Reflection column identifying and positioning method and system

Abstract

The invention discloses a reflective column recognition and positioning method and system, relates to the technical field of industrial-grade mobile robot navigation, and solves the problems that accumulative errors exist in an existing SLAM technology and the essential difference between a cylinder and a plane reflection strip cannot be distinguished. Dividing spatially adjacent points into point clusters by taking the radius of the reflective column as a clustering radius threshold, screening out the point clusters conforming to the features of the reflective column through a multi-geometric verification mechanism, calculating the center position and radius of the reflective column by adopting an improved algebraic fitting algorithm, and performing a three-stage management strategy on the newly detected features of the reflective column. The method is advantaged in that the reflection column can be accurately identified in a complex environment with scarce features or interference, false detection feature accumulation is prevented, system resource utilization is optimized, the false identification rate is reduced through a feature confirmation mechanism, the feature position of the confirmed features is smoothly updated through a dynamic learning rate mechanism, and the reflection column can be accurately identified in the complex environment with scarce features or interference.

Description

Technical Field

[0001] This application relates to the field of industrial-grade mobile robot navigation technology, and in particular to a method and system for identifying and locating reflective posts. Background Technology

[0002] In the field of industrial mobile robot navigation, reflective pillars, due to their 360° visible omnidirectional reflective properties, are increasingly becoming key artificial landmarks replacing traditional directional reflectors. Existing technologies primarily rely on LiDAR to identify highly reflective cylindrical objects for coarse localization, serving as a supplementary solution to SLAM systems in structured environments such as warehousing and logistics. Traditional methods construct a sparse localization network by deploying reflective pillars of specific dimensions, employing reflection point clustering and geometric center calculation principles to estimate robot pose.

[0003] However, this rudimentary application model has fundamental flaws: the reflective pillars are only used as discrete positioning reference points, failing to form a complete perception loop of multi-frame observation-confirmation-optimization; existing point cloud clustering algorithms cannot effectively distinguish between reflective pillars and wall reflective strips, resulting in a high misidentification rate in complex scenes; and the processing mechanism based on single-frame data causes significant jitter in the positioning results and cannot effectively suppress the cumulative error of the SLAM system.

[0004] In the field of Simultaneous Localization and Mapping (SLAM) in structured environments such as warehouses, existing technologies mainly rely on LiDAR scanning to acquire reflective point clouds of reflective pillars, and then form candidate point clusters through threshold filtering and distance clustering. A typical implementation process involves calculating the geometric center coordinates of the point clusters as an estimate of the reflective pillar's location, and directly incorporating the initial detection results into the global map. These reflective pillar coordinates are usually calculated per frame and, once confirmed, are not updated, forming a static landmark database. However, this traditional approach faces serious problems in real-world industrial scenarios: point cloud clustering does not consider geometric features, leading to reflective strips being misidentified as reflective pillars; the geometric center calculation method produces positioning errors as high as 10cm when partially occluded; and experimental data shows that landmark recognition stability in dynamic environments is less than 60%.

[0005] Main defects:

[0006] 1. The reflective column is only used as an auxiliary tool for initial positioning and is not deeply integrated into the SLAM optimization framework. It cannot establish a continuous constraint mechanism to correct the cumulative error during motion.

[0007] 2. The recognition of reflective pillars lacks a geometric feature verification mechanism. Existing solutions rely solely on distance clustering, which cannot distinguish the essential differences between cylindrical and planar reflective strips. The circle center localization algorithm has fundamental limitations. The traditional geometric center method does not consider the circular fitting model, resulting in systematic biases in scenarios with uneven point cloud distribution or partial occlusion. Summary of the Invention

[0008] The purpose of this application is to overcome the problems of accumulated errors and inability to distinguish the essential differences between cylindrical and planar reflective strips in existing SLAM technology, and to provide a reflective column identification and positioning method and system.

[0009] Firstly, a method for identifying and locating reflective pillars is provided, wherein reflective pillars are deployed at intervals within a site to ensure that the machine can observe at least two reflective pillars at any location within the site. The method includes:

[0010] Acquire point cloud data from the LiDAR on the machine;

[0011] From the point cloud data, extract potential reflective points that meet the following conditions: the point type is marked as REFLECTOR, and the Euclidean distance from the point to the lidar is less than the maximum effective detection distance set by the user.

