Data processing method, device and system, storage medium and program product

Abstract

The invention provides a data processing method, device and system, a storage medium and a program product, and relates to the field of data processing. The method comprises the following steps: selecting a first sample set from a target domain sample set; the second sample set is removed from the source domain sample set to obtain a third sample set, and a first area range corresponding to the first sample set is the same as a second area range corresponding to the second sample set; and fusing the first sample set and the third sample set to obtain a fused sample set for training a target detection model. The problem that the source domain sample set dominates network training can be reduced, in addition, due to the fact that fusion between the source domain sample set and the target domain sample set is achieved, the target detection model is made to learn domain-invariant universal features independent of specific configuration or environment, and therefore the adaptability of the model in the target domain is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to a data processing method, device, system, storage medium and program product. Background Art

[0002] 3D object detection based on LiDAR (Light Detection and Ranging) plays a vital role in autonomous driving systems. LiDAR is a technology that uses laser pulses to measure the distance of objects, which can provide three-dimensional information about the surrounding environment. This information is crucial for vehicle navigation, obstacle avoidance and decision-making.

[0003] In different application scenarios, LiDAR systems may encounter various configurations, such as different numbers of laser beams, different viewing angles, and different resolutions. Since the performance of the target detection model depends largely on the quality of the data and the ability to extract features, these differences can cause a 3D target detection model trained in one configuration to perform poorly in another configuration. It is very expensive to have sufficient training data in different domains at the same time. A more realistic scenario is to have sufficient training samples in the source domain and only limited training samples in the target domain. The source domain is the dataset where the model was originally trained, and the target domain is the new environment where the model will be applied. Researchers are beginning to explore domain adaptation techniques, which allow models to be trained in the source domain (such as 16-line data) and limited target domain data (such as 32-line data), and then adapted and promoted in the target domain. Summary of the invention

[0004] A technical problem to be solved by the present disclosure is to provide a data processing method, device, system, storage medium and program product, which can improve the adaptability of the target detection model in the target domain.

[0005] According to one aspect of the present disclosure, a data processing method is proposed, including: selecting a first sample set from a target domain sample set; removing a second sample set from a source domain sample set to obtain a third sample set, wherein a first area range corresponding to the first sample set and a second area range corresponding to the second sample set are the same; fusing the first sample set with the third sample set to obtain a fused sample set for training a target detection model.

[0006] In some embodiments, fusing the first sample set with the third sample set to obtain the fused sample set includes: fusing first data corresponding to the first sample set with second data corresponding to the third sample set to obtain fused data; fusing first label information corresponding to the first sample set with second label information corresponding to the third sample set to obtain fused label information; and using the fused label information as label information of the fused data.

[0007] In some embodiments, the target domain sample set is a first point cloud sample set obtained by scanning by a first radar sensor, and selecting the first sample set in the target domain sample set includes: selecting a point in the first point cloud sample set as a first center point; and based on a bird's-eye view of the target domain, selecting a point cloud in a first area around the first center point as the first sample set.

[0008] In some embodiments, the source domain sample set is a second point cloud sample set obtained by scanning by a second radar sensor, and removing the second sample set from the source domain sample set to obtain the third sample set includes: selecting a point in the second point cloud sample set as a second center point; and based on a bird's-eye view of the source domain, removing the point cloud within the second area around the second center point to obtain the third sample set.

[0009] In some embodiments, the second center point is obtained based on the selection of the first center point; or the first center point is obtained based on the selection of the second center point.

[0010] In some embodiments, the direction of the first center point relative to the first radar sensor is consistent with the direction of the second center point relative to the second radar sensor; the distance of the first center point relative to the first radar sensor is consistent with the distance of the second center point relative to the second radar sensor.

[0011] In some embodiments, a trained target detection model is used to detect targets within a target domain.

[0012] According to another aspect of the present disclosure, a data processing device is also proposed, including: a sample set acquisition module, configured to select a first sample set in a target domain sample set, remove a second sample set in a source domain sample set, and obtain a third sample set, wherein a first area range corresponding to the first sample set and a second area range corresponding to the second sample set are the same; a sample set fusion module, configured to fuse the first sample set with the third sample set to obtain a fused sample set for training a target detection model.

[0013] According to another aspect of the present disclosure, a data processing device is further provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the above-mentioned data processing method based on instructions stored in the memory.

[0014] According to another aspect of the present disclosure, an autonomous driving system is also proposed, comprising: the above-mentioned data processing device.

