A point cloud labeling type conversion method and device, a terminal device, and a storage medium

Abstract

The present application relates to the technical field of point cloud labeling, and particularly relates to a point cloud labeling type conversion method and device, terminal equipment and a storage medium, the method comprising: obtaining a labeled data set corresponding to a target detection task and a task flow label configuration corresponding to a semantic segmentation task; obtaining a preset label mapping, determining a semantic segmentation label corresponding to a target detection label of each point cloud in the labeled data set based on the label mapping; generating a labeling result of the semantic segmentation task based on the semantic segmentation label and the task flow label configuration, and publishing a point cloud labeling data set based on the labeling result. The present application can convert the labeling data of a point cloud target detection task into the labeling data of a semantic segmentation task, realize conversion of labeling types, and improve the efficiency of point cloud labeling.

Description

Technical Field

[0001] This invention relates to the field of point cloud annotation technology, and in particular to a point cloud annotation type conversion method, apparatus, terminal device and storage medium. Background Technology

[0002] Point cloud semantic segmentation and annotation is the process of assigning each point in a point cloud to a predefined semantic category. Point clouds are collections of points sampled in three-dimensional space, and because they can capture the true shape and geometry of objects, they are widely used in computer vision and robotics. The goal of point cloud semantic segmentation is to understand the objects in the point cloud and assign them semantic labels. This is particularly useful for autonomous driving; for example, in self-driving cars, point cloud semantic segmentation can help identify roads, vehicles, pedestrians, and other obstacles, thereby helping the vehicle plan and execute safe driving routes.

[0003] Currently, common point cloud semantic segmentation methods typically convert 3D point clouds into handcrafted voxel mesh features or multi-view image features, and then feed them into a deep learning network for feature extraction. This feature conversion method not only involves a large amount of data but also is computationally complex, resulting in low point cloud segmentation efficiency. Therefore, when analyzing the true shape and geometric structure of objects, the traditional point cloud semantic segmentation method is inefficient.

[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0005] The technical problem this invention aims to solve is to address the aforementioned deficiencies of existing technologies by providing a point cloud annotation type conversion method, apparatus, terminal device, and storage medium. This aims to resolve the issue of low efficiency in traditional point cloud semantic segmentation methods when analyzing the true shape and geometric structure of objects.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides a point cloud annotation type conversion method, wherein the method includes:

[0008] Obtain the labeled dataset for the object detection task and the task flow label configuration for the semantic segmentation task;

[0009] Obtain a preset label mapping, and determine the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping;

[0010] Based on the semantic segmentation labels and the task flow labels, the annotation results of the semantic segmentation task are generated, and the point cloud annotation dataset is published based on the annotation results.

[0011] Based on the aforementioned technical means, this embodiment of the application converts the labeled data of the point cloud object detection task into labeled data of the semantic segmentation task through a preset label mapping. The time required for point cloud object detection to realize the true shape and geometric structure of objects is significantly shorter than that required for point cloud semantic segmentation. Therefore, point cloud object detection is more efficient in point cloud annotation. Thus, this embodiment of the application utilizes the high annotation efficiency of the point cloud object detection task, and only requires converting the labeled data of the point cloud object detection task into labeled data of the semantic segmentation task, thereby saving time on point cloud annotation while meeting the data usage requirements of the semantic segmentation task.

[0012] In one embodiment of this application, determining the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping includes:

[0013] Based on the label mapping, the label mapping is used to reflect the correspondence between the source detection label and the source semantic label of each point cloud in the target detection task, wherein the source detection label is a label determined according to the target detection task, and the source semantic label is a label determined according to the semantic segmentation task;

[0014] The semantic segmentation label is obtained by matching the target detection label of each point cloud in the labeled dataset with the corresponding relationship.

[0015] Based on the above technical means, the embodiments of this application utilize label mapping to convert the labels of the object detection task into the labels of the semantic segmentation task, which is more accurate and faster.