[0012] Based on the spatial coordinates of each point in the extracted potential reflective point set, points whose spatial distance satisfies the proximity condition are aggregated into at least one point cluster according to a preset clustering radius threshold.

[0013] The point clusters are verified through a multi-geometric verification mechanism to select those that meet the characteristics of reflective pillars.

[0014] For the selected clusters of points that meet the characteristics of reflective pillars, an improved algebraic fitting algorithm is used to calculate the center position and radius of the reflective pillars;

[0015] Each new detected feature is divided into candidate features, and a unique identifier is assigned. Its initial position, observation count and first observation frame number are recorded. All candidate features are stored in a hash table structure, and candidate features that have not been observed again for more than a preset number of frames are cleaned up periodically.

[0016] Candidate features that have reached a certain threshold after multiple frames of observation are upgraded to confirmed features and added to the global map. The feature positions are then smoothly updated through a dynamic learning rate mechanism.

[0017] In some possible implementations, the installation height of the reflective column is adapted to the scanning plane of the lidar, and the reflective column adopts a 3M diamond-grade standard reflective column.

[0018] In some possible implementations, the value of the clustering radius threshold is determined based on the physical radius size of the reflective column.

[0019] In some possible implementations, the multiple geometric verification mechanism includes:

[0020] Point count verification: Filter out clusters with fewer than a first preset threshold of valid points;

[0021] Linearity verification: By calculating the covariance matrix and eigenvalues ​​of the point clusters, the geometric distribution characteristics of the point clusters are analyzed, and point clusters with linearity higher than the second preset threshold are screened out.

[0022] Radius consistency verification: For point clusters that pass the linearity verification, calculate the curvature distribution of the point clusters and filter out point clusters that do not conform to the geometric characteristics of the reflective column surface.

[0023] In some possible implementations, the linearity verification specifically includes:

[0024] Calculate the centroid of a point cluster :

[0025] ;

[0026] in, Let N be the i-th point in the point cluster, and N be the number of points in the point cluster.

[0027] Constructing the covariance matrix :

[0028] ;

[0029] in, Let be the covariance in the x-direction. , Let be the covariance in the xy direction. , Let be the covariance in the y-direction. ;

[0030] Calculate the eigenvalues ​​of the covariance matrix:

[0031] ;

[0032] ;

[0033] in, and Covariance matrix Two eigenvalues;

[0034] Calculate the linearity index L:

[0035] ;

[0036] Clusters of points whose linearity index L is higher than the third preset threshold are filtered out.

[0037] In some possible implementations, the radius consistency verification includes:

[0038] Calculate the distance from each point within the point cluster to the fitted circle center (x). c y cThe distance between ) and the expected radius r expected deviation :

[0039] ;

[0040] Among them, (x i y i Let be the i-th point in the point cluster, and r be the i-th point. expected The expected radius;

[0041] Calculate the average deviation :

[0042] ;

[0043] Where N is the number of points in the point cluster;

[0044] Screening average deviation Clusters of points exceeding the fourth preset threshold.

[0045] In some possible implementations, the improved algebraic fitting algorithm includes:

[0046] Calculate the centroid of a point cluster As an initial reference point: Where N is the number of points within the point cluster;

[0047] Construct the equation of the circle:

[0048] ;

[0049] Wherein, the coordinates of the i-th point within the point cluster are (x... i y i ), with center of mass Using the reference point, translate each point within the point cluster: , Similarly, the coordinates of the center of the circle are translated in the same way: ( , );

[0050] For the translated point The equation of a circle is:

[0051] ;

[0052] Construct a least squares problem for all points, with the objective of minimizing the error E:

[0053] ;

[0054] right and Take the partial derivatives and set them to zero to obtain the linear system of equations: Ax=b;

[0055] in, , , ;

[0056] By solving the system of linear equations, the optimal coordinates of the circle's center after translation can be obtained: ;

[0057] Transform the translated center coordinates back to the original coordinate system to obtain the fitted center: ( , );

[0058] Calculate the distance from each point to the center of the fitted circle, and take the average of these distances as the radius estimate: The difference between the estimated radius and the expected radius is calculated, and fitting results with a difference exceeding the fifth preset threshold are filtered out.

[0059] In some possible implementations, the improved algebraic fitting algorithm further includes:

[0060] Calculate the distance from the center of the fitted circle to the origin and the average distance from all points in the point cluster to the origin. Remove fitted results whose distance from the center of the fitted circle to the origin is less than or equal to the average distance from all points in the point cluster to the origin.

