CN115661213A

CN115661213A - Data processing method and device for high-precision map point cloud registration model

Info

Publication number: CN115661213A
Application number: CN202211344015.0A
Authority: CN
Inventors: 文永琨; 丁文东; 万国伟; 白宇
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-10-31
Filing date: 2022-10-31
Publication date: 2023-01-31

Abstract

The utility model provides a data processing method and device of a high-precision map point cloud registration model, which relates to the technical field of computers and data processing, in particular to the high-precision map technology, can be applied to automatic driving and intelligent transportation, and comprises the following steps: the method comprises the steps of obtaining target point clouds of a target high-precision map and target registration information corresponding to the target point clouds, carrying out restoration processing on the target registration information according to the target point clouds to obtain restoration registration information, calculating difference values between the restoration registration information and the target registration information, determining the target point clouds as training sets if the difference values are larger than preset difference threshold values, avoiding the defects of low efficiency and low reliability caused by collection of the training sets by relying on manual modes, achieving automation and intelligence of obtaining the training sets, improving the efficiency of obtaining the training sets, avoiding the influence of human factors, and improving the accuracy and reliability of the obtained training sets.

Description

Data processing method and device for high-precision map point cloud registration model

Technical Field

The disclosure relates to the technical field of computers and data processing, in particular to a high-precision map technology, which can be applied to automatic driving and intelligent transportation, and particularly relates to a data processing method and device of a high-precision map point cloud registration model.

Background

The high-precision map is an important element for realizing automatic driving, the high-precision map is usually required to be constructed firstly for realizing automatic driving, and point cloud registration is one of important links for constructing the high-precision map.

In some embodiments, the point cloud registration may be performed based on a neural network model, such as obtaining a training set, and training the neural network model based on the training set to obtain a point cloud registration model for performing the point cloud registration, where the training set is obtained based on a manual method.

However, the training set is obtained manually, which lacks automation and intelligence and may be affected by human factors, resulting in a technical problem of low accuracy.

Disclosure of Invention

The present disclosure provides a data processing method and apparatus for a high precision map point cloud registration model to address at least one of the above technical problems.

According to a first aspect of the present disclosure, a data processing method of a high-precision map point cloud registration model is provided, which includes:

acquiring target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, wherein the target high-precision map is a high-precision map which does not meet the preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model;

repairing the target registration information according to the target point cloud to obtain repaired registration information, and calculating a difference value between the repaired registration information and the target registration information;

and if the difference value is larger than a preset difference threshold value, determining the target point cloud as a training set, wherein the training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

According to a second aspect of the present disclosure, there is provided a data processing apparatus of a high-precision map point cloud registration model, including:

the system comprises a first acquisition unit, a second acquisition unit and a third acquisition unit, wherein the first acquisition unit is used for acquiring target point clouds of a target high-precision map and target registration information corresponding to the target point clouds, the target high-precision map is a high-precision map which does not reach a preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model;

the restoration unit is used for restoring the target registration information according to the target point cloud to obtain restored registration information;

a calculating unit, configured to calculate a difference value between the repair registration information and the target registration information;

and the determining unit is used for determining the target point cloud as a training set if the difference value is greater than a preset difference threshold value, wherein the training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

According to a third aspect of the present disclosure, there is provided an electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the first aspect.

According to a fourth aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method according to the first aspect.

According to a fifth aspect of the present disclosure, there is provided a computer program product comprising: a computer program, stored in a readable storage medium, from which at least one processor of an electronic device can read the computer program, execution of the computer program by the at least one processor causing the electronic device to perform the method of the first aspect.

The invention provides a data processing method and a data processing device for a high-precision map point cloud registration model, wherein the data processing method comprises the following steps: the method comprises the steps of obtaining a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, wherein the target high-precision map is a high-precision map which does not meet preset use requirements, the registration information is determined based on a pre-trained initial point cloud registration model, repairing the target registration information according to the target point cloud to obtain repaired registration information, calculating a difference value between the repaired registration information and the target registration information, and if the difference value is larger than a preset difference threshold value, determining the target point cloud as a training set, wherein the training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram according to a first embodiment of the present disclosure;

FIG. 2 is a schematic diagram according to a second embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a data processing method of a high precision map point cloud registration model according to the present disclosure;

FIG. 4 is a schematic illustration of the flow of an update process according to the present disclosure;

FIG. 5 is a schematic illustration of a third embodiment according to the present disclosure;

FIG. 6 is a schematic diagram of a target point cloud storage according to the present disclosure;

FIG. 7 is a schematic diagram according to a fourth embodiment of the present disclosure;

FIG. 8 is a schematic diagram according to a fifth embodiment of the present disclosure;

FIG. 9 is a schematic illustration according to a sixth embodiment of the present disclosure;

fig. 10 is a block diagram of an electronic device for implementing a data processing method of a high-precision map point cloud registration model according to an embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

To facilitate the reader's understanding of the principles of implementation of the present disclosure, at least some of the technical terms referred to in the present disclosure are now explained as follows:

electronic maps (Electronic maps), which may also be referred to as digital maps, refer to maps that are stored and referred to digitally using computer technology.

The electronic map comprises a common map and a high-precision map, wherein the high-precision map refers to a map relative to the common map. In contrast, high-precision maps provide higher precision, and map information with richer contents, and the high-precision maps mainly serve for automatic driving.

The point cloud refers to a point data set of the product appearance surface obtained by a measuring instrument. Accordingly, in the present embodiment, the point cloud may be understood as a point data set of the road outer surface obtained by the measuring instrument.

Illustratively, the point cloud includes points characterizing the road surface, and corresponding feature data, such as corresponding coordinates.

Point cloud registration refers to a process of obtaining coordinate transformation through calculation and uniformly integrating point clouds at different viewing angles to a specified coordinate system through rigid transformation such as rotation and translation. That is, two point clouds subjected to registration may completely coincide with each other through a position transformation such as rotational translation, that is, point cloud registration may be understood as solving a coordinate position transformation relationship between the two point clouds.

Point cloud registration is one of important links for constructing high-precision maps. For example, the method for constructing the high-precision map may include: and acquiring point clouds when the vehicle runs on the road, carrying out point cloud registration on the point clouds to obtain registration information, and constructing a high-precision map of the road according to the registration information.

In some embodiments, the Point cloud used for constructing the high-precision map may be subjected to Point cloud registration by using a geometric algorithm such as a 4-Points consistency Sets (4 PCS), an Iterative Closest Point (ICP) algorithm, a Discriminative Optimization (DO) algorithm, and the like, so as to obtain registration information.

However, the registration information is determined by adopting a geometric algorithm, and the defects of relatively complex algorithm and relatively low efficiency exist.

With the development of artificial intelligence technology and deep learning technology, in other embodiments, a neural network model may also be used to perform point cloud registration on a point cloud used for constructing a high-precision map to obtain registration information.

For example, a sample training set may be acquired that includes the sample point cloud to train the neural network model based on the sample point cloud such that the neural network model learns the ability to output registration information for the sample point cloud based on the sample point cloud.