[0016] In one embodiment of this application, matching the target detection label of each point cloud in the labeled dataset with the corresponding relationship to obtain the semantic segmentation label includes:

[0017] Obtain the label information of the target detection label for each point cloud in the labeled dataset, and determine the source detection label corresponding to the target detection label for each point cloud in the labeled dataset based on the label information;

[0018] Based on the correspondence, the source semantic label corresponding to the source detection label of each point cloud is determined, and the source semantic label is used as the semantic segmentation label.

[0019] Based on the above technical means, the embodiments of this application utilize the correspondence set in the label mapping to match the label information of the target detection label of each point cloud in the labeled dataset, which can determine the semantic segmentation label in sequence and quickly, thereby improving the efficiency and accuracy of label conversion.

[0020] In one embodiment of this application, the step of generating the annotation result of the semantic segmentation task based on the semantic segmentation label and the task flow label configuration includes:

[0021] Obtain several bounding boxes corresponding to the semantic segmentation labels, analyze the point cloud in the labeled dataset based on the several bounding boxes, and determine the relationship between the point cloud and the bounding boxes;

[0022] If the relationship between the point cloud and the bounding box is that the point cloud belongs to the bounding box, then the point cloud is associated with the semantic segmentation label, and a label index list is generated;

[0023] The annotation results of the semantic segmentation task are generated based on the label index list and the task flow label configuration.

[0024] Based on the above technical means, this application embodiment analyzes the point clouds in the labeled dataset and generates a label index list to associate the point clouds with semantic segmentation labels, thereby obtaining the labeling results of the semantic segmentation task. This ensures that the labeling type of each point cloud in the labeled dataset is converted, which facilitates obtaining the labeling results of the semantic segmentation task.

[0025] In one embodiment of this application, the step of analyzing the point cloud in the labeled dataset based on a plurality of the labeled bounding boxes to determine the relationship between the point cloud and the labeled bounding boxes includes:

[0026] Determine the convex hull corresponding to each of the aforementioned annotation boxes;

[0027] Traverse the point cloud in the labeled dataset to determine whether the point cloud belongs to any convex hull;

[0028] If the point cloud belongs to any convex hull, then the relationship between the point cloud and the annotation box is determined to be that the point cloud belongs to the annotation box;

[0029] If the point cloud does not belong to any convex hull, then the relationship between the point cloud and the annotation box is determined to be that the point cloud does not belong to the annotation box.

[0030] Based on the above technical means, the embodiments of this application analyze each point cloud in the labeled dataset by the convex hull corresponding to the labeled bounding box, and determine whether the point cloud belongs to the labeled bounding box based on the convex hull, thereby realizing the confirmation of the point cloud, so as to ensure that the labeling result of each point cloud meets the requirements of the semantic segmentation task.

[0031] In one embodiment of this application, generating the annotation result of the semantic segmentation task based on the tag index list and the task flow tag configuration includes:

[0032] Obtain the tag format address from the task flow tag configuration;

[0033] The semantic segmentation tags in the tag index list are associated with the tag format addresses to obtain the annotation results.

[0034] Based on the above technical means, the embodiments of this application generate annotation results by associating semantic segmentation tags with tag format addresses, thereby completing the conversion of annotation types.

[0035] In one embodiment of this application, publishing the point cloud annotation dataset based on the annotation results includes:

[0036] Based on the tag format address in the annotation results, determine the output path and the output directory corresponding to the output path;

[0037] Based on the output directory, the point cloud annotation dataset corresponding to the annotation results is published. The point cloud annotation dataset is the speech segmentation label corresponding to each point cloud of the semantic segmentation task.

[0038] Based on the above technical means, after obtaining the annotation results, this application embodiment publishes a point cloud annotation dataset so that when performing semantic segmentation tasks, the point cloud annotation dataset that has undergone type conversion can be used. This point cloud annotation dataset is converted from the labeled dataset corresponding to the object detection task, which takes advantage of the high annotation efficiency of the point cloud object detection task and meets the data usage requirements of point cloud semantic segmentation.

[0039] Secondly, embodiments of the present invention also provide a point cloud annotation type conversion device, wherein the device includes:

[0040] The data acquisition module is used to acquire the labeled dataset corresponding to the object detection task and the task flow label configuration corresponding to the semantic segmentation task.

