and ship unloading is a key means of improving work efficiency and ensuring personnel safety. Joint mapping of tank cleaning and ship unloaders is fundamental to achieving automated joint operations. However, existing equipment has drawbacks: the tank cleaning machine's lidar is easily obscured by coal piles, limiting its field of view to the interior of the hold and prone to error accumulation during pose estimation; and the point cloud acquired by the ship unloader during high-altitude scanning is relatively sparse, making it difficult to directly match the data from the tank cleaning machine.

[0003] For example, the Chinese patent with publication number CN118363032A relates to a pusher rake positioning method based on laser collaborative positioning of a ship unloader, which realizes collaborative positioning through point cloud downsampling and index marking. The specific technical solution is different from the present invention. Summary of the Invention

[0004] The present invention solves the problem that the point cloud obtained by the ship unloader during high-altitude scanning is relatively sparse and difficult to directly match with the data of the cargo hold cleaning machine. A joint mapping method based on coal pile height field matching and spatiotemporal global optimization is proposed, which significantly improves the robustness of the initial alignment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a joint mapping method based on coal pile height field matching and spatiotemporal global optimization, comprising the following steps: S1, using the multi-scale height field correlation matching algorithm of the coal pile point cloud of the tank cleaning machine and the ship unloader, using the coal pile as the common feature body, and finding the common area by raster matching its overall height distribution structure; S2, perform point cloud registration on the common area to determine the initial transformation relationship between the tank cleaning machine and the ship unloader; S3 builds a global constraint factor graph optimization model, inputs the key frame point cloud and map data of the tank cleaning machine and the point cloud data of the ship unloader, and outputs a joint map of the tank cleaning machine and the ship unloader.

[0006] In this technical solution, the initial transformation relationship between the cargo hold cleaning machine and the ship unloader is first constructed based on the matching of the coal pile height field. Then, the posture optimization is performed based on the obtained initial transformation relationship. After obtaining the optimized relationship, the entire joint mapping method based on global constraint factor graph optimization can be obtained.

[0007] The present invention is further configured as follows: Step S1 includes: S11, correcting the point cloud data of the ship unloader, and performing branch and bound on the coal pile data of the tank cleaning machine according to the size of the point cloud of the ship unloader; S12, rasterizing the coal pile data of the ship unloader and the cargo hold cleaning machine to construct a coal pile height field, including a high-resolution height grid map Hc and a low-resolution height grid map Hu; S13, calculating the similarity between Hc and Hu, and determining the common area of ​​the coal pile based on the similarity.

[0008] The multi-scale height field correlation matching algorithm is further limited to finally obtain the common area of ​​the coal pile.

[0009] The present invention is further configured as follows: Step S3 includes: S31, superimposing the key frame pose data and point cloud data of the tank cleaning machine to generate map data of the tank cleaning machine; S32, calculating the common area of ​​the map data of the tank cleaning machine and the point cloud data of the ship unloader according to step S1; S33, registering the shared area to obtain relative observation data between the tank cleaning machine and the ship unloader; S34, generating an optimization function based on historical relative observation data, and obtaining an optimized relative transformation relationship based on the optimization function.

[0010] Due to the sparsity of the point clouds of the tank cleaning machine and the ship unloader, the initial transformation relationship data is not accurate enough, which in turn has a negative impact on the mapping accuracy. To solve this problem, the transformation relationship is optimized through the above steps.

[0011] The present invention is further configured to correct the point cloud data of the ship unloader, including: The input point cloud of the ship unloader is subjected to PCA analysis, and the main direction of the ship unloader is made perpendicular to the main direction of the tank cleaning machine, and the point cloud of the ship unloader is rotated.

[0012] In this technical solution, the point cloud data of the ship unloader is corrected to lay the foundation for subsequent processing.

[0013] The present invention is further configured as follows: constructing the coal pile height field specifically includes: Downsample the point cloud data of the tank cleaning machine and then construct a high-resolution height grid map Hc; After PCA alignment of the point cloud data of the ship unloader, a low-resolution height grid map Hu is constructed.

