CN-121999044-A - Reflective column identification positioning method and system

CN121999044ACN 121999044 ACN121999044 ACN 121999044ACN-121999044-A

Abstract

The application discloses a method and a system for identifying and positioning a reflecting column, which relate to the technical field of industrial-grade mobile robot navigation and solve the problems of accumulated errors and incapability of distinguishing essential differences of a cylinder and a plane reflecting strip in the existing SLAM technology, wherein the method comprises the following steps: extracting potential reflection points from point cloud data according to a preset rule, dividing the points adjacent in space into point clusters by taking the radius of the reflection column as a clustering radius threshold, screening out the point clusters conforming to the characteristics of the reflection column through a multiple geometric verification mechanism, calculating the circle center position and the radius of the reflection column by adopting an improved algebraic fitting algorithm, carrying out a three-stage management strategy on the newly detected characteristics of the reflection column to prevent the accumulation of false detection characteristics, optimizing the utilization of system resources, reducing the false recognition rate through a characteristic confirmation mechanism, and smoothly updating the characteristic positions of the confirmed characteristics through a dynamic learning rate mechanism, thereby being capable of accurately identifying the reflection column in a complex environment with scarce characteristics or interference.

Inventors

ZHANG JIAHUI
LI WEIJUN
HU ZHIGUANG
Huang Lielie

Assignees

浙江迈睿机器人有限公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. A method of identifying and locating light reflecting columns, wherein the light reflecting columns are arranged at intervals in a field to ensure that at least two light reflecting columns can be observed at any position in the field by a machine, the method comprising: acquiring point cloud data of a laser radar on a machine; Extracting potential reflecting points from the point cloud data, wherein the point type is REFLECTOR, and the Euclidean distance from the point to the laser radar is smaller than the furthest effective detection distance set by a user; Based on the extracted space coordinates of each point in the potential reflecting point set, according to a preset clustering radius threshold, the points with the space distances meeting the adjacent conditions are aggregated into at least one point cluster; Verifying the point clusters through a multiple geometric verification mechanism to screen out the point clusters conforming to the characteristics of the reflective columns; Aiming at the screened point clusters which accord with the characteristics of the reflecting column, calculating the circle center position and the radius of the reflecting column by adopting an improved algebraic fitting algorithm; dividing each new detection feature into candidate features, allocating a unique identifier, recording the initial position, the observation count and the first observation frame number of each new detection feature, storing all candidate features by adopting a hash table structure, and periodically cleaning the candidate features which are not observed again beyond a preset frame number; And upgrading the candidate features which reach the determined threshold value through multi-frame observation to be the confirmed features, adding the confirmed features into a global map, and smoothly updating the feature positions through a dynamic learning rate mechanism.
2. The method for identifying and positioning the reflecting column according to claim 1, wherein the mounting height of the reflecting column is matched with a laser radar scanning plane, and the reflecting column adopts a standard reflecting column with a 3M diamond grade.
3. The method for identifying and positioning a reflecting column according to claim 1, wherein the value of the cluster radius threshold is determined based on the physical radius size of the reflecting column.
4. The method of claim 1, wherein the multiple geometry verification mechanism comprises: Point verification, namely screening out point clusters with effective points less than a first preset threshold value in the clusters; The linearity verification comprises the steps of analyzing geometric distribution characteristics of the point clusters by calculating covariance matrixes and characteristic values of the point clusters, and screening out the point clusters with linearity higher than a second preset threshold; And (3) radius consistency verification, namely calculating curvature distribution of the point clusters passing through linearity verification, and screening out the point clusters which do not accord with the geometric characteristics of the surface of the reflecting column.
5. The method for identifying and positioning a light reflecting column according to claim 4, wherein the linearity verification specifically comprises: Calculating centroid of point cluster : ; Wherein, the The point N is the number of points in the point cluster; constructing covariance matrix : ; Wherein, the As the covariance in the x-direction, , As the covariance in the xy-direction, , As the covariance in the y-direction, ; Calculating eigenvalues of covariance matrix: ; ; Wherein, the And As covariance matrix Is a function of the two characteristic values of (a); Calculating a linearity index L: ; And screening out the point clusters with the linearity index L higher than a third preset threshold value.
