A method and device for checking vehicle-road cooperative marking data and a storage medium

Abstract

The application provides a check method and device for vehicle-road cooperative marking data and a storage medium, wherein the method comprises the following steps: acquiring marking data, the marking data comprising continuous point cloud data which is marked by space and ID; determining all ID markings in the first frame point cloud data in the check window; tracking each ID marking by traversing all point cloud data in the check window; in response to the target ID marking not appearing in the nth frame point cloud data in the check window, judging the position of the space marking in the (n-1)th frame point cloud data based on the space marking corresponding to the target ID marking; and in response to any vertex of the space marking being located in the edge range of the (n-1)th frame point cloud data, determining that there is no missing marking. The application judges whether the missing reason of the ID object in the current frame is missing marking or driving out of the detection range of the detection radar by using the position of the space coordinates of the marking object in the last frame, and sets the check window to ensure the comprehensiveness of the missing marking check.

Description

Technical Field

[0001] This invention relates to the field of vehicle-road cooperative technology, and in particular to a method, device and storage medium for inspecting vehicle-road cooperative labeled data. Background Technology

[0002] When performing multi-target tracking in a vehicle-to-everything (V2X) system, radar point cloud data and image data need to be correlated through temporal alignment and spatial calibration. The image data comes from cameras installed on vehicles and / or roadsides. During spatial calibration, annotators can refer to information in the images to better label the point cloud data. However, even with this method, incorrect annotations still exist in the labeled point cloud data, causing significant inconvenience for subsequent model training. Therefore, it is often necessary to verify the labeled data before using it.

[0003] Currently, the methods for verifying labeled point cloud datasets only check for duplicate label IDs in the data. They cannot identify missing labels or incorrect labels in the images and can only rely on manual review. However, manual review is not only time-consuming and costly, but the review standards can also be inconsistent due to different reviewers, which is not conducive to using the data to train models later. Summary of the Invention

[0004] To address the aforementioned technical problems, in a first aspect of the present invention, a method for inspecting vehicle-road cooperative annotation data is proposed. The method includes acquiring annotation data, which comprises continuous point cloud data after spatial annotation and ID annotation; identifying all ID annotations in a first frame of point cloud data within an inspection window; traversing all point cloud data within the inspection window to track each ID annotation; in response to a target ID annotation not appearing in the nth frame of point cloud data within the inspection window, determining the position of the spatial annotation in the (n-1)th frame of point cloud data based on the spatial annotation corresponding to the target ID annotation; and in response to any vertex of the spatial annotation being located within the edge range of the (n-1)th frame of point cloud data, determining that no missing annotation exists.

[0005] In one or more embodiments, a method for inspecting vehicle-road cooperative annotation data according to the present invention further includes: sliding the inspection window backward by one frame; determining whether there is a new ID annotation in the first frame of point cloud data in the current inspection window compared to the previous frame of point cloud data; if so, traversing all point cloud data in the inspection window to track each new ID annotation; in response to the new ID annotation not appearing in the nth frame of point cloud data in the inspection window, determining the position of the spatial annotation in the (n-1)th frame of point cloud data based on the spatial annotation corresponding to the new ID annotation; in response to any vertex of the spatial annotation being located within the edge range of the (n-1)th frame of point cloud data, determining that there is no missing annotation.

[0006] In one or more embodiments, the size of the inspection window is determined based on the ratio of the longest time taken for the marked object to pass through the radar's scanning range to the radar's scanning cycle.

[0007] In one or more embodiments, a method for inspecting vehicle-road cooperative annotation data according to the present invention further includes: determining the displacement of all annotated objects in the (n-1)th frame point cloud data based on the spatial annotations in the point cloud data of the (n-1)th frame within the inspection window and the point cloud data of the previous frame; and taking the maximum displacement among all annotated objects as the edge range of the (n-1)th frame point cloud data.

[0008] In one or more embodiments, the method further includes performing a spatial annotation ground fit check before missing annotation checks, the ground fit check comprising: traversing the point cloud data of the consecutive multiple frames; traversing the ID annotations in the point cloud data of each frame and checking the bottom coordinates of the spatial annotations corresponding to each ID annotation; outputting the point cloud data of the corresponding frame and reporting an error in response to the Z-axis component of the bottom coordinates of the spatial annotations being non-zero; and performing the next check in response to all point cloud data passing the check.

