CN117894015B

CN117894015B - Point cloud annotation data optimization method and system

Info

Publication number: CN117894015B
Application number: CN202410297172.3A
Authority: CN
Inventors: 温志伟; 陈奇; 宋弯弯; 季航锋; 叶建标; 王强
Original assignee: Zhejiang Whyis Technology Co ltd
Current assignee: Zhejiang Whyis Technology Co ltd
Priority date: 2024-03-15
Filing date: 2024-03-15
Publication date: 2024-05-24
Anticipated expiration: 2044-03-15
Also published as: CN117894015A

Abstract

The invention discloses a point cloud annotation data optimization method and a point cloud annotation data optimization system, wherein the method comprises the following steps: s1, acquiring an original point cloud set; s2, carrying out background difference on original point clouds in the original point cloud set, filtering to obtain a difference point cloud set, and updating the original point cloud set according to a filtering result to obtain a latest point cloud set; s3, clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of the differential point cloud; s4, calculating information quantity of the latest point cloud, arranging the information quantity in a descending order, taking the latest point cloud corresponding to the first information quantity as a datum point cloud, respectively calculating comprehensive change rates of the latest point cloud corresponding to the subsequent information quantity and the datum point cloud according to the sequence, and marking the latest point cloud corresponding to the comprehensive change rate smaller than a preset change rate threshold; and (3) the marked latest point cloud does not participate in subsequent calculation, repeating the step (S4) until the marking is completed by the last datum point cloud, and taking all the unmarked latest point clouds as marked point clouds. The point cloud optimization efficiency is improved.

Description

Point cloud annotation data optimization method and system

Technical Field

The invention relates to the field of point cloud processing, in particular to a point cloud annotation data optimization method and system.

Background

The laser radar is adopted to scan the scene to obtain multi-frame continuous point clouds, the point clouds contain a lot of target information, different types of labels need to be carried out on the originally acquired point clouds according to different use conditions, for example, the types of targets, the number of targets and the like in the point clouds are labeled, and the labeled sample point clouds can be used in the scenes such as model training or object recognition and the like.

In the prior art, the labeling of sample point clouds mainly depends on manual labeling, and for a sequence of point clouds, the change of targets among the real part of point clouds is small, for example, only one person in a scene walks, but does not move to a certain place and stays for a long time, and under the condition that the labeling is not performed in advance but is performed in advance, a great deal of labor cost and time cost are consumed. In addition, it is obvious to a person training a model that he is faced with a sequence of originally acquired point clouds, which may consist of thousands of point clouds, and then looks frame by frame in the point cloud visualization software and then picks up labeled samples, which is a very time-consuming and laborious task. And the selection process is easily influenced by subjective judgment of people, and the difficulty of manual selection is extremely high.

Aiming at the problems of time and labor waste caused by manual selection in the prior art, no effective solution exists at present.

Disclosure of Invention

In order to solve the problems, the invention provides a point cloud labeling data optimizing method and system, which are characterized in that the original point clouds are differentiated, filtered and clustered to obtain directed bounding box information of a clustered target, then the change rate of the differential point clouds between two point clouds and the change rate of the clustered target are calculated, the two change rates are integrated to obtain an integrated change rate, and the point clouds of a sequence are optimized according to the integrated change rate in sequence to obtain more representative point clouds, so that the problems of time and labor waste in manual selection in the prior art are solved.

In order to achieve the above object, the present invention provides a point cloud labeling data optimization method, which includes: s1, acquiring an original point cloud set; wherein the original point cloud set comprises a continuous multi-frame original point cloud; s2, carrying out background difference on original point clouds in the original point cloud set, filtering to obtain a filtered difference point cloud set, and deleting original point clouds corresponding to the filtered point clouds from the original point cloud set to obtain a latest point cloud set; s3, clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud; s4, calculating information quantity of corresponding latest point clouds in the latest point cloud set according to the number of points of differential point clouds and the number of clustering targets, arranging the information quantity in a descending order, taking the latest point clouds corresponding to the first information quantity as reference point clouds, respectively calculating comprehensive change rates of the latest point clouds corresponding to the subsequent information quantity and the reference point clouds according to the point difference and the clustering target difference in sequence, and marking the latest point clouds corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step S4 until the last datum point cloud is labeled, and taking all the latest point clouds not labeled as labeled point clouds.

Further optionally, the calculating, according to the point difference and the clustering target difference, the comprehensive change rate of the latest point cloud and the reference point cloud corresponding to the subsequent information quantity in sequence includes: s401, determining a first differential point cloud corresponding to the datum point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating a differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud; s402, performing Hungary matching on the clustering target set of the first differential point cloud and the clustering target set of the second differential point cloud to obtain a single clustering target with successful matching and a single clustering target with unsuccessful matching, calculating the change rate of the matching target according to the directed bounding box information of the two clustering targets in the clustering target matching pair and the clustering target points, calculating the change rate of the non-matching target according to the points of the single clustering target and the points of the differential point cloud, and weighting and summing the change rates of all the matching targets and the change rates of the non-matching targets to obtain the change rate of the clustering target; and S403, carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

