A yard commodity vehicle rapid identification method based on feature evaluation function

Abstract

The present application belongs to the technical field of yard commodity vehicle rapid identification, and particularly relates to a yard commodity vehicle rapid identification method based on a feature evaluation function. A multi-line laser radar is used to scan a yard and obtain original point cloud data; the obtained original point cloud data is preprocessed; the preprocessed original point cloud data is subjected to reflection intensity threshold filtering to obtain vehicle reflection intensity features; the preprocessed original point cloud data is subjected to reflection intensity threshold filtering, and a motion compensation algorithm is used to obtain motion features; the point cloud data is subjected to clustering segmentation to extract the length, width and point cloud density of the vehicle; and a feature evaluation function is designed based on the length, width, reflection intensity features, motion features and point cloud density degree features of the object, which can effectively remove the interference of other false targets and improve the motion safety and system stability of a commodity vehicle transfer robot.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of yard commodity car rapid identification, and particularly relates to a yard commodity car rapid identification method based on a feature evaluation function. BACKGROUND

[0002] Commodity car handling robots play a key role in the field of automobile manufacturing and logistics. They are responsible for transferring commodity cars from one place to another, such as transferring commodity cars from the production line to the warehouse, or transporting vehicles. During this process, accurate identification and positioning of commodity cars are crucial for the operation and safety of the robot. However, traditional yard commodity car identification methods usually rely on visual recognition, extracting vehicle feature information such as color, shape, texture, etc. in advance. In uneven lighting or other conditions, the identification accuracy may decrease, leading to misidentification or missed identification. Therefore, a new method is needed to improve the accuracy and efficiency of yard commodity car identification. SUMMARY

[0003] In order to overcome the problems in the prior art, the present application proposes a yard commodity car rapid identification method based on a feature evaluation function.

[0004] The technical solution of the present application to solve the above technical problems is as follows:

[0005] The present application provides a yard commodity car rapid identification method based on a feature evaluation function, comprising the following steps:

[0006] Step 1: use a multi-line laser radar to scan the yard and obtain raw point cloud data;

[0007] Step 2: pre-process the obtained raw point cloud data;

[0008] Step 3: perform reflection intensity threshold filtering on the pre-processed three-dimensional point cloud data to obtain the reflection intensity feature of the vehicle;

[0009] Step 4: use a motion compensation algorithm on the pre-processed three-dimensional point cloud data to obtain the motion feature;

[0010] Step 5: perform clustering segmentation on the pre-processed point cloud data to extract the length, width and point cloud density of the object;

[0011] Step 6: based on the extracted length, width, reflection intensity feature, motion feature, and point cloud density feature, design a feature evaluation function to identify the vehicles in the yard.

[0012] Further, in step 2, the pre-processing of the obtained raw point cloud data includes removing outliers based on Gaussian filtering to obtain pre-processed three-dimensional point cloud data.

[0013] Further, in step 3, the preprocessed 3D point cloud data is subjected to reflection intensity threshold filtering to obtain the vehicle's reflection intensity characteristics, specifically including:

[0014] Iterate through each point in the preprocessed 3D point cloud data and check its reflection intensity value;

[0015] If the reflection intensity value at a certain point is lower than the preset reflection intensity threshold If so, this point will be considered noise and filtered out.

[0016] If the reflection intensity value at a certain point is greater than or equal to the preset reflection intensity threshold This point is then retained, and the average intensity of points with reflection intensity greater than or equal to the threshold is recorded as... I。

[0017] Further, in step 4, the preprocessed 3D point cloud data undergoes reflection intensity threshold filtering. The ICP-based motion compensation algorithm solves for the rigid body transformation relationship between two point clouds by minimizing the distance between them, including:

[0018] Define the objective function: Assume we have two point cloud datasets, source point cloud P and target point cloud Q, containing n and m points respectively. , ;

[0019] ;

[0020] In the above formula, T represents the rigid body transformation matrix. E Describe the objective function. This represents a point in the source point cloud P. The corresponding point in the target point cloud Q;

[0021] Corresponding point search: The ICP algorithm needs to find corresponding point pairs in two point clouds;

[0022] For each point in the source point cloud P, find the nearest point in the target point cloud Q. q i As corresponding points, this is achieved by calculating the Euclidean distance between all pairs of points and taking the minimum value;

[0023] The least squares method is used to solve for the rotation matrix and translation vector in order to minimize the objective function;

[0024] Repeat the above steps, using the transformation obtained from the previous iteration to update the position of the source point cloud P in each iteration, and re-search for the corresponding point; the iteration continues until the preset stopping condition is met.

