Method and device for determining accuracy of point cloud labeling, equipment and medium

Abstract

The invention relates to a method and device for determining accuracy of point cloud labeling, equipment and a medium. The method comprises the steps that after a point cloud data set with a labeling result is obtained, objects labeled by the point cloud data set are analyzed, and the kinematics quantity describing the motion state of each object in each point cloud data frame is determined; and carrying out kinematics quantity visualization processing on the same object in a time sequence to obtain a visualization result for displaying the kinematics quantity change of the same object so as to determine whether the point cloud labeling is accurate or not. According to the scheme provided by the invention, the marking accuracy of the point cloud data can be more intuitively determined through the visualization result, the probability of detecting the abnormal point cloud marking is further improved, and the reliability of the training data of automatic driving is further ensured.

Description

Technical Field

[0001] This application relates to the technical field of data annotation, and particularly to a method, apparatus, device, and medium for determining the accuracy of point cloud annotation. Background Art

[0002] In the field of autonomous driving technology, annotated data, as training data, can help understand and identify various traffic scenarios and environmental information. Among them, it is difficult to determine with the naked eye whether the annotation results of sparse point clouds of distant objects are accurate, and inaccurate annotation results will lead to a decline in the perception ability and decision-making errors of the autonomous driving system. Therefore, the annotation results need to be reviewed and verified. In the acquisition of point cloud data, there are limitations in the acquisition equipment and acquisition methods themselves. For example, sensors such as lidar and depth cameras on the market have factors such as signal attenuation and noise interference when capturing distant objects, so that the acquired point cloud data is sparse and the quality deteriorates. The point cloud data therein is discrete and unstructured data, and the distribution of sparse point clouds in the distance is irregular, and there is a lack of sufficient features and context information, which further increases the difficulty of accurate annotation and detection.

[0003] In the related art, the detection of annotated point cloud data is only by visual inspection. In practice, the point clouds in the middle and near distances are relatively dense, and the object contours are clear. It is easy for the naked eye to see whether the 3D box fits the object point cloud contour and whether the center point is correct. However, the point clouds in the distance are relatively sparse, and the point cloud contours of the objects are relatively blurred, and there are often problems such as the 3D box not fitting and the center point deviating too much.

[0004] Therefore, there are still many difficulties in detecting whether the annotation of distant targets is accurate. Summary of the Invention

[0005] To solve or partially solve the problems existing in the related art, this application provides a method, apparatus, device, and medium for determining the accuracy of point cloud annotation.

[0006] In a first aspect of this application, a method for determining the accuracy of point cloud annotation is provided, including: after obtaining a point cloud data set with annotation results, parsing the objects annotated in the point cloud data set to determine the kinematic quantities describing the motion states of each object in each point cloud data frame; performing visual processing on the kinematic quantities of the same object in time series to obtain a visual result showing the change of the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

[0007] In an optional embodiment, the parsing the objects annotated in the point cloud data set to determine the kinematic quantities describing the motion states of each object in each point cloud data frame includes: If the kinematic quantity includes position information, traverse all frames of the point cloud data, record the position information marked for each object, and each piece of the position information is associated with the timestamp of the point cloud data of the corresponding frame.

[0008] In an alternative embodiment, the visual processing of the kinematic quantities of the same object in time sequence to obtain a visual result showing the change of the kinematic quantities of the same object includes: Generate a position-time line graph according to the position information of the same object in consecutive frames, where the horizontal axis of the position-time line graph represents time and the vertical axis represents position information.

[0009] In an alternative embodiment, the generating of the position-time line graph according to the position information of the same object in consecutive frames includes: If the position information includes position coordinates, generate a position-time line graph of the corresponding component according to the position coordinate components of the same object in consecutive frames.

[0010] In an alternative embodiment, the position analysis of the objects involved in the point cloud dataset to determine the motion state of each object in each point cloud data frame further includes: If the kinematic quantity includes velocity information, determine the velocity information of the same object according to the temporal change of the position information marked for the same object, and each piece of the position information is associated with the timestamp of the point cloud data of the corresponding frame.

[0011] In an alternative embodiment, the visual processing of the kinematic quantities of the same object in time sequence to obtain a visual result showing the change of the kinematic quantities of the same object includes: Generate a velocity-time line graph according to the velocity information of the same object in consecutive frames, where the horizontal axis of the velocity-time line graph represents time and the vertical axis represents velocity information.