[0014] The present invention is further configured such that the similarity between Hc and Hu is calculated using a normalized cross-correlation algorithm.

[0015] The present invention is further configured as follows: after calculating the similarity between Hc and Hu, the grid resolution and the search position are iteratively optimized until the similarity is maximized, and then the common area of ​​the tank cleaning machine and the ship unloader is correspondingly found according to the maximized similarity.

[0016] The corresponding common regions are found by maximizing the similarity.

[0017] The present invention is further configured such that the left side of the optimization function equation is the optimized relative observation, and the right side of the optimization function equation is the minimum value of the cumulative sum of the transpose of the first parameter and the product of the first parameter.

[0018] The relative observation data is optimized by the above-mentioned optimization function, and the optimized relative transformation relationship is obtained by the optimized relative observation data.

[0019] The present invention is further configured such that: the first parameter is specifically a continuous product of the inverse of the state of the tank cleaning machine at time i, the inverse of the relative observation data, and the state of the ship unloader at time j.

[0020] The present invention is further configured such that the map data of the tank cleaning machine is the cumulative product of the posture data of the i-th key frame of the tank cleaning machine and the point cloud data of the i-th key frame of the tank cleaning machine.

[0021] The key frame's position data and point cloud data of the tank cleaning machine are read, and the data of multiple key frames are superimposed.

[0022] The present invention can bring the following beneficial effects: 1. This application involves a joint mapping method based on coal pile height field matching and spatiotemporal global optimization, which can effectively overcome the difficulties of direct matching of sparse and dense point clouds, provide reliable initial values ​​for subsequent optimization, and significantly improve the robustness of initial registration. 2. By optimizing the mapping method based on the spatiotemporal global constraint factor graph, it can achieve cross-spatiotemporal error compensation and improve the global consistency of the map. 3. The method of this application is significantly superior to traditional solutions in terms of unstructured scene registration success rate, long-term consistency, and computational efficiency, and can meet the high-precision mapping requirements of the entire operation cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of a joint mapping framework for a tank cleaning machine and a ship unloader in a joint mapping method based on coal pile height field matching and spatiotemporal global optimization in this application.

[0024] Figure 2 This is a schematic diagram of the coordinate transformation relationship of a joint mapping method based on coal pile height field matching and spatiotemporal global optimization in this application.

[0025] Figure 3 This is a schematic diagram of a multi-scale height field correlation matching algorithm of a joint mapping method based on coal pile height field matching and spatiotemporal global optimization in this application.

[0026] Figure 4 This is a schematic diagram of a joint mapping method based on coal pile height field matching and spatiotemporal global optimization based on a spatiotemporal global constraint factor graph optimization framework in the present application. DETAILED DESCRIPTION

[0027] Example 1 This embodiment proposes a joint mapping method based on coal pile height field matching and spatiotemporal global optimization, referring to Figure 1 、 Figure 2 、 Figure 3 and Figure 4 , which includes the following steps.

[0028] In step S1, first, the common area is determined based on the multi-scale height field correlation matching algorithm of the coal pile point cloud of the tank cleaning machine and the ship unloader. Specifically, the coal pile is used as the common feature body, and its overall height distribution structure is matched through rasterization to find the corresponding common area.

[0029] During the clearing operation, the clearing machine needs to pass through the common area of ​​the coal pile to determine the conversion relationship; Figure 2 The coordinate relationship between the tank cleaning machine and the ship unloader is shown in Figure 1. The tank cleaning machine is located in the ship's hold, and the ship unloader is located above the tank cleaning machine. Since most of the coal pile is located in the ship's hold, the data captured by the ship unloader's radar is less than that of the tank cleaning machine. If the iterative closest point method is used directly, it is difficult to find an effective conversion relationship. Therefore, it is necessary to detect the shared area of ​​the tank cleaning machine and the ship unloader's coal pile to ensure that the correct matching relationship is found. Figure 2 In the prior art, the conversion matrix Coordinate transformation is usually composed of a 3*3 rotation matrix R and a 3*1 translation vector t; the present invention uses multi-scale height field matching and point cloud registration to obtain the initial transformation relationship T between the tank cleaning machine and the ship unloader init , that is, the optimal transformation initial value T init .