[0025] Further, in the step 5, the pre-processed point cloud data is clustered and segmented to extract the length and width of the object, including:

[0026] Calculate the coordinate range of the points in the cluster;

[0027] For each cluster, traverse all the points in the cluster to find the maximum and minimum values of the three-dimensional coordinates; these maximum and minimum values define an axis-aligned bounding box of the cluster in three-dimensional space;

[0028] Calculate the size of the bounding box, the size of the object corresponds to the size of the bounding box; the length of the bounding box is x the difference between the maximum and minimum values of the coordinates; the width of the bounding box is y the difference between the maximum and minimum values of the coordinates; the height of the bounding box is z the difference between the maximum and minimum values of the coordinates.

[0029] Further, in the step 5, the pre-processed point cloud data is clustered and segmented to extract the point cloud density of the vehicle, including: Calculate the point cloud density

[0030] ;

[0031] wherein, is the number of points in the cluster, V is the volume of the cluster.

[0032] Further, in the step 6, based on the reflected average intensity I, the length of the object D, the width of the object, the position , the point cloud density , an evaluation function is designed, and the value of the evaluation function is :

[0033] ;

[0034] wherein, is the standard intensity value of the yard commodity vehicle, is the standard length of the yard commodity vehicle, is the standard width of the yard commodity vehicle, is k the position of the center of the yard commodity vehicle at time -1, is the position of the center of the yard commodity vehicle at time k; is the average point cloud density of the yard commodity vehicle; represents the weight of the reflected intensity; represents the weight of the length; represents the weight of the width; represents the weight of the center position of the commodity vehicle; represents the weight of the degree of point cloud density.

[0035] Compared with the prior art, the present application has the following technical effects:

[0036] The present application extracts specific parameters from laser radar data, comprehensively considers the length, width, reflection intensity, position, point cloud density and other characteristics of the object, and designs a function for evaluating the characteristics of the commodity vehicle in the yard. The method can improve the accuracy and efficiency of the commodity vehicle identification in the yard, and further increase the safety and performance of the transfer robot for the target vehicle transfer operation. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0038] Figure 1 The flowchart of the present application is shown in the figure.

[0039] Figure 2 The identification result of the present application is shown in the figure. DETAILED DESCRIPTION

[0040] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the technical solutions proposed according to the present application are described in detail below in combination with the drawings and preferred embodiments. The specific features, structures or characteristics in one or more embodiments can be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as understood by those skilled in the art of the technology to which the present application belongs.

[0041] Referring to Figure 1 The present application discloses a kind of based on feature evaluation function's yard commodity vehicle fast identification method, comprising the following steps:

[0042] Step 1: scan yard using multi-line laser radar, obtain original point cloud data;

[0043] Step 2: the original point cloud data obtained is preprocessed;

[0044] Step 3: the three-dimensional point cloud data after pre-processing is subjected to reflection intensity threshold filtering, and vehicle reflection intensity characteristics are obtained;

[0045] Step 4: the three-dimensional point cloud data after pre-processing is subjected to motion compensation algorithm, and motion characteristics are obtained;

[0046] Step 5: Cluster segmentation is performed on the preprocessed point cloud data to extract the length, width, and point cloud density of the object.

[0047] Step 6: Based on the extracted length, width, reflection intensity feature, motion feature, and point cloud density feature, a feature evaluation function is designed to identify the vehicles in the yard.

[0048] The following will be expanded in detail:

[0049] Step 1: Use a multi-line laser radar to scan the yard and obtain raw point cloud data.

