Three-dimensional environment reconstruction system for working area of scribing robot

Abstract

The invention relates to the technical field of three-dimensional environment reconstruction, in particular to a lineation robot working area three-dimensional environment reconstruction system which comprises a point cloud attribute calculation module, a point cloud attribute calculation module, a point cloud attribute calculation module, a point cloud attribute calculation module, a point cloud attribute calculation module and a point cloud attribute calculation module, and the point cloud attribute calculation module is used for conducting ground scanning on a lineation robot working area, obtaining original point cloud and conducting coordinate system alignment on the original point cloud; traversing all the data points and calculating the main curvature value and the main direction value of each data point; according to the method, after ground scanning is carried out on a working area of the scribing robot and an original point cloud is obtained, coordinate system alignment and main curvature and main direction calculation of each data point are carried out, and on the basis of establishing a curvature attribute point cloud with a local geometric attribute, a set threshold value is adopted to screen data points with relatively high main curvature values; the sensitive response capability to fine ground lineation traces is improved; the discrete lineation feature expression with higher spatial continuity and more accurate form recognition is constructed by combining a feature point cluster established by spatial proximity and geometric direction consistency.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional environment reconstruction, and in particular to a three-dimensional environment reconstruction system for a marking robot working area. Background Art

[0002] The field of three-dimensional environment reconstruction technology involves data collection, digital processing and three-dimensional model reconstruction of real physical environments to obtain high-precision spatial geometric structure and texture information.

[0003] In practice, 3D environment reconstruction technology often relies on static processing and geometric mapping of point cloud data. This can lead to fuzzy representation and unclear feature boundary recognition in areas with complex structures or poor texture continuity. Furthermore, curvature attribute calculations are often limited to global fitting or mean filtering, resulting in a delayed response to highly variable local areas and difficulty in identifying fine-grained line features. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a three-dimensional environment reconstruction system for the working area of ​​a marking robot.

[0005] In order to achieve the above-mentioned object, the present invention adopts the following technical solution: A three-dimensional environment reconstruction system for a marking robot working area includes:

[0006] The point cloud attribute calculation module performs a ground scan of the marking robot's working area to obtain the original point cloud, performs coordinate system alignment on the original point cloud, traverses all data points, and calculates the principal curvature value and principal direction value of each data point to establish a curvature attribute point cloud;

[0007] A line feature recognition module sets a curvature screening threshold based on the curvature attribute point cloud and traverses all data points to filter out points whose principal curvature values ​​are greater than the curvature screening threshold to obtain a trace candidate point set. Based on the trace candidate point set, a discrete line feature cluster is established;

[0008] A trace network construction module calculates the center position and main direction of each cluster based on the discrete line feature clusters, connects clusters whose spatial distance is below a threshold and whose main directions are consistent, generates a line connection path graph, and fits and establishes a line vectorization structure based on the line connection path graph;

[0009] A multi-resolution grid generation module divides the working area of ​​the marking robot into initial cells based on the curvature attribute point cloud, establishes an initial complexity grid, and obtains an adaptive terrain grid model based on the initial complexity grid.

[0010] Preferably, the steps of obtaining the curvature attribute point cloud are:

[0011] The marking robot performs ground laser scanning on the work area, collects and records the spatial three-dimensional position coordinates of the area, constructs a set of spatial coordinate points, and generates an original point cloud;

[0012] Based on the original point cloud, with a predefined reference coordinate system as a reference, by performing coordinate transformation on the set of spatial coordinate points point by point, the position of each spatial coordinate point is adjusted to the corresponding position in the reference coordinate system, thereby generating an aligned original point cloud;

[0013] Based on the aligned original point cloud, the spatial geometric distribution in the neighborhood of each spatial coordinate point is calculated point by point, the principal curvature value and the corresponding principal direction vector of each spatial coordinate point are extracted, and the calculated principal curvature value and principal direction vector are assigned to the corresponding spatial coordinate points in turn to generate a curvature attribute point cloud.

[0014] Preferably, the steps of obtaining the trace candidate point set are:

[0015] Based on the curvature attribute point cloud, the principal curvature values ​​of all spatial coordinate points in the curvature attribute point cloud are counted, and the mean of the principal curvature values ​​of all spatial coordinate points is calculated, and the mean is taken as the curvature screening threshold to generate the curvature screening threshold;

[0016] Based on the curvature screening threshold, the principal curvature value of each spatial coordinate point in the curvature attribute point cloud is traversed point by point, and whether the principal curvature value of the spatial coordinate point exceeds the curvature screening threshold is judged one by one. The spatial coordinate points whose principal curvature value exceeds the curvature screening threshold are marked one by one to generate a trace candidate point set.

[0017] Preferably, the step of obtaining the discrete line feature cluster is:

[0018] Based on the trace candidate point set, extract the three-dimensional coordinate value and local direction vector of each trace candidate point, traverse point by point to construct a combination of spatial position and direction relationships between two trace candidate points, and form a set of trace candidate point pairs;

[0019] Calculating the comprehensive morphological similarity between each pair of trace candidate points based on the set of trace candidate point pairs;

[0020] Based on the comprehensive morphological similarity between each pair of trace candidate points, the similarity between all trace candidate points is threshold-screened. The trace candidate point pairs with comprehensive morphological similarity higher than the set threshold are regarded as the connecting edges of the same clustering unit. All trace candidate points that form a connection relationship are aggregated into independent subsets, and discrete line feature clusters are obtained according to the aggregation relationship.

[0021] Preferably, the steps of obtaining the line connection path diagram are:

[0022] Based on the discrete line feature clusters, the three-dimensional spatial position coordinates of all trace candidate points in each discrete line feature cluster are traversed one by one, the spatial position coordinates of all trace candidate points are averaged, and the average value is defined as the center position of the discrete line feature cluster to generate a discrete line feature cluster center position set;

[0023] Based on the center position set of the discrete line feature cluster, all trace candidate points of the discrete line feature cluster are traversed one by one, and the principal component feature vector of each discrete line feature cluster is extracted by principal component analysis, and the principal component feature vector is defined as the principal direction vector of the discrete line feature cluster to generate a set of principal direction vectors of the discrete line feature cluster;

[0024] Based on the center position set of the discrete line feature clusters and the main direction vector set of the discrete line feature clusters, the spatial distance and the main direction angle between any two discrete line feature clusters are compared one by one, and the discrete line feature clusters whose spatial distance values ​​are less than a preset spatial threshold and whose main direction angles are less than a preset angle threshold are paired and connected to form a connection relationship and defined as a connection path, and a line connection path graph is generated.

[0025] Preferably, the steps of obtaining the line vectorization structure are:

[0026] Based on the line connection path graph, all discrete line feature clusters in each connection path in the line connection path graph are traversed one by one, the three-dimensional spatial position coordinates of all trace candidate points contained in each discrete line feature cluster are extracted, and all three-dimensional spatial position coordinates of each discrete line feature cluster are summarized to generate a three-dimensional spatial position coordinate set of the connection path;

[0027] According to the three-dimensional spatial position coordinate set of the connection path, linear fitting of the spatial point cloud is performed for each connection path, the direction vector and the center position of the fitting line segment are calculated by the least squares method, and the fitting line segment parameters of the connection path are obtained one by one to generate a connection path fitting line segment parameter set;

[0028] Based on the connection path fitting line segment parameter set, the fitting line segment parameters are converted into a standard vector representation format one by one, and the three-dimensional coordinates of the start and end endpoints and the vector direction of the fitting line segment are recorded in sequence to construct a line vectorization structure.