Further optionally, the calculating the change rate of the differential point cloud according to the point number of the first differential point cloud and the point number of the second differential point cloud includes: s4011, counting first points of the first differential point cloud and second points of the second differential point cloud; s4012, calculating absolute change rates of two differential point clouds according to the first points and the second points; s4013, defining a first statistical region in the first differential point cloud according to a preset statistical range, performing equidistant slicing on the first statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the third points in each point cloud slice; s4014, a second statistical region is defined in the second differential point cloud according to a preset statistical range, equidistant slicing is conducted on the second statistical region in the X direction and the Y direction, a plurality of point cloud slices are obtained, and fourth points in each point cloud slice are counted; s4015, calculating the distribution change rate of two differential point clouds according to the first point, the second point, the third point and the fourth point; s4016, carrying out weighted summation on the absolute change rate and the distribution change rate to obtain the change rate of the differential point cloud.

Further optionally, the calculating the change rate of the matching target according to the directed bounding box information of the two clustering targets in the clustering target matching pair and the clustering target point number is calculated by the following formula:

；

wherein, Weighting the rate of change of the height of the directed bounding box for a clustered object,/>Height value of directed bounding box of ith clustered object of first differential point cloud,/>A height value of a directed bounding box of a jth clustering target of the second differential point cloud; /(I)Weight of the length change rate of the directed bounding box for the clustered objects,/>Length value of directed bounding box of ith clustering target of first differential point cloud,/>A length value of a directed bounding box of a jth clustering target of the second differential point cloud; /(I)Weighting the width rate of change of the directed bounding box for a clustered object,/>Width value of directed bounding box of ith clustering result of first differential point cloud,/>A width value of a directed bounding box of a j-th clustering result of the second differential point cloud; /(I)Weighting the rotation angle change rate of the directed bounding box of the clustered object,/>Rotation angle of directional bounding box of ith clustering result of first differential point cloud,/>The rotation angle of the directed bounding box of the j-th clustering result of the second differential point cloud; /(I)Weighting of center point relative offset of directed bounding box for clustering target,/>，/>，/>X, y and z values of the central points of the directional bounding boxes of the ith clustering result of the first differential point cloud respectively; /(I)，/>，/>X, y and z values of the central points of the directed bounding boxes of the jth clustering result of the second differential point cloud respectively; /(I)Weight of point change rate of clustering result,/>Points of the ith clustering target of the first differential point cloud; /(I)And the number of the jth clustering target of the second differential point cloud.

Further optionally, the calculating the change rate of the unmatched targets according to the points of the single clustering targets and the points of the differential point cloud comprises the following steps:

；

wherein, Is the number of points of a single cluster target which is not successfully matched,/>Is the point number of the differential point cloud to which the single clustering target belongs.

Further optionally, the calculating the absolute change rate of the two differential point clouds according to the first point number and the second point number is calculated by the following formula:

；

wherein, Is absolute change rate,/>For the first point,/>Is the second point number.

Further optionally, the calculating the distribution change rate of the two differential point clouds according to the first point number, the second point number, the third point number and the fourth point number is calculated by the following formula:

；

wherein, For the distribution change rate,/>A third point number that is an i-th slice of the first differential point cloud; /(I)Fourth points of the ith slice of the second differential point cloud; /(I)For the first point,/>Is the second point number.

In another aspect, the present invention further provides a point cloud labeling data optimization system, including: the data acquisition module is used for acquiring an original point cloud set; wherein the original point cloud set comprises a continuous multi-frame original point cloud; the preprocessing module is used for carrying out background difference on original point clouds in the original point cloud set, filtering to obtain a filtered differential point cloud set, deleting original point clouds corresponding to the filtered point clouds from the original point cloud set, and obtaining a latest point cloud set; the clustering module is used for clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud; the optimization module is used for calculating the information quantity of the corresponding latest point cloud in the latest point cloud set according to the number of points of the differential point cloud and the number of clustering targets, arranging the information quantity in a descending order, taking the latest point cloud corresponding to the first information quantity as a datum point cloud, respectively calculating the comprehensive change rate of the latest point cloud corresponding to the subsequent information quantity and the datum point cloud according to the point difference and the clustering target difference in sequence, and marking the latest point cloud corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step of the optimization module until labeling is completed by the last datum point cloud, and taking all the latest point clouds not labeled as labeling point clouds.

Further optionally, the preferred module includes: the differential point cloud change rate calculation submodule is used for determining a first differential point cloud corresponding to the reference point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating the differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud; the clustering target change rate calculation submodule is used for carrying out Hungary matching on the clustering target set of the first differential point cloud and the clustering target set of the second differential point cloud to obtain successfully matched clustering target pairs and unmatched single clustering targets, calculating the matching target change rate according to directed bounding box information of the two clustering targets in the clustering target pairs and clustering target points, calculating unmatched target change rate according to the points of the single clustering targets and the points of the differential point clouds, and carrying out weighted summation on all the matching target change rates and unmatched target change rates to obtain clustering target change rate; and the comprehensive change rate calculation submodule is used for carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