[0041] The label conversion module is used to obtain a preset label mapping and determine the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping.

[0042] The result generation module is used to generate the annotation results of the semantic segmentation task based on the semantic segmentation label and the task flow label configuration, and to publish the point cloud annotation dataset based on the annotation results.

[0043] In one embodiment of this application, the tag conversion module includes:

[0044] A relationship determination unit is used to determine the relationship based on the label mapping, which reflects the correspondence between the source detection label and the source semantic label of each point cloud in the target detection task, wherein the source detection label is a label determined according to the target detection task, and the source semantic label is a label determined according to the semantic segmentation task.

[0045] The label matching unit is used to match the target detection label of each point cloud in the labeled dataset with the corresponding relationship to obtain the semantic segmentation label.

[0046] In one embodiment of this application, the tag matching unit includes:

[0047] The first label mapping unit is used to obtain the label information of the target detection label of each point cloud in the labeled dataset, and determine the source detection label corresponding to the target detection label of each point cloud in the labeled dataset based on the label information.

[0048] The second label mapping unit is used to determine the source semantic label corresponding to the source detection label of each point cloud based on the correspondence, and to use the source semantic label as the semantic segmentation label.

[0049] In one embodiment of this application, the result generation module includes:

[0050] The point cloud analysis unit is used to obtain several bounding boxes corresponding to the semantic segmentation labels, analyze the point cloud in the labeled dataset based on the several bounding boxes, and determine the relationship between the point cloud and the bounding boxes.

[0051] The label indexing unit is used to associate the point cloud with the semantic segmentation label and generate a label index list if the relationship between the point cloud and the annotation box is that the point cloud belongs to the annotation box.

[0052] The result generation unit is used to generate the annotation results of the semantic segmentation task based on the label index list and the task flow label configuration.

[0053] In one embodiment of this application, the point cloud analysis unit includes:

[0054] A convex hull determination unit is used to determine the convex hull corresponding to each of the annotation boxes;

[0055] The point cloud determination unit is used to traverse the point clouds in the labeled dataset and determine whether the point cloud belongs to any convex hull.

[0056] The first analysis unit is used to determine the relationship between the point cloud and the annotation box as the point cloud belonging to the annotation box if the point cloud belongs to any convex hull;

[0057] The second analysis unit is used to determine the relationship between the point cloud and the annotation box as follows: if the point cloud does not belong to any convex hull, the point cloud does not belong to the annotation box.

[0058] In one embodiment of this application, the result generation unit includes:

[0059] The address acquisition unit is used to acquire the tag format address in the task flow tag configuration;

[0060] The address association unit is used to associate the semantic segmentation tags in the tag index list with the tag format address to obtain the annotation result.

[0061] In one embodiment of this application, the result generation module includes:

[0062] The directory determination unit is used to determine the output path and the output directory corresponding to the output path based on the tag format address in the annotation result.

[0063] The dataset publishing unit is used to publish the point cloud annotation dataset corresponding to the standard results based on the output directory. The point cloud annotation dataset consists of the speech segmentation labels corresponding to each point cloud of the semantic segmentation task.

[0064] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and a point cloud annotation type conversion program stored in the memory and executable on the processor. When the processor executes the point cloud annotation type conversion program, it implements the steps of the point cloud annotation type conversion method of any of the above solutions.

[0065] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein a point cloud annotation type conversion program is stored on the computer-readable storage medium, and when the point cloud annotation type conversion program is executed by a processor, it implements the steps of the point cloud annotation type conversion method described in any of the above schemes.

[0066] Beneficial Effects: Compared with existing technologies, this application provides a point cloud annotation type conversion method. This application converts the annotation data of a point cloud object detection task into annotation data for a semantic segmentation task through a preset label mapping. Since the time required for point cloud object detection to realize the true shape and geometric structure of objects is significantly shorter than that required for point cloud semantic segmentation, point cloud object detection is more efficient in point cloud annotation. Therefore, this application utilizes the high annotation efficiency of point cloud object detection tasks, and only requires converting the annotation data of point cloud object detection tasks into annotation data for semantic segmentation tasks, thus saving point cloud annotation time while meeting the data usage needs of semantic segmentation tasks.