[0029] Preferably, the steps of obtaining the initial complexity grid are:

[0030] Based on the curvature attribute point cloud, the boundary range and spatial size parameters of the marking robot's working area are set, the entire marking robot's working area is divided into regular cubic cells according to a uniform side length, and each curvature attribute point in the curvature attribute point cloud is assigned to the cubic cell according to the three-dimensional position coordinate mapping to generate an initial cell set of the marking robot's working area;

[0031] Calculating the coupling morphological complexity of each initial cell according to the initial cell set of the working area of ​​the marking robot;

[0032] Based on the coupled morphological complexity of each initial cell, the initial cell set of the entire marking robot working area is traversed, and the spatial position indexes corresponding to all initial cells are paired with the coupled morphological complexity one by one to uniformly organize and form a complexity expression grid of the spatial structure changes in the area to generate the initial complexity grid.

[0033] Preferably, the steps of acquiring the adaptive terrain grid model are:

[0034] Based on the initial complexity grid, the coupling morphological complexity of each initial cell in the initial complexity grid is traversed one by one, and the numerical value is compared item by item with a pre-set coupling morphological complexity threshold. At the same time, whether the spatial position of each initial cell has a spatial geometric intersection relationship with the line vectorization structure is detected, and the initial cells that meet the coupling morphological complexity exceeding the threshold or having an intersection relationship are screened to generate a set of cells to be subdivided;

[0035] According to the set of cells to be subdivided, each cell to be subdivided is evenly divided along the midpoint position of the three-dimensional spatial coordinate axis to form eight smaller sub-cells, and the three-dimensional spatial position and boundary information of all sub-cells are recorded one by one to generate a first recursive subdivision cell set;

[0036] Based on the first recursively subdivided cell set, the steps of obtaining the cell set to be subdivided and obtaining the first recursively subdivided cell set are repeated until the coupled morphological complexity of all cells meets the set threshold and no longer intersects with the line vectorized structure, and then all cells are topologically connected in a grid to generate an adaptive terrain grid model.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are:

[0038] In the present invention, after scanning the ground of the working area of ​​the marking robot and obtaining the original point cloud, the coordinate system is aligned and the principal curvature and principal direction of each data point are calculated. On the basis of establishing a curvature attribute point cloud with local geometric properties, a threshold is set to filter data points with higher principal curvature values, thereby improving the sensitive response ability to subtle ground marking traces; the feature point clusters established by combining spatial proximity and geometric direction consistency are constructed to construct a discrete marking feature expression with stronger spatial continuity and more accurate morphological recognition; the center position and principal direction are judged between point clusters, and the point clusters with geometric coherence are connected and path fitting is completed, so that the marking information forms a structured vector expression in space, which is convenient for subsequent path extraction and task planning; through spatial grid division and introducing a coupled morphological complexity calculation based on the structural tensor determinant and the curvature average value for each cell, a complex grid that can dynamically respond to terrain changes is constructed, and the judgment and subdivision are based on whether it intersects with the marking structure, forming an adaptive terrain grid model with spatial adaptability and multi-scale expression capabilities, thereby enhancing the expression integrity of the marking area features and the accuracy of environmental modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] See also Figure 1 The present invention provides a technical solution: a three-dimensional environment reconstruction system for a marking robot working area includes:

[0042] The point cloud attribute calculation module performs a ground scan of the marking robot's working area to obtain the original point cloud, performs coordinate system alignment on the original point cloud, traverses all data points, and calculates the principal curvature value and principal direction value of each data point to establish a curvature attribute point cloud;

[0043] The line feature recognition module sets a curvature screening threshold based on the curvature attribute point cloud and traverses all data points. Points with a principal curvature value greater than the curvature screening threshold are screened out to obtain a trace candidate point set. Based on the trace candidate point set, a discrete line feature cluster is established.

[0044] The trace network construction module calculates the center position and main direction of each cluster based on discrete line feature clusters, connects clusters whose spatial distance is below a threshold and whose main directions are consistent, generates a line connection path graph, and then fits and establishes a line vectorization structure based on the line connection path graph.

[0045] The multi-resolution grid generation module divides the working area of ​​the marking robot into initial cells based on the curvature attribute point cloud, establishes an initial complexity grid, and obtains an adaptive terrain grid model based on the initial complexity grid.

[0046] The steps to obtain the curvature attribute point cloud are:

[0047] The marking robot performs ground laser scanning on the work area, collects and records the spatial three-dimensional position coordinates of the area, constructs a set of spatial coordinate points, and generates an original point cloud;

[0048] Based on the original point cloud, with the predefined reference coordinate system as a reference, the coordinate transformation of the spatial coordinate point set is performed point by point, and the position of each spatial coordinate point is adjusted to the corresponding position in the reference coordinate system to generate the aligned original point cloud;

[0049] Based on the aligned original point cloud, the spatial geometric distribution in the neighborhood of each spatial coordinate point is calculated point by point, the principal curvature value and the corresponding principal direction vector of each spatial coordinate point are extracted, and the calculated principal curvature value and principal direction vector are assigned to the corresponding spatial coordinate points in turn to generate a curvature attribute point cloud.

[0050] Specifically, the marking robot performs ground laser scanning on the working area. Specifically, the three-dimensional laser radar sensor carried by the marking robot is controlled, for example, a laser radar is used to scan the entire working area of ​​the marking robot without omissions according to a preset scanning path, such as a "bow" or "S" shaped path. During the scanning process, the laser radar continuously emits a laser beam and receives the signal reflected from the ground. The three-dimensional spatial position coordinates of each reflection point relative to the laser radar's own coordinate system are calculated based on the flight time or phase difference of the signal. ,in Represents the robot coordinate system), and at the same time, records these three-dimensional coordinate data and the corresponding high-precision timestamp information in real time, and gathers all the discrete three-dimensional spatial position coordinate points obtained during the entire scanning process to construct a spatial coordinate point set containing a large number of data points. This set is the original point cloud.

[0051] Based on the original point cloud, with a reference coordinate system pre-calibrated in the working area as a reference, the establishment of the reference coordinate system can be achieved by selecting at least three non-collinear fixed control points with clear physical identification in the working area, and using a total station or differential global positioning system to measure the three-dimensional coordinates of these control points in the global unified coordinate framework. For each spatial coordinate point in the original point cloud, its initial coordinate is relative to the sensor coordinate system of the marking robot itself during scanning. In order to unify all points to the reference coordinate system, coordinate transformation is required. First, based on the output data of the posture and position sensors (such as inertial measurement unit IMU and GPS-RTK) carried by the marking robot, combined with the sensor calibration parameters, the posture transformation matrix of the robot sensor coordinate system relative to the predefined reference coordinate system at each scanning moment can be solved. , which contains the rotation and translation information, then, this transformation matrix Applied to each spatial coordinate point in the original point cloud , through matrix multiplication operation (For homogeneous coordinates, and is a four-dimensional vector), the position of each spatial coordinate point is adjusted and converted from the robot sensor coordinate system to the corresponding position in the predefined reference coordinate system, thereby generating an aligned original point cloud.