[0030] like Figure 3 The figure shows a multi-scale height field correlation matching algorithm using the coal pile point cloud of the tank cleaning machine and the ship unloader. The coal pile is used as an unstructured but shared feature body, and the shared area is found by rasterizing and matching its overall height distribution structure (rather than local feature points).

[0031] refer to Figure 1 、 Figure 2 as well as Figure 3 , for step S1, it mainly includes the following sub-steps.

[0032] In step S11, the acquired point cloud data of the ship unloader is corrected, and then the coal pile data of the tank cleaning machine is branched and bounded by the size of the point cloud of the ship unloader.

[0033] In more detail, the above process mainly involves performing PCA analysis on the input point cloud of the ship unloader. PCA (principal components analysis) is a principal component analysis technique that can use the concept of dimensionality reduction to transform multiple indicators into a few comprehensive indicators. Specifically, the main direction of the ship unloader is perpendicular to the main direction of the tank cleaning machine, that is, the point cloud of the ship unloader is rotated to lay the foundation for subsequent data processing.

[0034] Afterwards, based on the size of the ship unloader point cloud, the coal pile data of the tank cleaning machine was branched and bounded to further clarify the data range. Then, appropriate coal pile data was extracted from both the ship unloader and tank cleaning machine point cloud data.

[0035] Step S12: After completing the above step S11, the coal pile data of the ship unloader and the tank cleaning machine are rasterized to construct a coal pile height field, wherein the constructed coal pile height field mainly includes a high-resolution height grid map Hc and a low-resolution height grid map Hu.

[0036] For the high-resolution height grid map Hc: it is constructed by downsampling the point cloud of the cabin cleaning machine.

[0037] For the low-resolution height grid image Hu: After PCA analysis of the ship unloader point cloud, a low-resolution height grid image Hu is constructed (resolution adaptive adjustment).

[0038] Step S13: After obtaining the high-resolution height grid map Hc and the low-resolution height grid map Hu in step S13, the similarity between the high-resolution height grid map Hc and the low-resolution height grid map Hu is calculated, and the common area of ​​the coal pile is determined according to the similarity.

[0039] In this embodiment, a correlation search is performed on the multi-scale heights of the coal pile, and the similarity between the high-resolution height grid image Hc and the low-resolution height grid image Hu is calculated by normalized cross correlation (NCC).

[0040] The above solution can solve the problem of sparse-dense point cloud registration and significantly improve the robustness of the initial registration. Specifically, it avoids reliance on local geometric features that are prone to failure and directly matches the overall height distribution structure of the coal pile. Brute force matching effectively overcomes the difficulties of direct sparse-dense point cloud matching through rasterization and volume difference calculation, providing a reliable initial value for subsequent optimization.

[0041] In step S2, point cloud registration is then performed on the common area, and the initial transformation relationship between the tank cleaning machine and the ship unloader is finally obtained.

[0042] By iteratively optimizing the grid resolution and search position until the NCC value is maximized, the corresponding common area of ​​the tank cleaning machine and the ship unloader is obtained, and the point cloud registration of the common area is performed to finally determine the optimal initial value of the transformation T init .

[0043] In step S3, a global constraint factor graph optimization model is constructed, key frame point cloud data and map data of the tank cleaning machine and point cloud data of the ship unloader are input, and finally a joint map of the tank cleaning machine and the ship unloader is output.

[0044] Step S3 includes the following sub-steps.

[0045] Step S31, superimposing the key frame pose data and point cloud data of the tank cleaning machine to generate map data of the tank cleaning machine; first read the key frame pose data and point cloud data from the tank cleaning machine, and then superimpose the position data and point cloud data of multiple key frames.

[0046] The map data of the tank cleaning machine can be obtained by superimposing the above data, that is, the accumulated data of the tank cleaning machine from the i-th to the i+m-th key frames. The map data of the tank cleaning machine is equal to the accumulation of the product of the position data of the tank cleaning machine in the i-th key frame and the point cloud data of the tank cleaning machine in the i-th key frame in the interval from the i-th to the i+m-th key frames, where i and m are both integers greater than 0.