[0067] Furthermore, this embodiment utilizes the correspondence set in the label mapping to match the label information of the target detection label of each point cloud in the labeled dataset, which can sequentially and quickly determine the semantic segmentation label, thereby improving the efficiency and accuracy of label conversion.

[0068] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0069] Figure 1 A flowchart illustrating a specific implementation of the point cloud annotation type conversion method provided in this embodiment of the invention;

[0070] Figure 2 This is a business process diagram of the point cloud annotation type conversion method provided in an embodiment of the present invention;

[0071] Figure 3 This is a flowchart illustrating the type conversion process of the point cloud annotation type conversion method provided in this embodiment of the invention.

[0072] Figure 4 This is a functional principle diagram of the point cloud annotation type conversion device provided in an embodiment of the present invention;

[0073] Figure 5 A schematic diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0074] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0075] Because existing technologies often employ inefficient point cloud semantic segmentation methods when analyzing the true shape and geometric structure of objects, this embodiment provides a point cloud annotation type conversion method. This method converts the annotation data from point cloud object detection tasks into annotation data for semantic segmentation tasks. Since point cloud object detection tasks take significantly less time to determine the true shape and geometric structure of objects than point cloud semantic segmentation, point cloud object detection is more efficient for point cloud annotation, thus improving the overall efficiency of point cloud annotation. In specific applications, this embodiment first obtains the labeled dataset corresponding to the object detection task and the task flow label configuration corresponding to the semantic segmentation task. Then, it obtains a preset label mapping and determines the semantic segmentation label corresponding to the object detection label of each point cloud in the labeled dataset based on the label mapping. Finally, it generates the annotation results of the semantic segmentation task based on the semantic segmentation labels and the task flow label configuration, and publishes the point cloud annotation dataset based on the annotation results. Therefore, this embodiment takes advantage of the high annotation efficiency of point cloud target detection task, and only needs to convert the annotation data of point cloud target detection task into annotation data of semantic segmentation task, which saves the time of point cloud annotation while meeting the data usage needs of semantic segmentation task.

[0076] The point cloud annotation type conversion method of this embodiment can be applied to a terminal device, which can be an in-vehicle controller, such as an in-vehicle central control computer. Alternatively, the terminal device can also be a user's mobile terminal, such as a mobile phone, which can connect to the in-vehicle terminal to receive data transmitted by the in-vehicle terminal and perform corresponding analysis and processing. Specifically, as shown... Figure 1 As shown in the figure, the point cloud annotation type conversion method in this embodiment includes the following steps:

[0077] Step S100: Obtain the labeled dataset corresponding to the object detection task and the task flow label configuration corresponding to the semantic segmentation task.

[0078] Since this embodiment aims to convert the labeled data corresponding to the object detection task into labeled data corresponding to the semantic segmentation task, it first obtains the labeled dataset corresponding to the object detection task, such as... Figure 2 Step 21 shown indicates that the labeled dataset is obtained after the object detection task has completed data labeling. Since point cloud labeling for object detection tasks is highly efficient, this embodiment can leverage the advantages of object detection tasks. It only requires converting the labeled data between the object detection task and the semantic segmentation task to meet the data usage requirements. Therefore, this embodiment also obtains the task flow label configuration, such as... Figure 2 As shown in step 22, the task flow label configuration is used to generate the annotation results of the semantic segmentation task in subsequent steps.

[0079] Step S200: Obtain a preset label mapping, and determine the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping.

[0080] This embodiment uses label mapping to achieve the conversion of labeled data between object detection and semantic segmentation tasks. In this embodiment, the label mapping reflects the correspondence between object detection labels and semantic segmentation labels. Therefore, based on this label mapping, the semantic segmentation label corresponding to the object detection label of each point cloud in the labeled dataset of the object detection task can be determined, thus realizing the conversion of labeled data between the object detection and semantic segmentation tasks. Figure 2 Step 23 in the process involves performing a data processing task to convert the format.

