SLAM (Simultaneous Localization and Mapping) optimization mapping method and device combining geometric verification and constraint of reflector

Abstract

The invention discloses an SLAM (Simultaneous Localization and Mapping) optimization mapping method and device combining geometric verification and constraint of a reflector, relates to the technical field of industrial-grade mobile robot navigation, and solves the problem that in the prior art, a continuous constraint mechanism cannot be established in a mapping process to correct accumulative errors in motion. The method comprises the following steps: acquiring an original laser radar point cloud containing a reflector, screening out reflection points with the distance smaller than a preset distance, and carrying out clustering and quintuple geometric verification on the screened points by adopting a density clustering algorithm, so as to obtain a multi-dimensional geometric verification mechanism based on PCA (Principal Component Analysis); local coordinates of the center of the reflector are calculated and converted into global coordinates, the reflector is used as a stable Landmark depth to be fused into an SLAM back-end graph optimization framework, and optimization nodes including a timestamp, a robot pose, a reflector ID and an observation pose are constructed based on a global coordinate system to participate in graph optimization; and the matching problem caused by insufficient natural characteristics or environment change in a long corridor scene can be solved in a targeted manner.

Description

Technical Field

[0001] This application relates to the field of industrial mobile robot navigation technology, and in particular to a SLAM graph optimization mapping method and apparatus that combines reflector geometric verification and constraints. Background Technology

[0002] In the field of industrial mobile robot navigation, reflectors are widely used as key artificial landmarks. Traditional solutions rely on LiDAR to identify highly reflective markers for basic localization, especially in long corridor scenarios with limited features, making it the mainstream approach. Existing technologies deploy reflectors as localization reference markers and perform pose estimation based on simple geometric feature matching. However, this superficial application has fundamental limitations: reflectors are only used for passive target identification and are not deeply integrated into the SLAM optimization framework; environmental interference causes a sharp drop in recognition rate; and accumulated errors cannot be effectively suppressed in long-distance scenarios.

[0003] In the field of Simultaneous Localization and Mapping (SLAM) in structured scenarios such as long corridors, existing technologies mainly rely on LiDAR scanning to acquire wall geometric features, while utilizing reflectors deployed in the environment as auxiliary positioning markers. A typical implementation process involves extracting straight-line features parallel to the walls (such as door frames and corners) for pose estimation, and then using reflectors as high-reflectivity objects for simple identification and matching. These reflectors are usually identified using basic clustering and distance judgment. However, this traditional approach faces significant challenges in long corridor scenarios: repetitive structures lead to feature matching ambiguity, and straight-line features in a single direction cause lateral positioning drift. Real-world data shows that mapping errors in corridors exceeding 20 meters can reach 5%, far exceeding the inherent error level of sensors.

[0004] Main defects:

[0005] 1. The reflector is only used as an auxiliary tool for initial positioning and has not been deeply integrated into the SLAM optimization framework. It cannot establish a continuous constraint mechanism to correct the cumulative error in motion and lacks a PCA principal direction analysis and multi-dimensional geometric verification mechanism.

[0006] 2. The entire process from data source to output lacks robust design. The feature extraction process lacks adaptive clustering, strength stability verification, and point distribution uniformity detection. The strategy of determining the pose on the first observation leads to irreversible error propagation and makes it impossible to achieve two-stage feature management of candidate-confirmation. Summary of the Invention

[0007] The purpose of this application is to overcome the problems in the existing technology that it is impossible to establish a continuous constraint mechanism to correct the cumulative error in motion during the mapping process, lack PCA principal direction analysis, multi-dimensional geometric verification mechanism, and robust design, and to provide a SLAM map optimization mapping method and device that combines reflector geometric verification and constraint.

[0008] Firstly, a SLAM graph optimization mapping method combining reflector geometric verification and constraints is provided, in which reflectors are deployed according to a strategy within a long corridor. The method includes:

[0009] Obtain the raw lidar point cloud containing the reflector;

[0010] An initial candidate set is formed by selecting points of type reflective point and distance less than a preset distance from the original lidar point cloud;

[0011] Density clustering algorithm is used to spatially group the initial candidate set, and each cluster is subjected to five-fold geometric verification of size, elongation, length, strength and uniformity.