[0061] In some possible implementations, feature positions are smoothly updated through a dynamic learning rate mechanism, including:

[0062] The backend uses a fixed learning rate for updates, while the frontend uses a dynamic learning rate. Update, in which, N obs For the total number of observations for this feature, the location update follows an exponentially weighted moving average method: , where P observed P represents the currently observed feature location. old For the feature location estimate before the update (i.e., the previous location), P new The updated feature location estimate is updated only when the feature location change exceeds the sixth preset threshold.

[0063] Secondly, a reflective column identification and positioning system is provided, in which reflective columns are deployed at intervals within a site to ensure that the machine can observe at least two reflective columns at any location within the site. The system includes:

[0064] The point cloud acquisition module is used to acquire point cloud data from the LiDAR on the machine.

[0065] The reflective point extraction module is used to extract potential reflective points from the point cloud data that meet the following conditions: the point type is marked as REFLECTOR, and the Euclidean distance from the point to the lidar is less than the farthest effective detection distance set by the user.

[0066] The aggregation module is used to aggregate points that meet the proximity condition into at least one point cluster based on the spatial coordinates of each point in the extracted potential reflective point set and according to a preset clustering radius threshold.

[0067] The verification module is used to verify the point clusters through a multi-geometric verification mechanism to filter out point clusters that meet the characteristics of reflective pillars;

[0068] The reflective column geometry fitting module is used to calculate the center position and radius of the reflective column for the selected point clusters that meet the reflective column characteristics using an improved algebraic fitting algorithm.

[0069] The reflective column feature management module is used to classify each newly detected feature into candidate features, assign a unique identifier, record its initial position, observation count and first observation frame number, use a hash table structure to store all candidate features, and periodically clean up candidate features that have not been observed again for more than a preset number of frames.

[0070] The feature confirmation and location update module is used to upgrade candidate features that have reached a certain threshold after multiple frames of observation to confirmed features and add them to the global map, and smoothly update the feature positions through a dynamic learning rate mechanism.

[0071] This application has the following beneficial effects: It provides a global absolute position reference for the SLAM process, fundamentally solving the problem of accumulated error caused by high feature repeatability in traditional laser SLAM in scenarios such as long corridors and large spaces. Secondly, the multi-level filtering and intelligent management mechanism designed in this application enables it to adapt to complex environments with scarce features or interference, exhibiting extremely high robustness. As a stable and persistent landmark, the reflective pillar is the key to building and maintaining a globally consistent and long-term effective map. Attached Figure Description

[0072] 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.

[0073] 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.

[0074] Figure 1 This is a flowchart of the reflective column identification and positioning method of Embodiment 1 of this application;

[0075] Figure 2 This is a physical image of the reflective column in Embodiment 1 of this application;

[0076] Figure 3(a) is a map construction effect diagram before optimization of the reflective column extraction algorithm in Embodiment 1 of this application;

[0077] Figure 3(b) is a mapping effect diagram after the optimization of the reflective column extraction algorithm in Embodiment 1 of this application;

[0078] Figure 3(c) is a map construction effect diagram of the optimized reflective column extraction algorithm and feature management mechanism in Embodiment 1 of this application;

[0079] Figure 4 This is a structural block diagram of the reflective column identification and positioning system of Embodiment 2 of this application.

[0080] Figure label:

[0081] 100. Point cloud acquisition module; 200. Reflection point extraction module; 300. Aggregation module; 400. Verification module; 500. Reflection column geometry fitting module; 600. Reflection column feature management module; 700. Feature confirmation and position update module. Detailed Implementation

[0082] 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.

[0083] Example 1

[0084] like Figure 2 The image shown is a physical picture of a reflective column. Before extracting the features of the reflective column, it is necessary to place the reflective columns in the area where the machine (referring to AGV, industrial mobile robot, etc.) operates according to the following principles:

[0085] 1. Density and distribution: In key areas of the site (such as main passage intersections, narrow corridors, and open area boundaries), reflective pillars are deployed at intervals of 5-8 meters (depending on the actual situation) to ensure that the robot can observe at least 2-3 reflective pillars at any position, forming a stable triangulation network.

[0086] 2. Height Specifications: The installation height of the reflector should match the scanning plane of the LiDAR (2D SLAM system is usually 0.2-0.5 meters above the ground) to ensure that the laser beam can stably hit the surface of the reflector.