[0081] In one implementation, this embodiment includes the following steps when implementing tag conversion:

[0082] Step S201: Based on the label mapping, the label mapping is used to reflect the correspondence between the source detection label and the source semantic label of each point cloud in the target detection task, wherein the source detection label is a label determined according to the target detection task, and the source semantic label is a label determined according to the semantic segmentation task;

[0083] Step S202: Match the target detection label of each point cloud in the labeled dataset with the corresponding relationship to obtain the semantic segmentation label.

[0084] In practical applications, this embodiment first sets source detection labels for each point cloud according to the object detection task and source semantic labels for each point cloud according to the semantic segmentation task when setting the label mapping. Then, the correspondence between the source detection labels and the source semantic labels is established to obtain the label mapping. This correspondence is the connection between the object detection task and the semantic segmentation task and is the basis for realizing the annotation data conversion. The label mapping is established based on this correspondence. In practical applications, the label mapping of this embodiment can be stored in the source input file, and the storage path can be as follows: / mnt / bos / autoai / matrix-workspace / matrix-process / task / task-1623303764691595266 / data / data_source.json. When performing label conversion, the label information of the object detection labels of each point cloud in the labeled dataset is first obtained. This label information can be label ID or other information used to identify the object detection labels. Then, based on the label information, the source detection label corresponding to the target detection label of each point cloud in the labeled dataset can be determined. That is, the preset source detection label corresponds to the target detection label obtained in the target detection task. Once the source detection label is found, this embodiment can find the source semantic label corresponding to the source detection label of each point cloud from the correspondence in the label mapping. This source semantic label is the semantic segmentation label in this embodiment. This embodiment uses the correspondence set in the label mapping to match the label information of the target detection label of each point cloud in the labeled dataset, which can sequentially and quickly determine the semantic segmentation label, improving the efficiency and accuracy of label conversion.

[0085] Step S300: Generate the annotation results of the semantic segmentation task based on the semantic segmentation labels and the task flow labels, and publish the point cloud annotation dataset based on the annotation results.

[0086] Once the semantic segmentation labels are determined, this embodiment can generate the annotation results of the semantic segmentation task according to the task flow label configuration. The task flow label configuration in this embodiment includes the label format address corresponding to the semantic segmentation labels. This label format address is preset in the task flow label configuration and can be used to generate the annotation results of the semantic segmentation task. After obtaining the annotation results, a point cloud annotation dataset can be published. This point cloud annotation dataset is the annotation dataset corresponding to the semantic segmentation task, as shown below. Figure 2 As shown in step 24, a new dataset is generated based on the conversion result, thus realizing the conversion of labeled data between the object detection task and the semantic segmentation task.

[0087] In one preferred embodiment, the following steps are included when generating the annotation results:

[0088] Step S301: Obtain several bounding boxes corresponding to the semantic segmentation labels, analyze the point cloud in the labeled dataset based on the several bounding boxes, and determine the relationship between the point cloud and the bounding boxes;

[0089] Step S302: If the relationship between the point cloud and the annotation box is that the point cloud belongs to the annotation box, then associate the point cloud with the semantic segmentation label and generate a label index list;

[0090] Step S303: Generate the annotation results of the semantic segmentation task based on the label index list and the task flow label configuration.

[0091] Specifically, since object detection and semantic segmentation tasks operate on different principles, object detection involves detecting objects in an input image and labeling their locations and classifications. Semantic segmentation, on the other hand, classifies the input image pixel-by-pixel, identifying objects at the pixel level. Therefore, the bounding boxes for these two tasks differ. Consequently, this embodiment, when generating the annotation results for the semantic segmentation task, requires analyzing and filtering each point cloud in the labeled dataset, removing those not belonging to the semantic segmentation task. In practical application, this embodiment obtains several bounding boxes corresponding to the semantic segmentation labels, and then analyzes the point clouds in the labeled dataset based on these bounding boxes to determine the relationship between the point clouds and the bounding boxes.