[0047] By reading the position data and point cloud data of the key frame of the tank cleaning machine, the data of multiple key frames are superimposed to obtain accumulated data.

[0048] In step S32, the shared area between the tank cleaning machine map data and the ship unloader point cloud data is calculated based on step S1. In this process, the ship unloader point cloud data can be directly read from the ship unloader's radar. After inputting the ship unloader point cloud data, the multi-scale height field correlation matching algorithm in step S1 is used to calculate the shared area between the tank cleaning machine map data and the ship unloader point cloud data.

[0049] Step S33, registering the shared area to obtain relative observation data between the tank cleaning machine and the ship unloader; specifically, registering the shared area of ​​the tank cleaning machine and the ship unloader to obtain relative observation data between the tank cleaning machine and the ship unloader.

[0050] Step S34: generating an optimization function based on historical relative observation data, and obtaining an optimized relative transformation relationship according to the optimization function.

[0051] Historical relative observation data between the tank cleaning machine and the ship unloader is obtained to generate a corresponding observation data set. The above optimization function is specifically as follows: the optimized relative observation is equal to the minimum value of the cumulative product of the transpose of the first parameter and the first parameter; where the first parameter is equal to the reciprocal of the tank cleaning machine state at time i, the reciprocal of the relative observation data, and the continuous product of the ship unloader state at time j.

[0052] The above optimization function is used to optimize the relative observation data between the tank cleaning machine and the ship unloader.

[0053] After optimization through the optimization function, the optimized relative transformation relationship can be obtained, that is, the optimized relative observation mentioned above. After obtaining the optimized relative transformation relationship, the entire joint mapping method based on global constraint factor graph optimization can be obtained.

[0054] The present invention realizes cross-temporal and spatial error compensation by constructing a factor graph optimization model that integrates multiple time-series key frames, including the integration of data such as the position of the tank cleaning machine, the position of the ship unloader, relative observation constraints, and historical observation constraints.

[0055] Through the above steps, it is possible to achieve cross-temporal and spatial error compensation and improve the global consistency of the map: the factor graph optimization framework integrates the data of the ship unloader-tank cleaning equipment and historical observation data at multiple times, constructs a cross-device and cross-time posture constraint network, effectively smooths the noise of single observations, significantly suppresses posture drift and error accumulation, and ultimately generates a globally consistent high-precision map.

[0056] The core steps of this invention are height field matching, multi-temporal factor graphs, and global map output. However, the Chinese patent application (CN118363032A) primarily employs downsampling, index tagging, single-frame GICP registration, and pose output. The two approaches address fundamentally different technical problems (localization vs. mapping) and implementation paths. Furthermore, the Chinese patent application (CN118363032A) suffers from the following limitations: Reliance on local features: Generalized Iterative Closest Point (GICP) registration is used to register downsampled point clouds, but coal pile scenes lack stable local features, making registration prone to failure; No historical optimization: Only the current frame data is processed, without establishing cross-temporal constraints, resulting in cumulative map drift over time; Data homogeneity assumption: The point clouds for the ship unloader and the pusher / rake have "similar data volumes," but the density differences between the dense (dense) and sparse (sparse) point clouds make this assumption unsatisfactory.

[0057] The main improvements of the present invention are reflected in the following aspects: abandoning structured feature matching: pioneering the "overall height distribution of coal piles" as an unstructured matching feature to avoid reliance on local geometric features that are prone to failure; fusing multiple time series data: integrating historical key frame observations through factor graph optimization to suppress long-term drift; being compatible with heterogeneous point clouds: directly designing a registration process for sparse-dense point cloud differences.

