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

Abstract

The present disclosure provides a data processing method and device for a high-precision map point cloud registration model, which relates to the fields of computer technology and data processing technology, and in particular to high-precision map technology. The method and device can be applied to autonomous driving and intelligent transportation, including: obtaining a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud, repairing the target registration information according to the target point cloud to obtain repaired registration information, and calculating the difference value between the repaired registration information and the target registration information. If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, avoiding the disadvantages of low efficiency and low reliability caused by relying on manual collection of training sets, realizing the automation and intelligence of obtaining training sets, improving the efficiency of obtaining training sets, avoiding the influence of human factors, and improving the accuracy and reliability of the obtained training sets.

Description

Technical Field

[0001] The present disclosure relates to the fields of computer technology and data processing technology, and in particular to high-precision map technology, which can be applied to autonomous driving and intelligent transportation, and more particularly to a data processing method and device for a high-precision map point cloud registration model. Background Art

[0002] High-precision maps are an important element in realizing autonomous driving. In order to achieve autonomous driving, it is usually necessary to build high-precision maps first, and point cloud registration is one of the important links in building high-precision maps.

[0003] In some embodiments, point cloud registration can 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 point cloud registration, wherein the training set is obtained manually.

[0004] However, obtaining training sets manually lacks automation and intelligence, and may be affected by human factors, resulting in technical problems such as low accuracy. Summary of the Invention

[0005] The present disclosure provides a data processing method and device for a high-precision map point cloud registration model for solving at least one of the above-mentioned technical problems.

[0006] According to a first aspect of the present disclosure, a data processing method for a high-precision map point cloud registration model is provided, comprising:

[0007] 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model;

[0008] Performing a repair process on 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;

[0009] If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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.

[0010] According to a second aspect of the present disclosure, a data processing device for a high-precision map point cloud registration model is provided, comprising:

[0011] a first acquisition unit, configured to acquire 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model;

[0012] a repairing unit, configured to perform repair processing on the target registration information according to the target point cloud to obtain repaired registration information;

[0013] a calculation unit, configured to calculate a difference value between the repaired registration information and the target registration information;

[0014] A determination unit is configured to determine the target point cloud as a training set if the difference value is greater than a preset difference threshold, wherein 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.

[0015] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect.

[0019] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method according to the first aspect.

[0020] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising: a computer program, wherein the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program so that the electronic device executes the method described in the first aspect.

[0021] The present disclosure provides a data processing method and device for a high-precision map point cloud registration model, including: 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 that does not meet preset usage requirements, the registration information is determined based on a pre-trained initial point cloud registration model, the target registration information is repaired according to the target point cloud to obtain repaired registration information, and a difference value between the repaired registration information and the target registration information is calculated. If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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. The technical feature of determining the training set based on the target point cloud and the target registration information by obtaining the target point cloud and the target registration information of the high-precision map that does not meet the preset usage requirements avoids the disadvantages of low efficiency and low reliability caused by relying on manual methods to collect training sets, realizes automation and intelligence in obtaining training sets, improves the efficiency of obtaining training sets, avoids the influence of human factors, and improves the accuracy and reliability of the obtained training sets.

[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0024] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0025] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;

[0026] Figure 3 Schematic diagram of the principle of the data processing method of the high-precision map point cloud registration model disclosed in the present invention;

[0027] Figure 4 is a schematic diagram of the principle of the update processing process according to the present disclosure;

[0028] Figure 5 is a schematic diagram according to a third embodiment of the present disclosure;

[0029] Figure 6 is a schematic diagram of the principle of target point cloud storage according to the present disclosure;

[0030] Figure 7 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0031] Figure 8is a schematic diagram according to a fifth embodiment of the present disclosure;

[0032] Figure 9 is a schematic diagram according to a sixth embodiment of the present disclosure;

[0033] Figure 10 It is a block diagram of an electronic device used to implement the data processing method of the high-precision map point cloud registration model of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0035] To facilitate readers' understanding of the implementation principles of the present disclosure, at least some of the technical terms involved in the present disclosure are explained as follows:

[0036] An electronic map, also known as a digital map, refers to a map that is stored and accessed digitally using computer technology.

[0037] Electronic maps include standard maps and high-precision maps. High-precision maps are higher-precision and richer in map information than standard maps. They primarily serve autonomous driving.

[0038] A point cloud refers to a collection of point data on the surface of a product obtained by a measuring instrument. Accordingly, in this embodiment, a point cloud can be understood as a collection of point data on the outer surface of a road obtained by a measuring instrument.

[0039] Exemplarily, the point cloud includes points used to represent the outer surface of the road, and feature data corresponding to the points, such as coordinates corresponding to the points.

[0040] Point cloud registration is the process of integrating point clouds from different perspectives into a specified coordinate system through rigid transformations such as rotation and translation, calculated by calculating coordinate transformations. In other words, the two point clouds being registered must be completely aligned with each other through these positional transformations. In other words, point cloud registration can be understood as determining the coordinate position transformation relationship between the two point clouds.

[0041] Point cloud registration is a crucial step in building high-precision maps. For example, a method for building a high-precision map may include acquiring a point cloud of a vehicle traveling on a road, performing point cloud registration on the point cloud to obtain registration information, and then building a high-precision map of the road based on the registration information.

[0042] In some embodiments, geometric algorithms such as the 4-Points Congruent Sets (4PCS), the Iterative Closest Point (ICP) algorithm, and the Discriminative Optimization (DO) algorithm can be used to perform point cloud registration on the point cloud used to construct the high-precision map to obtain registration information.

[0043] However, the use of geometric algorithms to determine registration information has the disadvantages of relatively complex algorithms and relatively low efficiency.

[0044] With the development of artificial intelligence technology and deep learning technology, in other embodiments, a neural network model can also be used to perform point cloud registration on the point cloud used to construct a high-precision map to obtain registration information.

[0045] Exemplarily, a sample training set may be collected, the sample training set including a sample point cloud, so as to train a neural network model based on the sample point cloud, so that the neural network model learns the ability to output registration information of the sample point cloud based on the sample point cloud.

[0046] However, the collection of sample point clouds is primarily done manually, for example, by staff either collecting or configuring them. This approach suffers from low efficiency in collecting sample point clouds and a relatively limited ability to scale, leading to technical issues such as low training efficiency and reliability of point cloud registration models.

[0047] After the point cloud registration model is trained, in order to make the point cloud registration model have higher accuracy and reliability, the point cloud registration model can be updated (also called optimized). However, how to automatically obtain a training set (which may include a sample training set) for training a new point cloud registration model, or obtain a training set (which may include a sample training set) for updating the point cloud registration model, has become an urgent problem to be solved.

[0048] In order to solve the above technical problems, the present disclosure proposes a technical concept after creative work: obtaining a high-precision map that does not meet the preset usage requirements, and obtaining the point cloud of the high-precision map, and the registration information corresponding to the point cloud of the high-precision map (determined based on a trained point cloud registration model), so as to determine a training set in combination with the obtained point cloud and registration information, wherein the training set can be used to train a new point cloud registration model, and can also update the already trained point cloud registration model.