[0009] In one or more embodiments, the method further includes performing an ID label duplication check before the spatial label ground fit check, the ID label duplication check comprising: traversing the point cloud data of the consecutive multiple frames; checking whether there are duplicate ID labels in the point cloud data of each frame; outputting the point cloud data of the corresponding frame and reporting an error in response to the existence of duplicate ID labels; and performing the next check in response to all point cloud data passing the check.

[0010] In one or more embodiments, the method further includes performing a coordinate anomaly check after the missing label check, the coordinate anomaly check including: fitting the trajectory of the labeled object based on the spatial labels in the point cloud data of multiple consecutive frames; predicting the spatial label of the same labeled object in the next frame of point cloud data based on the trajectory fitting result; obtaining the spatial label in the point cloud data of the next frame; determining whether the mean square error between the predicted coordinates and the actual coordinates of the spatial label of the same labeled object is greater than a preset value; if so, outputting the point cloud data of the corresponding frame and reporting an error.

[0011] In one or more embodiments, the method further includes performing an orientation check on the labeled object when orientation labels exist in the labeled data. The orientation check includes: classifying the orientation of the labeled object by label based on the orientation labels; label 0 represents the direction of movement of the labeled object towards the camera; label 1 represents the direction of movement of the labeled object away from the camera; label 2 represents the labeled object moving horizontally on the road with its head / face facing to the left; label 3 represents the labeled object moving horizontally on the road with its head / face facing to the right; in response to a change in the orientation label of the same labeled object in two adjacent frames of point cloud data, where label 0 changes to 1, 1 changes to 0, 2 changes to 3, or 3 changes to 2, the point cloud data of the corresponding frame is output and an error is reported.

[0012] In a second aspect of the invention, an inspection device for vehicle-road cooperative labeling data is provided, comprising: at least one processor; and a memory storing an executable computer program, which, when executed by the at least one processor, is used to implement the steps of the inspection method for vehicle-road cooperative labeling data as described in any one of claims 1-8.

[0013] In a third aspect of the invention, a readable storage medium is provided, comprising: an executable computer program, which, when executed by an executor, is used to implement the steps of the method for inspecting vehicle-road cooperative annotation data as described in any one of claims 1-8.

[0014] The beneficial effects of this invention include: This invention proposes using ID annotations to track annotated objects in consecutive data frames. When the ID annotation is absent in the nth frame, the position of the spatial annotation corresponding to the ID annotation in the previous frame (i.e., the (n-1)th frame) is used to determine whether the object's disappearance in the nth frame is due to a missed annotation or because it left the detection range of the corresponding radar (used to acquire point cloud data). The purpose of setting an inspection window is to define a data inspection range that includes all state data of a perceived object passing through a radar area, ensuring the comprehensiveness of the missed annotation check. Furthermore, based on the missed annotation check, this invention also proposes methods for checking other annotation problems. By utilizing the powerful computing capabilities of computers to rationally divide and optimize the inspection steps, the invention can effectively solve the aforementioned annotation problems, improving inspection efficiency while maintaining accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for inspecting vehicle-road cooperative annotation data according to the present invention.

[0017] Figure 2 This is a schematic diagram of the spatial labeling of the present invention;

[0018] Figure 3 This is a schematic diagram of the orientation markings of the present invention;

[0019] Figure 4 This is a schematic diagram of the structure of an inspection device for vehicle-road cooperative annotation data according to the present invention;

[0020] Figure 5 This is a schematic diagram of the structure of a readable storage medium according to the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0022] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.

[0023] Labeled point cloud data may contain issues such as duplicate IDs of different labeled objects, spatial labels not conforming to the ground, missing labels, spatial labeling errors, and incorrect vehicle heading labels. These problems can affect subsequent model training. To address these issues, this invention proposes a method for checking vehicle-road cooperative labeled data. Utilizing the powerful computing capabilities of computers, the checking steps are rationally divided and optimized, effectively resolving the aforementioned labeling problems and improving checking efficiency while maintaining accuracy. The method will be described in detail below with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart illustrating a method for checking vehicle-road cooperative annotation data according to the present invention. Figure 1 As shown, the workflow of the vehicle-road cooperative annotation data inspection method of the present invention includes the following steps executed by a computer: Step S1, acquiring annotation data, the annotation data including continuous point cloud data after spatial annotation and ID annotation; Step S2, determining all ID annotations in the first frame of point cloud data within the inspection window; Step S3, traversing all point cloud data within the inspection window to track each ID annotation; Step S4, in response to the target ID annotation not appearing in the nth frame of point cloud data within the inspection window, determining the position of the spatial annotation in the (n-1)th frame of point cloud data based on the spatial annotation corresponding to the target ID annotation; and Step S5, in response to any vertex of the spatial annotation being located within the edge range of the (n-1)th frame of point cloud data, determining that there are no missing annotations.

