A method and device for quickly identifying road elements using mobile laser radar

Abstract

The present invention relates to a method and device for quickly identifying road elements of a mobile laser radar, belonging to the technical field of road detection, the method comprising acquiring laser point cloud data, determining non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and the neighboring points; calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set; processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point; when the mean drop precision MSE value and the feature correlation coefficient meet preset conditions, classifying the feature point sample set to obtain the recognition result of the road element. The method realizes the rapid and accurate recognition of the complex scenes covered by the point cloud data, significantly improves the accuracy of data processing, reduces the need for manual operation, and effectively improves work efficiency.

Description

Technical Field

[0001] The present invention relates to the field of road detection technology, and in particular to a method and device for quickly identifying road elements using a mobile laser radar. Background Art

[0002] The core of the urban transportation network is the intricate road system, which is composed of a variety of road elements, including but not limited to the road surface, street trees, street lighting, safety guardrails, signal lights and various traffic signs. These road elements are interrelated and work together in a changing environment. Given the frequent flow of vehicles and pedestrians, the failure of any road element may cause the dysfunction of the entire road system. Therefore, monitoring these road elements is crucial to keep the road running smoothly.

[0003] In the related technology, through the mobile measurement system installed on the vehicle, laser scanning technology can be used to quickly collect high-precision, high-density and high-resolution three-dimensional spatial data, which is extremely beneficial for the rapid collection and instant update of road information.

[0004] However, the point cloud data acquired by mobile device laser scanning systems often cover complex scenes, diverse target objects, huge data volumes and uneven point density distribution, which makes the analysis and change detection of point cloud data quite challenging. At present, there is a lack of methods for automatically selecting samples and quickly identifying road elements. Summary of the invention

[0005] In view of this, it is necessary to provide a method and device for quickly identifying road elements using a mobile laser radar to solve the technical problem in the prior art of lacking a method for automatically selecting samples and quickly identifying road elements.

[0006] In order to solve the above problems, the present invention provides a method for quickly identifying road elements using a mobile laser radar, comprising:

[0007] Acquire laser point cloud data based on a mobile device, and determine non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point;

[0008] Calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set;

[0009] Processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point;

[0010] When the mean descending precision MSE value and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element.

[0011] In a possible implementation, determining the non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point includes:

[0012] Calculating the elevation difference according to the elevation of the current point and the elevation of the corresponding neighboring point;

[0013] If the elevation difference is less than a preset elevation difference threshold, determining that the current point does not belong to the non-ground point cloud data;

[0014] If the elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that the current point belongs to the non-ground point cloud data.

[0015] In a possible implementation, calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set includes:

[0016] Calculate the corresponding normal angle according to the normal vectors of each point in the non-ground point cloud data and the two points closest to it in the neighborhood;

[0017] And taking the preset M points in the non-ground point cloud data as a group, calculating the average value of the corresponding normal angles;

[0018] If the average value of the normal angle is greater than or equal to 0 and less than or equal to a preset first angle threshold, the corresponding point is determined as a feature point to obtain the feature point sample set;

[0019] If the average value of the normal angle is greater than the first angle threshold and less than a preset second angle threshold, the corresponding point is determined to be a transition point;

[0020] If the average value of the normal angles is greater than or equal to the second angle threshold and less than or equal to π, the corresponding point is determined to be a non-feature point.

[0021] In a possible implementation, the processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point includes:

[0022] Resampling the feature point sample set to obtain a sample subset, using the unsampled data as a validation sample set, and constructing a regression tree based on the sample subset;

[0023] The random forest model is trained by the regression tree in a loop, and the classification effect of the random forest model is verified by the verification sample set to obtain the mean descending precision (MSE) value of each feature point output by the random forest model;

[0024] By performing correlation analysis on the features of the classified sample set of the feature point sample set, a feature correlation coefficient of the classified sample set is obtained.

[0025] In a possible implementation, when the mean decreasing precision MSE value and the feature correlation coefficient meet a preset condition, classifying the feature point sample set to obtain a road element recognition result includes:

[0026] Determine the importance weight of the corresponding feature according to the MSE value;

[0027] Performing credibility analysis on the classified sample set to determine the credibility of the recognition result;

[0028] When the importance weight, the credibility and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element.

[0029] In a possible implementation, when the importance weight, the credibility and the feature correlation coefficient meet a preset condition, classifying the feature point sample set to obtain a road element recognition result includes:

[0030] Sort the features according to the importance weights and delete the features with the smallest importance weights until the number of features retained is a preset N, so as to generate an optimal feature set;

[0031] If the feature correlation coefficient is less than a preset correlation coefficient threshold and / or the credibility is less than a preset credibility threshold, a feature point sample set is obtained again based on the non-ground point cloud data, and then an optimal feature set is regenerated;

[0032] The optimal feature set is classified to obtain the recognition result of the road element.