6. The method of claim 4, wherein the radius uniformity verification comprises: Calculating the deviation of the distance from each point in the point cluster to the fitting circle center (x c ,y c ) and the expected radius r expected : ; Wherein (x i ,y i ) is the i-th point in the point cluster, and r expected is the expected radius; calculating average deviation : ; Wherein N is the number of points in the point cluster; Screening out average deviations Clusters of points exceeding a fourth preset threshold.
7. The method of claim 1, wherein the improved algebraic fitting algorithm comprises: Calculating centroid of point cluster As an initial reference point: Wherein N is the number of points in the point cluster; The equation for constructing a circle: ; Wherein the coordinate of the ith point in the point cluster is (x i ,y i ) and is represented by the mass center Translation is carried out on each point in the point cluster as a reference point , ) Likewise, the center coordinates are translated in the same way , ); For the point after translation The equation for a circle is: ; the least squares problem is built for all points, with the goal of minimizing the error E: ; For a pair of And Solving the bias guide and enabling the bias guide to be zero to obtain a linear equation set, wherein ax=b; Wherein, the , , ; Obtaining the optimal center coordinates after translation by solving a linear equation set: ; converting the translated circle center coordinates back to the original coordinate system to obtain a fitting circle center:. Times. , ); Calculating the distance from each point to the fitting circle center, and taking the average value as radius estimation: and calculating a difference value between the radius estimation and the expected radius, and screening out a fitting result of which the difference value exceeds a fifth preset threshold value.
8. The method of claim 7, wherein the improved algebraic fitting algorithm further comprises: And calculating the distance from the fitting circle center to the original point and the average distance from all the points in the point cluster to the original point, and screening out the fitting result that the distance from the fitting circle center to the original point is smaller than or equal to the average distance from all the points in the point cluster to the original point.
9. The method of claim 1, wherein smoothly updating the feature locations by a dynamic learning rate mechanism comprises: The back end position is updated by adopting a fixed learning rate, and the front end position is updated by adopting a dynamic learning rate Updating, wherein, N obs is the total number of observations of the feature, and the location update follows an exponentially weighted moving average: Where P observed is the currently observed feature position, P old is the pre-update feature position estimate, and P new is the post-update feature position estimate, updated only when the feature position change exceeds a sixth preset threshold.
10. A retroreflective column identification positioning system characterized by a spaced apart deployment of retroreflective columns within a venue to ensure that at least two retroreflective columns are observed by a machine at any location within the venue, the system comprising: The point cloud acquisition module is used for acquiring point cloud data of the laser radar on the machine; The reflecting point extraction module is used for extracting potential reflecting points from the point cloud data, wherein the point type mark is REFLECTOR, and the Euclidean distance from the point to the laser radar is smaller than the furthest effective detection distance set by a user; The aggregation module is used for aggregating the points with the space distance meeting the adjacent condition into at least one point cluster according to a preset clustering radius threshold value based on the space coordinates of the points in the extracted potential reflection point set; The verification module is used for verifying the point clusters through a multiple geometric verification mechanism to screen out the point clusters conforming to the characteristics of the reflective columns; the geometric fitting module of the reflecting column is used for calculating the circle center position and the radius of the reflecting column by adopting an improved algebraic fitting algorithm aiming at the screened point clusters conforming to the characteristics of the reflecting column; The reflective column feature management module is used for dividing each new detection feature into candidate features, allocating a unique identifier, recording the initial position, the observation count and the first observation frame number of the new detection feature, storing all the candidate features by adopting a hash table structure, and periodically cleaning the candidate features which are not observed again beyond a preset frame number; And the feature confirmation and position update module is used for upgrading the candidate features which reach the determined threshold value through multi-frame observation to confirm the features and adding the features into the global map, and smoothly updating the feature positions through a dynamic learning rate mechanism.