[0050] The multi-line laser radar is used to scan the yard and collect point cloud data in the three-dimensional space of the yard. The point cloud data includes the three-dimensional coordinates of each point and the reflection intensity information at that point. The reflection intensity information is related to the energy size of the laser beam reflected back at that point.

[0051] As an example, this step can include:

[0052] Step 11: Obtain a series of scans through the multi-line laser radar, each scan including distance and angle information of multiple laser beams, and analyze the distance and angle information of each laser beam.

[0053] The multi-line laser radar obtains SCAN data type, which is usually output by the multi-line laser radar in a series of scans (SCAN) stored in a data packet (pcap) in binary form. Each scan includes distance and angle information of multiple laser beams, which are emitted at different vertical angles and measure the distance to the nearest obstacle. Specifically, subscribe to the LaserScan message in the ROS (Robot Operating System) environment, and the laser radar usually publishes the / scan topic, which contains a series of laser beam scan data.

[0054] Parse the data packet to analyze the distance value and angle information of each laser beam and extract the distance values measured at different angles. By obtaining the point cloud data, the point cloud distribution of the entire yard can be obtained for subsequent processing.

[0055] Step 12: Convert the obtained distance and angle information of the laser beam into raw point cloud data.

[0056] Convert the obtained distance and angle information of the laser beam into three-dimensional point cloud data, where each point includes three-dimensional coordinates.

[0057] Where the laser radar is installed horizontally, assuming that the laser radar is located at the origin (0, 0, 0), the Z-axis represents the vertical direction (positive upward), and the direction towards the vehicle head is the x-axis direction, and the y-axis direction is obtained by rotating the x-axis counterclockwise by 90 degrees:

[0058] ;

[0059] wherein, is the distance data, representing the distance from the origin; is the elevation angle, the angle from the positive direction of the Z-axis (upward); is the azimuth angle, the angle from the positive x-axis counterclockwise to the line connecting the point and the positive x-axis.

[0060] Step 2: Preprocess the obtained raw point cloud data;

[0061] The raw point cloud is preprocessed to remove outliers based on Gaussian filtering. Gaussian filtering is a commonly used smoothing technique to remove noise and outliers in point cloud data, working by calculating the weighted average of each point and its neighboring points, where the weights are determined by a Gaussian function that assigns different weights to neighboring points based on their distance from the center point.

[0062] Specifically, the raw point cloud dataset p, where each point has coordinates , its corresponding weighted average point is . Then, for point , the coordinates of its weighted average point can be calculated by the following formula:

[0063] ;

[0064] In the above formula, is the number of neighboring points of point ; point is one of the neighboring points of point ; is the weight of point , usually calculated by a Gaussian function, related to the distance between point and point , which can be , where is the standard deviation of the Gaussian function, controlling the smoothing degree of the filtering.

[0065] Step 3: Perform reflection intensity threshold filtering on the preprocessed three-dimensional point cloud data to obtain the reflection intensity features of the vehicle.

[0066] The reflection intensity threshold is pre-set, and the selection of the reflection intensity threshold is based on the reflection characteristics of the target object, i.e. the commodity car, the environmental lighting conditions, and the expected noise level settings, aiming to retain high-intensity reflection points from the commodity car and remove low-intensity noise points generated by the background, ground, or other non-target sources.

[0067] Traverse each point in the point cloud dataset, check its reflection intensity value. If the reflection intensity value of a point is lower than the preset reflection intensity threshold , the point is considered as a noise point and is filtered out; if the reflection intensity value of a point is greater than or equal to the preset reflection intensity threshold , the point is retained, and the average reflection intensity of the part of points with reflection intensity greater than or equal to the threshold is recorded as I , these points correspond to the signals reflected by the surface of the commodity car. By setting the reflection intensity threshold , the reflection intensity information of the commodity car can be extracted from the original laser point cloud data, and the noise points with low intensity are filtered out.

[0068] Step 4: Use the motion compensation algorithm on the preprocessed three-dimensional point cloud data to obtain the motion features.