Further optionally, the differential point cloud change rate calculating submodule includes: the statistics unit is used for counting the first points of the first differential point cloud and the second points of the second differential point cloud; the absolute change rate calculation unit is used for calculating the absolute change rate of the two differential point clouds according to the first point number and the second point number; the first slice counting unit is used for defining a first counting area in the first differential point cloud according to a preset counting range, carrying out equidistant slicing on the first counting area in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the third point number in each point cloud slice; the second slice statistics unit is used for defining a second statistical region in the second differential point cloud according to a preset statistical range, carrying out equidistant slicing on the second statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the fourth point number in each point cloud slice; the distribution change rate calculation unit is used for calculating the distribution change rate of the two differential point clouds according to the first point, the second point, the third point and the fourth point; and the differential point cloud change rate calculation unit is used for carrying out weighted summation on the absolute change rate and the distribution change rate to obtain the differential point cloud change rate.

The technical scheme has the following beneficial effects: the difference point cloud and the corresponding clustering target of the point cloud are calculated, the difference of the difference point cloud and the difference of the clustering target are further quantized to obtain the comprehensive change rate between the two point clouds, the change condition between the two point clouds is measured, the batch of point clouds are optimized according to the comprehensive change rate, the selection efficiency of the marked point clouds is improved, and a large amount of manpower and time are saved; and the information quantity is calculated by the number of points of the differential point cloud and the number of clustering results, so that the information quantity of the point cloud which is preferably obtained is ensured.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flowchart of a preferred method for point cloud annotation data provided by an embodiment of the present invention;

FIG. 2 is a flowchart of a method for calculating an integrated change rate according to an embodiment of the present invention;

FIG. 3 is a flowchart of a method for calculating a change rate of a differential point cloud according to an embodiment of the present invention;

Fig. 4 is a schematic structural diagram of a point cloud labeling data optimization system according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a preferred module provided by an embodiment of the present invention;

Fig. 6 is a schematic structural diagram of a differential point cloud change rate calculation submodule according to an embodiment of the present invention.

Reference numerals: 100-a data acquisition module; 200-a pretreatment module; 300-a clustering module; 400-preference module; 4001-a differential point cloud change rate calculation submodule; 40011-a statistics unit; 40012-an absolute change rate calculation unit; 40013-a first slice statistics unit; 40014-a second slice statistics unit; 40015-a distribution change rate calculation unit; 40016-a differential point cloud change rate calculation unit; 4002-clustering target change rate calculation submodule; 4003—comprehensive rate of change calculation submodule.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In order to solve the problem that in the prior art, selecting point cloud is time consuming and labor consuming, the embodiment of the invention provides a point cloud labeling data optimizing method, and fig. 1 is a flowchart of the point cloud labeling data optimizing method provided by the embodiment of the invention, and as shown in fig. 1, the method comprises the following steps:

s1, acquiring an original point cloud set; wherein the original point cloud set comprises a continuous multi-frame original point cloud;

If the point cloud of a certain scene needs to be acquired, the laser radar is placed at a proper position to acquire a section of point cloud sequence in the target scene. Assume that there are t point clouds in the sequence, which is recorded as an original point cloud set { (Where)/(And (5) the original point cloud of the t frame.

S2, carrying out background difference on original point clouds in the original point cloud set, filtering to obtain a filtered difference point cloud set, and deleting original point clouds corresponding to the filtered point clouds from the original point cloud set to obtain a latest point cloud set;

collecting an original point cloud without a foreground object as a background point cloud . Sequentially combining each original point cloud in the original point cloud set with a background point cloud/>And carrying out difference to obtain a point cloud after the difference.

Specifically, first, the original point cloud of the first frame is obtainedAnd background Point cloud/>Performing difference to obtain a point cloud/>, after the difference. And then the second frame original point cloud/>And background Point cloud/>Performing difference to obtain a point cloud/>, after the difference. And so on, finally, the t frame point cloud/>Cloud of background points/>Performing difference to obtain a point cloud/>. Thus, a differential point cloud set {/> isobtained}。

After differencing, the obtained point cloud set {Filtering. Firstly counting the points of each point cloud in the set, comparing the points corresponding to each point cloud with a preset point threshold delta _pt, and if the points of the current point cloud are smaller than the preset point threshold delta _pt, considering that the foreground points of the current point cloud are fewer and have no marked meaning, and deleting the foreground points from the differential point cloud set. Each point cloud in the point cloud set performs the filtering operation to obtain a filtered differential point cloud set {/>(Where)/(. The original point cloud corresponding to the deleted point cloud is deleted from the original point cloud set while the differential point cloud set is filtered, so that a filtered original point cloud set { is obtainedI.e. the most recent point cloud set.

S3, clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud;

Clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud, wherein the clustering target set may comprise one or more targets, and naturally, the situation that no targets exist is also possible, and the clustering target set is an empty set. Each cluster target is represented by a directed bounding box, and one directed bounding box information corresponds to each cluster target.

Let the ith differential point cloud get M clustering targets altogether. Then the clustering target set of the ith differential point cloud is denoted as R _i {,/>,…,/>}. Wherein/>Directed bounding box (Oriented Bounding Box, OBB) information representing the jth clustered object of the ith differential point cloud.