[0015] According to another aspect of the present disclosure, a computer-readable storage medium is further provided, on which computer program instructions are stored, and when the instructions are executed by a processor, the data processing method as described above is implemented.

[0016] According to another aspect of the present disclosure, a computer program product is also proposed, including a computer program or instructions, and the computer program or instructions implement the above data processing method when executed by a processor.

[0017] In the disclosed embodiment, the third sample set in the source domain sample set is fused with the first sample set in the target domain sample set, and the fused sample set is used to train the target detection model, which can reduce the problem of the source domain sample set dominating the network training. In addition, since the fusion between the source domain sample set and the target domain sample set is achieved, the deviations between different fields can be effectively aligned. Moreover, although the source domain sample set and the target domain sample set are inconsistently distributed, they have the same migration information. Therefore, the learning domain invariant representation is achieved, so that the target detection model learns domain-invariant general features that are not dependent on specific configurations or environments, thereby improving the adaptability of the model in the target domain.

[0018] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0020] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0021] Figure 1 Schematic diagram of some embodiments of the data processing method disclosed in the present invention;

[0022] Figure 2 Schematic diagram of the flow chart of other embodiments of the data processing method disclosed in the present invention;

[0023] Figure 3 Schematic diagram of the flow chart of other embodiments of the data processing method disclosed in the present invention;

[0024] Figure 4 Schematic diagram of the flow chart of other embodiments of the data processing method disclosed in the present invention;

[0025] Figure 5 Schematic diagram of the flow chart of other embodiments of the data processing method disclosed in the present invention;

[0026] Figure 6 Schematic diagrams of some embodiments of model training disclosed herein;

[0027] Figure 7 It is a schematic diagram of the structure of some embodiments of the data processing device disclosed in the present invention;

[0028] Figure 8 Schematic diagrams of structures of other embodiments of the data processing device disclosed in the present invention. DETAILED DESCRIPTION

[0029] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0030] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0031] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0032] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0033] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0034] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0035] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0036] In related technologies, labeled source data and target data are trained simultaneously through a network. However, in practical applications, the amount of data in the source domain is often much larger than that in the target domain, which causes the feature distribution of the source domain to dominate the entire training process. This inappropriate domain mixing will cause the model to be disturbed by the source domain features when learning the features of the target domain, thereby affecting the performance of the model on the target domain.

[0037] The present disclosure provides a data processing solution that can effectively align data distribution between different fields, so that the trained model is more suitable for the target domain.

[0038] Figure 1 1 is a flowchart of some embodiments of the data processing method of the present disclosure, which includes steps S11-S13.

[0039] In step S11, a first sample set is selected from the target domain sample set.

[0040] In some embodiments, the target domain sample set is a first point cloud sample set obtained by scanning with a first radar sensor. The first radar sensor is, for example, a laser radar sensor, which forms a point cloud by sending a signal to a detection target.

[0041] For example, a first radar sensor is installed on an autonomous vehicle, and a target is scanned by the first radar sensor to obtain a first point cloud sample set. The target domain corresponds to a new data set or field to which the model will be applied, and there are usually fewer labeled samples available for learning in the target domain. That is, the target domain sample set may only have a small amount of sample data.

[0042] In some embodiments, the specific implementation of step S11 is as follows: Figure 2 , including steps S111-S112. In step S111, a point is selected in the first point cloud sample set as the first center point. The first center point can be a randomly selected point. In step S112, based on the bird's-eye view of the target domain, a point cloud in a first area is selected around the first center point as the first sample set. This step realizes the extraction of point cloud information to be fused from the target domain to the source domain.

[0043] In step S12, the second sample set is removed from the source domain sample set to obtain a third sample set, wherein the first area range corresponding to the first sample set is the same as the second area range corresponding to the second sample set.

[0044] In some embodiments, the source domain sample set is a second point cloud sample set obtained by scanning by a second radar sensor. The second radar sensor is, for example, a laser radar sensor, which forms a point cloud by sending signals to the detection target.

[0045] For example, a second radar sensor is installed on the autonomous vehicle, and the second radar sensor is used to scan the target to obtain a second point cloud sample set. The source domain corresponds to the data set or domain for pre-training the model. In the source domain, there are usually a large number of labeled samples available for learning. In other words, the source domain sample set is the sample data set for initial training, and the source domain sample set contains a large amount of sample data.

[0046] The source domain sample set and the target domain sample set may be sample sets of different line bundles. For example, the source domain sample set is a 16-line data sample set, and the target domain sample set is a 32-line data sample set. The number of samples in the source domain sample set may be much larger than the number of samples in the target domain sample set.