[0033] In a possible implementation, after the laser point cloud data is acquired based on the mobile device, the following steps are included:

[0034] Establishing a KD tree spatial index for the laser point cloud data;

[0035] According to the KD tree spatial index, a K nearest neighbor search is performed on each point in the laser point cloud data to determine the neighborhood point of each point;

[0036] Calculate the average and standard deviation of the Euclidean distance between each point in the laser point cloud data and the corresponding neighboring points;

[0037] Points whose Euclidean distance differs from the average value by more than a times the standard deviation are determined as noise points, and the noise points in the laser point cloud data are filtered.

[0038] In a possible implementation, before determining the non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and the neighboring points, the method includes:

[0039] Dividing the laser point cloud data into point cloud blocks according to a preset first scale, and taking the point cloud blocks whose internal point quantity is greater than a first quantity threshold as thinned point cloud blocks;

[0040] Dividing the thinned point cloud block into voxels according to a preset second scale, and taking voxels whose number of internal points is greater than a second number threshold as valid voxels;

[0041] Taking the point closest to the center of the valid voxel as the voxel center point, and taking the normal vector of the laser point cloud data at the voxel center point as the voxel direction vector of the valid voxel;

[0042] Calculating the vector angle between the direction vectors of each voxel in the thinned point cloud block, and calculating the proportion of the vector angle that is greater than a preset angle threshold;

[0043] The points in the corresponding thinned point cloud block are downsampled and thinned according to the ratio.

[0044] In a possible implementation, classifying the optimal feature set to obtain the recognition result of the road element includes:

[0045] Training the random forest model based on the optimal feature set to obtain a trained random forest model;

[0046] The validation sample set is classified by the trained random forest model to obtain the recognition result of the road element.

[0047] On the other hand, the present invention also provides a road element rapid identification device for a mobile laser radar, comprising:

[0048] A data determination module, used to obtain laser point cloud data based on a mobile device, and determine non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point;

[0049] An angle calculation module, used to calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set;

[0050] A sample processing module is used to process and perform correlation analysis on the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point;

[0051] The classification judgment module is used to classify the feature point sample set to obtain the recognition result of the road element when the mean drop precision MSE value and the feature correlation coefficient meet the preset conditions.

[0052] The beneficial effect of the present invention is that the present invention can first screen the laser point cloud data according to the elevation difference between each point and the neighboring points to obtain non-ground point cloud data, thereby reducing the amount of point cloud data. It is also possible to calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and then perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set, thereby performing feature clustering on unevenly distributed point cloud data. Furthermore, the feature point sample set can also be processed and correlated separately to obtain the average drop precision MSE value and feature correlation coefficient of each feature point; when the average drop precision MSE value and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element, thereby obtaining the recognition result of the road element according to the average drop precision MSE value and the feature correlation coefficient of each feature point scene, realizing fast and accurate recognition of the complex scenes covered by the point cloud data, significantly improving the accuracy of data processing, and significantly reducing the need for manual operation, thereby effectively improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic flow chart of an embodiment of a method for rapid identification of road elements by a mobile laser radar provided by the present invention;

[0054] Figure 2 A schematic diagram of a flow chart of an embodiment of filtering noise points in laser point cloud data provided by the present invention;

[0055] Figure 3 A schematic diagram of a flow chart of an embodiment of point thinning in laser point cloud data provided by the present invention;

[0056] Figure 4 For the present invention Figure 1 A schematic flow chart of an embodiment of step S102;

[0057] Figure 5 For the present invention Figure 1 A schematic flow chart of an embodiment of step S103;

[0058] Figure 6 For the present invention Figure 1A schematic flow chart of an embodiment of step S104;

[0059] Figure 7 A schematic structural diagram of an embodiment of a device for rapid identification of road elements by a mobile laser radar provided by the present invention;

[0060] Figure 8 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0062] like Figure 1 As shown, a specific embodiment of the present invention discloses a method for quickly identifying road elements using a mobile laser radar, comprising:

[0063] S101, obtaining laser point cloud data based on a mobile device, and determining non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point;

[0064] S102, calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set;

[0065] S103, processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point;

[0066] S104. When the mean descending precision (MSE) value and the feature correlation coefficient meet preset conditions, the feature point sample set is classified to obtain a recognition result of the road element.

[0067] It should be noted that the mobile device may be a vehicle or other movable device.

[0068] It should be understood that before determining the non-ground point cloud data based on the elevation difference between each point in the laser point cloud data and the neighboring points, it also includes: filtering the noise points in the point cloud data based on the Euclidean distance from each point in the collected point cloud data to each point in its neighborhood.