[0087] 3. Use standardized reflective columns (12cm in diameter, i.e., 6cm in radius), which must be 3M diamond grade, to ensure that the lidar can stably mark the points as REFLECTOR type points.

[0088] 4. Obstacle avoidance considerations: Reflective posts should be deployed in relatively fixed and unobstructed locations to avoid frequent obstruction by dynamic obstacles (such as shelves and AGVs) that could lead to interruptions in observation.

[0089] In this embodiment, the aforementioned reflector deployment principle enables the reflectors to form a sparse yet high-precision global reference network, providing reliable anchor points for subsequent precise positioning.

[0090] like Figure 1 As shown, the reflective column identification and positioning method involved in Embodiment 1 of this application includes:

[0091] S100. Obtain the point cloud data P={p1, p2, ..., pN} from the LiDAR on the machine.

[0092] S200. Extract potential reflective points from the point cloud data that meet the following conditions: the point type is marked as REFLECTOR (wherein, the point type is automatically determined by the LiDAR hardware based on the reflection intensity; a point cloud type of REFLECTOR indicates that the point is a reflective marker or a reflective pillar), and the Euclidean distance from the point to the LiDAR is less than the user-set maximum effective detection distance dmax, where dmax is the effective detection distance threshold, set according to the LiDAR performance and scene characteristics, for example, it can be set to any value between 8 meters and 15 meters. Reflective points beyond this distance suffer from reduced signal-to-noise ratio, resulting in decreased recognition reliability.

[0093] Construct a set of potential reflective points that meet the conditions. , among which, T i For point p i The type marker.

[0094] S300. Based on the spatial coordinates of each point in the extracted potential reflective point set, and according to the preset clustering radius threshold (wherein, the clustering radius threshold is the reflective column radius value, which is set according to the actual situation), the points whose spatial distance meets the proximity condition are aggregated into at least one point cluster.

[0095] S400. The point clusters are verified through a multi-geometric verification mechanism to select point clusters that meet the characteristics of reflective pillars.

[0096] The point clusters formed in step S300 are ensured to conform to the geometric characteristics of reflective columns through the following multiple geometric verification mechanisms:

[0097] Point count verification: Clusters with fewer than 4 valid points are filtered out. In this embodiment, the first preset threshold is 4, but it can be set to other values ​​according to user needs. The purpose of setting the first preset threshold for point cluster filtering is to ensure that there are enough points in the cluster to support the subsequent reflection column feature extraction. Clusters with too few points will be directly filtered out in this step to avoid unstable feature extraction.

[0098] Linearity verification: By calculating the covariance matrix and eigenvalues ​​of the point clusters, the geometric distribution characteristics of the point clusters are analyzed, and point clusters with linearity higher than a second preset threshold are filtered out. Specifically, the system first calculates the centroid of the point clusters. (The formula is as follows):

[0099] ;

[0100] Where, p i Let N be the i-th point in the point cloud, and N be the number of points in the point cluster.

[0101] Then construct the covariance matrix. (The formula is as follows):

[0102] ;

[0103] , , , Let be the variance in the x-direction. Let be the variance in the y-direction. Let be the covariance in the xy direction.

[0104] The formula for calculating the eigenvalues ​​of covariance is as follows:

[0105] ;

[0106] ;

[0107] in, and Covariance matrix Two eigenvalues;

[0108] The linearity index L is defined by the proportion of eigenvalues, and the formula is as follows:

[0109] ;

[0110] Clusters of points whose linearity index L is higher than the third preset threshold are filtered out.

[0111] It should be noted that the linearity index reflects the geometric characteristics of the point cluster—linear structures (such as reflective strips on a wall) have a linearity close to 1, while when... Approximately equal to When L is approximately equal to 0, to avoid being overly stringent, the linearity of circular structures (reflective pillars) is typically less than 0.9. Clusters of points with linearity higher than 0.9 are excluded, as are clusters with linearity exceeding the threshold: L > L threshold =0.9⇒ Exclude, where L threshold The value can be set according to your actual situation. In this embodiment, it is set to any value between 0.9 and 0.95, which can effectively avoid misjudging the wall reflective strip as a reflective column.