[0069] In the driving of the transfer robot, the motion compensation algorithm is used to correct the errors in the laser radar scanning data caused by the acceleration and deceleration of the robot turning, etc., to ensure accurate identification and positioning of the commodity car.

[0070] The mathematical expression of the ICP (Iterative Closest Point) algorithm mainly focuses on minimizing the distance measure between two point clouds, usually the sum of squares of Euclidean distances. This optimization problem can be formalized as finding an optimal rigid transformation matrix T, which can include a rotation matrix and a translation vector, such that the distance between the transformed point cloud and the target point cloud is minimized.

[0071] The motion compensation algorithm based on ICP solves the rigid transformation relationship between two point clouds by minimizing the distance between them, including:

[0072] Step 41: Set the objective function; the top of the window

[0073] Suppose there are two point cloud datasets, source point cloud P and target point cloud Q, including n and m points respectively, where, , ;

[0074] The goal of the ICP algorithm is to find a rigid transformation matrix T that makes the points in the source point cloud P closest to the corresponding points in the target point cloud Q , which can be achieved by minimizing the following objective function E :

[0075] ;

[0076] Step 42: Corresponding point search: the ICP algorithm needs to find the corresponding point pairs in the two point clouds;

[0077] The nearest point can be searched by Euclidean distance. For each point in one point cloud, the nearest point in the other point cloud can be found as the corresponding point.

[0078] For each point in the source point cloud P, find the nearest point in the target point cloud Q as the corresponding point q i As the corresponding point, this is achieved by calculating the Euclidean distance between all pairs of points and taking the minimum value.

[0079] Step 43: Use the least squares method to solve the rotation matrix and translation vector to minimize the objective function;

[0080] Based on the found corresponding point pairs, estimate the rigid body transformation relationship between the two point clouds, i.e. translation and rotation. The ICP algorithm usually uses the least squares method to solve this problem, which determines the optimal rigid body transformation by minimizing the distance between the corresponding points.

[0081] Step 44: Repeat the above steps, each iteration uses the transformation obtained in the last iteration to update the position of the source point cloud P, and re-search for the corresponding points; iteration continues until the preset stopping condition is met;

[0082] The preset stopping condition is, for example, the change of the objective function is less than a certain threshold, or the maximum number of iterations is reached.

[0083] Through the above steps, motion compensation of pre-processed three-dimensional point cloud data can be achieved. Motion compensation based on the ICP algorithm can correct the scanning data errors caused by robot motion, thereby improving the accuracy of subsequent commodity vehicle recognition and positioning.

[0084] Step 5: Cluster segmentation of pre-processed point cloud data to extract length, width and point cloud density.

[0085] Use the k-means clustering algorithm to segment the pre-processed point cloud data. The basic steps of the k-means clustering algorithm are as follows:

[0086] Initialize cluster centers: randomly select k points as initial cluster centers;

[0087] Assign points to the nearest cluster center: for each point in the point cloud, calculate the distance between it and each cluster center, and assign it to the nearest cluster center;

[0088] Update the nearest cluster center: recalculate the mean of all points in each cluster, and take the mean as the new cluster center;

[0089] Repeat the above two steps until the change of the cluster center is less than a certain threshold or the preset number of iterations is reached.

[0090] By k-menas clustering algorithm, the initial yard object clustering result is obtained through the steps of initializing the clustering center, assigning the point to the nearest clustering center, updating the nearest clustering center, repeating iteration, etc. After that, for each clustering cluster, the length and width dimensions of the vehicle are extracted. Step 51: For each clustering cluster, the length and width dimensions of the object are extracted;

[0091] Estimating the length and width dimensions of the object based on the distribution of the cluster center points: calculating the maximum and minimum values of the cluster center points, and then calculating the bounding box or bounding box of the point cloud. The size of these bounding boxes or bounding boxes is the length and width dimensions of the object.