[0069] It is worth noting that the laser point cloud data in this embodiment is collected by a mobile device laser scanner based on a multi-platform laser radar measurement system. During the collection process, noise points will be generated due to factors of the instrument itself and the road environment, and these noise points will affect the point cloud feature calculation, so it is necessary to filter the noise points in the laser point cloud data. In this embodiment, the overall idea of ​​filtering noise points in the laser point cloud data is to perform a statistical analysis of the Euclidean distance from each point in the laser point cloud data to each point in its neighborhood; then, the standard interval division is determined based on the two parameters of the standard deviation δ and mean μ of the Euclidean distance; finally, all points outside the standard interval are determined as noise points and removed from the original point cloud data.

[0070] Compared with the prior art, the present embodiment can first filter the laser point cloud data according to the elevation difference between each point and the neighboring points to obtain non-ground point cloud data, thereby reducing the amount of point cloud data. It is also possible to calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and then perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set, thereby performing feature clustering on unevenly distributed point cloud data. Furthermore, the feature point sample set can be processed and correlated separately to obtain the average drop precision MSE value and feature correlation coefficient of each feature point; when the average drop precision MSE value and feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element, thereby obtaining the recognition result of the road element according to the average drop precision MSE value and feature correlation coefficient of each feature point scene, realizing fast and accurate recognition of complex scenes covered by point cloud data, significantly improving the accuracy of data processing, and significantly reducing the need for manual operation, thereby effectively improving work efficiency.

[0071] In some embodiments of the present invention, Figure 2 As shown, after acquiring laser point cloud data based on a mobile device, it includes:

[0072] S201, establishing a KD tree (k-dimensional tree) spatial index for the laser point cloud data;

[0073] S202, performing a K nearest neighbor search on each point in the laser point cloud data according to the KD tree spatial index to determine the neighborhood points of each point;

[0074] S203, calculating the average value μ and standard deviation δ of the Euclidean distance between each point in the laser point cloud data and the corresponding neighborhood point;

[0075] S204, determining points whose Euclidean distance differs from the average value by more than a times the standard deviation as noise points, and filtering the noise points in the laser point cloud data.

[0076] In a specific embodiment of the present invention, a here can be taken as a value according to the denoising requirement. In this embodiment, the value of a is 3, that is, when the difference between the Euclidean distance of a point and the average value μ exceeds 3δ, the point is set as a noise point, and the noise points in the point cloud data are filtered, and the remaining point cloud data are saved and output.

[0077] Furthermore, after filtering the noise points in the point cloud data according to the Euclidean distance from each point in the collected point cloud data to each point in its neighborhood, it also includes: dividing the point cloud data into voxels according to a preset spatial three-dimensional grid scale, and downsampling the point cloud according to the angle between the direction vectors of each voxel.

[0078] It is worth noting that, unlike airborne and ground laser scanning, there is a large amount of repeated scanning data in the point cloud data of mobile device laser scanning. This is due to objective reasons such as repeated routes, red lights, and traffic jams during the driving process of the scanning mobile device. This type of redundant point cloud data is not helpful for data processing, and occupies a large amount of storage space, which seriously affects the data processing efficiency. Therefore, when denoising the point cloud, it is also necessary to remove the redundant point cloud data.

[0079] In some embodiments of the present invention, Figure 3 As shown, after step S204, before determining the non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and the neighboring points, the method further includes:

[0080] S301, dividing the laser point cloud data into point cloud blocks according to a preset first scale, and taking the point cloud blocks whose internal point quantity is greater than a first quantity threshold as thinned point cloud blocks;

[0081] In this embodiment, the first scale set is 10 cm, and the first quantity threshold is 2000 points, that is, a spatial three-dimensional grid with a side length of 10 cm is used to divide the point cloud data into point cloud blocks. If the number of point clouds in the block is greater than 2000, the point cloud block is used as a thinned point cloud block for subsequent calculations, otherwise no subsequent calculations are performed.

[0082] S302, dividing the thinned point cloud block into voxels according to a preset second scale, and taking the voxels whose internal point quantity is greater than a second quantity threshold as valid voxels;

[0083] In this embodiment, the second scale is set to 1 cm, and the second number threshold is 20 points, that is, the thinned point cloud block is divided into voxels with a 1 cm spatial three-dimensional grid, that is, each thinned point cloud block contains 10×10×10=800 voxels. If there are more than 20 laser points in a voxel, it is recorded as a valid voxel and participates in subsequent calculations, otherwise it does not participate in subsequent calculations. The number of valid voxels in the thinned point cloud block is recorded as m.

[0084] S303, taking the point closest to the center of the valid voxel as the voxel center point, and taking the normal vector of the laser point cloud data at the voxel center point as the voxel direction vector of the valid voxel;

[0085] S304, calculating the vector angle between the direction vectors of each voxel in the thinned point cloud block, and calculating the proportion of the vector angle greater than a preset angle threshold;

[0086] In this embodiment, the preset angle threshold of the vector angle is 5 degrees. For each thinned point cloud block, the set of voxel direction vectors is recorded as Calculate the vector angles between the voxel direction vectors in the set Q, and count the number of voxel direction vectors with angles greater than 5 degrees as p. Then the proportion of angles greater than 5 degrees is Where m is the number of valid voxels in the thinned point cloud block.