[0049] Based on the above technical concept, the present disclosure provides a data processing method and device for a high-precision map point cloud registration model, which relates to the fields of computer technology and data processing technology, specifically to high-precision map technology, and can be applied to autonomous driving and intelligent transportation to achieve the automation of collecting training sets.

[0050] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure, as shown in Figure 1 As shown, the data processing method of the high-precision map point cloud registration model of the embodiment of the present disclosure includes:

[0051] S101: Obtain a target point cloud of a target high-precision map and target registration information corresponding to the target point cloud. The target high-precision map is a high-precision map that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model.

[0052] Illustratively, the executing entity of this embodiment can be a data processing device of a high-precision map point cloud registration model (hereinafter referred to as a data processing device). The data processing device can be a server, a computer, a terminal device, a processor, a chip, etc., which are not listed one by one here.

[0053] If the data processing device is a server, the data processing device may be a local server, a cloud server, an independent server, or a server cluster, which is not limited in this embodiment.

[0054] The term "target" in the target point cloud in this embodiment is used to distinguish it from other point clouds in the following text, and should not be understood as limiting the content of the target point cloud. The target point cloud can be understood as the point cloud used to construct the target high-precision map.

[0055] Correspondingly, the target high-precision map can be understood as at least a portion of the high-precision map in the high-precision map, and the at least portion of the high-precision map is a high-precision map that does not meet the preset usage requirements.

[0056] The “target” in the target registration information is used to distinguish it from other registration information in the following text. The target registration information can be understood as the registration information obtained by performing point cloud registration on the target point cloud.

[0057] The "initial" in the initial point cloud registration model is used to distinguish it from other point cloud registration models in the following text. The initial point cloud registration model can be understood as a point cloud registration model that has been trained before building a high-precision map including the target high-precision map.

[0058] For example, in combination with the above analysis, the initial point cloud registration model can be understood as being obtained by training based on a sample training set including sample point clouds collected manually.

[0059] This embodiment does not limit the content of the preset usage requirements, which can be determined based on requirements, historical records, and experiments.

[0060] For example, for a high-precision map construction scenario with relatively high accuracy, the preset usage requirements are relatively high requirements; conversely, for a high-precision map construction scenario with relatively low accuracy, the preset usage requirements are relatively low requirements.

[0061] This step can be understood as: after constructing a high-precision map based on the acquired point cloud, determining the high-precision maps that do not meet the preset usage requirements from the high-precision maps, and the high-precision maps that do not meet the preset usage requirements can be called target high-precision maps; obtaining the point cloud of the target high-precision map, which can be called the target point cloud; obtaining the registration information output by the initial point cloud registration model input into the target point cloud, and this registration information can be called target registration information.

[0062] S102: Repairing the target registration information according to the target point cloud to obtain repaired registration information.

[0063] The repair process in this embodiment can be understood as calibrating the target registration information based on the target point cloud to avoid inaccuracy of the training set due to inaccurate target registration information.

[0064] S103: Calculate the difference between the restored registration information and the target registration information.

[0065] For example, the target registration information is the registration information determined by the initial point cloud registration model, the repaired registration information is the registration information obtained by repairing the target registration information, and calculating the difference value can be understood as determining the difference between the repaired registration information and the target registration information.

[0066] S104: If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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.

[0067] This embodiment does not limit the method of training a new point cloud registration model. Its implementation can refer to the principle of training an initial point cloud registration model, which will not be repeated here.

[0068] Similarly, the preset difference threshold can be determined based on demand, historical records, and experiments, and this embodiment does not limit this.

[0069] For example, the size of the difference value and the preset difference threshold can be judged. If the difference value is greater than the preset difference threshold, it means that the difference between the repaired registration information and the target registration information is large, which means that the reason for the large difference value is probably that the prediction reliability (or prediction accuracy) of the initial point cloud registration model is not high, which means that the target point cloud may be long-tail data. Accordingly, the target point cloud can be determined as a training set.

[0070] On the contrary, if the difference value is less than or equal to the preset difference threshold, it means that the difference between the repaired registration information and the target registration information is small, which means that the reason for the small difference value may not be the low prediction reliability (or prediction accuracy) of the initial point cloud registration model, but due to other reasons, such as the external parameters of the sensor that collects the target point cloud and the deviation of the image optimization, etc., which means that the target point cloud may not be long-tail data. Accordingly, the target point cloud can be determined as the training set.

[0071] Among them, long-tail data, also known as long-tail distribution data, is a skewed distribution, which means that a few categories (also called head categories) contain a large number of samples, while most categories (also called tail categories) have only a very small number of samples. In this embodiment, long-tail data can be understood as data that can be used as sample data.

[0072] For example, if the target point cloud may be long-tail data, it can be understood that the target point cloud is data that can be used as sample data. Conversely, if the target point cloud may not be long-tail data, it can be understood that the target point cloud is data that cannot be used as sample data.

[0073] Correspondingly, if the target point cloud is data that can be used as sample data, the target point cloud can be determined as a training set; if the target point cloud is data that cannot be used as sample data, the target point cloud may not be determined as a training set.

[0074] That is, in this embodiment, a training set can be determined based on the target point cloud and target registration information, wherein 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.

[0075] Combined with the above analysis, it can be seen that in this embodiment, the target point cloud and target registration information can be obtained in an automated manner without relying on manual implementation. Therefore, when the training set is determined based on the target point cloud and target registration information, the determination of the training set can be automated and intelligent, so as to avoid the disadvantages of low efficiency and low reliability caused by relying on manual methods to collect training sets in the above embodiment.

[0076] Therefore, when the initial point cloud registration model is updated using the training set determined by the method of this embodiment, the update process can be automated, and the efficiency and accuracy of the update process can be improved, so that the updated point cloud registration model has higher accuracy and reliability.

[0077] Accordingly, when a new point cloud registration model is trained using the training set determined by the method of this embodiment, the training can be automated, and the efficiency and reliability of the training can be improved, so that the new point cloud registration model obtained through training has higher accuracy and reliability.

[0078] This embodiment does not limit the method for determining the training set based on the target point cloud and the target registration information. For example, the target point cloud can be determined as the training set based on the target registration information, or a portion of the target point cloud can be determined as the training set based on the target registration information. Alternatively, other point clouds can be re-acquired based on the target registration information and the target point cloud so that the acquired point clouds can be determined as the training set, and so on. The examples are not listed here one by one.

[0079] Based on the above analysis, the present disclosure provides a data processing method for a high-precision map point cloud registration model, including: 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 that does not meet preset usage requirements, the registration information is determined based on a pre-trained initial point cloud registration model, the target registration information is repaired according to the target point cloud to obtain repaired registration information, and a difference value between the repaired registration information and the target registration information is calculated. If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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. In this embodiment, the technical feature of determining a training set based on the target point cloud and the target registration information by obtaining the target point cloud and the target registration information of the high-precision map that does not meet the preset usage requirements avoids the disadvantages of low efficiency and low reliability caused by relying on manual methods to collect training sets, realizes automation and intelligence in obtaining training sets, improves the efficiency of obtaining training sets, avoids the influence of human factors, and improves the accuracy and reliability of the obtained training sets.