However, the acquisition of the sample point cloud is mainly achieved manually, for example, the sample point cloud may be collected by a worker, or the sample point cloud may be configured by the worker. Therefore, the method has the defects that the efficiency of collecting the sample point cloud is low and the scale cannot be realized relatively, so that the technical problem that the training efficiency and the reliability of the point cloud registration model are low is caused.

After the point cloud registration model is obtained through training, in order to enable the point cloud registration model to have higher accuracy and reliability, the point cloud registration model may be updated (also referred to as optimization), and how to automatically acquire a training set (which may include a sample training set) for training a new point cloud registration model, or acquire a training set (which may include a sample training set) for updating the point cloud registration model becomes a problem to be solved urgently.

In order to solve the technical problems, the present disclosure proposes a technical idea after creative work: the method comprises the steps of obtaining a high-precision map which does not meet the preset use requirement, obtaining point clouds of the high-precision map and registration information (determined based on a point cloud registration model which is trained) corresponding to the point clouds of the high-precision map, and determining a training set by combining the obtained point clouds and the registration information, wherein the training set can be used for training a new point cloud registration model, and can also be used for updating the point cloud registration model which is trained.

Based on the technical concept, the present disclosure provides a data processing method and apparatus for a high-precision map point cloud registration model, which relates to the field of computer technology and data processing technology, in particular to a high-precision map technology, and can be applied to automatic driving and intelligent transportation to achieve automation of collecting a training set.

Fig. 1 is a schematic diagram according to a first embodiment of the present disclosure, and as shown in fig. 1, a data processing method of a high-precision map point cloud registration model according to an embodiment of the present disclosure includes:

s101: and acquiring target point cloud of the target high-precision map and target registration information corresponding to the target point cloud. The target high-precision map is a high-precision map which does not meet the preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model.

For example, the execution subject of the embodiment may be a data processing apparatus of the high-precision map point cloud registration model (hereinafter, simply referred to as a data processing apparatus), and the data processing apparatus may be a server, a computer, a terminal device, a processor, a chip, and the like, which are not listed here.

If the data processing apparatus is a server, the data processing apparatus may be a local server, a cloud server, an independent server, or a server cluster, and this embodiment is not limited.

The "target" of the target point cloud in this embodiment is used for distinguishing from other point clouds in the following, and cannot be understood as a limitation on the content of the target point cloud. The target point cloud can be understood as a point cloud used for constructing a high-precision map of the target.

Accordingly, the target high-precision map can be understood as at least part of the high-precision maps, and the at least part of the high-precision maps are high-precision maps which do not meet the preset use requirement.

The "target" in the target registration information is used to distinguish from other registration information in the following text, and the target registration information may be understood as registration information obtained by performing point cloud registration on a target point cloud.

The "initial" of the initial point cloud registration model is used to distinguish from other point cloud registration models in the following text, and the initial point cloud registration model can be understood as a point cloud registration model which is trained before a high-precision map including a target high-precision map is constructed.

Illustratively, in connection with the above analysis, the initial point cloud registration model may be understood as being trained based on a manually collected sample training set including sample point clouds.

The content of the preset use requirement is not limited in this embodiment, and may be determined based on the requirement, history, and experiment.

For example, for a high-precision map construction scene with relatively high precision, the use requirement is preset to be a relatively high requirement; conversely, for a high-precision map construction scene with relatively low precision, the preset use requirement is a relatively low requirement.

This step can be understood as: after a high-precision map is constructed and obtained based on the acquired point cloud, determining the high-precision map which does not meet the preset use requirement from the high-precision map, wherein the high-precision map which does not meet the preset use requirement can be called a target high-precision map; acquiring point cloud of a target high-precision map, wherein the point cloud can be called as target point cloud; and acquiring registration information input by the target point cloud and output by the initial point cloud registration model, wherein the registration information can be called target registration information.

S102: and repairing the target registration information according to the target point cloud to obtain repaired registration information.

The restoration process in this embodiment may be understood as calibrating the target registration information based on the target point cloud, so as to avoid inaccuracy of the training set due to inaccuracy of the target registration information.

S103: and calculating a difference value between the repair registration information and the target registration information.

Illustratively, the target registration information is registration information determined by the initial point cloud registration model, the repair registration information is registration information obtained by performing repair processing on the target registration information, and the difference value calculation can be understood as determining the difference between the repair registration information and the target registration information.

S104: and if the difference value is larger than a preset difference threshold value, determining the target point cloud as a training set. The training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

The method for training the new point cloud registration model is not limited in this embodiment, and the implementation thereof may refer to the principle of training the initial point cloud registration model, which is not described herein again.

Similarly, the preset difference threshold may be determined based on a demand, a history, a test, and the like, and this embodiment is not limited.

For example, the difference value and a preset difference threshold may be determined, if the difference value is greater than the preset difference threshold, it indicates that the difference between the repair registration information and the target registration information is large, and it indicates that the reason for the large difference value is very likely to be caused by low prediction reliability (or prediction accuracy) of the initial point cloud registration model, and it indicates that the target point cloud may be long-tailed data, and accordingly, the target point cloud may be determined as a training set.

On the contrary, if the difference value is less than or equal to the preset difference threshold, it indicates that the difference between the repair registration information and the target registration information is small, and indicates that the reason that the difference value is small may not be caused by low prediction reliability (or prediction accuracy) of the initial point cloud registration model, but is caused by other reasons, for example, external parameters of a sensor for acquiring the target point cloud and bias of graph optimization, and the like, it indicates that the target point cloud may not be long-tail data, and accordingly, the target point cloud may not be determined as the training set.

The long-tail data, which may also be referred to as long-tail distribution data, is a biased distribution, which means that several categories (also called head categories) contain a large number of samples, and most categories (also called tail categories) have only a very small number of samples. In the present embodiment, the long tail data can be understood as data that can be sample data.

For example, if the target point cloud may be long tail data, it can be understood as: the target point cloud is data that can be sample data. Conversely, if the target point cloud may not be long-tail data, it can be understood as: the target point cloud is data that may not be sample data.

Correspondingly, if the target point cloud is data which can be used as sample data, the target point cloud can be determined as a training set, and if the target point cloud is data which cannot be used as sample data, the target point cloud cannot be determined as the training set.

That is, in the present embodiment, a training set may be determined according to the target point cloud and the target registration information. The training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

As can be seen from the above analysis, in this embodiment, the target point cloud and the target registration information may be obtained in an automated manner without relying on a manual manner, so that when the training set is determined based on the target point cloud and the target registration information, automation and intelligence for determining the training set may be implemented, so as to avoid the disadvantages of low efficiency and low reliability caused by relying on a manual manner to collect the training set in the above embodiments.

Therefore, when the initial point cloud registration model is updated by combining the training set determined by the method of the embodiment, the automation of the updating process can be realized, and the efficiency and the accuracy of the updating process can be improved, so that the updated point cloud registration model has higher accuracy and reliability.

Correspondingly, when a new point cloud registration model is trained by combining the training set determined by the method of the embodiment, the automation of the training can be realized, and the efficiency and the reliability of the training can be improved, so that the new point cloud registration model obtained by the training has higher accuracy and reliability.