[0087] S305 , downsampling and thinning the points in the corresponding thinned point cloud block according to the ratio.

[0088] In a specific embodiment of the present invention, downsampling may be performed by adding a preset value based on R. For example, points of a ratio (R+0.1) are uniformly extracted from the thinned point cloud to complete downsampling of the point cloud.

[0089] It is worth noting that a large number of research experiments have shown that segmentation algorithms that simply adopt a single strategy have their own advantages and disadvantages and applicability limitations, and are insufficient in terms of accuracy, stability, and segmentation speed. The segmentation performance of the algorithm can be improved by integrating surface geometric features and smoothness, avoiding the problem of over-segmentation or under-segmentation caused by a single segmentation strategy. This embodiment comprehensively considers the constraints of surface geometric features and smoothness constraints, and provides a segmentation algorithm for road feature monitoring, which realizes accurate segmentation of point cloud data, thereby automatically generating a sample set of feature points of road features.

[0090] In some embodiments of the present invention, step S101 includes:

[0091] Calculate the elevation difference based on the elevation of the current point and the elevation of the corresponding neighboring point;

[0092] If the elevation difference is less than the preset elevation difference threshold, it is determined that the current point does not belong to non-ground point cloud data;

[0093] If the elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that the current point belongs to non-ground point cloud data.

[0094] In a specific embodiment of the present invention, non-ground point cloud data includes point cloud data of man-made objects on the road, such as buildings, street lights, traffic lights, and traffic signs. For a given height difference and a laser point with a large elevation difference, the greater the possibility that it belongs to a street light, a traffic light, and a traffic sign. Assuming A is the point cloud data set of the entire road, and Aggregate is the point set of buildings, street lights, traffic lights, and traffic signs, then the point that satisfies the detection function of formula (1) is the element of the non-ground point cloud data set:

[0095]

[0096] Where Aggregate is a collection of non-ground point cloud data, including point clouds of buildings, street lights, traffic lights, and traffic signs. i is the current point, p j is the neighborhood point of the current point, h pi is the elevation of point pi, For p j The elevation of the point, For p i to p j Elevation difference, It is the preset elevation difference threshold.

[0097] It is understandable that if for a given current point p i , if there is a point p in the neighborhood of this point j Satisfy the relationship Then, click p i That is, the feature points of buildings, street lights, traffic lights or traffic signs, thus completing the rough classification process of non-ground man-made objects.

[0098] Furthermore, through the above steps, feature points of non-ground man-made objects such as buildings, street lights, traffic lights, and traffic signs are extracted, and the surface geometric features are constrained. On this basis, the above-mentioned man-made objects need to be further refined, and then the point cloud set of the rod-shaped objects is extracted. The present invention proposes that the building segmentation based on normal vector feature clustering is based on the existence of a large number of planes in the building point cloud. First, the non-ground data extracted by the above algorithm is clustered by normal vector features, the plane objects in the scene are separated, and the maximum point count threshold is set to perform normal vector detection on the extracted plane objects.

[0099] In some embodiments of the present invention, Figure 4 As shown, step S102 includes:

[0100] S401, calculating the corresponding normal angle according to the normal vectors of each point in the non-ground point cloud data and the two points closest to it in the neighborhood;

[0101] S402, taking M points preset in the non-ground point cloud data as a group, calculating the average value of the corresponding normal angle, where M can be set according to the requirements;

[0102] S403: If the average value of the normal angle is greater than or equal to 0 and less than or equal to a preset first angle threshold, the corresponding point is determined as a feature point to obtain a feature point sample set;

[0103] S404: if the average value of the normal angle is greater than the first angle threshold and less than the preset second angle threshold, determine the corresponding point as a transition point;

[0104] S405: If the average value of the normal angle is greater than or equal to a second angle threshold and less than or equal to π, determine that the corresponding point is a non-feature point, wherein the angle of the second angle threshold is greater than the angle of the first angle threshold.

[0105] In a specific embodiment of the present invention, 5 points are taken as a group, and a two-dimensional point set is taken as an example. For any given 5 points to be classified {p1, p2, p3, p4, p5}, the two points closest to each point to be classified are found as its neighboring points, which are represented by [q1, q2], [q3, q4], [q5, q6], [q7, q8], and [q9, q10] respectively. Then, the normal vector at each point to be classified is estimated, and the angle between the normal line of each point to be classified and the neighboring point is calculated according to the formula As shown in formula (2):

[0106]

[0107] Among them, arccos() represents the arccosine calculation, n i Indicates p i The normal vector at n j Represents point p i In the field j Normal vector at the point.