[0080] In order to make readers more deeply understand the implementation principle of this disclosure, Figure 2 The data processing method of the high-precision map point cloud registration model disclosed in this disclosure is described in detail as follows. Figure 2 is a schematic diagram according to the second embodiment of the present disclosure, as shown in Figure 2 As shown in the figure, the data processing method of the high-precision map point cloud registration model includes:

[0081] S201: Obtaining the initial point cloud of the road.

[0082] It should be understood that, in order to avoid tedious description, the technical features of this embodiment that are the same as those of the above-mentioned embodiments will not be described in detail in this embodiment, for example, the execution subject of this embodiment.

[0083] In some embodiments, to construct a high-precision map of a road, a data collection vehicle can be driven along the road to collect an initial point cloud of the road. A communication link is established between the data collection vehicle and a data processing device. After collecting the initial point cloud, the data collection vehicle can transmit the initial point cloud to the processing device via the communication link. Accordingly, the processing device receives the initial point cloud transmitted by the data collection vehicle.

[0084] Similarly, the "initial" in the initial point cloud in this embodiment is used to distinguish it from other point clouds, such as from the target point cloud. The initial point cloud can be understood as the full point cloud of the road.

[0085] S202: Input the initial point cloud into a pre-trained initial point cloud registration model and output initial registration information.

[0086] Similarly, the initial registration information can be understood as the registration information corresponding to the initial point cloud, which is used to distinguish from registration information such as the target registration information.

[0087] In some embodiments, before executing S202 , the initial point cloud may be pre-processed, such as filtering, to improve the reliability and effectiveness of the pre-processed initial point cloud.

[0088] S203: Construct a full high-precision map of the road based on the initial registration information and the initial point cloud.

[0089] Since the initial point cloud is the full point cloud of the road, a high-precision map corresponding to the road is constructed by combining the full point cloud and the registration information corresponding to the full point cloud (i.e., the initial registration information). For the sake of distinction, this high-precision map can be called a full high-precision map. The full high-precision map is a high-precision map that relatively completely characterizes the characteristics of the road.

[0090] For example, the process of constructing a full HD map in S203 can be understood as the stage of map production. Figure 3 ( Figure 3 As shown in the schematic diagram of the principle of the data processing method of the high-precision map point cloud registration model disclosed in the present invention, the data processing device obtains the initial point cloud, inputs the initial point cloud into the initial point cloud registration model, outputs the initial registration information, and produces the map based on the initial registration information and the initial point cloud to obtain a full high-precision map.

[0091] S204: Obtain a target high-precision map from the full set of high-precision maps, wherein the target high-precision map is a high-precision map that does not meet preset usage requirements.

[0092] For example, Figure 3 As shown, after the processing device constructs the full high-precision map, it can control the full high-precision map to enter the map access stage. Map access can be understood as analyzing the full high-precision map to determine whether the full high-precision map can be put into use.

[0093] For example, a road may include multiple sections, each of which corresponds to a portion of the full HD map. Accordingly, the full HD map can be divided into multiple section HD maps (also called regional HD maps) based on the sections. For each section HD map, it can be determined whether the section HD map meets the preset usage requirements. If so, it means that the section HD map has passed the map access stage and can be put into use (such as Figure 3 On the contrary, if not, it means that the high-precision map of the road section has not passed the map access stage and cannot be put into use.

[0094] It should be understood that the division of road sections can be achieved based on demand, historical records, and experiments, and this embodiment does not limit this.

[0095] S205: Obtain a target point cloud of the target high-precision map and target registration information corresponding to the target point cloud.

[0096] For example, after obtaining the target high-precision map, the point cloud used to construct the target high-precision map can be obtained from the initial point cloud. The obtained point cloud is the target point cloud. Correspondingly, the registration information corresponding to the target point cloud can be obtained from the initial registration information. The obtained registration information is the target registration information.

[0097] S206: Repairing the target registration information according to the target point cloud to obtain repaired registration information.

[0098] For example, if the target high-precision map fails to pass the map access stage, and the target high-precision map is constructed based on the target point cloud and target registration information, therefore, it may be due to the target registration information that the target high-precision map does not meet the preset usage requirements, causing the target high-precision map to fail to pass the map access stage.

[0099] Correspondingly, such as Figure 3 As shown, the target high-precision map that has not passed the map access stage can enter the repair stage. In the repair stage, the target registration information can be repaired based on the target point cloud.

[0100] The repair process in this embodiment can be understood as calibrating the target registration information based on the target point cloud to avoid the target high-precision map failing to pass the map access stage due to inaccurate target registration information.

[0101] In some embodiments, S206 may include the following steps:

[0102] The first step is to obtain points representing the same object in the target point cloud.

[0103] Exemplarily, the same object may be a physical point on the road, and accordingly, a point representing the physical point may be obtained from the target point cloud.

[0104] The second step is to perform repair processing on the target registration information of the point cloud representing the same object to obtain the repair registration information corresponding to the points representing the same object.

[0105] Combined with the above analysis, after obtaining the point of the physical point, the target registration information of the point of the physical point can be obtained from the target registration information, and the obtained target registration information can be repaired to obtain the repaired registration information (which can be called repaired registration information) representing the point of the same object.

[0106] In this embodiment, by acquiring points representing the same object and repairing the target registration information of the points representing the same object, the pertinence and reliability of the repair can be achieved.

[0107] In some embodiments, the target point cloud includes multiple frames of point clouds, and the first step may include: dividing the multiple frames of point clouds into multiple groups of point clouds, each group of point clouds including at least two frames of point clouds, and obtaining points representing the same object in at least two frames of point clouds.

[0108] Correspondingly, the second step includes: performing point cloud registration on points representing the same object to obtain repair registration information of the points representing the same object.

[0109] For example, in the first step, the multi-frame point cloud of the target high-precision map that has not passed the map access stage can be divided into multiple groups (pairs), one group includes two frames of point clouds, and for the two frames of point clouds in one group, points representing the same object in the two frames of point clouds are sought, and point cloud registration is performed on the points representing the same object to obtain repair registration information of the points representing the same object.

[0110] Among them, this embodiment does not limit the grouping method. For example, multi-frame point clouds can be divided into multiple groups by random allocation, or multi-frame point clouds can be divided into multiple groups by a preset grouping strategy. The preset grouping strategy can be implemented based on needs, historical records, and experiments.

[0111] In this embodiment, the method of performing point cloud registration on points representing the same object can be implemented manually or by using the above-mentioned geometric algorithm, which is not limited in this embodiment.

[0112] In this embodiment, by first grouping and then performing point cloud registration on the points in the group that represent the same object to obtain repair registration information of the points that represent the same object, the resources for matching the points that represent the same object can be reduced, and the efficiency and reliability of determining the repair registration information can be improved.

[0113] In other embodiments, the repaired registration information representing points of the same object is the registration information representing points of the same object in two temporally adjacent frames of point clouds collected based on an odometer, and the target point cloud is collected by a collection vehicle, which includes an odometer.