The embodiment does not limit the way of determining the training set according to the target point cloud and the target registration information. For example, the target point cloud may be determined as a training set according to the target registration information, a part of the target point cloud may also be determined as a training set according to the target registration information, and other point clouds may also be re-acquired according to the target registration information and the target point cloud, so as to determine the acquired other point clouds as a training set, and so on, which are not listed here one by one.

Based on the analysis, the present disclosure provides a data processing method for a high-precision map point cloud registration model, which includes: the method comprises the steps of obtaining a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, wherein the target high-precision map is a high-precision map which does not meet preset use requirements, the registration information is determined based on a pre-trained initial point cloud registration model, repairing the target registration information according to the target point cloud to obtain repaired registration information, calculating a difference value between the repaired registration information and the target registration information, and if the difference value is larger than a preset difference threshold value, determining the target point cloud as a training set, wherein the training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

For the reader to more deeply understand the implementation principle of the present disclosure, the data processing method of the high-precision map point cloud registration model of the present disclosure is now explained in detail with reference to fig. 2 as follows. Fig. 2 is a schematic diagram according to a second embodiment of the present disclosure, and as shown in fig. 2, the data processing method of the high-precision map point cloud registration model includes:

s201: and acquiring an initial point cloud of the road.

It should be understood that, in order to avoid tedious statements, the present embodiment will not be described again with respect to the same technical features of the present embodiment as the above embodiments. For example, the execution body of the present embodiment, and the like.

In some embodiments, to construct a high-precision map of a road, the road may be traveled by a collection vehicle so that an initial point cloud of the road is collected by the collection vehicle. A communication link is established between the acquisition vehicle and the data processing device, and after the acquisition vehicle acquires the initial point cloud, the initial point cloud can be transmitted to the processing device based on the communication link. Correspondingly, the initial device acquires the initial point cloud transmitted by the acquisition vehicle.

Similarly, the "initial" in the initial point cloud is used to distinguish from other point clouds, such as from the target point cloud. The initial point cloud can be understood as the full amount of point cloud of the road.

S202: and inputting the initial point cloud into a pre-trained initial point cloud registration model, and outputting initial registration information.

Similarly, the initial registration information may be understood as registration information corresponding to the initial point cloud, and is used to distinguish from registration information such as target registration information.

In some embodiments, before performing S202, the initial point cloud may be preprocessed, such as filtering, to improve the reliability and effectiveness of the preprocessed initial point cloud.

S203: and constructing a full-scale high-precision map of the road according to the initial registration information and the initial point cloud.

Because the initial point cloud is the full-amount point cloud of the road, a high-precision map corresponding to the road is constructed according to the full-amount point cloud and the registration information (namely the initial registration information) corresponding to the full-amount point cloud, and for convenience of distinguishing, the high-precision map can be called as a full-amount high-precision map which is a high-precision map relatively completely representing the characteristics of the road.

Exemplarily, the process of constructing the full-scale high-precision map at S203 may be understood as a stage of map production. As shown in fig. 3 (fig. 3 is a schematic diagram of a data processing method of a high-precision map point cloud registration model according to the present disclosure), a data processing apparatus obtains an initial point cloud, inputs the initial point cloud to the initial point cloud registration model, outputs initial registration information, and performs map production based on the initial registration information and the initial point cloud to obtain a full-scale high-precision map.

S204: and acquiring a target high-precision map from the full-quantity high-precision map, wherein the target high-precision map is the high-precision map which does not meet the preset use requirement.

For example, as shown in fig. 3, after the full-amount high-precision map is constructed, the processing device may control the full-amount high-precision map to enter a map admission phase, where map admission may be understood as analyzing the full-amount high-precision map to determine whether the full-amount high-precision map can be put into use.

For example, a road may include a plurality of road segments, each road segment corresponding to a partial high-precision map of a full-scale high-precision map. Correspondingly, the full-scale high-precision map can be divided into a plurality of road section high-precision maps (also called regional high-precision maps) based on road sections, and whether the road section high-precision maps meet preset use requirements or not can be judged for each road section high-precision map, and if so, the road section high-precision maps can be put into use through a map admittance stage (as shown in fig. 3); and otherwise, if not, indicating that the high-precision map of the road section does not pass through the stage of map admission, and then the high-precision map of the road section cannot be used.

It should be understood that the division of the road segments in the road may be implemented based on the requirement, history, and experiment, and the present embodiment is not limited thereto.

S205: and acquiring target point cloud of the target high-precision map and target registration information corresponding to the target point cloud.

Illustratively, after the target high-precision map is obtained, point cloud used for building the target high-precision map may be obtained from the initial point cloud, and the obtained point cloud is the target point cloud.

S206: and repairing the target registration information according to the target point cloud to obtain repaired registration information.

For example, if the target high-precision map fails to pass through the stage of map admission, and the target high-precision map is constructed based on the target point cloud and the target registration information, the target high-precision map may not meet the preset use requirement due to the target registration information, so that the target high-precision map does not pass through the stage of map admission.

Accordingly, as shown in fig. 3, the target high-precision map that does not pass through the stage of map admission may enter the stage of repairing, and in the stage of repairing, the target registration information may be repaired based on the target point cloud.

The restoration processing in this embodiment may be understood as calibrating the target registration information based on the target point cloud, so as to avoid a stage in which the target high-precision map does not pass the map admission due to inaccuracy of the target registration information.

In some embodiments, S206 may include the steps of:

the first step is as follows: and acquiring points representing the same object in the target point cloud.

For example, the same object may be a physical point on a road, and accordingly, a point representing the physical point may be obtained from the target point cloud.

The second step is as follows: and repairing the target registration information of the point clouds representing the same object to obtain the corresponding repaired registration information representing the same object.

In combination with the above analysis, after the point of the physical point is obtained, the target registration information of the point of the physical point may be obtained from the target registration information, and the obtained target registration information is subjected to a repair process, so as to obtain repaired registration information (which may be referred to as repaired registration information) of the point representing the same object.

In this embodiment, by acquiring the points representing the same object to repair the target registration information of the points representing the same object, the pertinence and reliability of the repair can be achieved.

In some embodiments, the target point cloud comprises a plurality of frames of point clouds, and the first step may comprise: dividing a multi-frame point cloud into a plurality of groups of point clouds, wherein each group of point clouds comprises at least two frames of point clouds, and acquiring points representing the same object in the at least two frames of point clouds.

Correspondingly, the second step comprises: and carrying out point cloud registration on the points representing the same object to obtain the repair registration information of the points representing the same object.

For example, in the first step, multi-frame point clouds of a target high-precision map at a stage that does not pass through map admission may be divided into multiple groups (pairs), one group includes two frame point clouds, for the two frame point clouds in the group, points representing the same object in the two frame point clouds are sought, and point cloud registration is performed on the points representing the same object, so as to obtain repair registration information of the points representing the same object.