[0112] Radius consistency verification: For point clusters that pass the linearity verification, calculate the curvature distribution of the point clusters and filter out point clusters that do not conform to the geometric characteristics of the reflective column surface. This includes the following steps:

[0113] Calculate the distance from each point within the point cluster to the fitted circle center (x). c y c The distance between ) and the expected radius r expected deviation :

[0114] ;

[0115] Among them, (x i y i Let be the i-th point in the point cluster, and r be the i-th point. expected The expected radius usually refers to the radius of the reflector post.

[0116] Calculate the average deviation :

[0117] ;

[0118] Where N is the number of points in the point cluster;

[0119] Screening average deviation Clusters of dots exceeding a fourth preset threshold—that is, if the average deviation exceeds the fourth preset threshold (in this embodiment, the fourth preset threshold is 0.02m)—are considered to not conform to the geometric characteristics of a reflective column and are therefore excluded. δ avg >δ threshold ⇒ Exclude, where δ threshold It is usually set to 0.02m.

[0120] In this embodiment, the aforementioned multi-geometric verification mechanism can effectively distinguish reflective columns from other reflective objects in the environment, such as reflective strips on walls and reflective signs, reducing the false recognition rate by approximately 85%.

[0121] S500. For the selected cluster of points that meet the characteristics of reflective pillars, an improved algebraic fitting algorithm is used to calculate the center position and radius of the reflective pillars.

[0122] Specifically, for the point clusters that passed multiple geometric verifications in step S400, this embodiment uses an improved algebraic fitting algorithm to accurately calculate the center position and radius of the reflective pillars:

[0123] Calculate the centroid of a point cluster As an initial reference point: Where N is the number of points within the point cluster, p i =(x i y i The two-dimensional coordinates of the i-th laser point in the point cluster;

[0124] The system of equations is constructed based on algebraic equations of circles:

[0125]

[0126] After unfolding, we get:

[0127] ;

[0128] To simplify the calculation, the coordinates of the i-th point within the point cluster are (x... i y i ), with center of mass Using the reference point, translate each point within the point cluster: , (i.e., the coordinates of the i-th point after translation), and similarly, the coordinates of the center of the circle are translated in the same way: ( , );

[0129] For the translated point (Coordinates of the center of the circle after translation), the equation of the circle is: Expanding and organizing, we get:

[0130] ;

[0131] Construct a least squares problem for all points, with the objective of minimizing the error E:

[0132] ;

[0133] right and Taking the partial derivatives and setting them to zero, we obtain the linear equation system: Ax=b;

[0134] in, , , ;

[0135] That is, the above system of linear equations can be expressed as:

[0136] ;

[0137] Finding the center of the circle: By solving the system of linear equations, the optimal coordinates of the center of the circle after translation are obtained. ,Right now: Transform the coordinates of the translated center of the circle back to the original coordinate system to obtain the fitted center of the circle: ( , );

[0138] Radius Calculation and Verification: Calculate the distance from each point to the center of the fitted circle, and take the average of these distances as the radius estimate. The difference between the estimated radius and the expected radius is calculated, and fitting results with a difference exceeding a fifth preset threshold are filtered out. That is, the estimated radius r and the expected radius r... expected The difference exceeds the fifth preset threshold r threshold The fitted results will be excluded to ensure geometric consistency: Excluded, where r threshold In this embodiment, it is set to 0.01m.

[0139] In this embodiment, to eliminate erroneous fitting results in some special cases, an additional verification mechanism is introduced, which requires that the distance from the center of the circle to the origin be greater than the average distance from the point cluster to the origin: This additional constraint effectively eliminates erroneous fitting results in some special cases.

[0140] In this embodiment, the improved algebraic fitting algorithm described above can handle unevenly distributed point cloud data, accurately calculating the center position even when the reflector is partially obscured. Tests show that under normal observation conditions, the center positioning accuracy can reach ±5mm, thus meeting the requirements for high-precision positioning.

[0141] S600: Each newly detected feature is divided into candidate features, and a three-stage management strategy is implemented for newly detected reflective pillar features: 1) Assign a unique identifier to each new detected feature and record its initial position, observation count and first observation frame number; 2) Use an efficient hash table structure to store all candidate features to ensure fast retrieval and updating; 3) The system regularly cleans up candidate features that have not been observed again for more than 50 frames to effectively prevent the accumulation of false detection features and optimize the utilization of system resources.

[0142] S700 upgrades candidate features that have reached a certain threshold after multiple frames of observation to confirmed features and adds them to the global map, and smoothly updates the feature positions through a dynamic learning rate mechanism.