Directed bounding box informationCan be expressed as {/> , /> ,/> , /> , /> , />，/>}. Wherein/>The x value of the central point of the directed bounding box of the jth clustering target of the ith differential point cloud; /(I)A y value of a central point of a directed bounding box of a jth clustering target of the ith differential point cloud; /(I)Z value of central point of directional bounding box of jth clustering target of ith differential point cloud; /(I)The length of a directed bounding box of the jth clustering target of the ith differential point cloud; /(I)The width of the directed bounding box of the jth clustering target of the ith differential point cloud; /(I)The directed bounding box of the jth clustering target of the ith differential point cloud is high; /(I)The rotation angle of the directed bounding box of the jth clustering target of the ith differential point cloud.

As an alternative embodiment, the clustering method includes: the European clustering, density clustering and other methods are notable in that the same set of parameters needs to be executed in the whole algorithm operation process in the clustering process, and most targets can be clustered under the parameters.

S4, calculating information quantity of corresponding latest point clouds in the latest point cloud set according to the number of points of differential point clouds and the number of clustering targets, arranging the information quantity in a descending order, taking the latest point clouds corresponding to the first information quantity as reference point clouds, respectively calculating comprehensive change rates of the latest point clouds corresponding to the subsequent information quantity and the reference point clouds according to the point difference and the clustering target difference in sequence, and marking the latest point clouds corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step S4 until the last datum point cloud is labeled, and taking all the latest point clouds not labeled as labeled point clouds.

And for each latest point cloud in the latest point cloud set, obtaining the point number of the corresponding differential point cloud and the number of clustering targets in the clustering target set, calculating the information quantity corresponding to the latest point cloud, and calculating the information quantity of all the latest point clouds to obtain the information quantity set.

As an alternative embodiment, the information amount of a latest point cloud may be calculated by the following formula:

；

wherein, For the information content of the latest point cloud,/>For the point number of the point cloud corresponding to the latest point cloud,The point number of the point cloud is the point number of the point cloud corresponding to the latest point cloud. /(I)For the number of clustering targets corresponding to the latest point cloud,/>And the weight of the number of the clustering targets corresponding to the latest point cloud.

Ordering the information quantity sets according to the order of the information quantity from large to small to obtain an ordered point cloud information quantity set {Will have the largest information volume/>Corresponding latest point cloud of (a) is used as a datum point cloud, and/> iscalculated in sequenceCorresponding latest point cloud to/>Corresponding up-to-date point clouds, namely m-1 up-to-date point clouds andThe comprehensive change rate of the corresponding latest point clouds is calculated according to the change conditions of the two latest point clouds, and in the embodiment, the comprehensive change rate is calculated according to the point difference and the clustering target difference of the differential point clouds corresponding to the two latest point clouds, so that the change degree between the two latest point clouds is represented, if the change degree is low, the current latest point clouds compared with the reference point clouds are considered to be not changed greatly and are not used as the marked point clouds; otherwise, if the change degree is high, the current latest point cloud compared with the reference point cloud is considered to be changed greatly, and the current latest point cloud can be reserved.

In order to measure the change degree, in this embodiment, a change rate threshold δ _tcr is set, and the steps are sequentially determinedCorresponding latest point cloud to/>Corresponding latest point clouds-m-1 latest point clouds and/>Whether the integrated rate of change of the corresponding reference point cloud is greater than a set rate of change threshold δ _tcr. If there are AND/>, in m-1 most recent point cloudsIf the integrated change rate of the corresponding reference point clouds is smaller than the change rate threshold delta _tcr, marking the latest point clouds, and marking as non-marking. And repeating the steps for the latest point cloud corresponding to the next untagged information quantity, wherein the latest point cloud marked in the subsequent cycle cannot participate in the comprehensive change rate calculation with other latest point clouds. And after the circulation judgment of the latest point cloud corresponding to the whole ordered point cloud information amount set is completed, the latest point cloud marked with the label is not marked, and in addition, the latest point cloud marked with the label is the required marked point cloud data.

Now, an example will be described: if in order toWhen the comprehensive change rate is judged as the datum point cloud, the result is that only/>, of the m-1 latest point cloudsAnd/>Corresponding up-to-date point cloud sum/>The integrated rate of change of the corresponding reference point cloud of (c) is less than the rate of change threshold delta _tcr, then the information quantity/>And/>The corresponding latest point cloud is marked with a label which is not marked. Then continue to iterate to/>Corresponding latest point cloud, but due to information volume/>The corresponding latest point cloud is marked with a label which is not marked, so that the comprehensive change rate judgment of the latest point cloud corresponding to other information amounts is not carried out. Then continue to iterate to/>Corresponding up-to-date point cloud because of/>The corresponding latest point cloud is not labeled with unlabeled labels, and will therefore/>Corresponding latest point clouds are respectively summed/>Corresponding latest point cloud to/>Corresponding up-to-date point clouds these m-4 up-to-date point clouds (/ >)The corresponding latest point cloud is marked with a label which is not marked, so that the judgment of the comprehensive change rate is skipped). And labeling the latest point cloud with the comprehensive change rate smaller than the change rate threshold delta _tcr with unlabeled labels. Then, the process was repeated as above until/>The corresponding latest point cloud ends the cycle. The latest point cloud which is not marked with the label is the required marked point cloud.