The grouping manner is not limited in this embodiment, for example, the multi-frame point clouds may be randomly distributed and divided into multiple groups, or the multi-frame point clouds may be divided into multiple groups by using a preset grouping policy, and the preset grouping policy may be implemented based on requirements, history records, experiments, and the like.

In this embodiment, the point cloud registration of the points representing the same object may be implemented manually or by using the above geometric algorithm, which is not limited in this embodiment.

In the embodiment, the point cloud registration is performed on the points representing the same object in the group after the grouping to obtain the repair registration information of the points representing the same object, so that the resource for matching the points representing the same object can be reduced, and the efficiency and reliability for determining the repair registration information can be improved.

In other embodiments, the repair registration information of the points representing the same object is registration information of the points representing the same object in two frames of point clouds that are adjacent in time and collected by an odometer, the target point cloud is collected by a collection vehicle, and the collection vehicle includes the odometer.

For example, in combination with the above analysis, the target point cloud is collected from a road for a collection vehicle, and the collection vehicle includes a sensor, such as a vision sensor, a radar sensor, and a odometer, which is not limited in this embodiment.

The vision sensor may be an image acquisition device, such as a camera, and may be used to acquire an image of an environment where the vehicle is traveling on a road. The radar sensor may be a laser radar sensor, and may acquire an environmental point cloud (such as an initial point cloud in this embodiment) of the vehicle driving on the road. The odometer is a device for measuring and collecting the travel of the vehicle on the road.

The point cloud registration is to find out the coordinate position change relationship between two point clouds, and the odometer can measure the travel, which can represent the position coordinate change relationship of the vehicle, for example, the odometer can determine the travel based on two frames of point clouds before and after (i.e., before and after time).

Thus, repair registration information characterizing points of the same object may be determined based on the odometer. If the relative pose between the points representing the same object calculated by the odometer based on the two frames of point clouds before and after can be determined as the repair registration information of the points representing the same object.

Because the odometer can relatively accurately determine the representation, and the relative pose between the points of the same object in the front frame point cloud and the back frame point cloud is determined, in the embodiment, the repairing registration information of the points representing the same object is determined based on the odometer, so that the repairing registration information has higher accuracy and reliability, other resources are not required to be consumed to calculate the repairing registration information, and the convenience and the rapidness for determining the repairing registration information are improved.

In combination with the above analysis, the present disclosure provides at least two ways of performing the repair processing on the target registration information, thereby improving the flexibility and diversity of the repair processing.

In some embodiments, after obtaining the repair registration information, the target high-precision map may be repaired based on the repair registration information and the target point cloud, so as to obtain a repaired high-precision map.

Correspondingly, as shown in fig. 3, it can be determined whether the repaired high-precision map meets the preset use requirement, that is, the repaired high-precision map enters the stage of map admission again, and if so (that is, the preset use requirement is met), it can be determined that the repaired high-precision map passes the stage of map admission, and then the repaired high-precision map can be put into use.

S207: and calculating a difference value between the repair registration information and the target registration information.

Illustratively, the target registration information is registration information determined by the initial point cloud registration model, the repair registration information is registration information obtained by performing repair processing on the target registration information, and the difference value is calculated to be understood as determining the difference between the repair registration information and the target registration information.

S208: and if the difference value is larger than a preset difference threshold value, determining the target point cloud as a training set. The training set is used for updating the initial point cloud registration model, or the training set is used for training a new point cloud registration model.

On the contrary, if the difference value is less than or equal to the preset difference threshold, it indicates that the difference between the repair registration information and the target registration information is small, and indicates that the reason that the difference value is small may not be caused by low prediction reliability (or prediction accuracy) of the initial point cloud registration model, but is caused by other reasons, such as external reference of a sensor for acquiring the target point cloud and graph optimization bias, it indicates that the target point cloud may not be long-tail data, and accordingly, the target point cloud may not be determined as the training set.

In other embodiments, if the difference value is less than or equal to the preset difference threshold, the process may return to S201, or the repairing operation may be performed again, or the process is ended, and the like, which is not limited in this embodiment.

Therefore, in the embodiment, by determining the repair registration information so as to determine the target point cloud as the training set when the difference between the repair registration information and the target registration is large, automatic collection of the training set can be realized, and the training set has high pertinence and effectiveness.

The above-mentioned S207-S208 can be understood as a stage of data mining, as shown in fig. 3, to obtain a training set.

By combining the analysis, the data processing device can determine the repaired high-precision map based on the repair registration information, and can determine whether the repaired high-precision map meets the preset use requirement, and if so, can control the repaired high-precision map to enter the stage of putting into use.

It is worth explaining that if the repaired high-precision map meets the preset use requirement, the repaired registration information has high accuracy and reliability, and the target registration information has relatively low accuracy and reliability, so that the accuracy and reliability of the initial point cloud registration model can be determined to be relatively low, and therefore, the target point cloud can be determined as a training set.

For example, in other embodiments, S207-S208 may be replaced with: and generating a repaired high-precision map according to the repair registration information and the target point cloud, and determining the target point cloud as a training set if the repaired high-precision map meets the preset use requirement.

In this embodiment, the restored high-precision map is generated by combining the restoration registration information to determine whether the restored high-precision map meets the preset use requirement, and if so, it indicates that the target high-precision map possibly caused by the initial point cloud registration model does not meet the preset use requirement.

In combination with the above analysis, in this embodiment, different methods may be used to determine the training set, so as to improve the diversity and flexibility of determining the training set.

S209: and updating the initial point cloud registration model according to the target point cloud determined as the training set.

Based on the above analysis, it can be known that, in general, the updated training set of the initial point cloud registration model is collected manually, and the training set collected manually is interfered by human factors, which may have a disadvantage of low reliability.

In the embodiment, the training set (i.e. the target point cloud) is collected in an automatic manner, so that the training set has higher reliability and reality, and thus, when the initial point cloud registration model is updated based on the training set, the effectiveness and reliability of the updating process can be improved.

In combination with the above analysis, it can be known that the target point cloud can be determined as the training set under the condition that the accuracy of the initial point cloud registration model is relatively low, and the point cloud registration model after the update processing has higher reliability by performing the update processing on the initial point cloud registration model with relatively low accuracy.

In some embodiments, S209 may include the steps of:

the first step is as follows: and respectively determining target training data and target test data from the target point cloud.

For example, the target point cloud is a training set obtained through a data mining stage, and the first step may be understood as a data partitioning stage shown in fig. 3, so as to partition the target point cloud into two parts of data, where one part of data is target training data and one part of data is target test data.

The target training data may be understood as data used to complete a training procedure during the update process. The target test data may be understood as data for completing the test flow at the time of update processing.

The present embodiment does not limit the amount of the target training data and the target test data, and the target training data and the target test data may be determined based on a preset ratio, for example.

Similarly, the preset ratio may be determined based on a demand, a history, a test, and the like, and this embodiment is not limited.

In some embodiments, before the data processing device performs the first step, the target point cloud may be preprocessed, such as by filtering, to improve the reliability and effectiveness of the preprocessed target point cloud.