[0092] As an example, the step 51 can include the following sub-steps:

[0093] Step 511: Calculate the coordinate range of the center point of each clustering cluster;

[0094] For each clustering cluster, traverse all the points in the cluster to find the maximum and minimum values of the three-dimensional coordinates; these maximum and minimum values define an axis-aligned bounding box (AABB) of the cluster in three-dimensional space;

[0095] Step 512: Calculate the size of the bounding box:

[0096] The length of the bounding box is x The difference between the maximum and minimum values of the coordinates;

[0097] The width of the bounding box is y The difference between the maximum and minimum values of the coordinates;

[0098] The height of the bounding box is z The difference between the maximum and minimum values of the coordinates.

[0099] Step 52: For each clustering cluster, calculate the point density, i.e. the density degree extraction;

[0100] For the extraction of the density degree of the object, the point density of the object can be obtained by calculating the number of cluster center points divided by its volume. The point density represents the number of point clouds in a unit volume, which can be used to estimate the density degree of the object. The specific formula is as follows:

[0101] Step 521: Calculate the volume of the cluster:

[0102] Use the length, width, and height of the bounding box calculated in step 512 to calculate the volume of the cluster:

[0103] The volume V=length×width×height;

[0104] Step 522: Calculate the point density:

[0105] Point density It can be calculated as follows:

[0106] ;

[0107] Wherein, is the number of points in the cluster, V is the volume of the cluster.

[0108] This density value can reflect the tightness of the point cloud in the cluster, which helps to further distinguish different types of objects, such as vehicles, sundries, etc.

[0109] Step 6: Based on the extracted length and width size, reflection intensity feature, motion feature, point cloud density feature, design feature evaluation function, identify the vehicle in the yard.

[0110] Define a feature evaluation function, which comprehensively considers the length, width, reflection intensity, motion speed, and point cloud density of the object.

[0111] Based on the reflection intensity, length and width size, motion speed attribute and point cloud density degree obtained for the target object as a feature operator, the evaluation function is designed as follows, which comprehensively considers the reflection average intensity I, the length of the object, the width D of the object, the position , the point cloud density Five factors, the value of the smallest target is the final target with the highest confidence:

[0112] ;

[0113] Wherein, is the standard intensity value of the yard commodity car, is the standard length of the yard commodity car, is the standard width of the yard commodity car, is k the position of the center of the yard commodity car at time -1, is the position of the center of the yard commodity car at time k; is the average point cloud density of the yard commodity car. Indicates the weight of the reflection intensity; Indicates the weight of the length; Indicates the weight of the width; Indicates the weight of the center position of the commodity car; Indicates the weight of the point cloud density degree, which is used to adjust the contribution degree of each feature to the confidence. In the code debugging process, according to the actual operation situation, determine each weight to ensure the best effect of identifying the commodity car.

[0114] Referring to Figure 2 The object with the highest score is considered as the commodity vehicle by calculating the feature evaluation function score of all candidate objects in the yard, and the precise positioning of the commodity vehicle can be realized by using the position information of the object. The feature evaluation function is related to the length, width, reflection intensity, commodity vehicle center position and point cloud density of the object, and the weight setting is initially adjusted to 0.2:0.2:0.2:0.2:0.2 according to how to better identify the commodity vehicle in the actual engineering.

[0115] According to the feature evaluation function as the main information for the commodity vehicle transfer robot to identify the commodity vehicle, the commodity vehicle can be accurately identified and positioned, and the positioning process has strong robustness and stability.

[0116] The present application relates to multi-line laser radar acquisition point cloud data processing, clustering segmentation, motion compensation and feature evaluation function design. By comprehensively evaluating the length, width, reflection intensity, motion speed and point cloud density of the object, the commodity vehicle can be accurately identified. This method can improve the safety and performance of the vehicle.