[0012] Calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification;

[0013] Transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system;

[0014] The reflector is integrated as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, optimization nodes containing timestamps, robot poses, reflector IDs, and observation poses are constructed to participate in graph optimization.

[0015] In some possible implementations, the preset distance is any value between 8m and 16m.

[0016] In some possible implementations, a density-based clustering algorithm is used to spatially group the initial candidate set, and each cluster undergoes a five-fold geometric verification of size, elongation, length, strength, and uniformity, including:

[0017] Density clustering algorithm is used, with fixed parameters for clustering, including the neighborhood radius. =0.05, minimum number of cluster points minPts=3, neighborhood determination criterion is:

[0018] ;

[0019] Where pi represents the i-th point and pj represents the j-th point. This represents the distance between points pi and pj. If the neighborhood size of a point is greater than or equal to minPts, it is used as the core point to trigger recursive expansion to form a cluster Ck.

[0020] Apply size constraints to eliminate excessively small clusters: |Ck|≥mthresRfCnt;

[0021] Perform PCA principal direction analysis: construct the points within the cluster , , , , where P is the coordinate matrix of points within the cluster, μ is the mean vector of points within the cluster, and C is the covariance matrix;

[0022] Perform eigenvalue decomposition on the covariance matrix C Slenderness ratio :

[0023] ;

[0024] Where λ1 and λ2 are the eigenvalues ​​of the covariance matrix, v1 and v2 represent the eigenvectors corresponding to the eigenvalues, the principal axis is taken as a = v2, and the projection value of the principal axis is t. i With length L as:

[0025] ;

[0026] The constraint range for length L is:

[0027] 0.8L0 <L<1.2L0;

[0028] Strength stability:

[0029] ;

[0030] Among them, I i This represents the reflection intensity value at point i. Represents the mean intensity of points within the cluster. This represents the range of intensity fluctuations at points within the cluster.

[0031] Uniform distribution along the principal axis: sorting the projections Define the ideal spacing Where L represents the projected length along the principal axis, and the spacing variance V is:

[0032] .

[0033] In some possible implementations, mthresRfCnt=5.

[0034] In some possible implementations, the coordinates of the reflector center in the local coordinate system are calculated using the principal axis projection midpoint back substitution algorithm:

[0035] ;

[0036] Among them, t mid represents the midpoint value of the projection, and c represents the local coordinates of the reflector center in the local coordinate system.

[0037] In some possible implementations, the local coordinates of the reflector center in the local coordinate system are transformed to the global coordinate system, including:

[0038] Build a unique data structure for each valid reflector, including geometric center coordinates, original point cloud set, timestamp and unique identifier;

[0039] The coordinate transformation process achieves accurate mapping across multiple coordinate systems through a homogeneous transformation matrix: First, the point cloud position Plidar in the laser coordinate system is converted into the robot coordinate system position Probot, which is mathematically expressed as Probot=R(θlidar)×Plidar+Tlidar, where R is the rotation matrix and T is the translation vector.

[0040] Based on robot real-time pose [x] r ,y r ,θ r ], through transformation matrix The most discriminative transformation is performed to the global coordinate system, and the relative pose of the robot with respect to the reflector [Δx,Δy,Δθ] is calculated.

[0041] In some possible implementations, the reflector is integrated as a stable landmark depth into the SLAM backend graph optimization framework. Based on a global coordinate system, optimization nodes containing timestamps, robot poses, reflector IDs, and observation poses are constructed to participate in graph optimization, including:

[0042] Maintain the following two types of optimization nodes: trajectory nodes and reflector nodes. Trajectory nodes: one for each laser frame, variables [x, y, θ]; Reflector nodes: one for each reflector ID, variables [x, y, θ], recording the observation list {distance², node ID}.