[0047] In some embodiments, the specific implementation of step S12 is as follows: Figure 3 As shown, it includes steps S121-S122. In step S121, a point is selected in the second point cloud sample set as the second center point. In step S122, based on the bird's-eye view of the source domain, the point cloud within the second area around the second center point is removed to obtain a third sample set. Through this step, the extraction of point cloud information of the area to be fused with the target domain in the source domain is realized.

[0048] In some embodiments, the second center point is obtained based on the selection of the first center point; or the first center point is obtained based on the selection of the second center point, so that the data consistency of the two domains is improved.

[0049] In some embodiments, the direction of the first center point relative to the first radar sensor is consistent with the direction of the second center point relative to the second radar sensor; the distance of the first center point relative to the first radar sensor is consistent with the distance of the second center point relative to the second radar sensor.

[0050] By limiting the position of the first center point and the position of the second center point, the geometric features of the mixed point cloud can be maintained, which facilitates improving the consistency of migration information such as the same form of data in the two domains.

[0051] In step S13, the first sample set and the third sample set are fused to obtain a fused sample set for training the target detection model.

[0052] This step realizes the mixing of the source domain sample set and the target domain sample set. For example, part of the sample set in the source domain sample set is removed, and part of the sample set in the target domain sample set is used to replace the removed sample set in the source domain to obtain the intermediate domain sample set.

[0053] In some embodiments, the fused sample set includes, for example, fused data and fused label information.

[0054] For example, the detection network is trained using the fused data and the fused label information. The specific training process is an existing process and will not be further elaborated here.

[0055] In the above embodiment, the third sample set in the source domain sample set is fused with the first sample set in the target domain sample set, and the fused sample set is used to train the target detection model, which can reduce the problem of the source domain sample set dominating the network training. In addition, since the fusion between the source domain sample set and the target domain sample set is achieved, the deviations between different fields can be effectively aligned. Although the source domain sample set and the target domain sample set are inconsistent in distribution, they have the same migration information. Therefore, the learning domain invariant representation is achieved, so that the target detection model learns domain-invariant general features that are not dependent on specific configurations or environments, thereby improving the adaptability of the model in the target domain.

[0056] Figure 4 Schematic diagram of the flow chart of other embodiments of the data processing method disclosed in the present invention. Figure 4 As shown, step S13 includes steps S131-S133.

[0057] In step S131, the first data corresponding to the first sample set and the second data corresponding to the third sample set are fused to obtain fused data.

[0058] For example, the point clouds of the source domain and the target domain are fused to form a new point cloud.

[0059] In step S132, the first label information corresponding to the first sample set is fused with the second label information corresponding to the third sample set to obtain fused label information.

[0060] This step generates new label information.

[0061] In step S133, the fused label information is used as label information of the fused data.

[0062] In the above embodiment, the point cloud distributions between different fields are effectively aligned, and by generating a fused sample set, the model can learn domain-invariant knowledge, that is, general features that do not depend on specific configurations or environments, thereby significantly improving the adaptability of the model in the target field.

[0063] The following will take a specific embodiment as an example to introduce the solution of the present disclosure. Figure 5 As shown, Figure 5 Schematic diagram of the flow of other embodiments of the data processing method of the present disclosure, including steps S51-S57.

[0064] In step S51, the point cloud center of the target domain is selected.

[0065] For example, in the target domain point cloud x T A point is randomly selected as the center of the region c t .

[0066] The point cloud data of the target domain and the point cloud data of the source domain can be point clouds formed by a laser radar sensor scanning a target to be detected. For example, a laser point cloud refers to a data set of spatial points scanned by a three-dimensional laser radar device, and each point cloud includes three-dimensional coordinates (XYZ) and laser reflection intensity.

[0067] In step S52, a region where point clouds are mixed in the target domain is defined.

[0068] For example, based on the bird's-eye view (xy plane), around the center of the area c t Randomly select a rectangular area

[0069] In step S53, the point cloud within the target domain mixed area is extracted.

[0070] For example, select a rectangular area The point cloud inside is denoted as P t , and remove all points outside this region.

[0071] In step S54, the center of the hybrid region of the source domain point cloud is selected.

[0072] For example, in the source domain point cloud x s Select another center c s , where c s The distance and direction to the radar sensor installed on the vehicle and c t The distance and direction to the radar sensor installed on the vehicle are in the same range, thus maintaining the geometric characteristics of the hybrid point cloud. s The points with the same area size as the target domain, and the remaining area is recorded as

[0073] In step S55, the point clouds of the source domain and the target domain are fused.