[0012] In an alternative embodiment, the visual processing of the kinematic quantities of the same object in time sequence to obtain a visual result showing the change of the kinematic quantities of the same object for determining whether the point cloud annotation is accurate includes: the visual result includes a line graph. If no abnormal jump features appear on all the line graphs of the same object, it is determined that the point cloud annotation is accurate; otherwise, the point cloud annotation frame corresponding to the abnormal jump feature is abnormal.

[0013] The second aspect of the present application provides an apparatus for determining the accuracy of point cloud annotation, including: a determination module, configured to, after obtaining a point cloud data set with annotation results, parse the objects annotated in the point cloud data set and determine the kinematic quantities describing the motion states of each object in each point cloud data frame; a visualization module, configured to perform visualization processing on the kinematic quantities of the same object in time series to obtain a visualization result showing the change of the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

[0014] The third aspect of the present application provides a vehicle, including: a processor; and a memory storing executable code thereon, which, when executed by the processor, causes the processor to execute the method as described above.

[0015] The fourth aspect of the present application provides a computer-readable storage medium storing executable code thereon, which, when executed by a processor of a vehicle, causes the processor to execute the method as described above.

[0016] The fifth aspect of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to execute the method as described above.

[0017] The technical solution provided by the present application may include the following beneficial effects: In the annotation acceptance of point cloud data, the accuracy of the point cloud data annotation can be determined through visualizing the kinematic quantities of the same object in time series, and the visualization result is used to determine the accuracy of the point cloud data annotation, that is, the accuracy verification of the point cloud annotation result is realized. In particular, it is applicable to detecting the accuracy of the annotation result of distant point clouds, and this method is intuitive and simple, which can improve the probability of detecting abnormal point cloud annotations, and further ensure the reliability of the training data for autonomous driving.

[0018] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Brief Description of the Drawings

[0019] By describing the exemplary embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious, wherein, in the exemplary embodiments of the present application, the same reference numerals generally represent the same components.

[0020] Figure 1 is a flowchart of the method for determining the accuracy of point cloud annotation shown in the present application; Figure 2 is a position-time broken line graph of the target object in the longitudinal direction; Figure 3 is a position-time broken line graph of the target object in the horizontal direction; Figure 4 is another schematic flow chart for determining the accuracy of point cloud annotation shown in the present application; Figure 5 is a speed-time broken line graph of the target object; Figure 6 is a schematic structural diagram of the device for determining the accuracy of point cloud annotation shown in the present application; Figure 7 is a schematic structural diagram of the electronic device shown in the present application. Detailed implementation manners

[0021] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application will be more thorough and complete, and can fully convey the scope of the present application to those skilled in the art.

[0022] The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0023] It should be understood that although the terms "first", "second", "third", etc. may be used in the present application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality" means two or more unless otherwise specifically defined.

[0024] In the field of autonomous driving technology, the labeled point cloud data, as training data, can help understand and identify various traffic scenes and environmental information. Currently, the detection or verification of the labeled point cloud data is mainly by manual visual inspection. For long-distance point cloud data, the point cloud is relatively sparse, the point cloud contour of the object is relatively blurred, and problems such as non-fitting 3D boxes and large offsets of the center points often occur. Therefore, it is very necessary to determine whether the point cloud data annotation of long-distance targets is accurate.

[0025] Taking the labeled point cloud data as an example, through labeling, it is possible to determine which points in the three-dimensional point cloud data belong to specific target objects. For example, in the scenario of autonomous driving, different targets such as pedestrians, vehicles, buildings, roads, etc. in the point cloud can be distinguished, so as to be able to recognize and understand the objects in the scene, etc. However, lack of labeling accuracy may lead to subsequent target recognition errors or confusion in understanding, etc.

[0026] In view of the above problems, the present application provides a method for determining the accuracy of point cloud labeling, which can more intuitively determine the accuracy of point cloud data labeling through visualization results, further improve the probability of detecting abnormal point cloud labeling, and thus ensure the reliability of training data for autonomous driving.

[0027] The technical solution of the present application will be described in detail below with reference to the accompanying drawings.

[0028] Figure 1 It is a schematic flowchart of the method for determining the accuracy of point cloud labeling shown in the present application.

[0029] See Figure 1 , a method for determining the accuracy of point cloud labeling shown in the present application mainly includes steps S101 to S102.