[0092] Specifically, such as Figure 3 As shown in the figure, this embodiment executes Figure 3 Step 31 involves determining the convex hull corresponding to each of the labeled bounding boxes. This convex hull is the polygon formed by the labeled bounding boxes. Then, the point cloud in the labeled dataset (i.e., ...) is traversed. Figure 3 Step 32) determines whether the point cloud belongs to any convex hull (i.e. Figure 3 Step 33) If the point cloud belongs to any convex hull, then the relationship between the point cloud and the bounding box is determined to be that the point cloud belongs to the bounding box. If the point cloud does not belong to any convex hull, then the relationship between the point cloud and the bounding box is determined to be that the point cloud does not belong to the bounding box. Based on the above method, each point cloud is analyzed, that is, the following steps are performed: Figure 3In step 35, it is determined whether this is the last point cloud. If it is not the last point cloud, step 32 is repeated until the last point cloud is determined to belong to any convex hull. This embodiment analyzes each point cloud in the labeled dataset using the convex hull corresponding to the bounding box, and determines whether the point cloud belongs to the bounding box based on the convex hull, thereby confirming the point cloud and ensuring that the labeling result of each point cloud meets the requirements of the semantic segmentation task.

[0093] Once the relationship between the point cloud and the bounding box is determined to be that the point cloud belongs to the bounding box, this embodiment can associate the point cloud with the semantic segmentation label and generate a label index list (i.e., Figure 3 Step 34) involves associating the point cloud with the corresponding semantic segmentation label and adding it to a preset empty index list (e.g., an index list), thereby obtaining a label index list. The label index list in this embodiment can quickly locate the point cloud corresponding to each semantic segmentation label. Next, this embodiment can obtain the label format address (i.e., JSON address) in the task flow label configuration, and then associate the semantic segmentation labels in the label index list with the label format address, that is, execute... Figure 3 In step 36, constructing JSON based on the semantic segmentation annotation format yields the annotation results and the annotation type conversion.

[0094] After obtaining the annotation results, this embodiment can determine the output path and the corresponding output directory based on the label format address in the annotation results. Then, based on the output directory, the point cloud annotation dataset corresponding to the annotation results is published. The point cloud annotation dataset is the speech segmentation label corresponding to each point cloud of the semantic segmentation task. In specific applications, when publishing the point cloud annotation dataset, this embodiment can maintain the same output directory as the labeled dataset of the object detection task. A new file document needs to be written to the dataset_file collection in MongoDB (MongoDB is a scalable, high-performance, open-source, schema-free, document-oriented NoSQL database), and the annotation results should be written to MongoDB in the pre-annotated record format, such as... Figure 2 As shown in step 25, the new dataset is written to MongoDB, completing the publication of the point cloud annotation dataset. After the point cloud annotation dataset is published, labels can be configured and submitted for annotation on the annotation system (e.g., ...). Figure 2 As shown in step 26, the new dataset is sent for labeling. The semantic segmentation results after semantic segmentation annotation can then be viewed on the annotation system, allowing for further execution. Figure 2In step 27, a new round of annotation is performed. Therefore, this embodiment publishes a point cloud annotation dataset after obtaining the annotation results, so that a point cloud annotation dataset that has undergone type conversion can be used when performing semantic segmentation tasks. This point cloud annotation dataset is converted from the labeled dataset corresponding to the object detection task, taking advantage of the high annotation efficiency of the point cloud object detection task and meeting the data usage requirements of point cloud semantic segmentation.

[0095] Based on the above embodiments, the present invention also provides a point cloud annotation type conversion device, such as... Figure 4 As shown, the point cloud annotation type conversion device 100 specifically includes: a data acquisition module 10, a label conversion module 20, and a result generation module 30. Specifically, the data acquisition module 10 is used to acquire the labeled dataset corresponding to the object detection task and the task flow label configuration corresponding to the semantic segmentation task. The label conversion module 20 is used to acquire a preset label mapping and determine the semantic segmentation label corresponding to the object detection label of each point cloud in the labeled dataset based on the label mapping. The result generation module 30 is used to generate the annotation result of the semantic segmentation task based on the semantic segmentation label and the task flow label configuration, and publish the point cloud annotation dataset based on the annotation result.