As an optional implementation manner, fig. 2 is a flowchart of a comprehensive change rate calculating method provided by the embodiment of the present invention, and as shown in fig. 2, according to a point difference and a clustering target difference, the comprehensive change rates of a latest point cloud and a reference point cloud corresponding to a subsequent information amount are calculated in sequence, where the comprehensive change rate calculating method includes:

s401, determining a first differential point cloud corresponding to the reference point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating a differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud;

And taking the differential point cloud corresponding to the reference point cloud as a first differential point cloud, taking the differential point cloud corresponding to the current latest point cloud as a second differential point cloud, and calculating the differential point cloud change rate according to the points of the first differential point cloud and the second differential point cloud to measure the change of the point cloud global information between the reference point cloud and the current latest point cloud.

S402, performing Hungary matching on a clustering target set of the first differential point cloud and a clustering target set of the second differential point cloud to obtain a successfully matched clustering target pair and an unmatched single clustering target, calculating a matching target change rate according to directed bounding box information of two clustering targets in the clustering target pair and clustering target points, calculating an unmatched target change rate according to the points of the single clustering target and the points of the differential point cloud, and weighting and summing all the matching target change rates and the unmatched target change rates to obtain a clustering target change rate;

And carrying out global matching on the clustering target set corresponding to the first differential point cloud and the clustering target set of the second differential point cloud by adopting a Hungary algorithm. The distance between two clustering targets is calculated by adopting an intersection ratio (the intersection ratio between the directed bounding boxes of the two clustering targets of each pair to be matched). A distance threshold value Thr is preset, and if the cross ratio distance is larger than the distance threshold value Thr, the matching of the pair to be matched is considered to be failed. After the matching algorithm is completed, a plurality of successfully matched clustering target matching pairs and single successfully unmatched clustering targets are obtained.

And for successfully matched clustering target matching pairs, calculating the change rate of the matching targets according to the directed bounding box information of the successfully matched clustering target matching pairs and the point number of the clustering targets.

And for the single-clustering targets which are not successfully matched, calculating the change rate of the non-matched targets according to the points of the differential point cloud to which the single-clustering targets belong and the points of the clustering targets.

The clustering target change rate comprises two parts, namely a matching target change rate and an unmatched target change rate.

Assume that the number of clustering results of the first differential point cloud is U, and the clustering results are recorded as a clustering set R ₁ { , /> ,…, />V clustering results of the second differential point cloud are recorded as a clustering set R ₂ {/> , /> ,…, /> }。

Then, assuming U < V, matching of two clustering target sets is performed in the above manner, and S clustering target matching pairs are total, and then the unmatched single clustering targets are total U+V-2S. Rate of change of clustered targetsThe calculation formula of (2) is as follows:

；

wherein, For the matching target change rate of the ith clustering target matching pair,/>For the weight matching the sum of the target rates of change,/>Rate of change of unmatched targets for the ith single clustered target,/>Is the weight of the sum of the unmatched target change rates. The weights can be set by actual conditions.

S403, carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

For the datum point cloud and the current latest point cloud, the comprehensive change rate of the datum point cloud and the current latest point cloudCan be calculated by the following formula:

；

wherein, For the differential point cloud change rate of the two,/>Is the differential point cloud change rate/>Weights of (2); is the rate of change of clustering targets of the two/( Is the cluster target change rate/>Is a weight of (2). The weights can be set by actual conditions.

The change rate of the differential point clouds mainly reflects the change of global information among the point clouds, and the change rate of the clustering targets mainly reflects the change among individual targets among the point clouds, so that the change of local information among the point clouds is reflected.

As an optional implementation manner, fig. 3 is a flowchart of a differential point cloud change rate calculating method provided by an embodiment of the present invention, where, as shown in fig. 3, the differential point cloud change rate is calculated according to the number of points of the first differential point cloud and the number of points of the second differential point cloud, including:

S4011, counting first points of a first differential point cloud and second points of a second differential point cloud;

And counting the total points of the first differential point cloud, namely the first points. And counting the total points of the second differential point cloud, namely the second points.

S4012, calculating absolute change rates of two differential point clouds according to the first points and the second points;

And calculating the absolute change rate according to the absolute value of the difference between the first point and the second point and the sum of the first point and the second point.

The absolute change rate describes the change condition between two differential point clouds through the difference in the point number, and in general, the more the point number of the differential point clouds is, the more the foreground information is contained. The more points of the two differential point clouds differ, the greater the variation between the point clouds.

S4013, defining a first statistical region in a first differential point cloud according to a preset statistical range, performing equidistant slicing on the first statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the third points in each point cloud slice;

a statistical range is preset, wherein the minimum value in the Y direction is minY, the maximum value in the Y direction is maxY, the minimum value in the X direction is minX, and the maximum value in the X direction is maxX. Only the points in the first differential point cloud where the x value is in the range of [ minX, maxX ] and the y value is in the range of [ minY, maxY ], i.e., the first statistical region, are counted.