[0074] For example, using P t To repair the remaining area of ​​the source domain and obtain a new point cloud

[0075]

[0076] Among them, Concat is a function that connects multiple strings into one string, and ⊙ is the matrix dot product.

[0077] In step S56, a new label is generated.

[0078] For example, combined with the target domain x T Selected point cloud P t 3D Box label and source domain x S The 3DBox labels of the remaining point clouds are used to generate new labels

[0079]

[0080] Among them, y T is the label corresponding to the point cloud of the target domain, y s is the label corresponding to the point cloud in the source domain, and the newly fused point cloud And the newly generated tags Construct the intermediate domain point cloud sample set.

[0081] In step S57, the network is trained.

[0082] For example, using the blended point cloud and tags Train the detection network.

[0083] like Figure 6 As shown, the source domain point cloud sample set 610 and the target domain point cloud sample set 620 are mixed into an intermediate domain point cloud sample set 630, and the target detection model is trained using the intermediate domain point cloud sample set 630, wherein the target detection model includes a 3D detector and a 3D detection head, the 3D detector is used to extract features, and the 3D detection head is used for target prediction. The target detection model can be used to identify the target to be detected.

[0084] In the above embodiment, a local point cloud region in the source sample set is randomly removed, and then replaced with another region in the target sample set to reduce the difference in point cloud distribution in different domains. The newly formed point cloud sample set contains both the annotation information of the source domain and the sample set distribution information of the target domain, so that the newly formed sample set is closer to the target domain information, reducing domain bias and learning domain-invariant representations, which can significantly improve the adaptability of the model in the target domain even when there is only a small amount of labeled target data.

[0085] Applying the above solution in the field of autonomous driving or automatic operation of robots can make target prediction more consistent with real application scenarios and improve the safety of vehicle or robot operation.

[0086] Figure 7 Schematic diagram of the structure of some embodiments of the data processing device of the present disclosure. The data processing device includes a sample set acquisition module 710 and a sample set fusion module 720.

[0087] The sample set acquisition module 710 is configured to select a first sample set from the target domain sample set, remove a second sample set from the source domain sample set, and obtain a third sample set, wherein a first area range corresponding to the first sample set and a second area range corresponding to the second sample set are the same.

[0088] In some embodiments, the target domain sample set is a first point cloud sample set obtained by scanning by a first radar sensor, and the sample set acquisition module 710 selects a point in the first point cloud sample set as a first center point; and based on the bird's-eye view of the target domain, a point cloud in a first area range around the first center point is selected as the first sample set.

[0089] In some embodiments, the source domain sample set is a second point cloud sample set obtained by scanning by a second radar sensor, and the sample set acquisition module 710 selects a point in the second point cloud sample set as a second center point; and based on the bird's-eye view of the source domain, removes the point cloud within the second area around the second center point to obtain a third sample set.

[0090] In some embodiments, the second center point is obtained based on the selection of the first center point; or the first center point is obtained based on the selection of the second center point.

[0091] In some embodiments, the direction of the first center point relative to the first radar sensor is consistent with the direction of the second center point relative to the second radar sensor; the distance of the first center point relative to the first radar sensor is consistent with the distance of the second center point relative to the second radar sensor.

[0092] The sample set fusion module 720 is configured to fuse the first sample set with the third sample set to obtain a fused sample set for training the target detection model.

[0093] In some embodiments, the sample set fusion module 720 fuses the first data corresponding to the first sample set with the second data corresponding to the third sample set to obtain fused data; fuses the first label information corresponding to the first sample set with the second label information corresponding to the third sample set to obtain fused label information; and uses the fused label information as label information of the fused data.

[0094] In some embodiments, the sample set acquisition module and the sample set fusion module constitute a cross-domain point cloud mixing module.

[0095] In the above embodiment, the third sample set in the source domain sample set is fused with the first sample set in the target domain sample set, and the fused sample set is used to train the target detection model, which can reduce the problem of the source domain sample set dominating the network training. In addition, since the fusion between the source domain sample set and the target domain sample set is achieved, the deviations between different fields can be effectively aligned. Although the source domain sample set and the target domain sample set are inconsistent in distribution, they have the same migration information. Therefore, the learning domain invariant representation is achieved, so that the target detection model learns domain-invariant general features that are not dependent on specific configurations or environments.