[0108] After calculating the normal angle at each point to be classified, the average value η of the corresponding normal angle can be calculated. pi , and then obtain the feature point sample set according to the threshold.

[0109] Set the first angle threshold to T 1 , the second angle threshold is T 2 , it is understandable that T 1 is the lower limit of the angle threshold, T 2 is the lower limit of the angle threshold.

[0110] According to the threshold range of the normal angle, the required feature point sample set is obtained, as shown in formula (3):

[0111]

[0112] It should be noted that, in order to improve the classification accuracy, reduce the dimension of the point cloud data, and select the point cloud features that are beneficial to the monitoring and extraction of road elements, the present invention adopts a recursive feature elimination method, that is, the RFE-CV feature selection method, to select the three-dimensional point cloud features, so as to realize the classification and recognition of street lights, traffic lights, and traffic signs. Among them, the three-dimensional point cloud features include: the first eigenvalue 1st_eigenvalue, the second eigenvalue 2nd_eigenvalue, the third eigenvalue 3rd_eigenvalue, roughness, z_normal, x_normal, y_normal, ave_normal, position, color, curvature, point cloud density, intensity, etc.

[0113] In the RFE-CV feature selection of the point cloud, the final optimal feature set will be determined by calculating the importance of each feature point and sorting them. Random forest is a commonly used machine learning algorithm, and there are two common feature importance indicators in random forest: the average decrease MSE value and the Gini importance index. In the stage of constructing a decision tree in the random forest, the Gini importance index is used as the basis for judging the feature segmentation effect. If the index is larger, it means that the feature segmentation effect is worse. On the other hand, the average decrease MSE value is calculated by the root mean square error of the data prediction residuals when the random forest algorithm is executed. A higher average decrease MSE value represents a lower importance, while a lower average decrease MSE value represents a higher importance. Feature selection and evaluation are key steps in the application of random forest.

[0114] In some embodiments of the present invention, as Figure 5 shown, step S103 includes:

[0115] S501. Resample the feature point sample set to obtain a sample subset, use the unselected data as the validation sample set, and construct a regression tree according to the sample subset;

[0116] In this embodiment, it is assumed that the number of data samples in the feature point sample set is n. The bootstrapping resampling method is used to extract multiple samples from the n original samples to generate a sample subset, and a regression tree is constructed according to these sample subsets. The unselected data forms the validation sample set, and the number of samples in the sample subset can be set according to requirements.

[0117] As a preferred implementation manner, assume that there are p samples in the sample subset, randomly select x (x < p) as alternative values at each regression tree node, and iteratively split the samples into different subsets according to the feature rules to construct the corresponding regression tree.

[0118] S502, training the random forest model through the regression tree in a loop, and verifying the classification effect of the random forest model through the verification sample set, to obtain the mean drop precision MSE value of each feature point output by the random forest model;

[0119] In this embodiment, the generated regression trees are assembled into a random forest model and the prediction residual square MSE is used to evaluate the effect of the regression model. Then, the feature with the least importance is deleted by obtaining the mean decreasing precision MSE value. The calculation of the MSE value is shown in formula (4):

[0120]

[0121] Among them, MSE OOB is the MSE value, y i is the actual value of the dependent variable in the data, y i OOB is the predicted value of the random forest for the data.

[0122] S503 , performing correlation analysis on the features of the classified sample set of the feature point sample set to obtain a feature correlation coefficient of the classified sample set.

[0123] In a specific embodiment of the present invention, sample quality can generally be evaluated using methods such as manual visual inspection, sample credibility analysis, and sample correlation analysis. Among them, manual visual inspection has relatively high manpower and time costs, so the sample separability method is more commonly used in previous studies. In this embodiment, the sample quality evaluation index is composed of a combination of Person correlation coefficient and credibility analysis.

[0124] Among them, the sample feature correlation analysis method analyzes the sample quality by observing the correlation performance of the artificial non-ground object features between samples of the same type and samples of different types. Generally, the higher the degree of aggregation between samples of the same type and the lower the correlation between artificial objects of samples of different types, the higher the sample quality. Correlation indicators include Pearson (Person) correlation coefficient, Spearman correlation coefficient or chi-square test, etc. The Person correlation coefficient is often used to express the correlation of samples, and its definition is shown in formula (5):

[0125]

[0126] Among them, γ is the feature correlation coefficient, X and Y represent two pre-selected features, E is the mathematical expectation, cov represents the covariance, and N represents the number of feature values.

[0127] When the correlation coefficient is between 0.8 and 1.0, it means that the variables are highly correlated; when the correlation coefficient is between 0.5 and 0.8, the variables are moderately correlated; when the correlation coefficient is between 0.3 and 0.5, there is a low correlation between the variables; and when the correlation coefficient is between 0 and 0.3, there is a weak correlation or almost no correlation between the variables.