[0030] Step S101: After obtaining a point cloud data set with labeling results, parse the objects labeled in the point cloud data set to determine the kinematic quantities describing the motion states of each object in each point cloud data frame.

[0031] Among them, the point cloud data set can be obtained by collecting a set of point cloud data frames in sequence according to the time series when the collection vehicle is driving at a constant speed on the road. Then, by labeling each point cloud data frame in the point cloud data set, a point cloud data set with labeling results can be obtained. In this embodiment, the labeling results may include the center position of the labeling box, and the center position of the labeling box can be represented by specific three-dimensional coordinate values (x, y, z). The position of the labeling box in the point cloud space can be basically located through the center position of the labeling box. For example, in the point cloud data of a city street, if a car is to be labeled, the center coordinates of the labeling box can represent the position of the car in the street space.

[0032] In at least one embodiment, the kinematic quantity includes position information, and step S101 may include: Traverse all point cloud data frames, record the position information labeled for each object, and each position information is associated with the timestamp of the corresponding point cloud data frame.

[0033] In this embodiment, each point cloud data frame collected by the acquisition vehicle has a uniquely corresponding timestamp, and the point cloud data frames can be arranged in time series according to the timestamps. For the same object, the relevant point cloud data frames are screened out, and the positions of the object in each frame are recorded in sequence according to the time sequence of each frame. Similarly, the position records corresponding to all objects can be obtained. In order to more accurately determine the situation of point cloud annotation for distant objects, one or more target objects can be selected according to actual needs, and then the positions corresponding to each target object are recorded.

[0034] Step S102: Perform visual processing on the kinematic quantities of the same object in time series to obtain a visual result showing the change of the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

[0035] After obtaining the kinematic quantities of the same object in time series in step S101, the positions of the object at different times can be intuitively seen on the chart in the form of a statistical chart. By observing the abnormally changing parts of the positions, it can be determined whether there are abnormalities. Compared with directly observing the point cloud annotation with the naked eye, observing from the chart is more intuitive and can improve the detection rate of point cloud annotation abnormalities.

[0036] In at least one embodiment, according to the position information of the same object on consecutive frames, a position-time line graph is generated, where the horizontal axis of the position-time line graph represents time, and the vertical axis represents position information. It should be noted that the horizontal axis of the position-time line graph represents time, and each frame corresponds to a moment. The horizontal axis can be a specific moment, or other forms such as frame ID to represent time, etc., specifically subject to the actual situation. This embodiment is for illustrative purposes only and is not the only representation.

[0037] Among them, the position information is generally identified by three-dimensional coordinates. When the acquisition vehicle is driving on the road, the movement of the acquisition vehicle generally changes on the horizontal plane. Assuming that the three coordinate axes of the geodetic coordinate system are the x-axis, y-axis, and z-axis, the relative change of the z-axis position component of the acquisition vehicle during driving is less than that of the y-axis and z-axis. In this embodiment, the x-axis and y-axis in the geodetic coordinate system can be used as the observation reference quantities for the position components. In the geodetic coordinate system, according to the relative position relationship between the driving acquisition vehicle and the target object, it can be converted to process with the position components of the target object.

[0038] In this embodiment, for multiple point cloud data frames related to the same object, each frame corresponds to a moment, and the three-dimensional coordinates of the object can be solved on this frame. Then, three corresponding position-time line graphs can be obtained from the three coordinate axes. Among them, the position change along the vertical axis is relatively small. In order to save computational effort, it can be omitted. Such as Figure 2 And such as Figure 3As shown, taking the coordinates of the x-axis as the longitudinal distance and the coordinates of the y-axis as the lateral distance, two corresponding position-time line graphs can be obtained. In at least one embodiment, if the position information includes position coordinates, according to the position coordinate components of the same object on consecutive frames, a position-time line graph of the corresponding component is generated. It should be noted that whether the x-axis corresponds to the longitudinal or lateral direction is only for better explaining this embodiment, and it depends on the actual situation. This embodiment is for illustrative purposes only and is not the only representation. Also, as a vector, if the change of the position point is used for calculation, the calculation amount is larger. Observing by each component, the calculation amount is relatively simple, reducing the calculation load.