[0143] Specifically, when the candidate features in step S600 are observed continuously, a feature confirmation mechanism is implemented: the number of observations is accumulated, and the feature position is updated using a weighted average algorithm (weight formula: New observations are given higher weight, where N obs W represents the total number of times the feature was detected, where j represents the sequence number of the nth observation. j The weight of the j-th observation is represented by the weighting coefficient used in the weighted average. When the number of observations reaches the confirmation threshold (set to 3 in this embodiment), it is upgraded to a confirmed feature and added to the global map. This mechanism significantly reduces the false recognition rate (approximately 75%) while maintaining a high detection rate (>95%).

[0144] Dynamic location update:

[0145] For confirmed features, the system employs a two-layer position maintenance and dynamic learning rate mechanism: the back-end position is updated using a fixed learning rate (set to 0.2 in this embodiment), while the front-end position uses a dynamic learning rate. Update, in which, N obs For the total number of observations for this feature, the location update follows an exponentially weighted moving average method: , where P observed P represents the currently observed feature location. old For the feature location estimate before the update (i.e., the previous location), P new The updated feature position estimate is updated only when the feature position change exceeds a sixth preset threshold, effectively reducing positional jitter while maintaining environmental adaptability. A real-world image of the reflective column is shown below. Figure 2 As shown.

[0146] Figure 3(a) shows the mapping effect before the reflection pillar extraction algorithm is optimized; Figure 3(b) shows the mapping effect after the reflection pillar extraction algorithm is optimized; and Figure 3(c) shows the mapping effect after the reflection pillar extraction algorithm is optimized plus the feature management mechanism (i.e., the content in steps S600-S700 above). By comparing the mapping effects before and after the reflection pillar extraction algorithm optimization, the following conclusions can be drawn:

[0147] 1) There are a large number of misidentified features in Figure 3(a). Traditional methods have failed to effectively distinguish between reflective pillars and reflective strips on walls, resulting in multiple false road signs in the map;

[0148] 2) Figure 3(b) shows that the optimized extraction algorithm effectively eliminates interference such as wall reflective strips through multiple geometric verification mechanisms (linearity verification, radius consistency verification) and improved algebraic fitting algorithm;

[0149] 3) Figure 3(c) shows the effect of the complete scheme. After introducing a feature management mechanism on the basis of the extraction algorithm optimization, the occasional false detection features are further eliminated through the candidate-confirmation two-stage strategy. The feature position is smoothly updated through the dynamic learning rate mechanism, the jitter of the reflective column position is reduced, and the final mapping effect is stable and reliable, providing high-precision global constraints for the SLAM system.

[0150] In this embodiment, the "scientific deployment of reflective pillars - precise feature extraction and management" scheme, combined with the characteristics of industrial scenarios, is adopted: 1) For the characteristics of large open spaces, a reasonable deployment strategy for reflective pillars is designed to provide a 360° visible global reference as "omnidirectional artificial landmarks," fundamentally solving the problem of accumulated errors caused by high feature repeatability in traditional laser SLAM in scenarios such as long corridors and large spaces; 2) The reflective pillar feature extraction algorithm is optimized to achieve high-precision cylinder fitting and geometric verification; Subsequently, reflective pillar constraints and SLAM systems will be integrated, and the positioning accuracy and system robustness will be improved through multi-frame observation and dynamic weighting mechanisms, enabling it to adapt to complex environments with scarce features or interference, exhibiting extremely high robustness. Among them, reflective pillars, as stable and persistent landmarks, are the key to building and maintaining a globally consistent and long-term effective map.

[0151] In LiDAR-based SLAM systems, effectively utilizing reflective pillars can significantly improve positioning accuracy and mapping quality. The core of this approach lies in feature extraction, geometric verification, multi-frame fusion, and SLAM system integration. During feature extraction, points of type REFLECTOR and within a distance of less than 15m (distance can be set according to actual conditions) are selected from the LiDAR point cloud data to form a potential point set. These points are then clustered based on spatial distance using a clustering algorithm. Reflective pillars on walls are excluded through linearity verification, and those meeting geometric conditions are used to calculate the center position using an algebraic fitting algorithm. For data processing and multi-frame fusion, detected reflective pillar features are divided into candidate features and confirmed features. Candidate features are upgraded to confirmed features after reaching a confirmation threshold through multiple frames of observation, and their positions are smoothly updated using a dynamic learning rate mechanism. When integrating with the SLAM system, a data structure containing timestamps, robot pose, reflective pillar IDs, and global and relative pose information is created and incorporated into an optimizer for graph optimization. In practical applications, the system periodically removes candidate features that have not been observed for a long time through a feature cleaning mechanism and reduces feature position jitter through a dead zone mechanism, ensuring system stability and improving feature extraction accuracy.