[0043] Construct three types of constraint terms:

[0044] Adjacent frame constraints: For consecutive trajectory nodes i and i+1, calculate the relative positions given by local SLAM. The residual function is a three-dimensional vector:

[0045]

[0046] Huber robust loss function is used to suppress outliers, with a threshold set to δ=0.2m;

[0047] Reflector distance constraint: For the case where the reflector si is observed at trajectory node t, construct the radial distance residual: The reflector distance constraint provides radial distance information but does not constrain the tangential angle; the weight is dynamically set based on the observation quality.

[0048] Anchor point constraints: Stable reflectors detected in the initial mapping phase are marked as frozen anchor points. The criteria for determining a stable reflector are: a cumulative number of observation frames greater than or equal to 50, and a pose variance less than 0.005m. 2 And the distance from the starting point of the mapping is less than 10m;

[0049] Joint optimization: Solved using the Ceres nonlinear optimization library, the overall objective function is:

[0050]

[0051] Linear solver: SPARSE_NORMAL_CHOLESKY; Maximum iterations: 200; Convergence tolerance: 10 -12 Asynchronous execution: runs in an independent thread, with a single execution time of <200ms;

[0052] After optimization and convergence, perform the following operations:

[0053] Pose backfilling: Synchronously update the global pose of all trajectory nodes and reflector nodes;

[0054] Front-end feedback: Optimization results are fed back to the front-end via callback interfaces for: map display updates and location database updates;

[0055] Dynamic cropping: In positioning mode, when the number of sub-images exceeds 5, the oldest sub-image is deleted.

[0056] Secondly, a SLAM map optimization and mapping device combining reflector geometry verification and constraints is provided, including:

[0057] The acquisition module is used to acquire the original LiDAR point cloud.

[0058] The filtering module is used to filter out points of type reflective point and whose distance is less than a preset distance from the original lidar point cloud to form an initial candidate set;

[0059] The clustering and validation module is used to spatially group the initial candidate set using a density clustering algorithm and perform five-fold geometric validation on each cluster, including size, elongation, length, strength, and uniformity.

[0060] The calculation module is used to calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification.

[0061] The coordinate system transformation module is used to transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system;

[0062] The optimization module integrates the reflector as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, it constructs optimization nodes that include timestamps, robot poses, reflector IDs, and observation poses to participate in graph optimization.

[0063] Thirdly, a computer-readable storage medium is provided for program code executable by a device, the program code including steps for performing a method as described in any of the implementations of the first aspect above.

[0064] Fourthly, an electronic device is provided, the electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in any of the implementations of the first aspect above.

[0065] This application has the following beneficial effects: By strategically deploying reflectors in long corridors, these reflectors, acting as artificial anchors, possess stable reflective characteristics and fixed spatial positions, providing a reliable reference benchmark for matching. This effectively addresses matching problems caused by insufficient natural features or environmental changes in long corridor scenarios, ensuring the accuracy of mapping and positioning. Through engineering optimization measures such as distance filtering, motion distortion compensation, data quality closed-loop, and real-time assurance, the system achieves mapping accuracy better than 5% in long corridor scenarios, effectively solving the problem of accumulated errors in traditional solutions in feature-scarce environments. Attached Figure Description

[0066] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0067] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart of the SLAM graph optimization and mapping method combining reflector geometry verification and constraints according to Embodiment 1 of this application;

[0069] Figure 2 This is a rendering of the reflector used as a node in Embodiment 1 of this application.

[0070] Figure 3(a) is a rendering of the pre-constraint mapping in Embodiment 1 of this application;

[0071] Figure 3(b) is a diagram showing the effect of the constrained mapping in Embodiment 1 of this application;

[0072] Figure 4 This is a structural block diagram of the SLAM map optimization and mapping device combining reflector geometry verification and constraints according to Embodiment 2 of this application;

[0073] Figure 5 This is a schematic diagram of the internal structure of the electronic device according to Embodiment 4 of this application.