[0039] Further, the visualization result includes a position-time line graph. If no abnormal jump feature appears on all the position-time line graphs of the same object, then it is determined that the point cloud annotation is accurate; otherwise, the point cloud annotation frame corresponding to the abnormal jump feature is abnormal. As Figure 2 shown, an abnormal situation of a sharp drop in the longitudinal distance appears around time 4.8. Then, the point cloud data frame corresponding to the moment when the abnormality appears is determined as the part to be excluded, so as to avoid using the corresponding abnormal point cloud annotation as training data, etc., which may reduce the accuracy of the training model or cause other negative interferences.

[0040] In the position-time line graph, the position change of the object can be shown. However, in fact, there may be a situation where the position change is small in the short term but the speed change is large. Therefore, it is possible that the point cloud annotation is abnormal but not shown in the position-time line graph. Therefore, further, a speed-time line graph can also be generated, and the speed-time line graph and the position-time line graph are combined to obtain a more accurate observation conclusion.

[0041] The method for determining the accuracy of point cloud annotation in the present application will be further described below. Refer to Figure 4 A method for determining the accuracy of point cloud annotation shown in an embodiment of the present application includes: Step S401: Traverse all point cloud data frames, record the position information annotated for each object, and each position information is associated with the timestamp of the corresponding point cloud data frame.

[0042] Among them, the specific process of step S401 is the same as at least one embodiment of the above step S101, and will not be elaborated in detail here.

[0043] Step S402: Determine the speed information of the same object according to the sequential change of the position information annotated for the same object, and each position information is associated with the timestamp of the corresponding point cloud data frame.

[0044] In this embodiment, for the same object, relevant point cloud data frames are screened out, and the positions of the object in each frame are recorded in sequence according to the time sequence of each frame. Similarly, the speed information can be calculated based on the position changes of the target object in different frames, and then the speed records corresponding to all required target objects can be obtained, which can enrich the data dimension for analysis.

[0045] Step S403: Generate a position-time line graph based on the position information of the same object in consecutive frames. Among them, the horizontal axis of the position-time line graph represents time, and the vertical axis represents position information.

[0046] Among them, the specific process of step S403 is the same as at least one embodiment of the above step S102, and will not be elaborated in detail here. Specifically, the position information of the same object is represented in terms of the longitudinal distance dimension and the lateral distance dimension, as Figure 2 and Figure 3 shown.

[0047] Step S404: Generate a speed-time line graph based on the speed information of the same object in consecutive frames. Among them, the horizontal axis of the speed-time line graph represents time, and the vertical axis represents speed information.

[0048] Among them, speed can be understood as a form of presentation of position change. In this embodiment, the speed does not distinguish speed components, and the scalar of the speed is directly used as the speed value in the graph. Based on the speed information of the same object in consecutive frames, the speed-time line graph of the object is obtained. The horizontal axis of this line graph can be a specific moment or other forms such as frame ID, depending on the actual situation. This embodiment is for illustrative purposes and is not the only representation. The vertical axis represents the magnitude of the speed, as Figure 5 shown.

[0049] Step S405: If no abnormal jump features appear on all the line graphs of the same object, it is determined that the point cloud annotation is accurate; otherwise, the point cloud annotation frame corresponding to the abnormal jump feature is abnormal.

[0050] Among them, the visualization results include the position-time line graph and the speed-time line graph. As Figure 2 shown, an abnormal situation of a sharp drop in the longitudinal distance (i.e., a jump) appears around time 4.8. Then, the point cloud data frame corresponding to the moment when the abnormality appears is determined as the part to be excluded. As Figure 5 shown, at this moment, it can also be found that there is a large jump in the speed change in the speed-time line graph. Between time 5.0 and 5.4, there are certain fluctuations in the speed change. At this time, the position change is not obvious, and the speed change has fluctuations, which can indicate that this time period can be the emergency braking stage.

[0051] It should be noted that in the embodiments of the present application, the position-time line chart and the speed-time line chart are jointly used as the basis for determining whether the point cloud annotation is accurate. When only the speed-time line chart exists, the lack of position information easily leads to unclear trajectories. Suppose object A starts from the coordinate (0, 0) and object B starts from the coordinate (5, 5). The speeds of both increase by 1 m / s per second for a period of time. As time goes by, the movement trajectory of object A extends to the right along the x-axis starting from zero, and the trajectory of object B extends from (5, 5) to the upper right (assuming the movement range only considers the two-dimensional plane). If only the speed change is concerned, the actual position and movement trajectory of the object cannot be determined from the speed-time line chart. For example, two objects may have the same speed change pattern but different initial positions, and their actual movement paths may be completely different. An object starting from the origin and an object starting 10 meters away from the origin, even if their speed changes are exactly the same, there is always a gap in their positions in space. Moreover, the road driving conditions are complex, with normal severe acceleration / deceleration situations. Obviously, it is impossible to identify whether the object is driving normally based on the single speed-time line chart (speed curve). Therefore, by jointly using the position-time line chart and the speed-time line chart, when there is an obvious jump in any of the line charts, such as Figure 2 a part to the right of the time 4.8, Figure 3 the right part of the time 5.6, Figure 5 the part to the right of the time 4.8 and the part where the speed exceeds 80 in