[0052] Based on the aligned original point cloud, a neighborhood analysis is performed point by point for each spatial coordinate point to calculate its local spatial geometric distribution characteristics. First, for the currently processed spatial coordinate point Define a neighborhood. The neighborhood can be defined as a spherical neighborhood search based on a fixed radius. For example, if the average point spacing of the point cloud is 0.01 meters, the search radius can be set. The value is 0.05 meters, which is 5 times the average distance between points, to ensure that the neighborhood contains a sufficient number of points (for example, at least 15 to 20 points are expected) for robust geometric attribute estimation, or the K-nearest neighbor (KNN) search method is used to select the distance points. Recent points, of which The value can be selected based on the point cloud density and computational efficiency. For example, After collecting the point set in the neighborhood, the covariance matrix of the three-dimensional coordinates of these neighborhood points is calculated. Specifically, the centroid of the neighborhood points is calculated first, and then the deviation vector of each neighborhood point relative to the centroid is calculated, and then the covariance matrix of the three-dimensional coordinates of the neighborhood points is constructed. The covariance matrix of the covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues and their corresponding eigenvectors , the principal curvature value of the point can be derived from these eigenvalues. For example, a commonly used curvature measure is ,in The larger the value, the higher the degree of curvature or unevenness of the local area of ​​the point, and vice versa. The main direction value usually selects the eigenvector corresponding to the direction of maximum curvature change or the most significant geometric feature direction, such as the maximum eigenvalue. Associated eigenvector As the main direction, the main curvature value and the corresponding main direction vector of each spatial coordinate point calculated in this way are used as new attribute information, and are assigned to the corresponding spatial coordinate points in turn, thereby generating a curvature attribute point cloud.

[0053] The steps to obtain the trace candidate point set are:

[0054] Based on the curvature attribute point cloud, the principal curvature values ​​of all spatial coordinate points in the curvature attribute point cloud are counted, and the mean of the principal curvature values ​​of all spatial coordinate points is calculated. The mean is taken as the curvature screening threshold to generate the curvature screening threshold;

[0055] Based on the curvature screening threshold, the principal curvature value of each spatial coordinate point in the curvature attribute point cloud is traversed point by point, and whether the principal curvature value of the spatial coordinate point exceeds the curvature screening threshold is judged one by one. The spatial coordinate points whose principal curvature value exceeds the curvature screening threshold are marked one by one to generate a trace candidate point set.

[0056] Specifically, based on the curvature attribute point cloud, first, a comprehensive statistics of the principal curvature values ​​attached to each spatial coordinate point recorded in the point cloud is performed. This process includes reading the principal curvature values ​​of all spatial coordinate points in the curvature attribute point cloud one by one, and accumulating all the read principal curvature values ​​to obtain a sum, and at the same time counting the total number of spatial coordinate points involved in the accumulation. Then, the sum of all the calculated principal curvature values ​​is divided by the total number of spatial coordinate points to obtain the average value of these principal curvature values. The average value is directly adopted and set as the curvature screening threshold to identify points with significant geometric features in subsequent steps. For example, if a curvature attribute point cloud contains 500,000 spatial coordinate points, the principal curvature values ​​of each of these 500,000 points are extracted. For example, the range of these principal curvature values ​​is mainly concentrated between 0.05 and 0.7. The sum obtained by adding up all the 500,000 principal curvature values ​​is 150,000, so the mean of these principal curvature values ​​is ,Therefore, in this case, the curvature screening threshold is determined to be 0.30, generating the curvature screening threshold.

[0057] Based on the curvature screening threshold, the system then checks and screens each spatial coordinate point in the curvature attribute point cloud point by point. Specifically, the system sequentially accesses each spatial coordinate point in the curvature attribute point cloud, extracts the principal curvature value recorded at the point, and compares this principal curvature value with the curvature screening threshold previously calculated based on the mean of the principal curvature values ​​of all spatial coordinate points in the curvature attribute point cloud, to determine whether the principal curvature value of the current spatial coordinate point is strictly greater than the curvature screening threshold. For example, if the curvature screening threshold previously calculated is 0.30, for a spatial coordinate point with a principal curvature value of 0.45, since 0.45 is strictly greater than 0.30, the spatial coordinate point is The coordinate point will be judged to meet the screening conditions. If the principal curvature value of another spatial coordinate point is 0.25, since 0.25 is not greater than 0.30, the point does not meet the screening conditions. For the spatial coordinate point whose principal curvature value is exactly equal to 0.30, it will not be selected because it does not meet the condition of being strictly greater than. All spatial coordinate points whose principal curvature values ​​are strictly greater than the curvature screening threshold will be marked one by one by the system, such as setting a specific Boolean flag to true for these points in the data structure, or adding their indexes to a new list structure. All spatial coordinate points that have been screened and successfully marked are aggregated together to generate a trace candidate point set.

[0058] The steps to obtain discrete line feature clusters are:

[0059] Based on the trace candidate point set, the three-dimensional coordinate value and local direction vector of each trace candidate point are extracted, and the spatial position and direction relationship combination between each trace candidate point is constructed point by point to form a set of trace candidate point pairs;

[0060] According to the set of trace candidate point pairs, the comprehensive morphological similarity between each pair of trace candidate points is calculated. The calculation formula is:

[0061] ;

[0062] in, Trace candidate points With trace candidate points The comprehensive morphological similarity between Trace candidate points With trace candidate points The Euclidean distance between is the characteristic scale length, and Trace candidate points With trace candidate points The normalized local direction vector of is the dot product of two local direction vectors, The adjustment parameters are affected by the spatial distance. Adjust parameters for directional consistency effects;

[0063] Based on the comprehensive morphological similarity between each pair of trace candidate points, the similarity between all trace candidate points is threshold-screened. The trace candidate point pairs with comprehensive morphological similarity higher than the set threshold are regarded as the connecting edges of the same clustering unit. All trace candidate points that form a connection relationship are aggregated into independent subsets, and discrete line feature clusters are obtained according to the aggregation relationship.

[0064] Specifically, based on the trace candidate point set, each trace candidate point is previously obtained through the principal curvature screening and carries its original three-dimensional spatial coordinate information and the main direction vector obtained when calculating the curvature attribute. First, for each trace candidate point in the trace candidate point set, its stored three-dimensional coordinate value is directly extracted. And the local direction vector associated with it, this local direction vector is the main direction vector calculated by the point in its neighborhood, representing the main extension direction of the local geometric shape of the point. Then, in order to systematically analyze the relationship between these trace candidate points, all points in the trace candidate point set are combined pairwise, that is, for the points containing A set of trace candidate points, systematically generating all possible unordered point pairs, for example, if the trace candidate point set contains the point , it will build , ,..., ,Then ,..., , until , for each such trace candidate point pair (e.g. point and point ), record their respective three-dimensional coordinate values and the local direction vector ,This combination process is essentially to prepare basic data for the subsequent ,calculation of similarities between point pairs. By traversing ,all possible point pairs and recording their spatial position and ,orientation information, a set of trace candidate point pairs is formed.

[0065] formula: The benefit of the formula is that the comprehensive morphological similarity calculation method can effectively measure the spatial proximity and directional consistency of trace candidate points at the same time, which is crucial for accurately identifying lines with linear extension characteristics from point clouds. The exponential decay term is used to penalize the distance, ensuring that only points that are close enough in space can be considered similar, while the directional consistency term further filters out those points whose local directions are approximately parallel. The combination of the two can effectively exclude irrelevant noise points and enhance the recognition ability of real line segments. The parameters and The introduction of makes it possible to flexibly adjust the relative importance of distance and direction factors in similarity calculation according to the characteristics of actual line drawing and the quality of point cloud data, thereby improving the robustness and accuracy of line drawing feature extraction;

[0066] parameter The steps to obtain are:

[0067] Indicates trace candidate points With trace candidate points The Euclidean distance between them is calculated directly based on the coordinates of the two points in three-dimensional space. From the set of candidate trace points constructed in the previous step, extract the points The three-dimensional coordinates of and point The three-dimensional coordinates of , and then apply the standard Euclidean distance formula to calculate, for example, if the point The coordinates are meter, dot The coordinates are Meter, then rice.