[0114] For example, in combination with the above analysis, the target point cloud is collected by a collection vehicle driving on the road. The collection vehicle includes sensors such as visual sensors, radar sensors, odometers, etc., which are not limited in this embodiment.

[0115] The visual sensor can be an image acquisition device, such as a camera, that can be used to capture images of the environment as the vehicle travels on the road. The radar sensor can be a lidar sensor that can capture a point cloud of the environment as the vehicle travels on the road (such as the initial point cloud in this embodiment). The odometer is a device used to measure the distance traveled by the vehicle on the road.

[0116] Based on the above analysis, point cloud registration is to find the coordinate position change relationship between two point clouds, while the odometer can measure the distance traveled, which can represent the coordinate change relationship of the vehicle's position. For example, the odometer can determine the distance traveled based on two frames of point clouds (before and after in time).

[0117] Therefore, the odometry can be used to determine the repair and registration information of points representing the same object. For example, the relative poses of points representing the same object obtained by the odometry based on the cloud computing of the previous and next two frames of point cloud can be determined as the repair and registration information of points representing the same object.

[0118] Since the odometry can relatively accurately determine the relative pose between the points of the same object in the previous and next frames of point cloud, in this embodiment, by determining the repair registration information representing the points of the same object based on the odometry, the repair registration information can be repaired with high accuracy and reliability, and there is no need to consume other resources to calculate the repair registration information, thereby improving the convenience and speed of determining the repair registration information.

[0119] Based on the above analysis, it can be seen that the present disclosure provides at least two methods for repairing target registration information, thereby improving the flexibility and diversity of the repair process.

[0120] In some embodiments, after obtaining the repair registration information, the target high-precision map can be repaired based on the repair registration information and the target point cloud to obtain a repaired high-precision map.

[0121] Correspondingly, such as Figure 3 As shown, it can be determined whether the repaired high-precision map meets the preset usage requirements, that is, the repaired high-precision map enters the map access stage again. If it is (that is, it meets the preset usage requirements), it can be determined that the repaired high-precision map passes the map access stage and can be put into use.

[0122] S207: Calculate the difference between the restored registration information and the target registration information.

[0123] For example, the target registration information is the registration information determined by the initial point cloud registration model, the repaired registration information is the registration information obtained by repairing the target registration information, and calculating the difference value can be understood as determining the difference between the repaired registration information and the target registration information.

[0124] S208: If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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.

[0125] This embodiment does not limit the method of training a new point cloud registration model. Its implementation can refer to the principle of training an initial point cloud registration model, which will not be repeated here.

[0126] Similarly, the preset difference threshold can be determined based on demand, historical records, and experiments, and this embodiment does not limit this.

[0127] For example, the size of the difference value and the preset difference threshold can be judged. If the difference value is greater than the preset difference threshold, it means that the difference between the repaired registration information and the target registration information is large, which means that the reason for the large difference value is probably that the prediction reliability (or prediction accuracy) of the initial point cloud registration model is not high, which means that the target point cloud may be long-tail data. Accordingly, the target point cloud can be determined as a training set.

[0128] On the contrary, if the difference value is less than or equal to the preset difference threshold, it means that the difference between the repaired registration information and the target registration information is small, which means that the reason for the small difference value may not be the low prediction reliability (or prediction accuracy) of the initial point cloud registration model, but due to other reasons, such as the external parameters of the sensor that collects the target point cloud and the deviation of the image optimization, etc., which means that the target point cloud may not be long-tail data. Accordingly, the target point cloud can be determined as the training set.

[0129] Among them, long-tail data, also known as long-tail distribution data, is a skewed distribution, which means that a few categories (also called head categories) contain a large number of samples, while most categories (also called tail categories) have only a very small number of samples. In this embodiment, long-tail data can be understood as data that can be used as sample data.

[0130] For example, if the target point cloud may be long-tail data, it can be understood that the target point cloud is data that can be used as sample data. Conversely, if the target point cloud may not be long-tail data, it can be understood that the target point cloud is data that cannot be used as sample data.

[0131] Correspondingly, if the target point cloud is data that can be used as sample data, the target point cloud can be determined as a training set; if the target point cloud is data that cannot be used as sample data, the target point cloud may not be determined as a training set.

[0132] 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 repair operation may be performed again, or the process may end, etc. This embodiment does not limit this.

[0133] Therefore, in this embodiment, by determining the repair registration information and determining the target point cloud as the training set when the difference between the repair registration information and the target registration is large, the automatic collection of the training set can be achieved, and the training set can be made more targeted and effective.

[0134] The above S207-S208 can be understood as the stage of data mining, such as Figure 3 As shown, the training set is obtained by mining.

[0135] Combined with the above analysis, it can be seen that 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 usage requirements. If so, the repaired high-precision map can be controlled to enter the stage of being put into use.

[0136] It is worth noting that if the repaired high-precision map meets the preset usage requirements, it means that the repaired registration information has high accuracy and reliability, while the accuracy and reliability of the target registration information are relatively low. Therefore, it can be determined that the accuracy and reliability of the initial point cloud registration model are relatively low. Therefore, the target point cloud can be determined as the training set.

[0137] For example, in some other embodiments, S207-S208 can be replaced by: generating a repaired high-precision map based on the repaired registration information and the target point cloud; if the repaired high-precision map meets the preset usage requirements, the target point cloud is determined as a training set.

[0138] In this embodiment, a repaired high-precision map is generated by combining the repair registration information to determine whether the repaired high-precision map meets the preset usage requirements. If so, it means that the target high-precision map may not meet the preset usage requirements due to the initial point cloud registration model. Therefore, the target point cloud can be determined as a training set to realize the automatic collection of the training set and improve the pertinence and effectiveness of the training set.

[0139] Based on the above analysis, it can be seen that in this embodiment, different methods can be used to determine the training set to improve the diversity and flexibility of determining the training set.

[0140] S209: updating the initial point cloud registration model according to the target point cloud determined as the training set.

[0141] Based on the above analysis, it can be seen that, usually, the training set for updating the initial point cloud registration model is collected manually, and the training set collected manually is interfered by human factors and may have the disadvantage of low reliability.

[0142] In this embodiment, the training set (i.e., the target point cloud) is collected in an automated manner so that the training set has higher reliability and authenticity, so that when the initial point cloud registration model is updated based on the training set, the effectiveness and reliability of the update process can be improved.

[0143] Combined with the above analysis, it can be seen that when the accuracy of the initial point cloud registration model is low, the target point cloud can be determined as the training set, and by updating the initial point cloud registration model with low accuracy, the updated point cloud registration model can have higher reliability.

[0144] In some embodiments, S209 may include the following steps:

[0145] Step 1: Determine the target training data and target test data from the target point cloud.

[0146] For example, the target point cloud is a training set obtained in the data mining stage. The first step can be understood as follows: Figure 3 The data division stage shown is to divide the target point cloud into two parts of data, one part of the data is target training data, and the other part of the data is target testing data.

[0147] The target training data can be understood as the data used to complete the training process during the update process, and the target test data can be understood as the data used to complete the test process during the update process.