[0152] Example 2

[0153] like Figure 4 As shown, Embodiment 2 of this application relates to a reflective column identification and positioning system, which deploys reflective columns at intervals within a site to ensure that the machine can observe at least two reflective columns at any position within the site. The system includes:

[0154] Point cloud acquisition module 100 is used to acquire point cloud data from the LiDAR on the machine;

[0155] The reflective point extraction module 200 is used to extract potential reflective points from the point cloud data that meet the following conditions: the point type is marked as REFLECTOR, and the Euclidean distance from the point to the lidar is less than the farthest effective detection distance set by the user.

[0156] The aggregation module 300 is used to aggregate points that meet the proximity condition in space into at least one point cluster based on the spatial coordinates of each point in the extracted potential reflective point set and according to a preset clustering radius threshold.

[0157] The verification module 400 is used to verify the point clusters through a multi-geometric verification mechanism to filter out point clusters that meet the characteristics of reflective pillars.

[0158] The reflective column geometry fitting module 500 is used to calculate the center position and radius of the reflective column for the selected point clusters that meet the reflective column characteristics using an improved algebraic fitting algorithm.

[0159] The reflective column feature management module 600 is used to classify each newly detected feature into candidate features, assign a unique identifier, record its initial position, observation count and first observation frame number, store all candidate features using a hash table structure, and periodically clean up candidate features that have not been observed again for more than a preset number of frames.

[0160] The feature confirmation and location update module 700 is used to upgrade candidate features that have reached a certain threshold after multiple frames of observation to confirmed features and add them to the global map, and smoothly update the feature positions through a dynamic learning rate mechanism.

[0161] It should be noted that other specific implementations of the reflective column identification and positioning system in this embodiment can be found in the specific implementations of the reflective column identification and positioning method described above. To avoid redundancy, they will not be repeated here.

[0162] 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 method for identifying and locating reflective pillars, characterized in that, The method of deploying reflective pillars at intervals within the site to ensure that at least two reflective pillars can be observed from any position within the site includes: Acquire point cloud data from the LiDAR on the machine; From the point cloud data, extract potential reflective points that meet the following conditions: the point type is marked as REFLECTOR, and the Euclidean distance from the point to the lidar is less than the maximum effective detection distance set by the user. Based on the spatial coordinates of each point in the extracted potential reflective point set, points whose spatial distance satisfies the proximity condition are aggregated into at least one point cluster according to a preset clustering radius threshold. The point clusters are verified through a multi-geometric verification mechanism to select those that meet the characteristics of reflective pillars. For the selected clusters of points that meet the characteristics of reflective pillars, an improved algebraic fitting algorithm is used to calculate the center position and radius of the reflective pillars; Each new detected feature is divided into candidate features, and a unique identifier is assigned. Its initial position, observation count and first observation frame number are recorded. All candidate features are stored in a hash table structure. Candidate features that have not been observed again for more than a preset number of frames are cleaned up periodically. Candidate features that have reached a certain threshold after multiple frames of observation are upgraded to confirmed features and added to the global map. The feature positions are then smoothly updated through a dynamic learning rate mechanism.

2. The reflective column identification and positioning method according to claim 1, characterized in that, The installation height of the reflective column is adapted to the scanning plane of the lidar, and the reflective column adopts the standard reflective column of 3M diamond grade.

3. The reflective column identification and positioning method according to claim 1, characterized in that, The value of the clustering radius threshold is determined based on the physical radius of the reflective column.

4. The reflective column identification and positioning method according to claim 1, characterized in that, The multiple geometric verification mechanism includes: Point count verification: Filter out clusters with fewer than a first preset threshold of valid points; Linearity verification: By calculating the covariance matrix and eigenvalues ​​of the point clusters, the geometric distribution characteristics of the point clusters are analyzed, and point clusters with linearity higher than the second preset threshold are screened out. Radius consistency verification: For point clusters that pass the linearity verification, calculate the curvature distribution of the point clusters and filter out point clusters that do not conform to the geometric characteristics of the reflective column surface.