[0058] The present invention also includes a joint mapping experiment of a tank cleaning machine and a ship unloader: in the joint mapping experiment of a tank cleaning machine and a ship unloader, the tank cleaning machine is deployed under the ship unloader and started to operate. In the initial stage of the experiment, due to the posture deviation when the two are mapped independently, the maps generated by the tank cleaning machine and the ship unloader are spatially misaligned. There is an offset in the point cloud data of the two. This phenomenon essentially reflects the difference in the spatial positioning benchmark and data acquisition perspective of the equipment. Since the ship unloader is installed at a higher position than the tank cleaning machine, the center of its coordinate system is spatially misaligned with the tank cleaning machine. After the joint mapping, the posture deviation of the single-machine mapping can be effectively eliminated, and a globally unified spatial coordinate system can be constructed. At the same time, the collaborative mapping mechanism of the two makes up for the blind spots of the single-machine field of view, forming an integrated perception space covering the operation area, and providing an accurate global map for subsequent collaborative operations in complex scenarios.

[0059] Example 2 This embodiment proposes a joint mapping method based on coal pile height field matching and spatiotemporal global optimization, which includes the following steps.

[0060] In step S1, first, the common area is determined based on the multi-scale height field correlation matching algorithm of the coal pile point cloud of the tank cleaning machine and the ship unloader. Specifically, the coal pile is used as the common feature body, and its overall height distribution structure is matched through rasterization to find the corresponding common area.

[0061] During the clearing operation, the clearing machine needs to pass through the common area of ​​the coal pile to determine the conversion relationship; Figure 2 The coordinate relationship between the tank cleaner and the ship unloader is shown in Figure 1. The tank cleaner is located in the ship's hold, and the ship unloader is located above it. Because the coal pile is mostly located inside the ship's hold, the ship unloader's radar captures less data than the tank cleaner. Using the iterative closest point method directly would make it difficult to find an effective transformation relationship. Therefore, it is necessary to detect the shared area between the tank cleaner and the ship unloader's coal pile to ensure a correct matching relationship.

[0062] like Figure 3 The figure shows a multi-scale height field correlation matching algorithm using the coal pile point cloud of the tank cleaning machine and the ship unloader. The coal pile is used as an unstructured but shared feature body, and the shared area is found by rasterizing and matching its overall height distribution structure (rather than local feature points).

[0063] refer to Figure 1 、 Figure 2 as well as Figure 3 , for step S1, it mainly includes the following sub-steps.

[0064] In step S11, the acquired point cloud data of the ship unloader is corrected, and then the coal pile data of the tank cleaning machine is branched and bounded by the size of the point cloud of the ship unloader.

[0065] In more detail, the above process mainly involves performing PCA analysis on the input point cloud of the ship unloader. PCA (principal components analysis) is a principal component analysis technique that can use the concept of dimensionality reduction to transform multiple indicators into a few comprehensive indicators. Specifically, the main direction of the ship unloader is perpendicular to the main direction of the tank cleaning machine, that is, the point cloud of the ship unloader is rotated to lay the foundation for subsequent data processing.

[0066] Afterwards, based on the size of the ship unloader point cloud, the coal pile data of the tank cleaning machine was branched and bounded to further clarify the data range. Then, appropriate coal pile data was extracted from both the ship unloader and tank cleaning machine point cloud data.

[0067] Step S12: After completing the above step S11, the coal pile data of the ship unloader and the tank cleaning machine are rasterized to construct a coal pile height field, wherein the constructed coal pile height field mainly includes a high-resolution height grid map Hc and a low-resolution height grid map Hu.

[0068] For the high-resolution height grid map Hc: it is constructed by downsampling the point cloud of the cabin cleaning machine.

[0069] For the low-resolution height grid image Hu: After PCA analysis of the ship unloader point cloud, a low-resolution height grid image Hu is constructed (resolution adaptive adjustment).

[0070] Step S13: After obtaining the high-resolution height grid map Hc and the low-resolution height grid map Hu in step S13, the similarity between the high-resolution height grid map Hc and the low-resolution height grid map Hu is calculated, and the common area of ​​the coal pile is determined according to the similarity.

[0071] In this embodiment, a correlation search is performed on the multi-scale heights of the coal pile, and the similarity between the high-resolution height grid image Hc and the low-resolution height grid image Hu is calculated by normalized cross correlation (NCC).