[0068] parameter The steps to obtain are:

[0069] Represents the characteristic scale length. This parameter defines the characteristic distance at which a point is considered a "neighbor" in a local range. Its setting is based on the typical physical size of the line to be identified in the working area, such as the average width of the line or the typical point spacing of the laser scanning point cloud on such a target. The geometric characteristic parameters are obtained by conducting preliminary research and measurement on the expected line type in the working area of ​​the line marking robot. For example, if the target line is a white solid lane line commonly seen on urban roads, its standard width is usually 0.10 to 0.15 meters. Considering the density of the scanned point cloud and the need to capture the continuity of the line segment, It can be set to a value equivalent to or slightly larger than this width to ensure that neighboring points constituting the same line segment can be effectively associated. In this embodiment, by measuring multiple line samples, the average characteristic width of the line is about 0.12 meters. At the same time, considering the point cloud density, in order to ensure that enough points participate in the calculation and not over-smooth the details, the characteristic scale length is set to 0.12 meters. is set to 0.10 meters.

[0070] parameter and The steps to obtain are:

[0071] and Represent trace candidate points With trace candidate points The normalized local direction vectors of the point cloud are derived from the main direction vectors calculated for each original point cloud point in the previous step. When a point is selected as a trace candidate point, its main direction vector information is retained. The main direction vector describes the point or point The main direction of the point set distribution in the local neighborhood. Before substituting into the formula, it is necessary to ensure that these local direction vectors are normalized, that is, their modulus is 1. If the original main direction vector is , then its normalized form Calculated as , for example, point The local direction vector is , whose modulus is , then after normalization ,point The local direction vector is , whose modulus is , then after normalization .

[0072] parameter The steps to obtain are:

[0073] Adjustment parameters for spatial distance influence, used to control spatial distance Similarity The attenuation rate of the influence, this parameter is a positive real number, and its value is set to make the distance exceed a certain multiple of the characteristic scale length The similarity between the point pairs is significantly reduced to strengthen the local connectivity. The value of is usually determined by experimental tuning on sample point cloud data containing known line features. The goal is to find a balance point that can connect adjacent points on the same line and effectively distinguish different lines or noise points. For example, if we want to ), the contribution of the spatial distance term Decay to , then at this time Should be equal to , thus deriving In this embodiment, after multiple experiments on the test data set, it is found that When set to 1.0, it can better distinguish the inner and outer points of the line, so set .

[0074] parameter The steps to obtain are:

[0075] It is the direction consistency adjustment parameter used to adjust the contribution of the consistency of the local direction vectors of two trace candidate points to the overall similarity. This parameter is a positive real number. The larger the value, the heavier the penalty for directional deviation, and the higher the requirement of similarity on directional consistency. Its setting also depends on the analysis and experimental testing of the actual line drawing data characteristics. If the line drawing characteristics have a high degree of directional continuity, a larger value should be selected. On the contrary, if the line allows a certain degree of bending or direction change, then The value can be appropriately reduced. For example, it is necessary to evaluate the angle between the two direction vectors. When , the directional consistency term Attenuation of the direction angle ( ), the value of the directional consistency term is not less than 0.85, that is, ,like , the result is 0.933; if , the result is , meet the requirements, if , the result is , is too strict. In this embodiment, considering that the actual line may be slightly curved, in order to ensure the robustness of the connection and punish obvious direction inconsistencies, we choose .

[0076] Calculation process:

[0077] A pair of trace candidate points and Take the calculation as an example:

[0078] point The three-dimensional coordinates of meters, its normalized local direction vector .

[0079] point The three-dimensional coordinates of meters, its normalized local direction vector .

[0080] Set the parameter values ​​as follows:

[0081] Characteristic scale length rice.

[0082] Spatial distance affects adjustment parameters .

[0083] Directional consistency affects adjustment parameters .

[0084] First, calculate the Euclidean distance between two points :

[0085] ;

[0086] ;

[0087] rice.

[0088] Then, calculate the dot product of the two local direction vectors :

[0089] ;

[0090] .

[0091] Next, calculate the spatial distance term:

[0092] ;

[0093] .

[0094] Then, the directional consistency term is calculated:

[0095] ;

[0096] .

[0097] Finally, calculate the comprehensive morphological similarity :

[0098] .

[0099] The results show that the trace candidate points With trace candidate points The comprehensive morphological similarity between It is about 0.73338. This value is between 0 and 1. A higher value (close to 1) indicates that the two points are very similar in spatial position and local direction, while a lower value (close to 0) indicates poor similarity. The currently calculated 0.73338 is a relatively high similarity value, indicating that the points and point It is very likely that they belong to the same section of line features.

[0100] Based on the comprehensive morphological similarity between each pair of trace candidate points calculated based on the set of trace candidate point pairs , perform unified threshold screening on all these calculated similarity values. The "set threshold" here refers to the comprehensive morphological similarity threshold, which is set based on the known line feature point pairs and non-feature point pairs in a large number of sample data. Specifically, the value distribution characteristics of a series of typical line drawing scenarios are collected and calculated. Value, draw its histogram, and observe whether it can effectively distinguish the internal connection of the line from the noise or the connection between different lines. For example, in the test, it was found that the point pairs belonging to the same clear line have The values ​​are usually distributed between 0.6 and 0.9, while the random noise point pairs or the point pairs belonging to different lines are The values ​​are mostly lower than 0.4. In order to ensure the recall rate and improve the precision as much as possible, the comprehensive morphological similarity threshold can be set in the valley area of ​​the distribution or fine-tuned according to the expected clustering effect, for example, set to 0.55. Trace candidate point pairs (points) with values ​​higher than the set threshold (e.g. 0.55) and point ), they are considered to have a strong association and are marked as belonging to the same potential line segment. Conceptually, this is equivalent to establishing a connecting edge in a graph with trace candidate points as nodes and similarities between them above a threshold as edges. Subsequently, all trace candidate points that form a connection relationship in this way are aggregated, that is, all connected components in the graph are searched. All trace candidate points in each connected component together constitute an independent subset, which is divided according to this aggregation relationship. Finally, a series of discrete sets consisting of several trace candidate points with high morphological similarity are obtained, namely, discrete line feature clusters.

[0101] The steps for obtaining the line connection path diagram are as follows:

[0102] Based on the discrete line feature clusters, the three-dimensional spatial position coordinates of all trace candidate points in each discrete line feature cluster are traversed one by one. The spatial position coordinates of all trace candidate points are averaged, and the average value is defined as the center position of the discrete line feature cluster to generate the discrete line feature cluster center position set;

[0103] Based on the center position set of discrete line feature clusters, all trace candidate points of the discrete line feature clusters are traversed one by one, and the principal component feature vector of each discrete line feature cluster is extracted using principal component analysis. The principal component feature vector is defined as the principal direction vector of the discrete line feature cluster, and a set of principal direction vectors of the discrete line feature clusters is generated.

[0104] Based on the center position set of discrete line feature clusters and the main direction vector set of discrete line feature clusters, the spatial distance and main direction angle between any two discrete line feature clusters are compared one by one. The discrete line feature clusters whose spatial distance value is less than a preset spatial threshold and whose main direction angle is less than a preset angle threshold are paired and connected to form a connection relationship and defined as a connection path, and a line connection path graph is generated.