The second step: and updating the initial point cloud registration model according to the target training data, the target test data and the obtained initial training data set for training the initial point cloud registration model.

In this embodiment, the target training data and the target test data are data in the acquired training set, relatively speaking, the initial training data set may be understood as an original training data set, and the training set may be understood as a newly added training data set.

In some embodiments, the initial training data set comprises: initial training data for training the initial point cloud registration model, initial testing data for testing the initial point cloud registration model, the second step may comprise the sub-steps of:

the first substep: and combining the target training data and the initial training data to obtain a new training data set.

For example, the target training data and the initial training data are both used for training, and therefore, the target training data and the initial training data may be combined to obtain a new training data set, so that the new training data includes richer data.

The second substep: and updating the initial point cloud registration model according to the new training data set, the target test data and the initial test data.

In this embodiment, the data in the new training data set is relatively rich, and the effectiveness of the update processing can be improved by performing the update processing in combination with the relatively rich training data set.

In some embodiments, the second substep may comprise the following refinement steps:

a first thinning step: and training the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model.

The embodiment does not limit the training process of obtaining the target point cloud registration model through training. For example, a new training data set is input to the initial point cloud registration model, the predicted registration information is output, a loss function of the predicted registration information and preset calibrated registration information is calculated, and parameters of the initial point cloud registration model are adjusted based on the loss function until the iteration number reaches a number threshold or the loss function reaches a preset loss requirement, so that the target point cloud registration model is obtained.

A second refining step: and respectively adopting the target test data and the initial test data to test the target point cloud registration model to obtain respective corresponding target test results.

For example, the target point cloud registration model is tested by using the target test data to obtain a corresponding target test result, and for convenience of distinguishing, the target test result may be referred to as a first target test result.

And testing the target point cloud registration model by adopting the initial test data to obtain a corresponding target test result, wherein the target test result can be called as a second target test result for convenience of distinguishing.

A third refining step: and updating the initial point cloud registration model according to the corresponding target test result and the obtained initial test result. And the initial test result is obtained by testing the initial point cloud registration model based on the initial test data.

For example, after the initial point cloud registration model is obtained based on the initial training data, the initial point cloud registration model may be tested based on the initial test data, and the obtained test result may be referred to as an initial test result.

In this embodiment, by combining the first target test result, the second target test result, and the initial test result to perform the update processing, the reliability of the initial point cloud registration model is considered, the reliability of the target point cloud registration model in the target test data dimension is also considered, the reliability of the target point cloud registration model in the initial test data dimension is also considered, the update processing is performed by combining the reliabilities of a plurality of different dimensions, and the comprehensiveness and effectiveness of the update processing can be improved.

In some embodiments, the test results are used to characterize registration confidence; the third refinement step may include:

and if the registration confidence coefficient of the target test result representation of the target test data is greater than that of the target test result representation of the initial test data, and the registration confidence coefficient of the target test result representation of the initial test data is greater than that of the initial test result representation, replacing the initial point cloud registration model with the target point cloud registration model.

Wherein the registration confidence characterizes a degree of accuracy and/or reliability of the point cloud registration. In contrast, the greater the registration confidence, the greater the degree of accuracy and/or reliability of the point cloud registration.

For example, in combination with the above analysis, the first target test result may be characterized, and the target point cloud registration model is tested based on the target test data, so as to obtain a registration confidence of the target point cloud registration model, which may be referred to as a first registration confidence for convenience of distinguishing and describing. I.e., the first target test result may characterize the first registration confidence.

The second target test result may be characterized, the target point cloud registration model is tested based on the initial test data, and the registration confidence of the target point cloud registration model is obtained, and for convenience of distinguishing and describing, the registration confidence may be referred to as a second registration confidence. I.e. the second target test result may characterize the first registration confidence.

The initial test result can be characterized, the initial point cloud registration model is tested based on the initial test data, the registration confidence of the initial point cloud registration model is obtained, and the registration confidence can be called as a third registration confidence for convenience of distinguishing and describing. I.e., the initial test results may characterize the third registration confidence.

Correspondingly, if the first registration confidence coefficient is larger than the second registration confidence coefficient and is larger than or equal to the third registration confidence coefficient, the initial point cloud registration model is replaced by the target point cloud registration model, so that the registration information of the acquired point cloud is determined based on the target point cloud registration model.

In this embodiment, if the first registration confidence is greater than the second registration confidence and is not less than the third registration confidence, it indicates that, with respect to the initial point cloud registration model, the accuracy of determining the registration information of the point cloud based on the target point cloud registration model is relatively high, and therefore, the initial point cloud registration model may be replaced by the target point cloud registration model to update and optimize the initial point cloud registration model, so that the effectiveness and reliability of the update processing are improved.

In some embodiments, the target training data comprises first training data and first validation data; the initial training data includes second training data and second validation data.

Accordingly, the new training data set includes: a training data set obtained by combining the first training data and the second training data, and a verification data set obtained by combining the first verification data and the second verification data.

For example, the process of building the model may include three stages, namely a training stage, a verification stage, and a testing stage.

The training phase may be understood as a phase in which a model is trained. The verification stage may be understood as a stage of verifying the model obtained through the training stage, so as to optimize the trained model based on the verification result. The testing stage may be understood as a stage of testing the model obtained in the verification stage, that is, testing the optimized model to determine the prediction capability of the optimized model, and the prediction capability may be characterized based on the registration confidence.

Correspondingly, in this embodiment, the initial point cloud registration model may be trained based on the training data set to obtain an intermediate point cloud registration model, and the intermediate point cloud registration model may be verified based on the verification data set to optimize the intermediate point cloud registration model to obtain the target point cloud registration model.

Based on the above, the target point cloud registration model may be respectively tested based on the target test data and the initial test data, so as to obtain the prediction capabilities of the respective corresponding target point cloud registration models, such as the first registration confidence and the second registration confidence in the above embodiments.

In this embodiment, the new training data set includes a training data set and a verification data set, and the training phase can be divided into two sub-phases of training and verification, so that the update processing is performed from more dimensions, and the reliability and effectiveness of the update processing are improved.

For the reader to more deeply understand the implementation principle of the update process of the present disclosure, the flow of the update process of the present disclosure will now be described in detail with reference to fig. 4 as follows. Fig. 4 is a schematic diagram of the flow of the update process according to the present disclosure.

The target point cloud can be divided based on a preset proportion to obtain first training data, first verification data and target test data.

The initial training data set includes: second training data, second verification data, initial test data.

And combining the first training data and the second training data to obtain a training data set. And merging the first verification data and the second verification data to obtain a verification data set.

And training the initial point cloud registration model according to the training data set to obtain an intermediate point cloud registration model. And verifying and optimizing the intermediate point cloud registration model according to the verification data set to obtain a target point cloud registration model.

And testing the prediction performance (such as the registration confidence coefficient) of the target point cloud registration model according to the target test data to obtain a first target test result, wherein the registration confidence coefficient represented by the first target test result can be called as a first registration confidence coefficient.

And testing the prediction performance (such as the registration confidence coefficient) of the target point cloud registration model according to the initial test data to obtain a second target test result, wherein the registration confidence coefficient represented by the second target test result can be called as a second registration confidence coefficient.