[0052] etc., through this setting, the abnormal parts in the curve can be more comprehensively found from the line chart, and the point cloud data frame corresponding to the found abnormal part is deleted.

[0052] Obviously, compared with the related art that directly observes the point cloud annotation with the naked eye, the embodiments of the present application convert to observing the position change and speed change corresponding to the point cloud annotation, and judge whether the distant point cloud annotation is correct by analyzing whether there is a jump in the line chart. This judgment method based on graphic features is more intuitive and accurate, and can improve the accuracy of distant point cloud annotation verification.

[0053] Corresponding to the foregoing method embodiments for implementing application functions, the present application also provides a device, an electronic device, and corresponding embodiments for determining the accuracy of point cloud annotation.

[0054] Figure 6 is a schematic structural diagram of the device for determining the accuracy of point cloud annotation shown in the present application.

[0055] See Figure 6, a device for determining the accuracy of point cloud annotation shown in this application, which includes a determination module 601 and a visualization module 602. The determination module 601 is used to parse the objects annotated in the point cloud data set after obtaining the point cloud data set with annotation results, and determine the kinematic quantities describing the motion states of each object in each point cloud data frame; the visualization module 602 is used to perform visual processing on the kinematic quantities of the same object in time series, and obtain a visualization result showing the change of the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

[0056] Furthermore, the determination module 601 includes a position recording unit. If the kinematic quantity includes position information, the position recording unit is used to traverse all frame point cloud data, record the position information annotated for each object, and each position information is associated with the time stamp of the corresponding frame point cloud data.

[0057] In addition, the visualization module 602 is used to generate a position-time line graph according to the position information of the same object in consecutive frames, where the horizontal axis of the position-time line graph represents time, and the vertical axis represents position information.

[0058] Even further, if the position information includes position coordinates, the visualization module 602 is used to generate a position-time line graph of the corresponding component according to the position coordinate components of the same object in consecutive frames.

[0059] Further, the determination module 601 further includes a speed recording unit. If the kinematic quantity includes speed information, the speed recording unit is used to determine the speed information of the same object according to the time series change of the position information annotated for the same object, and each position information is associated with the time stamp of the corresponding frame point cloud data.

[0060] In addition, the visualization module 602 is used to generate a speed-time line graph according to the speed information of the same object in consecutive frames, where the horizontal axis of the speed-time line graph represents time, and the vertical axis represents speed information.

[0061] Preferably, the visualization result includes a line graph, and the visualization module 602 can make the following determination: if no abnormal jump feature appears on all the line graphs of the same object, it is determined that the point cloud annotation is accurate, otherwise the point cloud annotation frame corresponding to the abnormal jump feature is abnormal.

[0062] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0063] Figure 7 It is a schematic structural diagram of an electronic device shown in this application.

[0064] See Figure 7, the electronic device 700 includes a memory 701 and a processor 702.

[0065] The processor 702 can be a Central Processing Unit (CPU), or can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0066] The memory 701 can include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, the ROM can store static data or instructions required by the processor 702 or other modules of the computer. The permanent storage device can be a readable and writable storage device. The permanent storage device can be a non-volatile storage device that does not lose the stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device uses a mass storage device (such as a magnetic or optical disk, flash memory) as the permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, optical drive). The system memory can be a readable and writable storage device or a volatile readable and writable storage device, such as dynamic random access memory. The system memory can store some or all of the instructions and data required by the processor during operation. In addition, the memory 701 can include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and magnetic disks and / or optical disks can also be used. In some embodiments, the memory 701 can include a removable storage device that is readable and / or writable, such as a compact disc (CD), read-only digital versatile disc (such as DVD-ROM, dual-layer DVD-ROM), read-only Blu-ray disc, ultra density disc, flash memory card (such as SD card, min SD card, Micro-SD card, etc.), magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and instantaneous electronic signals transmitted wirelessly or wiredly.