The statistical width of the first statistical region is maxY-minY, and the statistical length is maxX-minX. Slicing the differential point cloud according to the distance dx and the distance dy in the X direction and the Y direction of the first differential point cloud, and dividing the differential point cloud into equal intervals dy in the Y directionIn the X direction, the partition is divided into/>In total, k point cloud slices can be obtained.

Wherein,And counting the number of points of each point cloud slice to obtain the third number of points of each slice in the first differential point cloud, wherein the number of points is { n ₁,n₂,…,n_k }.

S4014, defining a second statistical region in a second differential point cloud according to a preset statistical range, performing equidistant slicing on the second statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the fourth point number in each point cloud slice;

and obtaining the fourth point number of each slice in the second differential point cloud by adopting the same slice and the same statistical mode as the first differential point cloud.

S4015, calculating the distribution change rate of two differential point clouds according to the first point, the second point, the third point and the fourth point;

And calculating the distribution change rate between the first differential point cloud and the second differential point cloud by adopting the first point, the second point, the third point and the fourth point. The distribution change rate can make up the defect that the absolute change rate cannot describe the spatial distribution change of the point cloud, for example, the points of two differential point clouds are identical, but the actual position distribution of the points is different, and the larger change rate can be calculated through the distribution change rate.

S4016, carrying out weighted summation on the absolute change rate and the distribution change rate to obtain the change rate of the differential point cloud.

For the datum point cloud and the current latest point cloud, calculating the differential point cloud change rate of the datum point cloud and the current latest point cloud by the following formula:

；

wherein, Is the absolute rate of change of two differential point clouds,/>Is the distribution change rate of two differential point clouds,Is absolute rate of change/>Weights of/>Is the distribution change rate/>Is a weight of (2). The weights can be set by actual conditions.

As an alternative implementation manner, the change rate of the matched targets is calculated according to the directed bounding box information of two clustered targets in the clustered target matching pair and the clustered target points, and is calculated by the following formula:

；

The formula is calculated based on the fact that the ith clustering target of the first differential point cloud is successfully matched with the jth clustering target of the second differential point cloud.

As an alternative implementation manner, calculating the change rate of the unmatched targets according to the points of the single clustering targets and the points of the differential point cloud comprises the following steps:

；

The above formula is based on the calculation completed in the case that the j-th clustering target of the i-th differential point cloud is not successfully matched.

As an alternative embodiment, calculating the absolute rate of change of two differential point clouds from the first point number and the second point number is calculated by:

；

As an alternative embodiment, the distribution change rate of the two differential point clouds is calculated according to the first point, the second point, the third point and the fourth point by the following formula:

；

The embodiment of the invention also provides a point cloud labeling data optimizing system, and fig. 4 is a schematic structural diagram of the point cloud labeling data optimizing system provided by the embodiment of the invention, as shown in fig. 4, the system includes:

a data acquisition module 100, configured to acquire an original point cloud set; wherein the original point cloud set comprises a continuous multi-frame original point cloud;

The preprocessing module 200 is configured to perform background difference on original point clouds in the original point cloud set, filter the original point clouds to obtain a filtered differential point cloud set, and delete original point clouds corresponding to the filtered point clouds from the original point cloud set to obtain a latest point cloud set;

After differencing, the obtained point cloud set {Filtering. Firstly counting the points of each point cloud in the set, comparing the points corresponding to each point cloud with a preset point threshold delta _pt, and if the points of the current point cloud are smaller than the preset point threshold delta _pt, considering that the foreground points of the current point cloud are fewer and have no marked meaning, and deleting the foreground points from the differential point cloud set. Each point cloud in the point cloud set performs the filtering operation to obtain a filtered differential point cloud set {/>(Where)/(. The original point cloud corresponding to the deleted point cloud is deleted from the original point cloud set while the differential point cloud set is filtered, so that a filtered original point cloud set {/>I.e. the most recent point cloud set.

The clustering module 300 is configured to cluster each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud;

The optimization module 400 is configured to calculate, according to the number of points of the differential point clouds and the number of clustering targets, information amounts of corresponding latest point clouds in the latest point cloud set, arrange the information amounts in descending order, use the latest point clouds corresponding to the first information amount as reference point clouds, and respectively calculate, according to the point difference and the clustering target difference, a comprehensive change rate of the latest point clouds corresponding to the subsequent information amount and the reference point clouds in order, and mark the latest point clouds corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step of the optimization module until labeling is completed by the last datum point cloud, and taking all the latest point clouds not labeled as labeling point clouds.

；

wherein, For the information content of the latest point cloud,/>For the point number of the differential point cloud corresponding to the latest point cloud,/>The point number of the point cloud is the point number of the point cloud corresponding to the latest point cloud. /(I)For the number of clustering targets corresponding to the latest point cloud,/>And the weight of the number of the clustering targets corresponding to the latest point cloud.