And testing the prediction performance (such as the registration confidence coefficient) of the initial point cloud registration model according to the initial test data to obtain an initial test result, wherein the registration confidence coefficient represented by the initial test result can be called a third registration confidence coefficient.

And if the first registration confidence coefficient is greater than the second registration confidence coefficient and is greater than or equal to the third registration confidence coefficient, replacing the initial point cloud registration model with the target point cloud registration model.

In some embodiments, if the initial point cloud registration model is replaced with the target point cloud registration model, the initial test data and the target test data may be merged into a new test data set for use in subsequent tests.

It should be understood that the above examples are only for illustrative purposes, and the data processing method of the high-precision map point cloud registration model of the present disclosure may be implemented by any possible embodiments, and is not to be construed as limiting the embodiments.

For example, some technical features in the above embodiments may be combined to obtain a new embodiment, or new technical features may be added to the above embodiments to obtain a new embodiment, and for example, an execution subject for executing the method of the above embodiments may be a plurality of apparatuses, and the like.

By taking an example that some technical features in the above embodiments may be combined to obtain a new embodiment, S205-S208 may be combined to obtain a new embodiment, S205-S209 may be combined to obtain a new embodiment, and so on, which are not listed here.

Taking the example of adding a new technical feature to obtain a new embodiment in the above embodiment, a technical feature of training to obtain an initial point cloud registration model may be added to the above embodiment to obtain a new embodiment, a technical feature of training to obtain a new point cloud registration model based on a training set may also be added to the above embodiment, and the like, which are not listed here.

For example, the execution subject for determining the technical features of the training set may be one device, the execution subject for executing the technical features of the update process may be another device, and the execution subject for storing the training set may be yet another device.

For the convenience of the reader to understand, the implementation principle of the data processing method of the high-precision map point cloud registration model of the present disclosure is described in detail below with reference to fig. 5. Fig. 5 is a schematic diagram according to a third embodiment of the present disclosure, and as shown in fig. 5, the data processing method of the high-precision map point cloud registration model according to the embodiment of the present disclosure includes:

s501: the method comprises the steps that a local server obtains target point clouds of a target high-precision map and target registration information corresponding to the target point clouds, wherein the target high-precision map is a high-precision map which does not meet the preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model.

Similarly, in order to avoid tedious statements, the technical features of the present embodiment that are the same as those of the above embodiments are not repeated.

In this embodiment, a local server may be used to obtain the target point cloud and the target registration information.

S502: and the local server performs restoration processing on the target registration information according to the target point cloud to obtain restoration registration information.

For an exemplary implementation principle of this step, reference may be made to the second embodiment, which is not described herein again.

S503: the local server calculates a difference value between the repair registration information and the target registration information.

Similarly, for the implementation principle of this step, reference may be made to the second embodiment, which is not described herein again.

S504: and if the difference value is larger than a preset difference threshold value, the local server stores the target point cloud in the cloud server.

In an exemplary combination with the above analysis, if the difference value is greater than the preset difference threshold, the target point cloud may be determined as a training set, so as to train a new point cloud registration model based on the target point cloud as the training set, or update the initial point cloud registration model based on the target point cloud as the training set.

In this embodiment, the target point cloud determined as the training set may be stored in the cloud server, so as to facilitate migration and storage of the target point cloud and reduce storage space consumption of the local server.

S505: the local server generates an index text file for storing the target point cloud in the cloud server, and stores the index text file in the local server.

For example, in order to facilitate the local server to quickly obtain the target point cloud from the cloud server, the local server may generate and store an index text file.

The index text file comprises an index text, and address information of target point cloud stored in the cloud server is recorded in the index text.

Illustratively, as shown in fig. 6, an index text file is stored in the local server, and a target point cloud is stored in the cloud server. The local server transmits the target point cloud to the cloud server based on the communication link so as to store the target point cloud in the cloud server. The local server may also obtain the target point cloud from the cloud server based on the communication link.

Therefore, the local server stores the target point cloud in the cloud server and generates and stores the index text file, on one hand, migration and storage of the target point cloud can be facilitated, storage space consumption of the local server is reduced, and on the other hand, the local server can be used for conveniently and quickly calling the target point cloud.

S506: and the local server acquires the target point cloud from the cloud server according to the index text file and updates the initial point cloud registration model according to the target point cloud.

For example, if the local server has a need for updating the initial point cloud registration model, the target point cloud may be obtained from the cloud server by indexing the text file, so as to perform the updating process based on the obtained target point cloud, and the implementation principle may be referred to in the above embodiments.

Similarly, in some embodiments, the local server further stores an index text file of the initial training data set, and accordingly, if the local server has a need for updating the initial point cloud registration model, the target point cloud may be obtained from the cloud server through the index text file of the target point cloud, and the initial training data set is obtained from the cloud server according to the index text file of the initial training data set, so as to perform updating based on the obtained target point cloud and the initial training data set.

In this embodiment, the target point cloud can be quickly obtained from the cloud server by indexing the text file, so that flexibility and reliability of data calling are realized.

In other embodiments, the executing entity in S501-S505 may obtain the target point cloud and the target registration information for a cloud server, for example, the cloud server until the target point cloud is determined as a training set, and store the target point cloud that is the training set in the cloud server, and the cloud server may generate an index text file storing the target point cloud, and transmit the index text file to a local server.

Accordingly, the local server receives the index text file transmitted by the cloud server to perform S506.

Similarly, it should be understood that the first embodiment, the second embodiment and the third embodiment may be independent embodiments as described above, or may be combined with each other to obtain a new embodiment, and the present embodiment is not limited.

Fig. 7 is a schematic diagram of a data processing apparatus 700 of a high-precision map point cloud registration model according to a fourth embodiment of the present disclosure, as shown in fig. 7, including:

the first obtaining unit 701 is configured to obtain a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, where the target high-precision map is a high-precision map that does not meet a preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model.

And the repairing unit 702 is configured to perform repairing processing on the target registration information according to the target point cloud to obtain repaired registration information.

A calculating unit 703, configured to calculate a difference value between the repair registration information and the target registration information.

A determining unit 704, configured to determine the target point cloud as a training set if the difference value is greater than a preset difference threshold, where the training set is used to update the initial point cloud registration model, or the training set is used to train a new point cloud registration model.

Fig. 8 is a schematic diagram of a fifth embodiment according to the present disclosure, and as shown in fig. 8, a data processing apparatus 800 of a high-precision map point cloud registration model of the present disclosure includes:

the first obtaining unit 801 is configured to obtain a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, where the target high-precision map is a high-precision map that does not meet a preset use requirement, and the registration information is determined based on a pre-trained initial point cloud registration model.

And a repairing unit 802, configured to perform repairing processing on the target registration information according to the target point cloud to obtain repaired registration information.

In some embodiments, repair unit 802, includes:

the obtaining subunit 8021 is configured to obtain points representing the same object in the target point cloud.