[0025] This implementation proposes a solution to the problem of missing labels. Specifically, the difficulty in detecting missing labels lies in how to identify them, that is, how to determine if a labeled object that should appear in a certain frame of point cloud data is missing, which is difficult to achieve manually. To this end, this embodiment proposes to use ID labels to track labeled objects in consecutive data frames. When the ID label is not present in the nth frame, the position of the spatial label corresponding to the ID label in the previous data frame, i.e., the (n-1)th frame, is used to determine whether the disappearance of the labeled object in the nth frame is due to missing labels or because it has moved out of the detection range of the corresponding detection radar (used to acquire point cloud data). Specifically, there are two scenarios. First, if the spatial label corresponding to the ID is within the edge range of the point cloud data in frame (n-1), then it's highly probable that its disappearance in frame n is due to it leaving the detection range of the corresponding radar (used to acquire point cloud data). In this case, it's considered that there are no missing labels. More specifically, this can be determined by checking if any vertex of the aforementioned spatial coordinates is within the edge range of the point cloud data in frame (n-1). Conversely, if none of the vertices of the aforementioned spatial label are within the edge range of the point cloud data in frame (n-1), then it's considered that there are missing labels. The spatial label consists of the coordinates of the eight vertices of a hexahedron that encloses the labeled object. The purpose of setting up the inspection window is to define a data inspection range that includes all the state data of a sensing object passing through a radar area, ensuring the comprehensiveness of the missing label check.

[0026] In a further embodiment, the missing label checking step of the present invention further includes: sliding the checking window backward by one frame; determining whether there is a new ID label in the first frame of point cloud data in the current checking window compared to the previous frame of point cloud data; if so, traversing all point cloud data in the checking window to track each new ID label; in response to the new ID label not appearing in the nth frame of point cloud data in the checking window, determining the position of the spatial label in the (n-1)th frame of point cloud data based on the spatial label corresponding to the new ID label; in response to any vertex of the spatial label being located within the edge range of the (n-1)th frame of point cloud data, determining that there is no missing label.

[0027] Specifically, point cloud data exhibits temporal continuity. In this invention, "previous frame" and "next frame" refer to the point cloud data at the previous moment and the point cloud data at the next moment, respectively. Therefore, sliding the inspection window backward by one frame means sliding the inspection window towards the point cloud data at the next moment to include the point cloud data at the next moment. As seen in the previous embodiment, the inspection window of this invention should contain all states of a sensing object within the sensing area of ​​a detection radar. Therefore, under normal circumstances, an ID label appearing in the first frame of the inspection window should also appear in the last frame of the inspection window. Furthermore, during this process, new ID objects may appear at any time in each frame of point cloud data after the first frame, meaning that a new sensing object enters the sensing area of ​​the detection radar. The inspection window can only guarantee that all states of ID objects in the first frame are included within the inspection window. Therefore, when it is necessary to check whether a new ID object has been missed, it is first necessary to slide the inspection window backward by one frame so that the new ID object appears in the first frame of the inspection window. In addition, in order to reduce the amount of computation, this embodiment will also determine whether there are new ID labels in the first frame of point cloud data in the current inspection window compared with the previous frame of point cloud data in order to find the newly appearing ID objects. Then, only the newly appearing ID objects (i.e. ID labels) are tracked, thereby improving the inspection efficiency.

[0028] More specifically, since the inspection window can ensure that the entire state of the ID objects in the first frame is included in the inspection window, the benefit of setting an inspection window can be understood as enabling only a limited number of ID objects to be tracked in a single missing label check, ensuring the comprehensiveness of the missing label check, reducing the system's operating burden, and improving computational efficiency.