[0074] Figure label:

[0075] 100. Acquisition Module; 200. Filtering Module; 300. Clustering and Validation Module; 400. Calculation Module; 500. Coordinate System Transformation Module; 600. Optimization Module. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example 1

[0078] like Figure 1 As shown, Embodiment 1 of this application relates to a SLAM map optimization and mapping method that combines reflector geometry verification and constraints. This method involves deploying reflectors according to a strategy within a long corridor. The method includes:

[0079] S100. Acquire the raw LiDAR point cloud. Using reflectors strategically deployed along the long corridor, the raw point cloud output by the LiDAR at a frequency of 10Hz contains a large number of spatial points. These points, acting as artificial anchors, have stable reflection characteristics and fixed spatial positions, providing a reliable reference benchmark for matching.

[0080] S200. Select points of type reflective point and distance less than a preset distance from the original lidar point cloud to form an initial candidate set.

[0081] First, the raw point cloud is preprocessed, and core attributes are analyzed point by point: reflection intensity Ii, spatial coordinates pi=(xi,yi), scanning angle θi (scanning angle of the i-th point), etc., and subsequent processing is performed directly on the scanning sequence. To reduce data interference, basic filtering conditions are first applied: The initial candidate set is formed by selecting points of reflective type that are less than 12m away. This allows for the rapid elimination of distant and non-reflective points, ensuring the stability of subsequent geometric constraints. In the potential reflective point selection stage, spatial aggregation is prepared directly based on the above pre-selection results, without using additional constraints such as multi-threaded sector division or angle difference thresholds. Through this stage, a candidate point set with reflective characteristics can be separated from the original point cloud, laying the foundation for subsequent clustering.

[0082] S300. The initial candidate set is spatially grouped using a density clustering algorithm, and each cluster is subjected to a five-fold geometric verification of size, slenderness, length, strength, and uniformity.

[0083] Specifically, the clustering process employs the density-based clustering (DBSCAN) algorithm, using fixed parameters to ensure stability: neighborhood radius. =0.05, minimum number of cluster points minPts=3. Neighborhood criterion is:

[0084]

[0085] Where pi = (xi, yi) represents the two-dimensional coordinates of the i-th point, pj represents the j-th point, and pi represents the ith point. This represents the distance between points pi and pj.

[0086] When the neighborhood size of a point is ≥ minPts, it serves as the core point, triggering recursive expansion to form a cluster Ck, where Ck represents the k-th cluster. After clustering, a size constraint is first applied to remove excessively small clusters: |Ck| ≥ mthresRfCnt. For example, in this embodiment, mthresRfCnt = 5.

[0087] Cluster validity verification is performed using a joint criterion of geometric and strength criteria. First, principal direction analysis (PCA) is conducted: the points within each cluster are constructed as follows: , , , , where P is the coordinate matrix of points within the cluster, μ is the mean vector of points within the cluster, and C is the covariance matrix.

[0088] Perform eigenvalue decomposition on the covariance matrix C Slenderness ratio :

[0089] ;

[0090] Where λ1 and λ2 are the eigenvalues ​​of the covariance matrix, v1 and v2 represent the eigenvectors corresponding to the eigenvalues, and the slenderness ratio is... To determine whether it is a strip structure, we take the principal axis a = v2: the principal axis direction vector, and the principal axis projection value t.i The length L (where L represents the projected length along the principal axis) is:

[0091] ;

[0092] Reflector width constraint L0 = 10cm (adjusted according to the specific reflector size in actual applications):

[0093] 0.8L0 <L<1.2L0;

[0094] To suppress intensity and distribution anomalies, two more stability criteria are added: Intensity stability:

[0095] ;

[0096] Among them, I i This represents the reflection intensity value at point i. Represents the mean intensity of points within the cluster. This represents the range of intensity fluctuations within a cluster. As a criterion for strength and stability;

[0097] Uniform distribution along the principal axis: sorting the projections After projection sorting, t(i) represents the projection value of the i-th point along the principal axis, and the ideal spacing is defined. Where L represents the projected length along the principal axis, and the spacing variance V is:

[0098] ,in, This is a criterion for uniform distribution.