[0067] An executable code is stored on the memory 701, and when the executable code is processed by the processor 702, it can cause the processor 702 to execute some or all of the methods described above.

[0068] In addition, the method according to the present application can also be implemented as a computer program or a computer program product, which includes computer program code instructions for performing some or all of the steps in the above-mentioned method of the present application.

[0069] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium), on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or a server, etc.), the processor executes some or all of the steps of the above-mentioned method according to the present application.

[0070] The embodiments of the present application also provide a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method described in any one of the above embodiments.

[0071] The computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), solid state drive (SSD, Solid State Drives), or optical disc, etc. Among them, the random access memory may include resistive random access memory (ReRAM, Resistance RandomAccess Memory) and dynamic random access memory (DRAM, Dynamic Random Access Memory). The above serial numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0072] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be read-only memory, magnetic disk or optical disc, etc.

[0073] The embodiments of the present application have been described above. The above description is exemplary and not exhaustive, and is also not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for determining the accuracy of point cloud annotation, characterized in that: include: After obtaining the point cloud data set with the annotation results, the objects annotated in the point cloud data set are parsed to determine the kinematic quantities describing the motion state of each object in each point cloud data frame; The kinematic quantities of the same object are visualized in time series to obtain visualization results showing the changes in the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

2. The method for determining the accuracy of point cloud annotation according to claim 1, characterized in that: The parsing of the objects annotated in the point cloud data set to determine the kinematic quantities describing the motion state of each object in each point cloud data frame includes: If the kinematic quantity includes position information, the point cloud data of all frames are traversed to record the position information marked by each object, and each position information is associated with the timestamp of the point cloud data of the corresponding frame.

3. The method for determining the accuracy of point cloud annotation according to claim 2, characterized in that: The visualization processing of the kinematic quantities of the same object in time series to obtain a visualization result showing the change of the kinematic quantities of the same object includes: A position-time line graph is generated according to the position information of the same object in the continuous frames, wherein the horizontal axis of the position-time line graph represents time and the vertical axis represents position information.

4. The method for determining the accuracy of point cloud annotation according to claim 3, characterized in that: The step of generating a position-time line graph according to the position information of the same object in the continuous frames includes: If the position information includes position coordinates, a position-time line graph of corresponding components is generated according to position coordinate axis components of the same object in consecutive frames.

5. The method for determining the accuracy of point cloud annotation according to any one of claims 2 to 4, characterized in that: The performing of position analysis on the objects involved in the point cloud data set to determine the motion state of each object in each point cloud data frame also includes: If the kinematic quantity includes speed information, the speed information of the same object is determined according to the temporal changes of the position information marked by the same object, and each position information is associated with the timestamp of the point cloud data of the corresponding frame.

6. The method for determining the accuracy of point cloud annotation according to claim 5, characterized in that: The visualization processing of the kinematic quantities of the same object in time series to obtain a visualization result showing the change of the kinematic quantities of the same object includes: A speed-time line graph is generated according to the speed information of the same object in continuous frames, wherein the horizontal axis of the speed-time line graph represents time, and the vertical axis represents speed information.

7. The method for determining the accuracy of point cloud annotation according to claim 1, characterized in that: The visualization processing of the kinematic quantities of the same object in time series to obtain a visualization result showing the change of the kinematic quantities of the same object for determining whether the point cloud annotation is accurate includes: The visualization result includes a line graph. If no abnormal jump feature appears on all the line graphs under the same object, it is determined that the point cloud annotation is accurate. Otherwise, the point cloud annotation frame corresponding to the abnormal jump feature is abnormal.

8. A device for determining the accuracy of point cloud annotation, characterized in that: include: A determination module is used to parse the objects annotated in the point cloud data set after acquiring the point cloud data set with the annotation results, and determine the kinematic quantities describing the motion state of each object in each point cloud data frame; The visualization module is used to perform visualization processing on the kinematic quantities of the same object in time sequence to obtain visualization results showing the changes in the kinematic quantities of the same object, so as to determine whether the point cloud annotation is accurate.

9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium having executable codes stored thereon, which, when executed by a processor of a vehicle, causes the processor to execute the method according to any one of claims 1 to 7.

Images