[0096] In one embodiment of this application, the tag conversion module includes:

[0097] A relationship determination unit is used to determine the relationship based on the label mapping, which reflects the correspondence between the source detection label and the source semantic label of each point cloud in the target detection task, wherein the source detection label is a label determined according to the target detection task, and the source semantic label is a label determined according to the semantic segmentation task.

[0098] The label matching unit is used to match the target detection label of each point cloud in the labeled dataset with the corresponding relationship to obtain the semantic segmentation label.

[0099] In one embodiment of this application, the tag matching unit includes:

[0100] The first label mapping unit is used to obtain the label information of the target detection label of each point cloud in the labeled dataset, and determine the source detection label corresponding to the target detection label of each point cloud in the labeled dataset based on the label information.

[0101] The second label mapping unit is used to determine the source semantic label corresponding to the source detection label of each point cloud based on the correspondence, and to use the source semantic label as the semantic segmentation label.

[0102] In one embodiment of this application, the result generation module includes:

[0103] The point cloud analysis unit is used to obtain several bounding boxes corresponding to the semantic segmentation labels, analyze the point cloud in the labeled dataset based on the several bounding boxes, and determine the relationship between the point cloud and the bounding boxes.

[0104] The label indexing unit is used to associate the point cloud with the semantic segmentation label and generate a label index list if the relationship between the point cloud and the annotation box is that the point cloud belongs to the annotation box.

[0105] The result generation unit is used to generate the annotation results of the semantic segmentation task based on the label index list and the task flow label configuration.

[0106] In one embodiment of this application, the point cloud analysis unit includes:

[0107] A convex hull determination unit is used to determine the convex hull corresponding to each of the annotation boxes;

[0108] The point cloud determination unit is used to traverse the point clouds in the labeled dataset and determine whether the point cloud belongs to any convex hull.

[0109] The first analysis unit is used to determine the relationship between the point cloud and the annotation box as the point cloud belonging to the annotation box if the point cloud belongs to any convex hull;

[0110] The second analysis unit is used to determine the relationship between the point cloud and the annotation box as follows: if the point cloud does not belong to any convex hull, the point cloud does not belong to the annotation box.

[0111] In one embodiment of this application, the result generation unit includes:

[0112] The address acquisition unit is used to acquire the tag format address in the task flow tag configuration;

[0113] The address association unit is used to associate the semantic segmentation tags in the tag index list with the tag format address to obtain the annotation result.

[0114] In one embodiment of this application, the result generation module includes:

[0115] The directory determination unit is used to determine the output path and the output directory corresponding to the output path based on the tag format address in the annotation result.

[0116] The dataset publishing unit is used to publish the point cloud annotation dataset corresponding to the standard results based on the output directory. The point cloud annotation dataset consists of the speech segmentation labels corresponding to each point cloud of the semantic segmentation task.

[0117] The working principle of each module in the point cloud annotation type conversion device 100 of this embodiment is the same as that of each step in the above method embodiment, and will not be repeated here.

[0118] The point cloud annotation type conversion device 100 in this embodiment converts the annotation data of the point cloud object detection task into the annotation data of the semantic segmentation task through a preset label mapping. Since the point cloud object detection task takes significantly less time to realize the true shape and geometric structure of an object than the point cloud semantic segmentation task, point cloud object detection is more efficient in point cloud annotation. Therefore, this embodiment utilizes the high annotation efficiency of the point cloud object detection task, and only requires converting the annotation data of the point cloud object detection task into the annotation data of the semantic segmentation task, thus saving point cloud annotation time while meeting the data usage needs of the semantic segmentation task.

[0119] Figure 5 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. The terminal device may include: a memory 501, a processor 502, and a computer program stored in the memory 501 and executable on the processor 502. When the processor 502 executes the program, it implements the point cloud annotation type conversion method provided in the above embodiment.

[0120] Furthermore, the terminal equipment also includes:

[0121] Communication interface 503 is used for communication between memory 501 and processor 502.

[0122] The memory 501 is used to store computer programs that can run on the processor 502.