[0105] Specifically, based on the discrete line feature cluster, the following operations are performed on each discrete line feature cluster: first, the system accesses all the trace candidate points contained in the discrete line feature cluster and extracts the three-dimensional spatial position coordinates of these trace candidate points one by one. , if a discrete line feature cluster contains trace candidate points, whose coordinates are , then the system will be these spatial position coordinates 、 、 The components are summed up and then divided by the total number of trace candidate points , in order to calculate the discrete line feature cluster in 、 、 The average coordinate value in three dimensions, that is, the center position of the cluster , , For example, a discrete line feature cluster contains three trace candidate points with coordinates of P1 (1.0, 2.0, 0.5), P2 (1.1, 2.1, 0.6), and P3 (1.2, 2.0, 0.4). The x-coordinate of its center position is , the y coordinate is , the z coordinate is , so the center position of the discrete line feature cluster is (1.1, 2.033, 0.5). The calculated average three-dimensional spatial position coordinates are defined as the center position of the discrete line feature cluster, and the center positions of all discrete line feature clusters are organized to generate a discrete line feature cluster center position set.

[0106] Based on the discrete line feature clusters generated in the previous step, the system will process each discrete line feature cluster one by one to determine its main direction. For the currently processed discrete line feature cluster, the three-dimensional spatial position coordinates of all trace candidate points contained in it are first collected. Using the coordinate data of these points, the principal component analysis method is used to extract the main direction of the point set. When implementing PCA, the covariance matrix of the three-dimensional coordinates of all trace candidate points in the discrete line feature cluster is first calculated. This covariance matrix describes the degree of dispersion of the distribution of the point set in various directions. Then, the covariance matrix is ​​subjected to eigenvalue decomposition to obtain three eigenvalues ​​and their corresponding eigenvectors. These eigenvectors are mutually orthogonal and point to the direction with the largest data variance. Among them, the eigenvector associated with the largest eigenvalue, namely the first principal component, represents the most important direction of the data point set distribution, that is, the main trend direction of the discrete line feature cluster extending in space. The first principal component eigenvector (usually a three-dimensional unit vector) is defined as the main direction vector of the discrete line feature cluster. This process is repeated for all discrete line feature clusters, and the calculated main direction vectors are collected to generate a set of main direction vectors of discrete line feature clusters.

[0107] Based on the discrete line feature cluster center position set and the discrete line feature cluster main direction vector set obtained in the previous step, the system starts to compare any two different discrete line feature clusters in pairs to determine whether there is a connection relationship between them. For any two selected discrete line feature clusters, such as cluster A and cluster B, the center position coordinates of each of them are first extracted from the discrete line feature cluster center position set. and , and calculate the Euclidean distance between these two center locations At the same time, the main direction vectors of the discrete line feature clusters are extracted from their respective main direction vector sets. and , and calculate the angle between these two main direction vectors , Next, the calculated spatial distance Compare with a preset spatial threshold and calculate the main direction angle Compared with a preset angle threshold, the "preset spatial threshold" here is set according to the continuity characteristics of the actual line and the point cloud density. For example, if the maximum break length allowed between line segments is 0.5 meters, the spatial threshold can be set to 0.5 meters, or slightly larger than this value, such as 0.6 meters, to tolerate a certain measurement error and data sparsity. The "preset angle threshold" is set according to the degree of curvature allowed by the line. For example, for a relatively straight line, the angle threshold can be set to a smaller value, such as 15 degrees (about 0.26 radians), to ensure that the connected clusters have good consistency in direction. If the spatial distance Less than a preset space threshold (e.g. m) and the main direction angle Less than a preset angle threshold (e.g. ), it is considered that there is a potential connection between the two discrete line feature clusters. At this time, the system will establish a connecting edge between the nodes representing the two clusters and define this connection relationship as a connection path. By repeating this comparison and connection process for all possible pairs of discrete line feature clusters, a network structure containing all connected discrete line feature clusters and the connection paths between them is finally constructed, generating a line connection path graph.

[0108] The steps to obtain the line vectorization structure are:

[0109] Based on the line connection path graph, all discrete line feature clusters in each connection path in the line connection path graph are traversed one by one, the three-dimensional spatial position coordinates of all trace candidate points contained in each discrete line feature cluster are extracted, and all three-dimensional spatial position coordinates of each discrete line feature cluster are summarized to generate a set of three-dimensional spatial position coordinates of the connection path;

[0110] Based on the three-dimensional spatial position coordinate set of the connection path, linear fitting of the spatial point cloud is performed for each connection path. The direction vector and center position of the fitting line segment are calculated using the least squares method. The fitting line segment parameters of the connection path are obtained one by one to generate a connection path fitting line segment parameter set.

[0111] Based on the connection path fitting line segment parameter set, the fitting line segment parameters are converted into a standard vector representation format one by one, and the three-dimensional coordinates of the starting and ending points and the vector direction of the fitting line segment are recorded in sequence to construct a line vectorization structure.

[0112] Specifically, based on the line connection path graph, each connection path in the graph is processed independently. First, the system will traverse the currently processed connection path and identify all discrete line feature clusters that constitute the path. For each discrete line feature cluster in the path, the system will further extract the three-dimensional spatial position coordinates of all trace candidate points contained in the cluster. For example, a connection path may be composed of discrete line feature cluster A, discrete line feature cluster B and discrete line feature cluster C connected in sequence. The system will first collect the coordinates of all trace candidate points in discrete line feature cluster A, then collect the coordinates of all trace candidate points in discrete line feature cluster B, and finally collect the coordinates of all trace candidate points in discrete line feature cluster C. All three-dimensional spatial position coordinates extracted from all discrete line feature clusters in the connection path are merged to form a large point set for the current connection path that contains the coordinates of all relevant trace candidate points. This point set is the three-dimensional spatial position coordinate set of the connection path.

[0113] According to the three-dimensional spatial position coordinate set of the connection path, for each three-dimensional spatial position coordinate set corresponding to the connection path, a linear fitting operation of the spatial point cloud is performed respectively. Specifically, for a specific connection path and its corresponding three-dimensional spatial position coordinate set, the least squares method is used to determine a three-dimensional straight line that can best fit these spatial points. The goal of the least squares method is to find a straight line such that the sum of the squares of the perpendicular distances of all points to the straight line is minimized. In three-dimensional space, a straight line can be parameterized by a point (such as the center position of the straight line or any point on the line) and the direction vector of the straight line. By applying the least squares algorithm to the three-dimensional coordinate data of the point set (for example, The covariance matrix of the decentralized point set can be processed by singular value decomposition (SVD), where the eigenvector corresponding to the maximum eigenvalue is the direction vector of the fitting line, and the average coordinates of all points are the center position of the fitting line). The direction vector and center position representing the best fitting segment are calculated. This direction vector describes the overall direction of the connection path, and the center position calibrates the approximate position of the path in space. This linear fitting process is repeated for each connection path in the line connection path diagram, and the fitting line segment parameters of each connection path, including its direction vector and center position, are obtained one by one. These parameters are organized to generate a set of connection path fitting line segment parameters.

[0114] Based on the connection path fitting line segment parameter set, which includes the direction vector and center position of each connection path obtained by least squares fitting, each fitting line segment parameter in this set is processed one by one to convert it into a standardized vector representation format. The standard format usually requires the explicit recording of the three-dimensional coordinates of the start and end points of the line segment, as well as the possible line segment direction (although it can sometimes be calculated from the start and end points). For each connection path fitting line segment parameter (i.e., direction vector and central location ), first, we need to determine the actual coverage of the line segment on the original point cloud data. This can be achieved by projecting the original trace candidate points contained in the connection path onto the fitting line, and then finding the farthest two ends of the projection point distribution on the fitting line. The three-dimensional coordinates of these two farthest endpoints are defined as the three-dimensional coordinates of the start and end points of the fitting line segment. For example, if the direction vector of the fitting line of a connection path is , the center position is By analyzing the projection of the points in the three-dimensional space coordinate set of the corresponding connection path on the fitting line, the coordinates of the starting point can be determined. and the end point coordinates At the same time, the vector direction of the line segment can be directly adopted by the direction vector obtained by fitting , or by The information (three-dimensional coordinates of the starting and ending points, vector direction) is calculated and normalized, and recorded in sequence according to a predetermined data structure or file format (for example, a commonly used GIS vector data format or a custom format). This conversion and recording process is repeated for all connection paths, and finally the overall vectorized representation of the linework, i.e., the linework vectorized structure, is constructed.