Description

Reflective column identification positioning method and system Technical Field The application relates to the technical field of industrial mobile robot navigation, in particular to a method and a system for identifying and positioning a reflecting column. Background In the field of industrial-grade mobile robot navigation, a reflecting column is becoming a key artificial road sign for replacing a traditional directional reflecting plate due to the characteristic of 360-degree visible omnidirectional reflection. The prior art scheme mainly relies on the identification of a laser radar on a high-reflectivity cylindrical object to realize rough positioning, and is used as a supplementary solution of an SLAM system in structural environments such as warehouse logistics and the like. In the traditional method, a sparse positioning network is constructed by deploying reflective columns with specific specifications, and the pose of the robot is estimated by adopting the reflective point clustering and geometric center calculation principle. However, the primary application mode has the intrinsic defects that the reflecting column is only used as a discrete positioning reference point, a multi-frame observation-confirmation-optimized complete sensing closed loop cannot be formed, the existing point cloud clustering algorithm cannot effectively distinguish the reflecting column from a wall surface reflecting strip, the false recognition rate in a complex scene is high, the positioning result is obviously dithered due to a single-frame data-based processing mechanism, and the accumulated error of the SLAM system cannot be effectively restrained. In the field of synchronous positioning and mapping (SLAM) of structured environments such as storage, the prior art scheme mainly relies on laser radar scanning to obtain reflection point clouds of a reflection column, and candidate point clusters are formed through threshold screening and distance clustering. The typical implementation flow is to calculate the geometric center coordinates of the point clusters as the reflective column position estimation, and directly incorporate the first detection result into the global map. These retroreflective cylinder coordinates are typically calculated in a single frame and once confirmed are no longer updated, forming a static landmark database. However, the traditional scheme faces serious problems in actual industrial scenes that the point cloud clustering does not consider geometric features to cause misjudgment of the reflective strips as reflective columns, the geometric center calculation method generates positioning deviation of up to 10cm when the geometric center calculation method is partially shielded, and measured data show that the road sign recognition stability in the dynamic environment is less than 60%. Major drawbacks: 1. The reflection column is only used as an auxiliary tool for initial positioning, cannot be deeply integrated into the SLAM optimization framework, and cannot establish a continuous constraint mechanism to correct accumulated errors in movement. 2. The existing scheme can not distinguish the essential difference between a cylinder and a plane reflection strip only according to distance clustering, the circle center positioning algorithm has fundamental limitation, the traditional geometric center method does not consider a circular fitting model, and systematic deviation is generated under the condition of uneven distribution of point cloud or partial shielding. Disclosure of Invention The application aims to solve the problems of accumulated errors and incapability of distinguishing essential differences of a cylinder and a plane reflection strip in the existing SLAM technology and provides a method and a system for identifying and positioning a reflection column. In a first aspect, a method for identifying and positioning reflective columns is provided, wherein reflective columns are deployed at intervals in a field to ensure that at least two reflective columns can be observed at any position in the field by a machine, and the method comprises the following steps: acquiring point cloud data of a laser radar on a machine; Extracting potential reflecting points from the point cloud data, wherein the point type is REFLECTOR, and the Euclidean distance from the point to the laser radar is smaller than the furthest effective detection distance set by a user; Based on the extracted space coordinates of each point in the potential reflecting point set, according to a preset clustering radius threshold, the points with the space distances meeting the adjacent conditions are aggregated into at least one point cluster; Verifying the point clusters through a multiple geometric verification mechanism to screen out the point clusters conforming to the characteristics of the reflective columns; Aiming at the screened point clusters which accord with the characteristics of the reflecting column, calculating