The repairing subunit 8022 is configured to perform repairing processing on the target registration information of the points representing the same object, so as to obtain repairing registration information of the points representing the same object.

In some embodiments, the target point cloud includes a plurality of frame point clouds; the obtaining subunit 8021 is configured to divide the multiple frames of point clouds into multiple groups of point clouds, where each group of point clouds includes at least two frames of point clouds, and obtain points representing the same object in the at least two frames of point clouds.

And the repairing subunit 8022 is configured to perform point cloud registration on the points representing the same object to obtain repairing registration information of the points representing the same object.

In some embodiments, the repair registration information characterizing the points of the same object is registration information characterizing the points of the same object in two frames of point clouds collected by the odometer and adjacent in time.

The target point cloud is acquired by an acquisition vehicle, and the acquisition vehicle comprises a speedometer.

A calculating unit 803, configured to calculate a difference value between the repair registration information and the target registration information.

The determining unit 804 is configured to determine the target point cloud as a training set if the difference value is greater than a preset difference threshold.

The first storage unit 805 is configured to store the target point cloud in the cloud server if the difference value is greater than a preset difference threshold.

Wherein the apparatus is applied to the local server.

The generating unit 806 is configured to generate an index text file storing the target point cloud in the cloud server.

A second storage unit 807 for storing the index text file in the local server.

The second obtaining unit 808 is configured to obtain the target point cloud from the cloud server according to the index text file.

And the updating unit 809 is configured to update the initial point cloud registration model according to the target point cloud determined as the training set.

In some embodiments, as can be seen in fig. 8, the updating unit 809 includes:

the determining subunit 8091 is configured to determine target training data and target test data from the target point cloud, respectively.

And the updating subunit 8092 is configured to update the initial point cloud registration model according to the target training data, the target test data, and the acquired initial training data set used for training the initial point cloud registration model.

In some embodiments, the initial training data set comprises: initial training data used for training the initial point cloud registration model and initial testing data used for testing the initial point cloud registration model; the update subunit 8092, including:

and the merging module is used for merging the target training data and the initial training data to obtain a new training data set.

And the updating module is used for updating the initial point cloud registration model according to the new training data set, the target test data and the initial test data.

In some embodiments, the update module comprises:

and the training submodule is used for training the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model.

And the test sub-module is used for testing the target point cloud registration model by respectively adopting the target test data and the initial test data to obtain respective corresponding target test results.

And the updating submodule is used for updating the initial point cloud registration model according to the corresponding target test result and the obtained initial test result, wherein the initial test result is a test result obtained by testing the initial point cloud registration model based on the initial test data.

In some embodiments, the test result is used for characterizing the registration confidence, and the test result includes a target test result and an initial test result which respectively correspond to the test result; and the updating sub-module is used for replacing the initial point cloud registration model with the target point cloud registration model if the registration confidence coefficient of the target test result representation of the target test data is greater than the registration confidence coefficient of the target test result representation of the initial test data and the registration confidence coefficient of the target test result representation of the initial test data is greater than or equal to the registration confidence coefficient of the target test result representation of the initial test data.

The new training data set includes: the training data set is obtained by combining the first training data and the second training data, and the verification data set is obtained by combining the first verification data and the second verification data.

According to another aspect of the present disclosure, the present disclosure also provides a data processing system of a high-precision map point cloud registration model, including a local server and a cloud server, wherein,

the local server is used for determining the training set and transmitting the training set to the cloud server.

And the cloud server is used for storing the training set.

In some embodiments, the local server is further configured to generate and store an index text file of the training set stored in the cloud server.

Correspondingly, the local server is further used for acquiring a training set from the cloud server based on the index text file and updating the initial point cloud registration model based on the training set.

In other embodiments, the cloud server is configured to determine the training set, generate an index text file that stores the training set at the cloud server, and transmit the index text file to the local server.

The local server is used for acquiring and storing the index text file, acquiring a training set from the cloud server according to the index text file, and updating the initial point cloud registration model according to the training set.

Fig. 9 is a schematic diagram according to a sixth embodiment of the present disclosure, and as shown in fig. 9, an electronic device 900 in the present disclosure may include: a processor 901 and a memory 902.

A memory 902 for storing programs; the Memory 902 may include a volatile Memory (RAM), such as a Static Random Access Memory (SRAM), a Double Data Rate Synchronous Dynamic Random Access Memory (DDR SDRAM), and the like; the memory may also include a non-volatile memory, such as a flash memory. The memory 902 is used to store computer programs (e.g., applications, functional modules, etc. that implement the above-described methods), computer instructions, etc., which may be stored in one or more of the memories 902 in a partitioned manner. And the above-described computer programs, computer instructions, data, and the like can be called by the processor 901.

The computer programs, computer instructions, etc. described above may be stored in one or more memories 902 in partitions. And the above-mentioned computer program, computer instruction, etc. can be called by the processor 901.

A processor 901 for executing the computer program stored in the memory 902 to implement the steps of the method according to the above embodiments.

Reference may be made in particular to the description relating to the preceding method embodiment.

The processor 901 and the memory 902 may be separate structures or may be an integrated structure integrated together. When the processor 901 and the memory 902 are separate structures, the memory 902 and the processor 901 may be coupled by a bus 903.

The electronic device of this embodiment may execute the technical solution in the method, and the specific implementation process and technical principle are the same, which are not described herein again.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

According to an embodiment of the present disclosure, the present disclosure also provides a computer program product comprising: a computer program, stored in a readable storage medium, from which at least one processor of the electronic device can read the computer program, the at least one processor executing the computer program causing the electronic device to perform the solution provided by any of the embodiments described above.

FIG. 10 illustrates a schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 10, the apparatus 1000 includes a computing unit 1001 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a Random Access Memory (RAM) 1003. In the RAM 1003, various programs and data necessary for the operation of the device 1000 can also be stored. The calculation unit 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. An input/output (I/O) interface 1005 is also connected to bus 1004.

A number of components in device 1000 are connected to I/O interface 1005, including: an input unit 1006 such as a keyboard, a mouse, and the like; an output unit 1007 such as various types of displays, speakers, and the like; a storage unit 1008 such as a magnetic disk, an optical disk, or the like; and a communication unit 1009 such as a network card, a modem, a wireless communication transceiver, or the like. The communication unit 1009 allows the device 1000 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.

Computing unit 1001 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 1001 executes the respective methods and processes described above, such as the data processing method of the high-precision map point cloud registration model. For example, in some embodiments, the data processing method of the high precision map point cloud registration model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of a computer program may be loaded onto and/or installed onto device 1000 via ROM 1002 and/or communications unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the data processing method of the high precision map point cloud registration model described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured by any other suitable means (e.g. by means of firmware) to perform the data processing method of the high precision map point cloud registration model.