[0099] In this embodiment, clusters that meet the five criteria of "size + elongation + length + intensity + uniformity" will be accepted as candidates for reflective strips.

[0100] S400: Calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification.

[0101] The validated clusters require the calculation of their geometric center points as reflector feature positions. This embodiment uses the geometric center obtained by substituting the midpoint of the principal axis projection. ;

[0102] Among them, t mid c represents the midpoint value of the projection, c represents the local coordinates of the reflector center in the local coordinate system, and c_W represents the coordinates of the reflector center in the world coordinate system.

[0103] Subsequently, based on the current pose, it is mapped to the world coordinate system. In the mapping mode (modeMap), to avoid adopting false features that are too close to the reflector, a proximity filter is added: features >1 m are retained, otherwise ignored; in other modes, they are directly retained. In summary, the reflector center is calculated using the principal axis projection midpoint back substitution algorithm. Combined with DBSCAN clustering with fixed parameters, PCA elongation, and joint constraints of length / intensity / uniformity, the reflector center feature points are robustly extracted. The final output is the world coordinate system center that satisfies all criteria, which is used for subsequent optimization and localization.

[0104] S500: Transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system.

[0105] The successfully extracted feature point set enters the structured processing stage. A dedicated data structure is constructed for each valid reflector, including the geometric center coordinates, the original point cloud set, a timestamp, and a unique identifier (generated using "timestamp + sequence number" encoding). The coordinate transformation stage achieves accurate mapping across multiple coordinate systems through a homogeneous transformation matrix: firstly, the point cloud position Plidar in the laser coordinate system is converted to the robot coordinate system position Probot, the mathematical expression of which is Probot=R(θlidar)×Plidar+Tlidar, where R is the rotation matrix and T is the translation vector;

[0106] Based on robot real-time pose [x] r ,y r ,θ r ], through transformation matrix The most discriminative transformation is performed to the global coordinate system, and the robot's relative pose [Δx, Δy, Δθ] with respect to the reflector is calculated. This parameter provides the optimizer with key spatial constraints. In long corridors, robots often move at a constant velocity in a straight line. The robot's displacement during the LiDAR scanning process will cause point cloud distortion (the offset is more obvious for distant points).

[0107] The S600 integrates the reflector as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, it constructs optimization nodes that include timestamps, robot poses, reflector IDs, and observation poses to participate in graph optimization.

[0108] Maintain the following two types of optimization nodes:

[0109] Trajectory node: One for each laser frame, with variables [x,y,θ].

[0110] Reflector node: Each reflector ID corresponds to one node, with variables [x, y, θ], and records the observation list {distance², node ID}.

[0111] Constraint type:

[0112] Construct three types of constraint terms:

[0113] Adjacent frame constraints: For consecutive trajectory nodes i and i+1, calculate the relative positions given by local SLAM. The residual function is a three-dimensional vector:

[0114]

[0115] Huber robust loss function is used to suppress outliers, with the threshold set to δ=0.2m.

[0116] Reflector distance constraint: For the case where the reflector si is observed at trajectory node t, construct the radial distance residual (one-dimensional constraint):

[0117]

[0118] The reflector distance constraint only provides radial distance information and does not constrain the tangential angle, thus adapting to the characteristics of corridor geometric degradation. The weights are dynamically set based on observation quality.

[0119] Standard observation: w=1.0;

[0120] High confidence observation (intensity stability and number of observations > 20 frames): w = 10.0.

[0121] Anchor point constraints: Stable reflectors detected in the initial mapping stage are marked as frozen anchor points. The criteria for determining stable reflectors are as follows:

[0122] Cumulative number of observation frames ≥ 50;

[0123] Position variance Less than 0.005m 2 ;

[0124] The distance from the starting point of the mapping is less than 10m.