5. The reflective column identification and positioning method according to claim 4, characterized in that, The linearity verification specifically includes: Calculate the centroid of a point cluster : ； in, Let N be the i-th point in the point cluster, and N be the number of points in the point cluster. Constructing the covariance matrix : ； in, Let be the covariance in the x-direction. , Let be the covariance in the x and y directions. , Let be the covariance in the y-direction. ; Calculate the eigenvalues ​​of the covariance matrix: ； ； in, and Covariance matrix Two eigenvalues; Calculate the linearity index L: ； Clusters of points whose linearity index L is higher than the third preset threshold are filtered out.

6. The reflective column identification and positioning method according to claim 4, characterized in that, The radius consistency verification includes: Calculate the distance from each point within the point cluster to the fitted circle center (x). c y c The distance between ) and the expected radius r expected deviation : ； Among them, (x i y i Let be the i-th point in the point cluster, and r be the i-th point. expected The expected radius; Calculate the average deviation : ； Where N is the number of points in the point cluster; Screening average deviation Clusters of points exceeding the fourth preset threshold.

7. The reflective column identification and positioning method according to claim 1, characterized in that, The improved algebraic fitting algorithm includes: Calculate the centroid of a point cluster As an initial reference point: Where N is the number of points within the point cluster; Construct the equation of the circle: ； Wherein, the coordinates of the i-th point within the point cluster are (x... i y i ), with center of mass Using the reference point, translate each point within the point cluster: , Similarly, the coordinates of the center of the circle are translated in the same way: ( , ); For the translated point The equation of a circle is: ； Construct a least squares problem for all points, with the objective of minimizing the error E: ； right and Take the partial derivatives and set them to zero to obtain the linear system of equations: Ax=b; in, , , ; By solving the system of linear equations, the optimal coordinates of the circle's center after translation can be obtained: ; Transform the translated center coordinates back to the original coordinate system to obtain the fitted center: ( , ); Calculate the distance from each point to the center of the fitted circle, and take the average of these distances as the radius estimate: The difference between the estimated radius and the expected radius is calculated, and fitting results with a difference exceeding the fifth preset threshold are filtered out.

8. The method for identifying and locating reflective columns according to claim 7, characterized in that, The improved algebraic fitting algorithm also includes: Calculate the distance from the center of the fitted circle to the origin and the average distance from all points in the point cluster to the origin. Remove fitted results whose distance from the center of the fitted circle to the origin is less than or equal to the average distance from all points in the point cluster to the origin.

9. The method for identifying and locating reflective columns according to claim 1, characterized in that, The feature positions are smoothly updated through a dynamic learning rate mechanism, including: The backend uses a fixed learning rate for updates, while the frontend uses a dynamic learning rate. Update, in which, N obs For the total number of observations for this feature, the location update follows an exponentially weighted moving average method: Among them, P observed P represents the currently observed feature location. old For the feature location estimation before the update, P new The updated feature location estimate is updated only when the feature location change exceeds the sixth preset threshold.

10. A reflective column identification and positioning system, characterized in that, The system comprises: Reflective posts are spaced apart within the site to ensure that at least two reflective posts are visible from any position within the site; The point cloud acquisition module is used to acquire point cloud data from the LiDAR on the machine. The reflective point extraction module is used to extract potential reflective points from the point cloud data that meet the following conditions: the point type is marked as REFLECTOR, and the Euclidean distance from the point to the lidar is less than the farthest effective detection distance set by the user. The aggregation module is used to aggregate points that meet the proximity condition into at least one point cluster based on the spatial coordinates of each point in the extracted potential reflective point set and according to a preset clustering radius threshold. The verification module is used to verify the point clusters through a multi-geometric verification mechanism to filter out point clusters that meet the characteristics of reflective pillars; The reflective column geometry fitting module is used to calculate the center position and radius of the reflective column for the selected point clusters that meet the reflective column characteristics using an improved algebraic fitting algorithm. The reflective column feature management module is used to classify each newly detected feature into candidate features, assign a unique identifier, record its initial position, observation count and first observation frame number, use a hash table structure to store all candidate features, and periodically clean up candidate features that have not been observed again for more than a preset number of frames. The feature confirmation and location update module is used to upgrade candidate features that have reached a certain threshold after multiple frames of observation to confirmed features and add them to the global map, and smoothly update the feature positions through a dynamic learning rate mechanism.

Images