As an alternative embodiment, the preferred module 400 includes:

The differential point cloud change rate calculation submodule 4001 is used for determining a first differential point cloud corresponding to the reference point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating the differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud;

The clustering target change rate calculation submodule 4002 is used for performing hungarian matching on a clustering target set of the first differential point cloud and a clustering target set of the second differential point cloud to obtain a single clustering target with successfully matched clustering target matching pair and an unmatched single clustering target, calculating the matching target change rate according to directed bounding box information of the two clustering targets in the clustering target matching pair and the clustering target points, calculating the unmatched target change rate according to the points of the single clustering target and the points of the differential point cloud, and performing weighted summation on all the matching target change rates and the unmatched target change rates to obtain the clustering target change rate;

；

The comprehensive change rate calculation submodule 4003 is used for carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

；

As an alternative embodiment, the differential point cloud change rate calculation submodule 4001 includes:

a statistics unit 40011, configured to count a first point number of the first differential point cloud and a second point number of the second differential point cloud;

An absolute change rate calculation unit 40012, configured to calculate an absolute change rate of two differential point clouds according to the first point number and the second point number;

The first slice statistics unit 40013 is configured to define a first statistics area in a first differential point cloud according to a preset statistics range, perform equidistant slicing on the first statistics area in an X direction and a Y direction, obtain a plurality of point cloud slices, and count a third point number in each point cloud slice;

And counting the number of points of each point cloud slice to obtain the third number of points of each slice in the first differential point cloud, wherein the number of points is { n ₁,n₂,…,n_k }.

A second slice statistics unit 40014, configured to define a second statistical region in a second differential point cloud according to a preset statistical range, perform equidistant slicing in an X direction and a Y direction on the second statistical region, obtain a plurality of point cloud slices, and count a fourth point number in each point cloud slice;

A distribution change rate calculation unit 40015, configured to calculate a distribution change rate of two differential point clouds according to the first point, the second point, the third point, and the fourth point;

The differential point cloud change rate calculation unit 40016 is configured to obtain a differential point cloud change rate by performing weighted summation on the absolute change rate and the distribution change rate.

；

The foregoing description of the embodiments of the present invention further provides a detailed description of the objects, technical solutions and advantages of the present invention, and it should be understood that the foregoing description is only illustrative of the embodiments of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements, etc. that fall within the spirit and principles of the present invention should be included in the scope of the present invention.

Claims

1. The point cloud labeling data optimization method is characterized by comprising the following steps of:

S4, calculating information quantity of corresponding latest point clouds in the latest point cloud set according to the number of points of differential point clouds and the number of clustering targets, arranging the information quantity in a descending order, taking the latest point clouds corresponding to the first information quantity as reference point clouds, respectively calculating comprehensive change rates of the latest point clouds corresponding to the subsequent information quantity and the reference point clouds according to the point difference and the clustering target difference in sequence, and marking the latest point clouds corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step S4 until the last datum point cloud is labeled, and taking all the latest point clouds not labeled as labeled point clouds; according to the point difference and the clustering target difference, respectively calculating the comprehensive change rate of the latest point cloud and the reference point cloud corresponding to the subsequent information according to the sequence, wherein the comprehensive change rate comprises the following steps:

s401, determining a first differential point cloud corresponding to the datum point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating a differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud;

S402, performing Hungary matching on the clustering target set of the first differential point cloud and the clustering target set of the second differential point cloud to obtain a single clustering target with successful matching and a single clustering target with unsuccessful matching, calculating the change rate of the matching target according to the directed bounding box information of the two clustering targets in the clustering target matching pair and the clustering target points, calculating the change rate of the non-matching target according to the points of the single clustering target and the points of the differential point cloud, and weighting and summing the change rates of all the matching targets and the change rates of the non-matching targets to obtain the change rate of the clustering target;

And S403, carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

2. The method according to claim 1, wherein calculating the differential point cloud change rate according to the number of points of the first differential point cloud and the number of points of the second differential point cloud comprises:

S4011, counting first points of the first differential point cloud and second points of the second differential point cloud;

S4013, defining a first statistical region in the first differential point cloud according to a preset statistical range, performing equidistant slicing on the first statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the third points in each point cloud slice;

s4014, a second statistical region is defined in the second differential point cloud according to a preset statistical range, equidistant slicing is conducted on the second statistical region in the X direction and the Y direction, a plurality of point cloud slices are obtained, and fourth points in each point cloud slice are counted;

3. The point cloud labeling data optimization method according to claim 1, wherein the calculating the change rate of the matching target according to the directed bounding box information of two clustering targets in the clustering target matching pair and the clustering target point number is calculated by the following formula:

Wherein b ₁ is the weight of the height change rate of the directed bounding box of the clustered object, Height value of directed bounding box of ith clustered object of first differential point cloud,/>A height value of a directed bounding box of a jth clustering target of the second differential point cloud; b ₂ is the weight of the length change rate of the directed bounding box of the clustering target,/>Length value of directed bounding box of ith clustering target of first differential point cloud,/>A length value of a directed bounding box of a jth clustering target of the second differential point cloud; b ₃ is the weight of the width change rate of the directed bounding box of the clustering target, W _i ¹ is the width value of the directed bounding box of the ith clustering result of the first differential point cloud, and W _j ² is the width value of the directed bounding box of the jth clustering result of the second differential point cloud; b ₄ is the weight of the rotation angle change rate of the directed bounding box of the clustered object,/>Rotation angle of directional bounding box of ith clustering result of first differential point cloud,/>The rotation angle of the directed bounding box of the j-th clustering result of the second differential point cloud; b ₅ is the weight of the relative offset of the center point of the directed bounding box of the clustered object,/>Y_i ¹，/>X, y and z values of the central points of the directional bounding boxes of the ith clustering result of the first differential point cloud respectively; /(I)Y_j ²，/>X, y and z values of the central points of the directed bounding boxes of the jth clustering result of the second differential point cloud respectively; b ₆ is the weight of the point change rate of the clustering result,/>Points of the ith clustering target of the first differential point cloud; /(I)And the number of the jth clustering target of the second differential point cloud.