[0029] In a further embodiment, the size of the inspection window is determined based on the ratio of the longest time taken for the labeled object to pass through the radar's scanning range to the radar's scanning cycle. Specifically, since the movement speed of the sensed objects varies in reality, in order to ensure that the entire movement state of the slowest sensed object is included within the inspection window, this invention determines the size of the inspection window based on the ratio of the longest time taken for the labeled object to pass through the radar's scanning range to the radar's scanning cycle. The size of the inspection window determines how many frames of point cloud data will be included within the inspection window.

[0030] In a further embodiment, the missing label checking step of the present invention further includes: determining the displacement of all labeled objects in the (n-1)th frame point cloud data based on the spatial annotations in the point cloud data of the (n-1)th frame within the inspection window and the point cloud data of the previous frame; and taking the maximum displacement among all labeled objects as the edge range of the (n-1)th frame point cloud data. Specifically, as can be seen from the principle explanation of the foregoing embodiments, the size of the edge range will affect the accuracy of the missing label judgment. Therefore, this embodiment uses the instantaneous displacement of the sensing object with the fastest moving speed at the current moment to determine the edge range, which can ensure that the distance moved by all sensing objects in the current data frame at the current moment within one radar detection cycle will not exceed the width of the edge range, so that as long as it has not entered the edge range, its ID label should definitely appear in the point cloud data of the next frame.

[0031] In an alternative embodiment, the edge range can be given in advance, such as based on the maximum driving speed allowed for that road segment.

[0032] In a preferred embodiment, as in the previous embodiment, the information is provided in real time at each site where a missing label needs to be determined, thereby ensuring the accuracy of the missing label check.

[0033] In one embodiment, the method of the present invention further includes performing a spatial annotation ground alignment check before missing annotation check. The ground alignment check includes: traversing point cloud data of multiple consecutive frames; traversing the ID annotations in the point cloud data of each frame and checking the bottom coordinates of the spatial annotations corresponding to each ID annotation; outputting the point cloud data of the corresponding frame and reporting an error in response to the Z-axis component of the bottom coordinates of the spatial annotations being non-zero; and performing the next check in response to all point cloud data passing the check.

[0034] For details, please see Figure 2 , Figure 2 This is a schematic diagram of the spatial annotation of the present invention. Each spatial annotation is a hexahedron composed of 8 spatial coordinate points and 6 annotation frames (rectangles), with the annotation object located within the hexahedron. Each spatial coordinate point is a three-dimensional spatial coordinate point (x, y, z), where the x-axis is parallel to the positive direction of road travel, the y-axis is collinear with the road width and perpendicular to the x-axis, and the z-axis points towards the sky. The 8 vertices are sequentially labeled as (x... i ,y i ,z i If z3 = z4 = z7 = z8 = 0, then there is no error; otherwise, the system will mark the current ID of the current data frame, output the current frame and report the error that the bottom of the annotation box does not fit the ground.

[0035] In one embodiment, the method of the present invention further includes performing an ID label duplication check before the spatial label ground fit check. The ID label duplication check includes: traversing point cloud data of multiple consecutive frames; checking whether there are duplicate ID labels in the point cloud data of each frame; outputting the point cloud data of the corresponding frame and reporting an error in response to the existence of duplicate ID labels; and performing the next check in response to all point cloud data passing the check.

[0036] In one embodiment, the method of the present invention further includes performing coordinate anomaly checking after missing label checking. The coordinate anomaly checking includes: fitting the trajectory of the labeled object based on the spatial labels in the point cloud data of multiple consecutive frames; predicting the spatial label of the same labeled object in the next frame of point cloud data based on the trajectory fitting result; obtaining the spatial label in the point cloud data of the next frame; determining whether the mean square error between the predicted coordinates and the actual coordinates of the spatial label of the same labeled object is greater than a preset value; if so, outputting the point cloud data of the corresponding frame and reporting an error.

[0037] Specifically, coordinate anomaly detection targets spatial annotations. For coordinate anomaly detection of the same ID object, this invention chooses not to process the first n frames (assuming that the coordinates of the annotated object are correct in the first n frames). Then, based on the first n coordinates, the predicted value of the corresponding coordinate in the next frame is predicted using trajectory linear fitting. An error threshold is set, and the mean square error between the predicted coordinate and the actual coordinate is calculated. When the mean square error is greater than the set threshold, the system can report a coordinate anomaly error.