[0115] The steps to obtain the initial complexity grid are:

[0116] Based on the curvature attribute point cloud, the boundary range and spatial size parameters of the marking robot's working area are set. The entire marking robot's working area is divided into regular cubic cells with a uniform side length. Each curvature attribute point in the curvature attribute point cloud is assigned to the cubic cell according to the three-dimensional position coordinate mapping to generate the initial cell set of the marking robot's working area.

[0117] According to the initial cell set of the marking robot working area, the coupling morphological complexity of each initial cell is calculated. The calculation formula is:

[0118] ;

[0119] in, For the The coupled morphological complexity of the initial cells, For the The determinant of the direction structure tensor of the initial cells, For the The maximum eigenvalue of the initial unit cell direction structure tensor, For the The average curvature value of all curvature attribute points in the initial cell, is the characteristic curvature scale constant, is the normal distribution response index, is the curvature response index, , For the The direction structure tensor of the initial cells, For the The number of curvature attribute points contained in the initial cells, For the The first cell in the The unit normal direction vector of the curvature attribute point, is the transpose of the unit normal direction vector, It is the matrix product of the unit normal direction vector and its own transpose, used to represent its direction contribution tensor;

[0120] Based on the coupled morphological complexity of each initial cell, the initial cell set of the entire marking robot working area is traversed, and the spatial position indexes corresponding to all initial cells are paired with the coupled morphological complexity one by one to uniformly organize and form a complexity expression grid of the spatial structure changes in the area to generate the initial complexity grid.

[0121] Specifically, based on the curvature attribute point cloud, it is first necessary to accurately define the boundary range of the entire three-dimensional working area where the marking robot operates, which is usually achieved by the minimum and maximum values ​​in the global coordinate system. Coordinates are used to define, for example, a working area is defined as rice, rice, Meters, and at the same time set the spatial size parameters, which mainly refer to the uniform side length of the regular cubic cells used in the subsequent division. The selection of this side length requires a balance between computational efficiency and detail capture capabilities. For example, if the average density of the point cloud is high and it is necessary to identify finer terrain changes, a smaller side length such as 0.2 meters can be selected. If the point cloud is sparse or mainly focuses on macro-terrain, a larger side length such as 0.5 meters can be selected. In this embodiment, according to the size of the typical marking area and the density of the laser scanning point cloud, the uniform side length is set to 0.25 meters. Next, according to this uniform side length, the entire defined marking robot working area is cut into a series of tightly arranged, non-overlapping regular cubic cells in three-dimensional space. For each curvature attribute point in the curvature attribute point cloud, its three-dimensional position coordinates in the reference coordinate system are extracted. , and compare the coordinates with the spatial range of each cube cell, accurately map and assign them to the only cube cell to which they belong. This mapping process can be achieved through simple integer division operations. For example, if the cell side length is , the starting point of the working area is , then point The index of the cell to which it belongs Can be achieved through , , It is calculated that after all curvature attribute points are assigned, all cubic cells containing at least one curvature attribute point together constitute the initial cell set of the marking robot's working area.

[0122] formula: , the formula is beneficial in that the coupled morphological complexity Taking into account the anisotropy of the normal distribution of the point cloud inside the cell and the size of the average curvature, the geometric complexity of the local area can be more comprehensively characterized. It reflects the distribution consistency of the normal direction: when the normal direction is highly consistent (such as a plane) or linearly distributed, this term approaches 0; when the normal direction is isotropically distributed (such as a corner or a highly irregular surface), this term approaches 1. This is directly related to the curvature of the surface. The greater the average curvature, the greater the value. By combining these two aspects of information with the adjustable parameter and Combined, it can identify areas containing significant topographic changes or complex geometric features;

[0123] parameter The steps to obtain are:

[0124] For the The direction structure tensor of the initial cells is calculated as follows: ,in It is The number of curvature attribute points contained in the initial cells is obtained by traversing the initial cell set in the working area of ​​the marking robot and calculating the curvature attribute points of the cells. The curvature attribute points in the cell are counted. For example, if the cell Contains 20 curvature attribute points, then ,parameter It is The first cell in the The unit normal direction vector of the curvature attribute point, for the cell Each curvature attribute point within , whose unit normal vector By analyzing the local neighborhood geometry in the curvature attribute point cloud, it is estimated. Specifically, Select the three-dimensional space nearest neighbors (e.g. , search from the entire curvature attribute point cloud), for this points (including The principal component analysis (PCA) of the three-dimensional coordinates of itself is performed, and the eigenvector corresponding to the minimum eigenvalue is the normal direction of the local point set, which is obtained after normalization. , yes The matrix product (outer product) with its own transpose, representing its directional contribution tensor, is a The symmetric matrix of All The sum of the direction contribution tensors of the points and the average are obtained. .

[0125] parameter The steps to obtain are:

[0126] For the Direction structure tensor of initial cells The determinant of matrix After that, its determinant can be directly obtained by the standard determinant calculation formula. For example, if ,but .

[0127] parameter The steps to obtain are:

[0128] For the Initial cell direction structure tensor The maximum eigenvalue of Then, by solving its characteristic equation (in is the identity matrix, is the eigenvalue) and get three eigenvalues , the largest of which is .

[0129] parameter The steps to obtain are:

[0130] For the The average curvature value of all curvature attribute points in the initial cell, for the cell Each curvature attribute point within (common points), extract the calculated principal curvature values ​​from the curvature attribute point cloud , and then calculate the average of these curvature values, for example, if the cell There are 3 points in the triangle, whose principal curvature values ​​are 0.2, 0.3, and 0.4 respectively. .

[0131] parameter The steps to obtain are:

[0132] is the characteristic curvature scale constant, which is used to adjust the mean curvature Normalization is performed, and its value should be set according to the curvature range of typical objects or line features in the work area. By analyzing the known scene data, the curvature value distribution of the features of interest such as the edge of the line and the undulation of the road surface is unified, and a representative curvature value is selected as For example, if analysis shows that the curvature values ​​of important terrain changes are generally between 0.1 and 0.8, you can Set to the median or mean of the range, such as 0.4. In this embodiment, by statistically analyzing the sample point cloud containing various road features, it is found that the average principal curvature value of the edge of the line and the slightly damaged area of ​​the road surface is usually around 0.35, so it is set .

[0133] parameter The steps to obtain are:

[0134] is the normal distribution response index, which is a positive real number used to adjust the sensitivity of the coupled morphological complexity to the degree of anisotropy of the normal distribution. Its value is determined experimentally and is calculated on multiple sample cells containing different normal distribution features (such as planes, edges, corners). Item, and adjust The value of is used to observe its contribution to distinguishing these features. The goal is to make this item effectively amplify the influence of complex normal distribution. The value range is generally between 0.5 and 2.0. In this embodiment, through testing, when , it can better reflect the complexity of normal distribution, so we set .