[0096] In some embodiments, the data processing device further includes a detection module (not shown), which is configured to detect targets within the target domain using a trained target detection model, thereby improving the accuracy of target detection within the target domain.

[0097] Figure 8 Schematic diagram of the structure of some other embodiments of the data processing device of the present disclosure. The data processing device 800 includes a memory 810 and a processor 820. Among them: the memory 810 can be a disk, a flash memory or any other non-volatile storage medium. The memory is used to store the instructions in the above embodiments. The processor 820 is coupled to the memory 810 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 820 is used to execute the instructions stored in the memory.

[0098] In some embodiments, the processor 820 is coupled to the memory 810 via a BUS bus 830. The data processing device 800 can also be connected to an external storage device 850 via a storage interface 840 to call external data, and can also be connected to a network or another computer system (not shown) via a network interface 860. No further detailed description will be given here.

[0099] In this embodiment, the adaptability of the model in the target domain can be improved by storing data instructions in a memory and processing the instructions through a processor.

[0100] In other embodiments of the present disclosure, an autonomous driving system is protected, which includes the data processing device in the above-mentioned embodiment to improve the safety of autonomous driving vehicle driving.

[0101] In other embodiments, a computer-readable storage medium stores computer program instructions, which implement the steps of the methods in the above-mentioned embodiments when executed by a processor. It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, devices, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transient storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0102] In some embodiments, a computer program product is protected, including a computer program or an instruction, which implements the above-mentioned driving strategy determination method or driving control method when executed by a processor. The computer program product includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by the CPU, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0103] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.

[0107] The method and apparatus of the present disclosure may be implemented in many ways. For example, the method and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

[0108] Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A data processing method, comprising: Selecting a first sample set from the target domain sample set; Removing the second sample set from the source domain sample set to obtain a third sample set, wherein the first area range corresponding to the first sample set is the same as the second area range corresponding to the second sample set; The first sample set and the third sample set are fused to obtain a fused sample set for training a target detection model.

2. The data processing method according to claim 1, wherein: The fusing the first sample set and the third sample set to obtain a fused sample set includes: Fusing the first data corresponding to the first sample set with the second data corresponding to the third sample set to obtain fused data; Merging the first label information corresponding to the first sample set with the second label information corresponding to the third sample set to obtain fused label information; and The fused label information is used as label information of the fused data.

3. The data processing method according to claim 1 or 2, wherein: The target domain sample set is a first point cloud sample set obtained by scanning with a first radar sensor, and selecting the first sample set from the target domain sample set includes: Selecting a point in the first point cloud sample set as a first center point; and Based on the bird's-eye view of the target domain, a point cloud within the first area is selected around the first center point as the first sample set.

4. The data processing method according to claim 3, wherein: The source domain sample set is a second point cloud sample set obtained by scanning with a second radar sensor, and removing the second sample set from the source domain sample set to obtain the third sample set includes: Selecting a point in the second point cloud sample set as a second center point; and Based on the bird's-eye view of the source domain, the point cloud within the second area around the second center point is removed to obtain the third sample set.

5. The data processing method according to claim 4, wherein: The second center point is obtained based on the selection of the first center point; or The first center point is obtained based on the selection of the second center point.

6. The data processing method according to claim 5, wherein: The direction of the first center point relative to the first radar sensor is consistent with the direction of the second center point relative to the second radar sensor; The distance of the first center point relative to the first radar sensor is consistent with the distance of the second center point relative to the second radar sensor.

7. The data processing method according to claim 1 or 2, further comprising: Use the trained target detection model to detect targets in the target domain.

8. A data processing device, comprising: a sample set acquisition module, configured to select a first sample set from the target domain sample set, remove a second sample set from the source domain sample set, and obtain a third sample set, wherein a first area range corresponding to the first sample set is the same as a second area range corresponding to the second sample set; The sample set fusion module is configured to fuse the first sample set with the third sample set to obtain a fused sample set for training a target detection model.

9. A data processing device, comprising: Memory; as well as A processor coupled to the memory, wherein the processor is configured to execute the data processing method according to any one of claims 1 to 7 based on instructions stored in the memory.

10. An automatic driving system comprising: The data processing device according to claim 8 or 9.

11. A computer-readable storage medium having computer program instructions stored thereon, wherein the instructions, when executed by a processor, implement the data processing method according to any one of claims 1 to 7.

12. A computer program product, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the data processing method according to any one of claims 1 to 7 is implemented.

Images