[0038] In one embodiment, the method of the present invention further includes performing an orientation check on the labeled object when orientation labels exist in the labeled data. The orientation check on the labeled object includes: classifying the orientation of the labeled object by label based on the orientation labels; label 0 represents the direction of the labeled object moving towards the camera; label 1 represents the direction of the labeled object moving away from the camera; label 2 represents the labeled object moving horizontally on the road and the head / face of the object facing to the left; label 3 represents the labeled object moving horizontally on the road and the head / face of the object facing to the right; in response to a change in the orientation label of the same labeled object in the point cloud data of two adjacent frames, where label 0 to 1, 1 to 0, 2 to 3, or 3 to 2, the point cloud data of the corresponding frame is output and an error is reported.

[0039] For details, please see Figure 3 , Figure 3 This is a schematic diagram of the orientation marking of the present invention. The specific process of checking the orientation anomaly of the marked object includes: assuming the marked object is a vehicle, extracting the orientation label values ​​of vehicles with the same ID in consecutive frames and placing them in a list, and marking values ​​from [0,1,2,3] according to the vehicle's front orientation;

[0040] 0 represents the front of the car facing the camera (including all angles except 0 degrees and 180 degrees), such as southeast and south.

[0041] 1 represents the front of the car facing away from the camera (including all angles except 0 degrees and 180 degrees), such as northeast and north.

[0042] 2 represents a vehicle completely lying across the road, with its front (or face if it is a pedestrian) facing directly to the left;

[0043] 3 represents a vehicle completely lying across the road, with its front (or face if it is a pedestrian) facing directly to the right;

[0044] When a vehicle label in consecutive frames shows a vehicle's heading as 1 in the previous frame and 0 in the next frame (meaning the vehicle was facing the camera in the previous frame and facing away from the camera in the next), the likelihood of the vehicle's heading direction reversing directly between two frames is low due to the small time interval between adjacent frames. In this case, it is considered a labeling error in vehicle heading direction. Similarly, the probability of the vehicle's heading direction reversing left or right between two adjacent frames is also very small. In this case, the current frame needs to be output for subsequent manual review. The objects or labels mentioned above include pedestrians and / or vehicles.

[0045] In one embodiment, errors reported in each of the above inspection steps need to be manually verified and corrected in the output point cloud data frame.

[0046] As described in the above embodiments, this invention proposes five types of detection for data annotation. The above-described inspection order is a preferred inspection order scheme. Except for ID duplicate checking, which should be placed first, the other inspection orders can be arbitrarily changed, and the corresponding technical solutions should also be within the protection scope of this invention. Through the optimization of the inspection method and inspection order in the above embodiments, the method of this invention can improve the accuracy and efficiency of annotation inspection.

[0047] In a second aspect of the invention, an inspection device for vehicle-road cooperative annotation data is proposed. For example... Figure 4 As shown, Figure 4 This is a schematic diagram of a device for inspecting vehicle-road cooperative labeling data according to the present invention. The device includes: at least one processor 100; and a memory 200 storing an executable computer program. When executed by the at least one processor 100, the computer program is used to implement the steps of a method for inspecting vehicle-road cooperative labeling data in any of the above embodiments.

[0048] In a third aspect of the invention, a readable storage medium is provided. For example... Figure 5 As shown, Figure 5This is a schematic diagram of the structure of a readable storage medium according to the present invention. The readable storage medium 400 of the present invention includes: an executable computer program 401, which, when executed by an executor, is used to implement the steps of a method for checking vehicle-road cooperative annotation data in any of the above embodiments.

[0049] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.

[0050] It should be understood that, as used herein, the singular form “a” is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, “and / or” refers to any and all possible combinations of one or more of the associated listed items.

[0051] The embodiment numbers disclosed in the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0052] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A method for inspecting vehicle-road cooperative annotation data, characterized in that, The method includes: Acquire labeled data, which includes continuous point cloud data that has undergone spatial labeling and ID labeling; All ID labels in the first frame of point cloud data within the inspection window are identified, wherein the size of the inspection window is determined based on the ratio of the longest time taken for the labeled object to pass through the radar's scanning range to the radar's scanning cycle. Traverse all point cloud data within the inspection window to track each ID label; In response to the fact that the target ID label does not appear in the nth frame of point cloud data within the inspection window, the position of the spatial label in the (n-1)th frame of point cloud data is determined based on the spatial label corresponding to the target ID label; In response to any vertex of the spatial annotation being located within the edge range of the point cloud data of the (n-1)th frame, it is determined that there are no missing annotations.