[0135] parameter The steps to obtain are:

[0136] is the curvature response index, which is a positive real number used to adjust the sensitivity of the coupled morphological complexity to the average curvature size. Its setting method is the same as Similarly, we experimentally tested the sample cells with different average curvatures and adjusted value, so that the curvature term It can properly reflect the complexity of the high curvature area, and the value range is generally between 0.5 and 2.0. In this embodiment, in order to make the complexity respond appropriately to the change of the average curvature, it is set .

[0137] Calculation process:

[0138] With an initial cell For example, the cell contains curvature attribute points.

[0139] Their unit normal direction vectors are: , , .

[0140] The curvature values ​​at these points are: , , .

[0141] The default parameters are: , , .

[0142] Calculate the directional structure tensor :

[0143] , , .

[0144] .

[0145] calculate and :

[0146] .

[0147] because is a diagonal matrix whose eigenvalues ​​are the diagonal elements: .

[0148] therefore, .

[0149] Calculate the normal distribution term:

[0150] .

[0151] Calculating mean curvature :

[0152] .

[0153] Compute the curvature term:

[0154] .

[0155] Calculating coupled morphological complexity :

[0156] .

[0157] This result shows that the The coupled morphological complexity of the initial cells It is about 5.196. The larger the value, the more complex the normal distribution and average curvature of the local area represented by the cell are. For example, a cell with sharp edges or corners and severe surface curvature has a The value will be relatively high, while a cell representing a flat, uniform surface will have The value will be low (for example, if the normal distribution term is 0, the curvature term is ,but ).

[0158] Based on the previously calculated coupled morphological complexity of each initial unit cell , the system will perform a complete traversal operation on the initial cell set of the entire working area of ​​the marking robot. During the traversal process, for each initial cell, the system will record its unique spatial position index in the three-dimensional grid. For example, if the entire working area is divided into a If the grid is a three-dimensional grid, each cell can be represented by its integer coordinates in the grid (in , , ) to uniquely identify the cell. At the same time, the system will also read the coupled morphological complexity value calculated for the cell. , then, index this spatial position The corresponding coupling morphological complexity value One-to-one pairing is performed and this pairing information is stored. For example, a data structure can be constructed in which each entry contains the index of a cell and its complexity value. In this way, all initial cells and their respective coupled morphological complexities are organized uniformly to form a complexity expression grid that can reflect the degree of change in the spatial structure within the region. This grid is actually based on the initial regular cubic cells, and each cell is given an attribute value that quantifies its geometric complexity, ultimately generating an initial complexity grid.

[0159] The steps to obtain the adaptive terrain grid model are:

[0160] Based on the initial complexity grid, the coupled morphological complexity of each initial cell in the initial complexity grid is traversed one by one, and the numerical value is compared item by item with the pre-set coupled morphological complexity threshold. At the same time, the spatial position of each initial cell is checked to see if there is a spatial geometric intersection relationship with the line vectorization structure. The initial cells that meet the coupling morphological complexity exceeding the threshold or have an intersection relationship are selected to generate a set of cells to be subdivided.

[0161] Based on the set of cells to be subdivided, each cell to be subdivided is evenly divided along the midpoint of the three-dimensional spatial coordinate axis to form eight smaller sub-cells. The three-dimensional spatial position and boundary information of all sub-cells are recorded one by one to generate the first recursive subdivision cell set;

[0162] Based on the first recursively subdivided cell set, the steps of obtaining the set of cells to be subdivided and obtaining the first recursively subdivided cell set are repeated until the coupled morphological complexity of all cells meets the set threshold and no longer intersects with the line vectorized structure. Then, all cells are topologically connected to generate an adaptive terrain grid model.

[0163] Specifically, based on the initial complexity grid, the system will check and evaluate each initial cell in the grid one by one. For the initial cell currently traversed, the coupled morphological complexity value previously calculated is first extracted. , and then compare this value with a "pre-set coupling morphological complexity threshold". The setting of this threshold is based on the statistical analysis of the distribution of the coupling morphological complexity of the initial cells in a large number of different scenarios, as well as the trade-off between the accuracy of the final grid model and the computational efficiency. Specifically, by analyzing the expression effect of the grid models generated under different complexity thresholds on the known terrain features, a critical value that can effectively distinguish between areas that need to be refined (i.e., complex terrain or containing important features) and areas that do not need to be refined (i.e., relatively flat and simple terrain) can be selected. For example, if statistics show that most of the flat areas Values ​​below 2.0, but containing areas with significant topographic changes or lined structures The value is usually higher than 3.5, then the coupled morphological complexity threshold can be set to 3.0. When compared with this threshold, the system will also detect whether the spatial position of the initial cell has a spatial geometric intersection relationship with the previously constructed line vector structure. This detection process includes checking whether the bounding box of the cell intersects with any line segment (defined by the start and end points) in the line vector structure. If the coupled morphological complexity value of an initial cell is Strictly greater than a pre-set coupling morphological complexity threshold (e.g. ), or if the spatial range of the initial cell geometrically intersects or overlaps with any line segment in the line vectorization structure, then the initial cell is determined to need further refinement. All initial cells that meet one of these two conditions (i.e., complexity exceeds the threshold or intersects with the line) will be screened out and aggregated to form a set of cells to be refined.

[0164] According to the set of cells to be subdivided, each initial cell in the set marked as to be subdivided is processed. The specific operation is to move the currently processed cell to be subdivided along its axis in the three-dimensional space. 、 、 The midpoints of the three coordinate axes are evenly divided. For example, a side length of The cubic unit cell of The range of the axis direction is (in ), then its The axis midpoint is , and similarly calculate Axis and The midpoint of the axis, through the three mutually orthogonal planes formed by these three midpoints, accurately cuts the original cell to be subdivided into eight small cells of exactly the same size and shape as cubes. The side lengths of these eight sub-cells are half of the side length of the original cell, that is, For each newly generated child cell, the system records its position in three-dimensional space (for example, through its minimum corner coordinates or center coordinates) and its new boundary range (that is, the minimum and maximum values ​​on each coordinate axis), and collects all the child cells generated after such an octree partitioning of all the cells to be subdivided to form the first recursive subdivision cell set.

[0165] Based on the first recursive subdivision cell set, the system will start an iterative refinement process. The core of this process is to repeat the same steps as generating the cell set to be subdivided and generating the first recursive subdivision cell set. Specifically, for each sub-cell in the first recursive subdivision cell set, the coupled morphological complexity of the curvature attribute points contained in it will be recalculated (if there are still curvature attribute points in the sub-cell), and its coupled morphological complexity will be checked again to see if it exceeds the pre-set coupled morphological complexity threshold (this threshold remains unchanged during the recursive process, for example, it is still 3.0). At the same time, the spatial position of the sub-cell will be re-checked to see if it exists with the line vectorization structure. In the spatial geometric intersection relationship, if a sub-cell meets one of the two conditions, it will be added to the new "set of cells to be subdivided". Then, each cell in this new set of cells to be subdivided will be octree-divided again to generate a smaller set of sub-cells. This cycle of "calculating complexity, checking intersections, screening subdivisions, and subdividing subdivisions" will continue. Each round of iteration will make finer divisions of the areas that need to be further refined until the following two termination conditions are met: First, the coupled morphological complexity of all current cells (regardless of their size) no longer exceeds the pre-set coupled morphological complexity threshold (for example, the coupled morphological complexity of all cells is greater than the pre-set coupled morphological complexity threshold). ); second, the spatial positions of all current cells no longer geometrically intersect with the line vectorization structure. When these two conditions are met at the same time, the recursive subdivision process stops. At this time, all cells in the system (including the initial cells that have not been subdivided and the sub-cells that reach the final size after multiple levels of subdivision) together constitute the final cell set. Finally, all cells in this final cell set are meshed topologically. For example, the MarchingCubes algorithm can be used to generate triangular facets connecting adjacent cells based on whether each cell corner is inside the object (judged by checking whether there are curvature attribute points or the relationship with the line vectorization structure), thereby constructing a three-dimensional grid model that can adaptively reflect the complexity of the terrain and accurately express the line position, that is, an adaptive terrain grid model.