[0128] In some embodiments of the present invention, Figure 6 As shown, step S104 includes:

[0129] S601, determining the importance weight of the corresponding feature according to the MSE value;

[0130] S602, performing credibility analysis on the classified sample set to determine the credibility of the recognition result;

[0131] S603: When the importance weight, credibility and feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element.

[0132] In a specific embodiment of the present invention, for credibility analysis, this embodiment uses an inconsistency measurement function and a sequence randomness test P value function to perform credibility analysis on the classifier prediction output result. n+1 Add a class label y w After that, together with the training samples, they constitute the test sample sequence Z (n+1) ={z 1 ,z 2 ,…,z n ,z n+1 ,}, for Z n+1 , in Z (n+1) The randomness P value in the test sequence is calculated. The P value is calculated as follows:

[0133]

[0134] Where i is the inconsistency measurement value sequence a 1 , a 2 , …a n , a n+1 Greater than a n+1 The number of measurement values.

[0135] Among them, the inconsistency measure a of each sample i It is obtained by constructing the sample inconsistency measurement function, as shown in formula (7):

[0136]

[0137] in, Represents x iThe sum of the distances of the K nearest neighbors of the same category of samples, Represents x i The sum of the K nearest neighbor distances of different categories of samples. The P value describes the similarity between the current test sample sequence and independent and identical distribution. The higher the similarity, the higher the test sample belongs to y w The greater the possibility, the credibility is defined as C = P × 100%, where C is the credibility.

[0138] Through the above method, street lights, traffic lights, and traffic signs in road elements can be quickly and accurately classified and identified, and accurate classification results can be output.

[0139] In some embodiments of the present invention, when the importance weight, credibility and feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element, including:

[0140] Sort the features according to their importance weights and delete the features with the smallest importance weights until the number of features retained is a preset N, so as to generate an optimal feature set, where N can be set according to requirements;

[0141] If the feature correlation coefficient is less than a preset correlation coefficient threshold and / or the credibility is less than a preset credibility threshold, a feature point sample set is obtained again based on the non-ground point cloud data, and then an optimal feature set is regenerated;

[0142] The optimal feature set is classified to obtain the recognition result of road elements.

[0143] In a specific embodiment of the present invention, features are sorted according to importance weights. After the feature importance weights are sorted, features with low importance are deleted, and this process is repeated. The number of features to be retained is determined according to actual conditions, and a new feature set is obtained. The selected feature set is cross-validated, and the number of features with the highest average score is finally determined as the optimal feature set. If the feature correlation coefficient is less than the preset correlation coefficient threshold and / or the credibility is less than the preset credibility threshold, the process can return to step S102, and the subsequent steps can be repeated to obtain a feature point sample set based on the non-ground point cloud data, and then the optimal feature set is regenerated until the correlation coefficient and credibility are higher than the corresponding thresholds to obtain the optimal feature set.

[0144] In some embodiments of the present invention, the optimal feature set is classified to obtain the recognition result of the road element, including:

[0145] The random forest model is trained based on the optimal feature set to obtain a trained random forest model;

[0146] The validation sample set is classified by the trained random forest model to obtain the recognition results of road elements.

[0147] In a specific embodiment of the present invention, the random forest model can be trained based on the optimal feature set to obtain a trained random forest model, and then the validation sample set can be classified by the trained random forest model to obtain the recognition result of the road element. The above method can quickly and accurately classify and recognize the street lights, traffic lights, and traffic signs in the road elements, and output accurate classification results.

[0148] Furthermore, in this embodiment, the point cloud data of the same location at different times can be classified and processed using the above random forest algorithm to obtain the classification results of road elements, and then the point cloud data of the same location at different times can be classified and identified with the road environment as a constraint condition, which can meet the visualization, interaction and scheduling processing of point cloud data, and can be quickly classified by different categories, so as to realize the rapid visualization and identification of road elements of mobile device laser point cloud. The specific steps include:

[0149] Different man-made non-ground objects are classified by color attributes, thereby identifying different man-made non-ground objects.

[0150] The data from the two phases are superimposed, with the light-colored portion representing areas with obvious change characteristics and the dark-colored portion representing areas that have not changed, thereby enabling rapid identification of the changing status of road elements.

[0151] The characteristic attributes of man-made non-ground objects are marked by setting different labels, and the attribute features of different categories are distinguished based on the geometric features and feature descriptions of the object's attribute information.