[0148] This embodiment does not limit the amount of target training data and target test data. For example, the target training data and target test data may be determined based on a preset ratio.

[0149] Similarly, the preset ratio can be determined based on demand, historical records, and experiments, and this embodiment does not limit this.

[0150] In some embodiments, before the data processing device executes the first step, the target point cloud may be pre-processed, such as filtering, to improve the reliability and effectiveness of the pre-processed target point cloud.

[0151] The second step: 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.

[0152] In this embodiment, the target training data and the target test data are the data in the acquired training set. Relatively speaking, the initial training data set can be understood as the original training data set, and the training set can be understood as the newly added training data set. By combining the original training data set and the newly added training data set for update processing, the updated data set can be richer, thereby making the update processing comprehensive and highly reliable.

[0153] In some embodiments, the initial training data set includes: initial training data for training the initial point cloud registration model, initial test data for testing the initial point cloud registration model, and the second step may include the following sub-steps:

[0154] The first sub-step: merge the target training data and the initial training data to obtain a new training data set.

[0155] Exemplarily, both the target training data and the initial training data are data used for training. Therefore, the target training data and the initial training data can be merged to obtain a new training data set, so that the new training data includes richer data.

[0156] The second sub-step: update the initial point cloud registration model based on the new training data set, target test data, and initial test data.

[0157] In this embodiment, the data in the new training data set is relatively rich. By performing the update process in combination with the relatively rich training data set, the effectiveness of the update process can be improved.

[0158] In some embodiments, the second sub-step may include the following refinement steps:

[0159] The first refinement step: train the initial point cloud registration model based on the new training dataset to obtain the target point cloud registration model.

[0160] This embodiment does not limit the training process for obtaining the target point cloud registration model. For example, a new training dataset is input into the initial point cloud registration model, predicted registration information is output, a loss function is calculated between the predicted registration information and the preset calibration registration information, and the parameters of the initial point cloud registration model are adjusted based on the loss function until the number of iterations reaches a threshold or the loss function meets a preset loss requirement, thereby obtaining the target point cloud registration model.

[0161] The second refinement step: Use the target test data and the initial test data to test the target point cloud registration model respectively to obtain the corresponding target test results.

[0162] Exemplarily, the target point cloud registration model is tested using target test data to obtain a corresponding target test result. For ease of distinction, the target test result may be referred to as a first target test result.

[0163] The target point cloud registration model is tested using the initial test data to obtain a corresponding target test result. For ease of distinction, the target test result may be referred to as a second target test result.

[0164] The third refinement step is to update the initial point cloud registration model based on the corresponding target test results and the obtained initial test results, wherein the initial test results are test results obtained by testing the initial point cloud registration model based on the initial test data.

[0165] For example, after an initial point cloud registration model is obtained by training 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.

[0166] In this embodiment, by combining the first target test results, the second target test results, and the initial test results for update processing, the reliability of the initial point cloud registration model, the reliability of the target point cloud registration model in the target test data dimension, and the reliability of the target point cloud registration model in the initial test data dimension are taken into consideration. By combining the reliabilities of multiple different dimensions for update processing, the comprehensiveness and effectiveness of the update processing can be improved.

[0167] In some embodiments, the test results are used to characterize the registration confidence; the third refinement step may include:

[0168] If the registration confidence represented by the target test result of the target test data is greater than the registration confidence represented by the target test result of the initial test data, and the registration confidence represented by the target test result of the initial test data is greater than the registration confidence represented by the initial test result, then the initial point cloud registration model is replaced with the target point cloud registration model.

[0169] The registration confidence represents the accuracy and / or reliability of the point cloud registration. Relatively speaking, the greater the registration confidence, the greater the accuracy and / or reliability of the point cloud registration.

[0170] For example, in combination with the above analysis, the first target test result can represent the registration confidence of the target point cloud registration model obtained by testing the target point cloud registration model based on the target test data. For ease of distinction and description, this registration confidence can be referred to as the first registration confidence. In other words, the first target test result can represent the first registration confidence.

[0171] The second target test result can represent the registration confidence of the target point cloud registration model obtained by testing it based on the initial test data. For ease of distinction and description, this registration confidence can be referred to as the second registration confidence. In other words, the second target test result can represent the first registration confidence.

[0172] The initial test results can represent the registration confidence of the initial point cloud registration model obtained by testing it based on the initial test data. For ease of distinction and description, this registration confidence can be referred to as the third registration confidence. In other words, the initial test results can represent the third registration confidence.

[0173] Correspondingly, if the first registration confidence > the second registration confidence ≥ the third registration confidence, the initial point cloud registration model is replaced with the target point cloud registration model, so that the registration information of the acquired point cloud is subsequently determined based on the target point cloud registration model.

[0174] In this embodiment, if the first registration confidence > the second registration confidence ≥ the third registration confidence, it means that relative to the initial point cloud registration model, the accuracy of the point cloud registration information determined based on the target point cloud registration model is relatively high. Therefore, the initial point cloud registration model can be replaced with the target point cloud registration model to achieve the update and optimization of the initial point cloud registration model, thereby improving the effectiveness and reliability of the update process. Moreover, since the target point cloud registration model has relatively higher reliability and accuracy, when the target point cloud registration model is subsequently combined for point cloud registration, the accuracy and reliability of the point cloud registration can be improved.

[0175] In some embodiments, the target training data includes first training data and first verification data; and the initial training data includes second training data and second verification data.

[0176] Accordingly, the new training data set includes: a training data set obtained by merging the first training data and the second training data, and a verification data set obtained by merging the first verification data and the second verification data.

[0177] Exemplarily, the process of building a model may include three stages: a training stage, a validation stage, and a testing stage.

[0178] The training phase can be understood as the phase in which a model is obtained through training. The validation phase can be understood as the phase in which the model obtained through the training phase is verified, with the trained model optimized based on the validation results. The testing phase can be understood as the phase in which the model obtained through the validation phase is tested, i.e., the optimized model is tested to determine its predictive capability, which can be characterized by the registration confidence.

[0179] Accordingly, in this embodiment, the initial point cloud registration model can be trained based on the training data set to obtain an intermediate point cloud registration model, and the intermediate point cloud registration model can be verified based on the verification data set to optimize the intermediate point cloud registration model to obtain the target point cloud registration model.

[0180] Based on this, the target point cloud registration model can be tested based on the target test data and the initial test data respectively to obtain the prediction capabilities of the corresponding target point cloud registration models, such as the first registration confidence and the second registration confidence in the above embodiment.

[0181] In this embodiment, the new training data set includes a training data set and a verification data set, which can divide the training phase into two sub-phases of training and verification, so as to perform update processing from more dimensions and improve the reliability and effectiveness of the update processing.

[0182] In order to make readers more deeply understand the implementation principle of the update process disclosed in this disclosure, Figure 4 The update process of the present disclosure is described in detail below. Figure 4 It is a schematic diagram of the principle of the update processing flow according to the present disclosure.

[0183] The target point cloud may be divided and processed based on a preset ratio to obtain first training data, first verification data, and target test data.