4. The point cloud labeling data optimization method according to claim 1, wherein the calculation of the unmatched target change rate according to the number of points of the single clustering target and the number of points of the differential point cloud is calculated by the following formula:

Wherein, C _j is the point number of the single clustering target which is not successfully matched, and N _i is the point number of the differential point cloud to which the single clustering target belongs.

5. The point cloud labeling data optimization method of claim 2, wherein the calculating the absolute change rate of two differential point clouds according to the first point number and the second point number is calculated by the following formula:

Wherein ACR is an absolute rate of change, N _A is a first point number, and N _B is a second point number.

6. The method for optimizing point cloud labeling data according to claim 2, wherein the calculating the distribution change rate of the two differential point clouds according to the first point number, the second point number, the third point number and the fourth point number is calculated by:

wherein, SCR is the distribution change rate, A third point number that is an i-th slice of the first differential point cloud; /(I)Fourth points of the ith slice of the second differential point cloud; n _A is the first point and N _B is the second point.

7. A point cloud annotation data optimization system, comprising:

The data acquisition module is used for acquiring an original point cloud set; wherein the original point cloud set comprises a continuous multi-frame original point cloud;

the preprocessing module is used for carrying out background difference on original point clouds in the original point cloud set, filtering to obtain a filtered differential point cloud set, deleting original point clouds corresponding to the filtered point clouds from the original point cloud set, and obtaining a latest point cloud set;

the clustering module is used for clustering each differential point cloud in the differential point cloud set to obtain a clustering target set of each differential point cloud;

The optimization module is used for calculating the information quantity of the corresponding latest point cloud in the latest point cloud set according to the number of points of the differential point cloud and the number of clustering targets, arranging the information quantity in a descending order, taking the latest point cloud corresponding to the first information quantity as a datum point cloud, respectively calculating the comprehensive change rate of the latest point cloud corresponding to the subsequent information quantity and the datum point cloud according to the point difference and the clustering target difference in sequence, and marking the latest point cloud corresponding to the comprehensive change rate smaller than a preset change rate threshold; taking the latest point cloud corresponding to the next unlabeled information quantity as a datum point cloud, respectively calculating the comprehensive change rate of each subsequent latest point cloud not labeled and the datum point cloud according to the point difference and the clustering target difference and the information quantity arrangement sequence, labeling the latest point cloud corresponding to the comprehensive change rate smaller than the preset change rate threshold, repeating the step of the optimization module until labeling is completed by the last datum point cloud, and taking all the latest point clouds not labeled as labeling point clouds; wherein the preference module comprises:

the differential point cloud change rate calculation submodule is used for determining a first differential point cloud corresponding to the reference point cloud and a second differential point cloud corresponding to the current latest point cloud, and calculating the differential point cloud change rate according to the points of the first differential point cloud and the points of the second differential point cloud;

The clustering target change rate calculation submodule is used for carrying out Hungary matching on the clustering target set of the first differential point cloud and the clustering target set of the second differential point cloud to obtain successfully matched clustering target pairs and unmatched single clustering targets, calculating the matching target change rate according to directed bounding box information of the two clustering targets in the clustering target pairs and clustering target points, calculating unmatched target change rate according to the points of the single clustering targets and the points of the differential point clouds, and carrying out weighted summation on all the matching target change rates and unmatched target change rates to obtain clustering target change rate;

And the comprehensive change rate calculation submodule is used for carrying out weighted summation on the change rate of the differential point cloud and the change rate of the clustering target to obtain the comprehensive change rate.

8. The point cloud annotation data optimization system of claim 7, wherein the differential point cloud change rate calculation sub-module comprises:

the statistics unit is used for counting the first points of the first differential point cloud and the second points of the second differential point cloud;

the absolute change rate calculation unit is used for calculating the absolute change rate of the two differential point clouds according to the first point number and the second point number;

The first slice counting unit is used for defining a first counting area in the first differential point cloud according to a preset counting range, carrying out equidistant slicing on the first counting area in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the third point number in each point cloud slice;

The second slice statistics unit is used for defining a second statistical region in the second differential point cloud according to a preset statistical range, carrying out equidistant slicing on the second statistical region in the X direction and the Y direction to obtain a plurality of point cloud slices, and counting the fourth point number in each point cloud slice;

The distribution change rate calculation unit is used for calculating the distribution change rate of the two differential point clouds according to the first point, the second point, the third point and the fourth point;

And the differential point cloud change rate calculation unit is used for carrying out weighted summation on the absolute change rate and the distribution change rate to obtain the differential point cloud change rate.