[0125] In this step, freezing anchor points fixes the parameter block during optimization, providing an absolute reference for the global coordinate system and effectively suppressing long-distance cumulative drift.

[0126] Joint optimization:

[0127] Solving using the Ceres nonlinear optimization library, the overall objective function is:

[0128]

[0129] Linear solver: SPARSE_NORMAL_CHOLESKY;

[0130] Maximum iterations: 200, convergence tolerance: 10 -12 ;

[0131] Asynchronous execution: Runs in an independent thread, with a single execution time of <200ms.

[0132] Results updated:

[0133] After optimization and convergence, perform the following operations:

[0134] Pose backfilling: Synchronously update the global pose of all trajectory nodes and reflector nodes;

[0135] Front-end feedback: Optimization results are fed back to the front-end via callback interfaces for: map display updates and location database updates;

[0136] Dynamic cropping: In positioning mode, when the number of sub-images exceeds 5, the oldest sub-image is deleted.

[0137] Images of reflectors used as nodes in a graph, such as... Figure 2 As shown, the blue-and-yellow rectangle is the reflector, and the red square is the trajectory node. The relevant process is as follows:

[0138] Front-end: Reflection point filtering → DBSCAN clustering → Five-fold verification → Center calculation → Coordinate transformation → Distortion compensation.

[0139] Submission: Generate observation data → Create trajectory node + reflector node → Record distance constraints.

[0140] Backend: Construct factor graph (pose constraint + distance constraint + anchor point) → Ceres asynchronous optimization → backfill pose → feedback to frontend.

[0141] The comparative effects of introducing reflectors are shown in Figures 3(a) and 3(b). Figure 3(a) shows the matching results before the improved constraints. When relying solely on the natural features of the LiDAR for matching, point cloud matching fails in some areas, the proportion of unmatched points increases, and it is difficult to output stable and accurate positioning results. Figure 3(b) shows the matching effect after introducing reflector constraints. By strategically deploying reflectors in the long corridor, they serve as artificial anchor points with stable reflection characteristics and fixed spatial positions, providing a reliable reference benchmark for matching. Even if temporary environmental changes occur in the corridor, reflectors can still be stably identified and provide strong constraints. The matching logic combined with reflector features can effectively avoid the problem of natural feature failure. At the same time, the global pose consistency of reflectors further improves the stability of matching, significantly reduces unmatched points, and significantly improves the accuracy of matching results. This result shows that the introduction of reflectors can specifically solve the matching problems caused by insufficient natural features or environmental changes in long corridor scenarios, ensuring the accuracy of mapping and positioning.

[0142] In this embodiment, reflectors strategically deployed within the long corridor serve as artificial anchor points with stable reflective characteristics and fixed spatial positions, providing a reliable reference benchmark for matching. This effectively addresses matching issues caused by insufficient natural features or environmental changes in long corridor scenarios, ensuring the accuracy of mapping and positioning. Through engineering optimization measures such as distance filtering, motion distortion compensation, data quality closed-loop, and real-time assurance, the system achieves mapping accuracy better than 5% in long corridor scenarios, effectively solving the cumulative error problem of traditional solutions in feature-scarce environments.

[0143] Example 2

[0144] like Figure 4 As shown, Embodiment 2 of this application relates to a SLAM map optimization and mapping device that combines reflector geometry verification and constraints, comprising:

[0145] Acquisition module 100 is used to acquire the original lidar point cloud;

[0146] The filtering module 200 is used to filter out points of type reflective point and whose distance is less than a preset distance from the original lidar point cloud to form an initial candidate set;

[0147] Clustering and validation module 300 is used to spatially group the initial candidate set using a density clustering algorithm and to perform five-fold geometric validation of each cluster in terms of size, slenderness, length, strength and uniformity.

[0148] Calculation module 400 is used to calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification;

[0149] The coordinate system transformation module 500 is used to transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system.

[0150] The optimization module 600 is used to integrate the reflector as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, it constructs optimization nodes that include timestamps, robot poses, reflector IDs, and observation poses to participate in graph optimization.