[0184] The initial training data set includes: second training data, second verification data, and initial test data.

[0185] The first training data and the second training data are combined to obtain a training data set. The first validation data and the second validation data are combined to obtain a validation data set.

[0186] The initial point cloud registration model is trained based on the training dataset to obtain the intermediate point cloud registration model. The intermediate point cloud registration model is verified and optimized based on the validation dataset to obtain the target point cloud registration model.

[0187] The prediction performance (such as registration confidence) of the target point cloud registration model is tested according to the target test data to obtain a first target test result. The registration confidence represented by the first target test result can be called the first registration confidence.

[0188] The prediction performance (such as registration confidence) of the target point cloud registration model is tested based on the initial test data to obtain a second target test result. The registration confidence represented by the second target test result can be called the second registration confidence.

[0189] The prediction performance (such as registration confidence) of the initial point cloud registration model is tested based on the initial test data to obtain an initial test result. The registration confidence represented by the initial test result can be called the third registration confidence.

[0190] If the first registration confidence > the second registration confidence ≥ the third registration confidence, the initial point cloud registration model is replaced with the target point cloud registration model.

[0191] 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.

[0192] It should be understood that the above examples are only used to illustrate possible embodiments of the data processing method of the high-precision map point cloud registration model disclosed in the present invention, and should not be understood as limiting the embodiments.

[0193] For example, some technical features in the above embodiments may be combined to obtain new embodiments, or new technical features may be added to the above embodiments to obtain new embodiments. For example, the execution subject of the method for executing the above embodiments may be multiple devices, and so on.

[0194] Taking the example that some technical features in the above embodiments may be combined to form new embodiments, S205-S208 can be combined to form a new embodiment, S205-S209 can be combined to form a new embodiment, and so on, which will not be listed one by one here.

[0195] Taking the example of adding new technical features to the above embodiments to obtain a new embodiment, the technical features of training to obtain an initial point cloud registration model can be added to the above embodiments to obtain a new embodiment. The technical features of training based on a training set to obtain a new point cloud registration model can also be added to the above embodiments to obtain a new embodiment, and so on. They are not listed one by one here.

[0196] Taking the example that the execution entity for executing the method of the above embodiment can be multiple devices, for example, the execution entity for determining the technical features of the training set can be one device, the execution entity for performing the technical features of the update processing can be another device, and the execution entity for storing the training set can be yet another device.

[0197] To facilitate the understanding of readers, the implementation principle of the data processing method of the high-precision map point cloud registration model disclosed in this disclosure is now combined with the following examples under different execution entities. Figure 5 The details are as follows. Figure 5 is a schematic diagram according to the third embodiment of the present disclosure, as shown in Figure 5 As shown, the data processing method of the high-precision map point cloud registration model of the embodiment of the present disclosure includes:

[0198] S501: The local server obtains 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model.

[0199] Similarly, in order to avoid tedious description, the technical features of this embodiment that are the same as those of the above embodiments will not be repeated in this embodiment.

[0200] In this embodiment, a local server may be used to obtain the target point cloud and target registration information.

[0201] S502: The local server repairs the target registration information according to the target point cloud to obtain repaired registration information.

[0202] For example, regarding the implementation principle of this step, please refer to the second embodiment, which will not be described again here.

[0203] S503: The local server calculates the difference between the repaired registration information and the target registration information.

[0204] Similarly, regarding the implementation principle of this step, please refer to the second embodiment and will not be repeated here.

[0205] S504: If the difference value is greater than the preset difference threshold, the local server stores the target point cloud in the cloud server.

[0206] Exemplarily, in combination with the above analysis, if the difference value is greater than the preset difference threshold, the target point cloud can be determined as a training set, so as to train a new point cloud registration model based on the target point cloud of the training set, or update the initial point cloud registration model based on the target point cloud of the training set.

[0207] In this embodiment, the target point cloud determined as the training set can be stored in a cloud server to facilitate the migration and preservation of the target point cloud and reduce the storage space consumption of the local server.

[0208] 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.

[0209] 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.

[0210] The index text file is a file including an index text, and the index text records the address information of the target point cloud stored in the cloud server.

[0211] For example, Figure 6 As shown, the local server stores an index text file, and the cloud server stores the target point cloud. A communication link exists between the local server and the cloud server. The local server transmits the target point cloud to the cloud server via this communication link, so that the target point cloud is stored on the cloud server. The local server can also retrieve the target point cloud from the cloud server via this communication link.

[0212] Therefore, the local server stores the target point cloud on the cloud server and generates and stores the index text file. On the one hand, it can facilitate the migration and preservation of the target point cloud and reduce the storage space consumption of the local server. On the other hand, it can also enable the local server to retrieve the target point cloud conveniently and quickly.

[0213] S506: The local server obtains 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.

[0214] For example, if the local server needs to update the initial point cloud registration model, it can obtain the target point cloud from the cloud server through the index text file, and perform update processing based on the obtained target point cloud. The implementation principle can be found in the above embodiment.

[0215] Similarly, in some embodiments, the local server also stores an index text file of the initial training data set. Accordingly, if the local server needs to update the initial point cloud registration model, the target point cloud can be obtained from the cloud server through the index text file of the target point cloud, and the initial training data set can be obtained from the cloud server according to the index text file of the initial training data set, so as to perform update processing based on the obtained target point cloud and the initial training data set. The implementation principle can be referred to the above embodiment.

[0216] In this embodiment, the target point cloud can be quickly obtained from the cloud server by indexing the text file, so as to achieve flexibility and reliability of data calling.

[0217] In other embodiments, the execution entity of S501-S505 can be a cloud server, such as the cloud server obtains the target point cloud and target alignment information until the target point cloud is determined as a training set, and stores the target point cloud of the training set in the cloud server, and the cloud server can generate an index text file for storing the target point cloud, and transmit the index text file to the local server.

[0218] Correspondingly, the local server receives the index text file transmitted by the cloud server to execute S506.

[0219] 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 new embodiments, which is not limited in this embodiment.

[0220] Figure 7 is a schematic diagram according to a fourth embodiment of the present disclosure, as shown in Figure 7 As shown, the data processing device 700 of the high-precision map point cloud registration model disclosed in the present invention includes:

[0221] The first acquisition unit 701 is used to obtain 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model.

[0222] The repairing unit 702 is configured to perform repair processing on the target registration information according to the target point cloud to obtain repaired registration information.

[0223] The calculation unit 703 is configured to calculate a difference between the restored registration information and the target registration information.

[0224] The determination unit 704 is configured to determine the target point cloud as a training set if the difference value is greater than a preset difference threshold, wherein 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.

[0225] Figure 8 is a schematic diagram according to the fifth embodiment of the present disclosure, as shown in Figure 8 As shown, the data processing device 800 of the high-precision map point cloud registration model disclosed in the present invention includes:

[0226] The first acquisition unit 801 is used to obtain 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model.

[0227] The repair unit 802 is configured to perform repair processing on the target registration information according to the target point cloud to obtain repaired registration information.

[0228] In some embodiments, the repair unit 802 includes:

[0229] The acquisition subunit 8021 is used to acquire points representing the same object in the target point cloud.