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The Server can be a cloud Server, also called a cloud computing Server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server may also be a server of a distributed system, or a server incorporating a blockchain.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and the present disclosure is not limited herein.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A data processing method of a high-precision map point cloud registration model comprises the following steps:

acquiring target point clouds of a target high-precision map and target registration information corresponding to the target point clouds, wherein the target high-precision map is a high-precision map which does not meet preset use requirements, and the registration information is determined based on a pre-trained initial point cloud registration model;

2. The method of claim 1, wherein the repairing the target registration information according to the target point cloud to obtain repaired registration information comprises:

acquiring points representing the same object in the target point cloud;

and repairing the target registration information of the points representing the same object to obtain the repaired registration information of the points representing the same object.

3. The method of claim 2, wherein the target point cloud comprises a plurality of frames of point clouds; the acquiring of the points representing the same object in the target point cloud comprises:

dividing the multi-frame point cloud into a plurality of groups of point clouds, wherein each group of point clouds comprises at least two frames of point clouds, and acquiring points representing the same object in the at least two frames of point clouds;

and the repairing the target registration information of the points representing the same object to obtain the repaired registration information of the points representing the same object comprises: and carrying out point cloud registration on the points representing the same object to obtain the repair registration information of the points representing the same object.

4. The method according to claim 3, wherein the repair registration information characterizing the points of the same object is registration information characterizing the points of the same object in two frames of point clouds temporally adjacent to each other and collected based on an odometer;

the target point cloud is collected by a collection vehicle, which includes the odometer.

5. The method of any of claims 1-4, further comprising:

and updating the initial point cloud registration model according to the target point cloud determined as the training set.

6. The method of claim 5, wherein the updating the initial point cloud registration model according to the target point cloud determined as the training set comprises:

respectively determining target training data and target test data from the target point cloud;

and updating the initial point cloud registration model according to the target training data, the target test data and the acquired initial training data set for training the initial point cloud registration model.

7. The method of claim 6, wherein the initial training data set comprises: initial training data used for training the initial point cloud registration model, and initial testing data used for testing the initial point cloud registration model; the updating process of the initial point cloud registration model according to the target training data, the target test data and the acquired initial training data set for training the initial point cloud registration model comprises the following steps:

merging the target training data and the initial training data to obtain a new training data set;

and updating the initial point cloud registration model according to the new training data set, the target test data and the initial test data.

8. The method of claim 7, wherein the updating the initial point cloud registration model from the new training dataset, the target test data, and the initial test data comprises:

training the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model;

respectively adopting the target test data and the initial test data to test the target point cloud registration model to obtain respective corresponding target test results;

and updating the initial point cloud registration model according to the corresponding target test result and the obtained initial test result, wherein the initial test result is obtained by testing the initial point cloud registration model based on the initial test data.

9. The method of claim 8, wherein test results are used to characterize registration confidence, the test results including the respective corresponding target test result, the initial test result; the updating the initial point cloud registration model according to the respective corresponding target test result and the obtained initial test result comprises:

and if the registration confidence coefficient of the target test result representation of the target test data is greater than that of the target test result representation of the initial test data, and the registration confidence coefficient of the target test result representation of the initial test data is greater than or equal to that of the initial test result representation, replacing the initial point cloud registration model with the target point cloud registration model.

10. The method of any of claims 7-9, wherein the target training data includes first training data and first validation data; the initial training data comprises second training data and second validation data;

the new training data set comprises: a training data set obtained by combining the first training data and the second training data, and a verification data set obtained by combining the first verification data and the second verification data.

11. The method according to any one of claims 5-8, applied to a local server; after the calculating a difference value between the repair registration information and the target registration information, the method further comprises:

if the difference value is larger than a preset difference threshold value, storing the target point cloud in a cloud server;

and generating an index text file for storing the target point cloud in the cloud server, and storing the index text file in the local server.

12. The method of claim 11, prior to the updating the initial point cloud registration model according to the target point clouds determined to be the training set, further comprising:

and acquiring the target point cloud from the cloud server according to the index text file.

13. A data processing device of a high-precision map point cloud registration model comprises:

the system comprises a first acquisition unit, a second acquisition unit and a third acquisition unit, wherein the first acquisition unit is used for acquiring target point clouds of a target high-precision map and target registration information corresponding to the target point clouds, the target high-precision map is a high-precision map which does not meet preset use requirements, and the registration information is determined based on a pre-trained initial point cloud registration model;

14. The apparatus of claim 13, wherein the repair subunit comprises:

the acquisition subunit is used for acquiring points representing the same object in the target point cloud;

and the repairing subunit is used for repairing the target registration information of the points representing the same object to obtain the repaired registration information of the points representing the same object.

15. The apparatus of claim 14, wherein the target point cloud comprises a plurality of frame point clouds; the acquisition subunit is configured to divide the multi-frame point cloud into multiple groups of point clouds, where each group of point clouds includes at least two frames of point clouds, and acquire points representing the same object from the at least two frames of point clouds;

and the repairing subunit is used for performing point cloud registration on the points representing the same object to obtain repairing registration information of the points representing the same object.

16. The apparatus according to claim 14, wherein the repair registration information characterizing the points of the same object is registration information characterizing the points of the same object in two frames of point clouds temporally adjacent to each other and collected based on an odometer;

17. The apparatus of any one of claims 13-16, the apparatus further comprising:

and the updating unit is used for updating the initial point cloud registration model according to the target point cloud determined as the training set.

18. The apparatus of claim 17, wherein the update unit comprises:

the determining subunit is used for respectively determining target training data and target test data from the target point cloud;

and the updating subunit is used for updating the initial point cloud registration model according to the target training data, the target test data and the acquired initial training data set used for training the initial point cloud registration model.

19. The apparatus of claim 18, wherein the initial training data set comprises: initial training data used for training the initial point cloud registration model, and initial testing data used for testing the initial point cloud registration model; the update subunit includes:

a merging module, configured to merge the target training data and the initial training data to obtain a new training data set;

20. The apparatus of claim 19, wherein the update module comprises:

the training submodule is used for training the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model;

the test sub-module is used for testing the target point cloud registration model by respectively adopting the target test data and the initial test data to obtain respective corresponding target test results;

21. The apparatus of claim 20, wherein test results are used to characterize registration confidence, the test results including the respective corresponding target test result, the initial test result; the updating sub-module is used for replacing the initial point cloud registration model with the target point cloud registration model if the registration confidence degree of the target test result representation of the target test data is larger than the registration confidence degree of the target test result representation of the initial test data and the registration confidence degree of the target test result representation of the initial test data is larger than or equal to the registration confidence degree of the target test result representation of the initial test data.

22. The apparatus of any of claims 19-21, wherein the target training data comprises first training data and first validation data; the initial training data comprises second training data and second validation data;

23. The apparatus according to any of claims 17-20, applied to a local server; the device further comprises:

the first storage unit is used for storing the target point cloud in a cloud server if the difference value is larger than a preset difference threshold value;

the generating unit is used for generating an index text file for storing the target point cloud in the cloud server;

and the second storage unit is used for storing the index text file in the local server.

24. The apparatus of claim 23, the apparatus further comprising:

and the second acquisition unit is used for acquiring the target point cloud from the cloud server according to the index text file.

25. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-12.

26. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-12.

27. A computer program product comprising a computer program which, when executed by a processor, carries out the steps of the method of any one of claims 1 to 12.