[0123] The memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0124] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0125] In practical implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface. The processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0126] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the point cloud annotation type conversion method described above.

[0127] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0128] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0129] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0130] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0131] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0132] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0133] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0134] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A point cloud annotation type conversion method, characterized in that, The method includes: Obtain the labeled dataset for the object detection task and the task flow label configuration for the semantic segmentation task; Obtain a preset label mapping, and determine the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping; Based on the semantic segmentation labels and the task flow labels, the annotation results of the semantic segmentation task are generated, and the point cloud annotation dataset is published based on the annotation results; The step of determining the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping includes: Based on the label mapping, the label mapping is used to reflect the correspondence between the source detection label and the source semantic label of each point cloud in the target detection task, wherein the source detection label is a label determined according to the target detection task, and the source semantic label is a label determined according to the semantic segmentation task; Obtain the label information of the target detection label for each point cloud in the labeled dataset, and determine the source detection label corresponding to the target detection label for each point cloud in the labeled dataset based on the label information; Based on the correspondence, the source semantic label corresponding to the source detection label of each point cloud is determined, and the source semantic label is used as the semantic segmentation label.

2. The point cloud annotation type conversion method according to claim 1, characterized in that, The step of generating the annotation result for the semantic segmentation task based on the semantic segmentation label and the task flow label configuration includes: Obtain the bounding boxes corresponding to the semantic segmentation labels, analyze the point cloud in the labeled dataset based on the bounding boxes, and determine the relationship between the point cloud and the bounding boxes; If the relationship between the point cloud and the bounding box is that the point cloud belongs to the bounding box, then the point cloud is associated with the semantic segmentation label, and a label index list is generated; The annotation results of the semantic segmentation task are generated based on the label index list and the task flow label configuration.

3. The point cloud annotation type conversion method according to claim 2, characterized in that, The step of analyzing the point cloud in the labeled dataset based on the labeled bounding boxes to determine the relationship between the point cloud and the labeled bounding boxes includes: Determine the convex hull corresponding to each of the aforementioned annotation boxes; Traverse the point cloud in the labeled dataset to determine whether the point cloud belongs to any convex hull; If the point cloud belongs to any convex hull, then the relationship between the point cloud and the annotation box is determined to be that the point cloud belongs to the annotation box; If the point cloud does not belong to any convex hull, then the relationship between the point cloud and the annotation box is determined to be that the point cloud does not belong to the annotation box.

4. The point cloud annotation type conversion method according to claim 3, characterized in that, The step of generating the annotation results for the semantic segmentation task based on the label index list and the task flow label configuration includes: Obtain the tag format address from the task flow tag configuration; The semantic segmentation tags in the tag index list are associated with the tag format addresses to obtain the annotation results.

5. The point cloud annotation type conversion method according to claim 4, characterized in that, The process of publishing a point cloud annotation dataset based on the annotation results includes: Based on the tag format address in the annotation results, determine the output path and the output directory corresponding to the output path; Based on the output directory, the point cloud annotation dataset corresponding to the annotation results is published. The point cloud annotation dataset is the speech segmentation label corresponding to each point cloud of the semantic segmentation task.

6. A point cloud annotation type conversion device, characterized in that, The apparatus is used to implement the steps of the point cloud annotation type conversion method as described in any one of claims 1-5, the apparatus comprising: The data acquisition module is used to acquire the labeled dataset corresponding to the object detection task and the task flow label configuration corresponding to the semantic segmentation task. The label conversion module is used to obtain a preset label mapping and determine the semantic segmentation label corresponding to the target detection label of each point cloud in the labeled dataset based on the label mapping. The result generation module is used to generate the annotation results of the semantic segmentation task based on the semantic segmentation label and the task flow label configuration, and to publish the point cloud annotation dataset based on the annotation results.

7. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a point cloud annotation type conversion program stored in the memory and executable on the processor. When the processor executes the point cloud annotation type conversion program, it implements the steps of the point cloud annotation type conversion method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a point cloud annotation type conversion program, which, when executed by a processor, implements the steps of the point cloud annotation type conversion method as described in any one of claims 1-5.

Images