[0230] The repair subunit 8022 is used to perform repair processing on the target registration information representing the points of the same object to obtain repaired registration information representing the points of the same object.

[0231] In some embodiments, the target point cloud includes multiple frames of point clouds; the acquisition subunit 8021 is used to divide the multiple frames of point clouds into multiple groups of point clouds, each group of point clouds includes at least two frames of point clouds, and obtain points representing the same object in at least two frames of point clouds.

[0232] Furthermore, the restoration subunit 8022 is used to perform point cloud registration on points representing the same object to obtain restoration registration information of the points representing the same object.

[0233] In some embodiments, the repaired registration information representing points of the same object is registration information representing points of the same object in two temporally adjacent frames of point clouds collected based on the odometer.

[0234] Among them, the target point cloud is collected by the collection vehicle, which includes an odometer.

[0235] The calculation unit 803 is configured to calculate a difference between the restored registration information and the target registration information.

[0236] 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.

[0237] The first storage unit 805 is configured to store the target point cloud in a cloud server if the difference value is greater than a preset difference threshold.

[0238] Wherein, the device is applied to the local server.

[0239] The generating unit 806 is configured to generate an index text file for storing the target point cloud in the cloud server.

[0240] The second storage unit 807 is used to store the index text file in the local server.

[0241] The second acquiring unit 808 is configured to acquire the target point cloud from the cloud server according to the index text file.

[0242] 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.

[0243] In some embodiments, combined Figure 8 It can be seen that the updating unit 809 includes:

[0244] The determination subunit 8091 is used to determine target training data and target test data from the target point cloud.

[0245] The updating subunit 8092 is used 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 for training the initial point cloud registration model.

[0246] In some embodiments, the initial training data set includes: initial training data for training the initial point cloud registration model, and initial test data for testing the initial point cloud registration model; the updating subunit 8092 includes:

[0247] The merging module is used to merge the target training data and the initial training data to obtain a new training data set.

[0248] The updating module is used to update the initial point cloud registration model according to the new training data set, target test data and initial test data.

[0249] In some embodiments, the update module includes:

[0250] The training submodule is used to train the initial point cloud registration model based on the new training data set to obtain the target point cloud registration model.

[0251] The test submodule is used to test the target point cloud registration model using target test data and initial test data respectively, and obtain the corresponding target test results.

[0252] The update submodule is used to update the initial point cloud registration model according to the corresponding target test results and the obtained initial test results, wherein the initial test results are the test results obtained by testing the initial point cloud registration model based on the initial test data.

[0253] In some embodiments, the test results are used to characterize the registration confidence, and the test results include the corresponding target test results and initial test results; the update submodule is used to replace the initial point cloud registration model with the target point cloud registration model if the registration confidence represented by the target test result of the target test data is greater than the registration confidence represented by the target test result of the initial test data, and the registration confidence represented by the target test result of the initial test data is greater than or equal to the registration confidence represented by the initial test result.

[0254] In some embodiments, the target training data includes first training data and first verification data; and the initial training data includes second training data and second verification data.

[0255] The new training data set includes: a training data set obtained by merging the first training data and the second training data, and a verification data set obtained by merging the first verification data and the second verification data.

[0256] According to another aspect of the present disclosure, the present disclosure also provides a data processing system for a high-precision map point cloud registration model, including a local server and a cloud server, wherein:

[0257] The local server is used to determine the training set and transmit the training set to the cloud server.

[0258] The cloud server is used to store the training set.

[0259] In some embodiments, the local server is further used to generate and store an index text file of the training set stored in the cloud server.

[0260] Correspondingly, the local server is also used to obtain a training set from the cloud server based on the index text file, and update the initial point cloud registration model based on the training set.

[0261] In other embodiments, the cloud server is used to determine a training set, generate an index text file for storing the training set on the cloud server, and transmit the index text file to the local server.

[0262] The local server is used to obtain and store the index text file, obtain the training set from the cloud server according to the index text file, and update the initial point cloud registration model according to the training set.

[0263] Figure 9 is a schematic diagram according to a sixth embodiment of the present disclosure, as shown in Figure 9 As shown, the electronic device 900 in the present disclosure may include: a processor 901 and a memory 902 .

[0264] Memory 902 is used to store programs. Memory 902 may include volatile memory (volatile memory), such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), etc. Memory may also include non-volatile memory (non-volatile memory), such as flash memory. Memory 902 is used to store computer programs (such as applications and functional modules that implement the above-mentioned methods), computer instructions, etc. The above-mentioned computer programs and computer instructions may be partitioned and stored in one or more memories 902. Furthermore, the above-mentioned computer programs, computer instructions, data, etc. may be called by processor 901.

[0265] The aforementioned computer programs, computer instructions, etc. may be partitioned and stored in one or more memories 902 . Furthermore, the aforementioned computer programs, computer instructions, etc. may be called by the processor 901 .

[0266] The processor 901 is configured to execute the computer program stored in the memory 902 to implement the various steps in the method involved in the above embodiment.

[0267] For details, please refer to the relevant description in the previous method embodiment.

[0268] The processor 901 and the memory 902 may be independent structures or integrated structures. When the processor 901 and the memory 902 are independent structures, the memory 902 and the processor 901 may be coupled via a bus 903 .

[0269] The electronic device of this embodiment can execute the technical solution in the above method. Its specific implementation process and technical principles are the same and will not be repeated here.

[0270] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0271] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0272] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0273] like Figure 10 As shown, the device 1000 includes a computing unit 1001, which 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. Various programs and data required for the operation of the device 1000 can also be stored in the RAM 1003. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0274] Various components in device 1000 are connected to I / O interface 1005, including an input unit 1006, such as a keyboard, mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a magnetic disk, optical disk, etc.; and a communication unit 1009, such as a network card, modem, wireless communication transceiver, etc. The communication unit 1009 allows device 1000 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0275] The computing unit 1001 can be any general-purpose and / or specialized processing component 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 specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the data processing method for the HD map point cloud registration model. For example, in some embodiments, the data processing method for the HD map point cloud registration model can 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 the computer program can be loaded and / or installed onto the device 1000 via the ROM 1002 and / or the communication 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 for the HD map point cloud registration model described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to execute the data processing method of the high-precision map point cloud registration model in any other appropriate manner (for example, by means of firmware).

[0276] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0277] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0278] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, 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 portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0279] To provide interaction with a user, the systems and techniques described herein 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 the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 can be received in any form (including acoustic input, voice input, or tactile input).

[0280] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, 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 a local area network (LAN), a wide area network (WAN), and the Internet.

[0281] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. This client-server relationship is established by computer programs running on the respective computers, establishing a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosts and VPS services ("Virtual Private Servers" or simply "VPS"). The server may also be a server in a distributed system or a server integrated with blockchain.

[0282] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.

[0283] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A data processing method for a high-precision map point cloud registration model, comprising: 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model; Performing a repair process on 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; If the difference value is greater than a preset difference threshold, the target point cloud is determined as a training set, wherein 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.