[0152] The embodiment of the present application provides a method for rapid identification of road elements based on laser point cloud of mobile device, which can effectively reduce the influence of noise points and redundant points in laser point cloud data on cloud point feature calculation by filtering noise points in collected laser point cloud data and then downsampling laser point cloud data. Then, non-ground point cloud data in laser point cloud data is identified by elevation, and by feature clustering of non-ground point cloud data, buildings therein are identified and filtered, and feature point sample sets including street lights, traffic lights, and traffic signboard road elements are obtained, so as to realize automatic generation of road element feature point set sample sets, and then street lights, traffic lights, and traffic signboard roads in road elements are specifically classified by random forest model, and the accuracy of classification results is determined by calculating the feature correlation coefficient and credibility of the classified sample set, so as to realize accurate identification of point cloud data covering complex scenes, diverse target objects, and huge data volume and uneven point density distribution. The method for rapid identification of road elements based on laser point cloud of mobile device provided in the present embodiment not only improves the accuracy of data processing, but also significantly reduces the need for manual operation, thereby effectively improving work efficiency.

[0153] It is worth noting that the basic idea of ​​adaptive adjustment of object-level sample parameters based on sample quality is:

[0154] In order to better implement the road element rapid identification method of the mobile laser radar in the embodiment of the present invention, based on the road element rapid identification method of the mobile laser radar, the embodiment of the present invention also provides a road element rapid identification device of the mobile laser radar, such as Figure 7 As shown, the road element rapid identification device 700 of the mobile laser radar includes:

[0155] The data determination module 701 is used to obtain laser point cloud data based on a mobile device, and determine non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point;

[0156] Angle calculation module 702, used to calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set;

[0157] The sample processing module 703 is used to process and perform correlation analysis on the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point;

[0158] The classification judgment module 704 is used to classify the feature point sample set to obtain the recognition result of the road element when the mean drop precision MSE value and the feature correlation coefficient meet the preset conditions.

[0159] The road element rapid identification device 700 of the mobile laser radar provided in the above embodiment can implement the technical solution described in the above mobile laser radar road element rapid identification method embodiment. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above mobile laser radar road element rapid identification method embodiment, which will not be repeated here.

[0160] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802 and a display 803. Figure 8 Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0161] In some embodiments, the memory 802 may be an internal storage unit of the electronic device 800, such as a hard disk or memory of the electronic device 800. In other embodiments, the memory 802 may also be an external storage device of the electronic device 800, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 800.

[0162] Furthermore, the memory 802 may include both an internal storage unit of the electronic device 800 and an external storage device. The memory 802 is used to store application software installed in the electronic device 800 and various data.

[0163] In some embodiments, the processor 801 can be a central processing unit (CPU), a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory 802, such as the road element rapid identification method based on mobile device laser point cloud in the present invention.

[0164] In some embodiments, the display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 803 is used to display information of the electronic device 800 and to display a visual user interface. The components 801-803 of the electronic device 800 communicate with each other through a system bus.

[0165] In some embodiments of the present invention, when the processor 801 executes the road element rapid identification program of the mobile laser radar in the memory 802, the following steps may be implemented:

[0166] Obtain laser point cloud data based on mobile devices, and determine non-ground point cloud data based on the elevation difference between each point in the laser point cloud data and the neighboring points;

[0167] Calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set;

[0168] Process and analyze the correlation of feature point sample sets to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point;

[0169] When the mean descending precision (MSE) value and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element.

[0170] It should be understood that: when the processor 801 executes the road element rapid identification program based on the mobile device laser point cloud in the memory 802, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0171] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 800 mentioned, and the electronic device 800 may be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable electronic device may also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 800 may not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0172] Accordingly, an embodiment of the present application also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, the method steps or functions of quickly identifying road elements based on laser point clouds of mobile devices provided in the above-mentioned method embodiments can be implemented.

[0173] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0174] The above is a detailed introduction to the method and device for rapid identification of road elements based on mobile device laser point cloud provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for quickly identifying road elements using a mobile laser radar, characterized in that: include: Acquire laser point cloud data based on a mobile device, and determine non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point; Calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set; Processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point; When the mean decreasing precision MSE value and the feature correlation coefficient meet the preset conditions, classifying the feature point sample set to obtain the recognition result of the road element; Before determining the non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and the neighboring points, the method includes: Dividing the laser point cloud data into point cloud blocks according to a preset first scale, and taking the point cloud blocks whose internal point quantity is greater than a first quantity threshold as thinned point cloud blocks; Dividing the thinned point cloud block into voxels according to a preset second scale, and taking voxels whose number of internal points is greater than a second number threshold as valid voxels; Taking the point closest to the center of the valid voxel as the voxel center point, and taking the normal vector of the laser point cloud data at the voxel center point as the voxel direction vector of the valid voxel; Calculating the vector angle between the direction vectors of each voxel in the thinned point cloud block, and calculating the proportion of the vector angle that is greater than a preset angle threshold; The points in the corresponding thinned point cloud block are downsampled and thinned according to the ratio.

2. The method for rapid identification of road elements by mobile laser radar according to claim 1, characterized in that: The step of determining the non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and the neighboring points includes: Calculating the elevation difference according to the elevation of the current point and the elevation of the corresponding neighboring point; If the elevation difference is less than a preset elevation difference threshold, determining that the current point does not belong to the non-ground point cloud data; If the elevation difference is greater than or equal to the preset elevation difference threshold, it is determined that the current point belongs to the non-ground point cloud data.