Claims

1. A three-dimensional environment reconstruction system for a marking robot working area, characterized in that: The system comprises: The point cloud attribute calculation module performs a ground scan of the marking robot's working area to obtain the original point cloud, performs coordinate system alignment on the original point cloud, traverses all data points, and calculates the principal curvature value and principal direction value of each data point to establish a curvature attribute point cloud; A line feature recognition module sets a curvature screening threshold based on the curvature attribute point cloud and traverses all data points to filter out points whose principal curvature values ​​are greater than the curvature screening threshold to obtain a trace candidate point set. Based on the trace candidate point set, a discrete line feature cluster is established; A trace network construction module calculates the center position and main direction of each cluster based on the discrete line feature clusters, connects clusters whose spatial distance is below a threshold and whose main directions are consistent, generates a line connection path graph, and fits and establishes a line vectorization structure based on the line connection path graph; A multi-resolution grid generation module divides the working area of ​​the marking robot into initial cells based on the curvature attribute point cloud, establishes an initial complexity grid, and obtains an adaptive terrain grid model based on the initial complexity grid.

2. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the curvature attribute point cloud are: The marking robot performs ground laser scanning on the work area, collects and records the spatial three-dimensional position coordinates of the area, constructs a set of spatial coordinate points, and generates an original point cloud; Based on the original point cloud, with a predefined reference coordinate system as a reference, by performing coordinate transformation on the set of spatial coordinate points point by point, the position of each spatial coordinate point is adjusted to the corresponding position in the reference coordinate system, thereby generating an aligned original point cloud; Based on the aligned original point cloud, the spatial geometric distribution in the neighborhood of each spatial coordinate point is calculated point by point, the principal curvature value and the corresponding principal direction vector of each spatial coordinate point are extracted, and the calculated principal curvature value and principal direction vector are assigned to the corresponding spatial coordinate points in turn to generate a curvature attribute point cloud.

3. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the trace candidate point set are: Based on the curvature attribute point cloud, the principal curvature values ​​of all spatial coordinate points in the curvature attribute point cloud are counted, and the mean of the principal curvature values ​​of all spatial coordinate points is calculated, and the mean is taken as the curvature screening threshold to generate the curvature screening threshold; Based on the curvature screening threshold, the principal curvature value of each spatial coordinate point in the curvature attribute point cloud is traversed point by point, and whether the principal curvature value of the spatial coordinate point exceeds the curvature screening threshold is judged one by one. The spatial coordinate points whose principal curvature value exceeds the curvature screening threshold are marked one by one to generate a trace candidate point set.

4. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the discrete line feature cluster are: Based on the trace candidate point set, extract the three-dimensional coordinate value and local direction vector of each trace candidate point, traverse point by point to construct a combination of spatial position and direction relationships between two trace candidate points, and form a set of trace candidate point pairs; Calculating the comprehensive morphological similarity between each pair of trace candidate points based on the set of trace candidate point pairs; Based on the comprehensive morphological similarity between each pair of trace candidate points, the similarity between all trace candidate points is threshold-screened. The trace candidate point pairs with comprehensive morphological similarity higher than the set threshold are regarded as the connecting edges of the same clustering unit. All trace candidate points that form a connection relationship are aggregated into independent subsets, and discrete line feature clusters are obtained according to the aggregation relationship.

5. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the line connection path diagram are as follows: Based on the discrete line feature clusters, the three-dimensional spatial position coordinates of all trace candidate points in each discrete line feature cluster are traversed one by one, the spatial position coordinates of all trace candidate points are averaged, and the average value is defined as the center position of the discrete line feature cluster to generate a discrete line feature cluster center position set; Based on the center position set of the discrete line feature cluster, all trace candidate points of the discrete line feature cluster are traversed one by one, and the principal component feature vector of each discrete line feature cluster is extracted by principal component analysis, and the principal component feature vector is defined as the principal direction vector of the discrete line feature cluster to generate a set of principal direction vectors of the discrete line feature cluster; Based on the center position set of the discrete line feature clusters and the main direction vector set of the discrete line feature clusters, the spatial distance and the main direction angle between any two discrete line feature clusters are compared one by one, and the discrete line feature clusters whose spatial distance values ​​are less than a preset spatial threshold and whose main direction angles are less than a preset angle threshold are paired and connected to form a connection relationship and defined as a connection path, and a line connection path graph is generated.

6. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1, characterized in that: The steps for obtaining the line vectorization structure are as follows: Based on the line connection path graph, all discrete line feature clusters in each connection path in the line connection path graph are traversed one by one, the three-dimensional spatial position coordinates of all trace candidate points contained in each discrete line feature cluster are extracted, and all three-dimensional spatial position coordinates of each discrete line feature cluster are summarized to generate a three-dimensional spatial position coordinate set of the connection path; According to the three-dimensional spatial position coordinate set of the connection path, linear fitting of the spatial point cloud is performed for each connection path, the direction vector and the center position of the fitting line segment are calculated by the least squares method, and the fitting line segment parameters of the connection path are obtained one by one to generate a connection path fitting line segment parameter set; Based on the connection path fitting line segment parameter set, the fitting line segment parameters are converted into a standard vector representation format one by one, and the three-dimensional coordinates of the start and end endpoints and the vector direction of the fitting line segment are recorded in sequence to construct a line vectorization structure.

7. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the initial complexity grid are: Based on the curvature attribute point cloud, the boundary range and spatial size parameters of the marking robot's working area are set, the entire marking robot's working area is divided into regular cubic cells according to a uniform side length, and each curvature attribute point in the curvature attribute point cloud is assigned to the cubic cell according to the three-dimensional position coordinate mapping to generate an initial cell set of the marking robot's working area; Calculating the coupling morphological complexity of each initial cell according to the initial cell set of the working area of ​​the marking robot; Based on the coupled morphological complexity of each initial cell, the initial cell set of the entire marking robot working area is traversed, and the spatial position indexes corresponding to all initial cells are paired with the coupled morphological complexity one by one to uniformly organize and form a complexity expression grid of the spatial structure changes in the area to generate the initial complexity grid.

8. The three-dimensional environment reconstruction system of the working area of ​​the marking robot according to claim 1 is characterized in that: The steps for obtaining the adaptive terrain grid model are: Based on the initial complexity grid, the coupling morphological complexity of each initial cell in the initial complexity grid is traversed one by one, and the numerical value is compared item by item with a pre-set coupling morphological complexity threshold. At the same time, whether the spatial position of each initial cell has a spatial geometric intersection relationship with the line vectorization structure is detected, and the initial cells that meet the coupling morphological complexity exceeding the threshold or having an intersection relationship are screened to generate a set of cells to be subdivided; According to the set of cells to be subdivided, each cell to be subdivided is evenly divided along the midpoint position of the three-dimensional spatial coordinate axis to form eight smaller sub-cells, and the three-dimensional spatial position and boundary information of all sub-cells are recorded one by one to generate a first recursive subdivision cell set; Based on the first recursively subdivided cell set, the steps of obtaining the cell set to be subdivided and obtaining the first recursively subdivided cell set are repeated until the coupled morphological complexity of all cells meets the set threshold and no longer intersects with the line vectorized structure, and then all cells are topologically connected in a grid to generate an adaptive terrain grid model.

Images