2. The method for inspecting vehicle-road cooperative annotation data according to claim 1, characterized in that, The method further includes: Slide the inspection window back one frame; Determine whether the first frame of point cloud data in the current inspection window has a new ID label compared to the previous frame of point cloud data; Then, iterate through all point cloud data within the inspection window and track each new ID label; In response to the fact that the new ID label does not appear in the nth frame of point cloud data within the inspection window, the position of the spatial label in the (n-1)th frame of point cloud data is determined based on the spatial label corresponding to the new ID label. In response to any vertex of the spatial annotation being located within the edge range of the point cloud data of the (n-1)th frame, it is determined that there are no missing annotations.

3. The method for inspecting vehicle-road cooperative annotation data according to claim 2, characterized in that, The method further includes: The displacement of all labeled objects in the (n-1)th frame point cloud data is determined based on the spatial annotations in the point cloud data of the previous frame and the point cloud data of the inspection window. The maximum displacement among all labeled objects is taken as the edge range of the point cloud data in the (n-1)th frame.

4. The method for inspecting vehicle-road cooperative annotation data according to claim 1, characterized in that, The method further includes performing a spatial labeling ground alignment check prior to the missing label check, the ground alignment check including: Traverse the point cloud data of the consecutive frames; Iterate through the ID labels in the point cloud data of each frame and check the bottom coordinates of the spatial label corresponding to each ID label; In response to the Z-axis component of the bottom coordinate of the spatial annotation being non-zero, the point cloud data of the corresponding frame is output and an error is reported; In response to all point cloud data, the next check is performed.

5. The method for inspecting vehicle-road cooperative annotation data according to claim 4, characterized in that, The method further includes performing an ID label duplication check before the spatial label ground fit check, the ID label duplication check including: Traverse the point cloud data of the consecutive frames; Check if there are duplicate ID labels in the point cloud data of each frame; In response to the existence of duplicate ID labels, output the point cloud data of the corresponding frame and report an error; In response to all point cloud data, the next check is performed.

6. The method for inspecting vehicle-road cooperative annotation data according to claim 1, characterized in that, The method further includes performing a coordinate anomaly check after the missing marker check, the coordinate anomaly check including: Trajectory fitting of labeled objects based on spatial annotations in continuous multi-frame point cloud data; Predict spatial annotations for the same labeled object in the next frame of point cloud data based on trajectory fitting results; Obtain the spatial annotations in the point cloud data of the next frame; Determine whether the mean square error between the predicted coordinates and the actual coordinates of the spatial annotation of the same labeled object is greater than a preset value; If so, output the point cloud data of the corresponding frame and report an error.

7. The method for inspecting vehicle-road cooperative annotation data according to claim 1, characterized in that, The method further includes performing an orientation check on the labeled object when orientation labels exist in the labeled data, wherein the orientation check on the labeled object includes: Classify the orientation of the labeled objects based on the orientation annotations; Label 0 indicates that the object is moving towards the camera; Label 1 indicates that the object is moving away from the camera; Label 2 indicates that the object is lying horizontally on the road, and the head / face of the object is facing directly to the left; Label 3 indicates that the object is lying horizontally on the road, and the head / face of the object is facing directly to the right; In response to a change in the orientation label of the same labeled object in two adjacent frames of point cloud data (label 0 to 1, 1 to 0, 2 to 3, or 3 to 2), the point cloud data of the corresponding frame is output and an error is reported.

8. An inspection device for vehicle-road cooperative annotation data, characterized in that, include: At least one processor; as well as A memory storing an executable computer program, which, when executed by the at least one processor, is used to implement the steps of the method for checking vehicle-road cooperative annotation data as described in any one of claims 1-7.

9. A readable storage medium, characterized in that, include: An executable computer program, when executed by an executor, is used to implement the steps of the method for checking vehicle-road cooperative annotation data as described in any one of claims 1-7.

Images