3. The method for rapid identification of road elements by mobile laser radar according to claim 1, characterized in that: The calculating the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and performing feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set, includes: Calculate the corresponding normal angle according to the normal vectors of each point in the non-ground point cloud data and the two points closest to it in the neighborhood; And taking the preset M points in the non-ground point cloud data as a group, calculating the average value of the corresponding normal angles; If the average value of the normal angle is greater than or equal to 0 and less than or equal to a preset first angle threshold, the corresponding point is determined as a feature point to obtain the feature point sample set; If the average value of the normal angle is greater than the first angle threshold and less than a preset second angle threshold, the corresponding point is determined to be a transition point; If the average value of the normal angles is greater than or equal to the second angle threshold and less than or equal to π, the corresponding point is determined to be a non-feature point.

4. The method for rapid identification of road elements by mobile laser radar according to claim 1, characterized in that: The processing and correlation analysis of the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point includes: Resampling the feature point sample set to obtain a sample subset, using the unsampled data as a validation sample set, and constructing a regression tree based on the sample subset; The random forest model is trained by the regression tree in a loop, and the classification effect of the random forest model is verified by the verification sample set to obtain the mean descending precision (MSE) value of each feature point output by the random forest model; By performing correlation analysis on the features of the classified sample set of the feature point sample set, a feature correlation coefficient of the classified sample set is obtained.

5. The method for rapid identification of road elements by mobile laser radar according to claim 4 is characterized in that: When the mean decreasing precision MSE value and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element, including: Determine the importance weight of the corresponding feature according to the MSE value; Performing credibility analysis on the classified sample set to determine the credibility of the recognition result; When the importance weight, the credibility and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element.

6. The method for rapid identification of road elements by mobile laser radar according to claim 5, characterized in that: When the importance weight, the credibility and the feature correlation coefficient meet the preset conditions, the feature point sample set is classified to obtain the recognition result of the road element, including: Sort the features according to the importance weights and delete the features with the smallest importance weights until the number of features retained is a preset N, so as to generate an optimal feature set; If the feature correlation coefficient is less than a preset correlation coefficient threshold and / or the credibility is less than a preset credibility threshold, a feature point sample set is obtained again based on the non-ground point cloud data, and then an optimal feature set is regenerated; The optimal feature set is classified to obtain the recognition result of the road element.

7. The method for rapid identification of road elements by mobile laser radar according to claim 1, characterized in that: After the laser point cloud data is obtained based on the mobile device, the method includes: Establishing a KD tree spatial index for the laser point cloud data; According to the KD tree spatial index, a K nearest neighbor search is performed on each point in the laser point cloud data to determine the neighborhood point of each point; Calculate the average and standard deviation of the Euclidean distance between each point in the laser point cloud data and the corresponding neighboring points; Points whose Euclidean distance differs from the average value by more than a times the standard deviation are determined as noise points, and the noise points in the laser point cloud data are filtered.

8. The method for rapid identification of road elements by mobile laser radar according to claim 6, characterized in that: The classifying the optimal feature set to obtain the recognition result of the road element includes: Training the random forest model based on the optimal feature set to obtain a trained random forest model; The verification sample set is classified by the trained random forest model to obtain the recognition result of the road element.

9. A road element rapid identification device using a mobile laser radar, characterized in that: include: A data determination module, used to obtain laser point cloud data based on a mobile device, and determine non-ground point cloud data according to the elevation difference between each point in the laser point cloud data and a neighboring point; An angle calculation module, used to calculate the normal angle between each point in the non-ground point cloud data and two points in the neighborhood, and perform feature clustering on the non-ground point cloud data according to the normal angle to obtain a feature point sample set; A sample processing module is used to process and perform correlation analysis on the feature point sample set to obtain the mean drop precision MSE value and feature correlation coefficient of each feature point; A classification judgment module, used for classifying the feature point sample set to obtain a road element recognition result when the mean drop precision MSE value and the feature correlation coefficient meet a preset condition; The data determination module is also used to divide the laser point cloud data into point cloud blocks according to a preset first scale, and use the point cloud blocks with a number of internal points greater than a first number threshold as thinned point cloud blocks; divide the thinned point cloud blocks into voxels according to a preset second scale, and use the voxels with a number of internal points greater than a second number threshold as valid voxels; use the point closest to the center of the valid voxel as the voxel center point, and use the normal vector of the laser point cloud data at the voxel center point as the voxel direction vector of the valid voxel; calculate the vector angle between each voxel direction vector in the thinned point cloud block, and calculate the proportion of the vector angle greater than a preset angle threshold; and downsample and thin out the points in the corresponding thinned point cloud block according to the proportion.

Images