[0151] It should be noted that other specific implementations of the SLAM map optimization and mapping device combining reflector geometric verification and constraints in this embodiment can be found in the specific implementations of the SLAM map optimization and mapping method combining reflector geometric verification and constraints described above. To avoid redundancy, they will not be repeated here.

[0152] Example 3

[0153] This application relates to a computer-readable storage medium in Embodiment 3, which is used for program code executed by a device, the program code including steps for performing a method as described in any implementation of Embodiment 1 of this application;

[0154] The computer-readable storage medium may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM); the computer-readable storage medium may store program code, and when the program stored in the computer-readable storage medium is executed by a processor, the processor is used to perform the steps of the method in any of the implementations of Embodiment 1 of this application.

[0155] Example 4

[0156] like Figure 5 As shown, an electronic device according to Embodiment 4 of this application includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the method in any of the implementations in Embodiment 1 of this application.

[0157] The processor can be a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), graphics processing unit (GPU), or one or more integrated circuits, used to execute related programs to implement the method in any of the implementations of Embodiment 1 of this application.

[0158] The processor can also be an integrated circuit electronic device with signal processing capabilities. In implementation, each step of the method in any of the implementations of Embodiment 1 of this application can be completed by the integrated logic circuitry in the processor's hardware or by software instructions.

[0159] The aforementioned processor can also be a general-purpose processor, digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the functions required by the units included in the data processing apparatus of the embodiments of this application, or executes the methods in any implementation of Embodiment 1 of this application.

[0160] The above are merely preferred embodiments of this application; however, the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and its improved concept, should be covered within the scope of protection of this application.

Claims

1. A SLAM graph optimization and mapping method combining reflector geometric verification and constraints, characterized in that, The method of strategically deploying reflectors within a long corridor includes: Obtain the original LiDAR point cloud; An initial candidate set is formed by selecting points of type reflective point and distance less than a preset distance from the original lidar point cloud; Density clustering algorithm is used to spatially group the initial candidate set, and each cluster is subjected to five-fold geometric verification of size, elongation, length, strength and uniformity. Calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification; Transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system; The reflector is integrated as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, optimization nodes containing timestamps, robot poses, reflector IDs, and observation poses are constructed to participate in graph optimization.

2. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 1, characterized in that, The preset distance is any value between 8m and 16m.

3. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 1, characterized in that, Density clustering algorithm is used to spatially group the initial candidate set, and each cluster undergoes a five-fold geometric verification of size, elongation, length, strength, and uniformity, including: Density clustering algorithm is used, with fixed parameters for clustering, including the neighborhood radius. =0.05, minimum number of cluster points minPts=3, neighborhood determination criterion is: ； Where pi represents the i-th point and pj represents the j-th point. This represents the distance between points pi and pj. If the neighborhood size of a point is greater than or equal to minPts, it is used as the core point to trigger recursive expansion to form a cluster Ck. Apply size constraints to eliminate excessively small clusters: |Ck|≥mthresRfCnt; Perform PCA principal direction analysis: construct the points within the cluster , , , , where P is the coordinate matrix of points within the cluster, μ is the mean vector of points within the cluster, and C is the covariance matrix; Perform eigenvalue decomposition on the covariance matrix C Slenderness ratio : ； Where λ1 and λ2 are the eigenvalues ​​of the covariance matrix, v1 and v2 represent the eigenvectors corresponding to the eigenvalues, the principal axis is taken as a = v2, and the projection value of the principal axis is t. i With length L as: ； The constraint range for length L is: ； Where L0 is the actual length of the reflector; Strength stability: ； Among them, I i This represents the reflection intensity value at point i. Represents the mean intensity of points within the cluster. This represents the range of intensity fluctuations at points within the cluster. Uniform distribution along the principal axis: sorting the projections Define the ideal spacing Where L represents the projected length along the principal axis, and the spacing variance V is: 。 4. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 3, characterized in that, mthresRfCnt=5.

5. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 3, characterized in that, The coordinates of the reflector center in the local coordinate system are calculated using the principal axis projection midpoint substitution algorithm: ； Among them, t mid represents the midpoint value of the projection, and c represents the local coordinates of the reflector center in the local coordinate system.

6. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 1 or 5, characterized in that, Transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system, including: Build a unique data structure for each valid reflector, including geometric center coordinates, original point cloud set, timestamp and unique identifier; The coordinate transformation process achieves accurate mapping across multiple coordinate systems through a homogeneous transformation matrix: First, the point cloud position Plidar in the laser coordinate system is converted into the robot coordinate system position Probot, which is mathematically expressed as Probot=R(θlidar)×Plidar+Tlidar, where R is the rotation matrix and T is the translation vector. Based on robot real-time pose [x] r ,y r ,θ r ], through transformation matrix The most discriminative transformation is performed to the global coordinate system, and the relative pose of the robot with respect to the reflector [Δx,Δy,Δθ] is calculated.

7. The SLAM map optimization and mapping method combining reflector geometry verification and constraints as described in claim 6, characterized in that, The reflector is integrated as a stable landmark depth into the SLAM backend graph optimization framework. Based on a global coordinate system, optimization nodes containing timestamps, robot poses, reflector IDs, and observation poses are constructed to participate in graph optimization, including: Maintain the following two types of optimization nodes: trajectory nodes and reflector nodes. Trajectory nodes: one for each laser frame, variables [x, y, θ]; Reflector nodes: one for each reflector ID, variables [x, y, θ], recording the observation list {distance², node ID}. Construct three types of constraint terms: Adjacent frame constraints: For consecutive trajectory nodes i and i+1, calculate the relative positions given by local SLAM. The residual function is a three-dimensional vector: ； Huber robust loss function is used to suppress outliers, with a threshold set to δ=0.2m; Reflector distance constraint: For the case where the reflector si is observed at trajectory node t, construct the radial distance residual: The reflector distance constraint provides radial distance information but does not constrain the tangential angle; the weight is dynamically set based on the observation quality. Anchor point constraints: Stable reflectors detected in the initial mapping phase are marked as frozen anchor points. The criteria for determining a stable reflector are: a cumulative number of observation frames greater than or equal to 50, and a pose variance less than 0.005m. 2 And the distance from the starting point of the mapping is less than 10m; Joint optimization: Solved using the Ceres nonlinear optimization library, the overall objective function is: ； Linear solver: SPARSE_NORMAL_CHOLESKY; Maximum iterations: 200; Convergence tolerance: 10 -12 Asynchronous execution: runs in an independent thread, with a single execution time of <200ms; After optimization and convergence, perform the following operations: Pose backfilling: Synchronously update the global pose of all trajectory nodes and reflector nodes; Front-end feedback: Optimization results are fed back to the front-end via callback interfaces for: map display updates and location database updates; Dynamic cropping: In positioning mode, when the number of sub-images exceeds 5, the oldest sub-image is deleted.

8. A SLAM map optimization and mapping device combining reflector geometric verification and constraints, characterized in that, include: The acquisition module is used to acquire the original LiDAR point cloud. The filtering module is used to filter out points of type reflective point and whose distance is less than a preset distance from the original lidar point cloud to form an initial candidate set; The clustering and validation module is used to spatially group the initial candidate set using a density clustering algorithm and perform five-fold geometric validation on each cluster, including size, elongation, length, strength, and uniformity. The calculation module is used to calculate the local coordinates of the center of the light plate in the local coordinate system for the clusters that have passed the five-fold geometric verification. The coordinate system transformation module is used to transform the local coordinates of the reflector center in the local coordinate system to the global coordinate system; The optimization module integrates the reflector as a stable landmark into the SLAM backend graph optimization framework. Based on the global coordinate system, it constructs optimization nodes that include timestamps, robot poses, reflector IDs, and observation poses to participate in graph optimization.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is for program code executed by the device, the program code including steps for performing the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in any one of claims 1-7.

Images