[0072] The above solution can solve the problem of sparse-dense point cloud registration and significantly improve the robustness of the initial registration. Specifically, it avoids reliance on local geometric features that are prone to failure and directly matches the overall height distribution structure of the coal pile. Brute force matching effectively overcomes the difficulties of direct sparse-dense point cloud matching through rasterization and volume difference calculation, providing a reliable initial value for subsequent optimization.

[0073] In step S2, point cloud registration is then performed on the common area, and the initial transformation relationship between the tank cleaning machine and the ship unloader is finally obtained.

[0074] By iteratively optimizing the grid resolution and search position until the NCC value is maximized, the corresponding common area of ​​the tank cleaning machine and the ship unloader is obtained, and the point cloud registration of the common area is performed to finally determine the optimal initial value of the transformation T init .

[0075] In step S3, a global constraint factor graph optimization model is constructed, key frame point cloud data and map data of the tank cleaning machine and point cloud data of the ship unloader are input, and finally a joint map of the tank cleaning machine and the ship unloader is output.

[0076] Step S3 includes the following sub-steps.

[0077] Step S31, superimposing the key frame pose data and point cloud data of the tank cleaning machine to generate map data of the tank cleaning machine; first read the key frame pose data and point cloud data from the tank cleaning machine, and then superimpose the position data and point cloud data of multiple key frames.

[0078] The map data of the tank cleaning machine can be obtained by superimposing the above data, that is, the accumulated data of the tank cleaning machine from the i-th to the i+m-th key frames. The map data of the tank cleaning machine is equal to the accumulation of the product of the position data of the tank cleaning machine in the i-th key frame and the point cloud data of the tank cleaning machine in the i-th key frame in the interval from the i-th to the i+m-th key frames, where i and m are both integers greater than 0.

[0079] By reading the position data and point cloud data of the key frame of the tank cleaning machine, the data of multiple key frames are superimposed to obtain accumulated data.

[0080] In step S32, the shared area between the tank cleaning machine map data and the ship unloader point cloud data is calculated based on step S1. In this process, the ship unloader point cloud data can be directly read from the ship unloader's radar. After inputting the ship unloader point cloud data, the multi-scale height field correlation matching algorithm in step S1 is used to calculate the shared area between the tank cleaning machine map data and the ship unloader point cloud data.

[0081] Step S33, registering the shared area to obtain relative observation data between the tank cleaning machine and the ship unloader; specifically, registering the shared area of ​​the tank cleaning machine and the ship unloader to obtain relative observation data between the tank cleaning machine and the ship unloader.

[0082] Step S34: generating an optimization function based on historical relative observation data, and obtaining an optimized relative transformation relationship according to the optimization function.

[0083] Historical relative observation data between the tank cleaning machine and the ship unloader is obtained to generate a corresponding observation data set. The above optimization function is specifically as follows: the optimized relative observation is equal to the minimum value of the cumulative product of the transpose of the first parameter and the first parameter; where the first parameter is equal to the reciprocal of the tank cleaning machine state at time i, the reciprocal of the relative observation data, and the continuous product of the ship unloader state at time j.

[0084] The above optimization function is used to optimize the relative observation data between the tank cleaning machine and the ship unloader.

[0085] After optimization through the optimization function, the optimized relative transformation relationship can be obtained, that is, the optimized relative observation mentioned above. After obtaining the optimized relative transformation relationship, the entire joint mapping method based on global constraint factor graph optimization can be obtained.

[0086] The present invention realizes cross-temporal and spatial error compensation by constructing a factor graph optimization model that integrates multiple time-series key frames, including the integration of data such as the position of the tank cleaning machine, the position of the ship unloader, relative observation constraints, and historical observation constraints.

[0087] Through the above steps, it is possible to achieve cross-temporal and spatial error compensation and improve the global consistency of the map: the factor graph optimization framework integrates the data of the ship unloader-tank cleaning equipment and historical observation data at multiple times, constructs a cross-device and cross-time posture constraint network, effectively smooths the noise of single observations, significantly suppresses posture drift and error accumulation, and ultimately generates a globally consistent high-precision map.