[0117] The above examples are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A yard commodity vehicle rapid identification method based on a feature evaluation function, characterized in that, The method comprises the following steps: Step 1: scanning the yard using a multi-line laser radar to obtain raw point cloud data; Step 2: preprocessing the obtained raw point cloud data; Step 3: performing reflection intensity threshold filtering on the preprocessed three-dimensional point cloud data to obtain reflection intensity features of the vehicle, specifically including: Pre-set reflection intensity threshold The selection of the reflection intensity threshold is based on the reflection characteristics of the target object, i.e. the car, the ambient lighting conditions and the expected noise level setting, with the aim of preserving high intensity reflection points originating from the car and removing low intensity noise points due to the background, ground or other non-target sources; traversing each point in the preprocessed three-dimensional point cloud data to check its reflection intensity value; If the reflection intensity value of a certain point is lower than the preset reflection intensity threshold value , the point is regarded as a noise point and filtered out; if the reflection intensity value of a certain point is greater than or equal to the preset reflection intensity threshold value , the point is retained, and the average reflection intensity of the points with reflection intensity greater than or equal to the threshold value is recorded as I ; Step 4: obtaining motion features by using a motion compensation algorithm on the preprocessed three-dimensional point cloud data; performing reflection intensity threshold filtering on the preprocessed three-dimensional point cloud data, and using an ICP-based motion compensation algorithm to solve the rigid body transformation relationship between two point clouds by minimizing the distance between them, including: Setting the objective function: the top of the window assumes that there are two point cloud data sets, source point cloud P and target point cloud Q, which contain n and m points respectively, wherein, , ; ； In the above formula, T represents a rigid transformation matrix, E represents a target function, represents a point in the source point cloud P, a corresponding point in the target point cloud Q; corresponding point search: the ICP algorithm needs to find a corresponding point pair in two point clouds; For each point in the source point cloud P, find the closest point in the target point cloud Q q i As a correspondence, this is achieved by computing the Euclidean distance between all pairs of points and taking the minimum value; using the least squares method to solve the rotation matrix and translation vector to minimize the objective function; repeating the above steps, each iteration uses the transformation obtained in the last iteration to update the position of the source point cloud P and re-search for corresponding points; the iteration continues until the preset stopping condition is met; Step 5: clustering and segmenting the preprocessed point cloud data to extract the length, width and point cloud density of the object; Step 6: based on the extracted length, width, reflection intensity features, motion features and point cloud density features, designing a feature evaluation function to identify the vehicles in the yard; Based on the average reflection intensity I, the length of the object , the width D of the object, the position , the point cloud density , the design evaluation function is designed, and the value of the substitution is: ； wherein, is the standard strength value of the yard commodity car, is the standard length of the yard commodity car, is the standard width of the yard commodity car, is k the position of the center of the yard commodity car at time -1, is the position of the center of the yard commodity car at time k; is the average point cloud density of the yard commodity car; denotes the weight of the reflection intensity; denotes the weight of the length; denotes the weight of the width; denotes the weight of the center position of the commodity car; denotes the weight of the degree of point cloud density.

2. The yard commodity car rapid identification method based on the feature evaluation function according to claim 1, characterized in that, In step 2, the obtained raw point cloud data is preprocessed, including removing outliers based on Gaussian filtering to obtain preprocessed three-dimensional point cloud data.

3. The yard commodity car rapid identification method based on the feature evaluation function according to claim 1, characterized in that, In step 5, the preprocessed point cloud data is clustered and segmented to extract the length and width of the object, including: calculating the coordinate range of the cluster center; for each cluster, traverse all points in the cluster to find the maximum and minimum values of the three-dimensional coordinates; these maximum and minimum values define an axis-aligned bounding box of the cluster in three-dimensional space; The size of the bounding box is calculated, and the size of the object corresponds to the size of the bounding box; the length of the bounding box is x The difference between the maximum and minimum values of the coordinates; the width of the bounding box is y The difference between the maximum and minimum values of the coordinates; the height of the bounding box is z The difference between the maximum and minimum values of the coordinates.

4. The yard commodity car rapid identification method based on the feature evaluation function according to claim 3, characterized in that, In step 5, the pretreated point cloud data is clustered and segmented to extract the point cloud density of the vehicle, including: point cloud density is calculated using the following formula: ； wherein, is the number of points in the cluster, V is the volume of the cluster.

Images