2. The method according to claim 1, wherein The repairing process of the target registration information according to the target point cloud to obtain the repaired registration information includes: Obtaining points representing the same object in the target point cloud; The target registration information representing the points of the same object is repaired to obtain the repaired registration information representing the points of the same object.

3. The method according to claim 2, wherein: The target point cloud includes multiple frame point clouds; and obtaining points representing the same object in the target point cloud includes: Dividing the multiple frame point clouds into multiple groups of point clouds, each group of point clouds includes at least two frames of point clouds, and obtaining points representing the same object in the at least two frames of point clouds; And, the repairing processing of the target registration information of the points representing the same object to obtain the repaired registration information of the points representing the same object includes: performing point cloud registration on the points representing the same object to obtain the repaired registration information of the points representing the same object.

4. The method according to claim 3, wherein: The repair registration information of the points representing the same object is the registration information of the points representing the same object in two temporally adjacent frames of point clouds collected based on the odometer; The target point cloud is collected by a collection vehicle, and the collection vehicle includes the odometer.

5. The method according to any one of claims 1 to 4, further comprising: The initial point cloud registration model is updated according to the target point cloud determined as the training set.

6. The method according to claim 5, wherein: The updating process of the initial point cloud registration model according to the target point cloud determined as the training set includes: Determining target training data and target test data from the target point cloud respectively; The initial point cloud registration model is updated 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 according to claim 6, wherein: The initial training data set includes: initial training data for training the initial point cloud registration model, and initial test data for testing the initial point cloud registration model; the updating process of the initial point cloud registration model based on the target training data, the target test data, and the acquired initial training data set for training the initial point cloud registration model includes: Merging the target training data and the initial training data to obtain a new training data set; The initial point cloud registration model is updated according to the new training data set, the target test data, and the initial test data.

8. The method according to claim 7, wherein: The updating process of the initial point cloud registration model according to the new training data set, the target test data, and the initial test data includes: Training the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model; Using the target test data and the initial test data respectively, the target point cloud registration model is tested to obtain respective corresponding target test results; The initial point cloud registration model is updated according to the respective corresponding target test results and the obtained initial test results, wherein the initial test results are test results obtained by testing the initial point cloud registration model based on the initial test data.

9. The method according to claim 8, wherein The test results are used to characterize the registration confidence, and the test results include the corresponding target test results and the initial test results; The updating process of the initial point cloud registration model according to the respective corresponding target test results and the obtained initial test results includes: If the registration confidence represented by the target test result of the target test data is greater than the registration confidence represented by the target test result of the initial test data, and the registration confidence represented by the target test result of the initial test data is greater than or equal to the registration confidence represented by the initial test result, then the initial point cloud registration model is replaced with the target point cloud registration model.

10. The method according to any one of claims 7 to 9, wherein: The target training data includes first training data and first verification data; The initial training data includes second training data and second verification data; The new training data set includes: a training data set obtained by merging the first training data and the second training data, and a verification data set obtained by merging the first verification data and the second verification data.

11. The method according to any one of claims 5 to 8, wherein the method is applied to a local server; After calculating the difference value between the repaired registration information and the target registration information, the method further includes: If the difference value is greater than a preset difference threshold, the target point cloud is stored in a cloud server; An index text file for storing the target point cloud is generated in the cloud server, and the index text file is stored in the local server.

12. The method according to claim 11, before updating the initial point cloud registration model based on the target point cloud determined as the training set, the method further comprises: The target point cloud is obtained from the cloud server according to the index text file.

13. A data processing device for a high-precision map point cloud registration model, comprising: a first acquisition unit, configured to acquire 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 that does not meet preset usage requirements, and the registration information is determined based on a pre-trained initial point cloud registration model; a repairing unit, configured to perform repair processing on the target registration information according to the target point cloud to obtain repaired registration information; a calculation unit, configured to calculate a difference value between the repaired registration information and the target registration information; A determination unit is configured to determine the target point cloud as a training set if the difference value is greater than a preset difference threshold, wherein 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.

14. The device according to claim 13, wherein The repair unit comprises: an acquisition subunit, configured to acquire points representing the same object in the target point cloud; The repair subunit is configured to perform repair processing on 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 device according to claim 14, wherein The target point cloud includes multiple frames of point clouds; the acquisition subunit is used to divide the multiple frames of point clouds into multiple groups of point clouds, each group of point clouds includes at least two frames of point clouds, and acquire points representing the same object in the at least two frames of point clouds; Furthermore, the restoration subunit is used to perform point cloud registration on the points representing the same object to obtain restoration registration information of the points representing the same object.

16. The device according to claim 14, wherein The repair registration information of the points representing the same object is the registration information of the points representing the same object in two temporally adjacent frames of point clouds collected based on the odometer; The target point cloud is collected by a collection vehicle, and the collection vehicle includes the odometer.

17. The apparatus according to any one of claims 13 to 16, further comprising: An updating unit is used to update the initial point cloud registration model according to the target point cloud determined as the training set.

18. The device according to claim 17, wherein The updating unit includes: a determination subunit, configured to determine target training data and target test data from the target point cloud; An updating subunit is used 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 for training the initial point cloud registration model.

19. The device according to claim 18, wherein The initial training data set includes: initial training data for training the initial point cloud registration model and initial test data for testing the initial point cloud registration model; the updating subunit includes: A merging module, configured to merge the target training data and the initial training data to obtain a new training data set; An updating module is used to update the initial point cloud registration model according to the new training data set, the target test data, and the initial test data.

20. The device according to claim 19, wherein The update module includes: A training submodule, configured to train the initial point cloud registration model based on the new training data set to obtain a target point cloud registration model; A testing submodule, configured to test the target point cloud registration model using the target test data and the initial test data respectively, and obtain corresponding target test results; An updating submodule is used to update the initial point cloud registration model according to the respective corresponding target test results and the obtained initial test results, wherein the initial test results are test results obtained by testing the initial point cloud registration model based on the initial test data.

21. The device according to claim 20, wherein The test results are used to characterize the registration confidence, and the test results include the corresponding target test results and the initial test results; The update submodule is used to replace the initial point cloud registration model with the target point cloud registration model if the registration confidence represented by the target test result of the target test data is greater than the registration confidence represented by the target test result of the initial test data, and the registration confidence represented by the target test result of the initial test data is greater than or equal to the registration confidence represented by the initial test result.

22. The device according to any one of claims 19 to 21, wherein: The target training data includes first training data and first verification data; The initial training data includes second training data and second verification data; The new training data set includes: a training data set obtained by merging the first training data and the second training data, and a verification data set obtained by merging the first verification data and the second verification data.

23. The apparatus according to any one of claims 17 to 20, wherein the apparatus is applied to a local server; the apparatus further comprising: a first storage unit, configured to store the target point cloud in a cloud server if the difference value is greater than a preset difference threshold; a generating unit, configured to generate an index text file for storing the target point cloud in the cloud server; The second storage unit is used to store the index text file in the local server.

24. The apparatus according to claim 23, further comprising: The second acquisition unit is configured to acquire the target point cloud from the cloud server according to the index text file.

25. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12.

26. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

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

Images