[0088] Based on Example 1, this embodiment also proposes a joint mapping system based on coal pile height field matching and spatiotemporal global optimization, which mainly includes an initial relationship construction module and an optimization module. The initial relationship construction module can execute the processes of the above steps S1 and S2, and the optimization module can execute the process of the above step S3 based on the spatiotemporal global constraint factor graph optimization mapping method. The optimization module can be connected to the initial relationship construction module, and the joint mapping task is completed through the collaborative operation of the optimization module and the initial relationship construction module.

[0089] Example 3 Based on Example 1, the factor graph optimization used can also be replaced by pose graph optimization, but factor graph optimization is more flexible and supports multiple constraint types; the rasterized height difference calculation process used can also be replaced by voxel density matching, but in this embodiment, the height difference of the coal pile is more distinguishable.

Claims

1. A joint mapping method based on coal pile height field matching and spatiotemporal global optimization, characterized in that: The following steps are involved: S1, using the multi-scale height field correlation matching algorithm of the coal pile point cloud of the tank cleaning machine and the ship unloader, using the coal pile as the common feature body, and finding the common area by raster matching its overall height distribution structure; S2, perform point cloud registration on the common area to determine the initial transformation relationship between the tank cleaning machine and the ship unloader; S3 builds a global constraint factor graph optimization model, inputs the key frame point cloud and map data of the tank cleaning machine and the point cloud data of the ship unloader, and outputs a joint map of the tank cleaning machine and the ship unloader.

2. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 1 is characterized in that: The step S1 comprises: S11, correcting the point cloud data of the ship unloader, and performing branch and bound on the coal pile data of the tank cleaning machine according to the size of the point cloud of the ship unloader; S12, rasterizing the coal pile data of the ship unloader and the cargo hold cleaning machine to construct a coal pile height field, including a high-resolution height grid map Hc and a low-resolution height grid map Hu; S13, calculating the similarity between Hc and Hu, and determining the common area of ​​the coal pile based on the similarity.

3. A joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 1 or 2, characterized in that: The step S3 comprises: S31, superimposing the key frame pose data and point cloud data of the tank cleaning machine to generate map data of the tank cleaning machine; S32, calculating the common area of ​​the map data of the tank cleaning machine and the point cloud data of the ship unloader according to step S1; S33, registering the shared area to obtain relative observation data between the tank cleaning machine and the ship unloader; S34, generating an optimization function based on historical relative observation data, and obtaining an optimized relative transformation relationship based on the optimization function.

4. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 2 is characterized in that: Correcting the point cloud data of the ship unloader includes: The input point cloud of the ship unloader is subjected to PCA analysis, and the main direction of the ship unloader is made perpendicular to the main direction of the tank cleaning machine, and the point cloud of the ship unloader is rotated.

5. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 2 is characterized in that: The construction of the coal pile height field specifically includes: Downsample the point cloud data of the tank cleaning machine and then construct a high-resolution height grid map Hc; After PCA alignment of the point cloud data of the ship unloader, a low-resolution height grid map Hu is constructed.

6. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 2 is characterized in that: The similarity between Hc and Hu is calculated using a normalized cross-correlation algorithm.

7. A joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 2 or 6, characterized in that: After calculating the similarity between Hc and Hu, the grid resolution and search position are iteratively optimized until the similarity is maximized. At this time, the common area of ​​the tank cleaning machine and the ship unloader is found according to the maximized similarity.

8. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 3 is characterized in that: The left side of the optimization function equation is the optimized relative observation, and the right side of the optimization function equation is the minimum value of the cumulative sum of the transpose of the first parameter and the product of the first parameter.

9. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 8, characterized in that: The first parameter is specifically a continuous product of the inverse of the state of the tank cleaning machine at time i, the inverse of the relative observation data, and the state of the ship unloader at time j.

10. The joint mapping method based on coal pile height field matching and spatiotemporal global optimization according to claim 3 is characterized in that: The map data of the tank cleaning machine is the accumulation of the product of the posture data of the i-th key frame of the tank cleaning machine and the point cloud data of the i-th key frame of the tank cleaning machine.

Images