A dynamic target labeling method and related device

Abstract

The application discloses a dynamic target labeling method and related device, and carries out timestamp alignment on millimeter wave radar data sets and camera image data sets obtained in a preset time period to obtain target millimeter wave radar data sets and target camera image data sets, screens out potential millimeter wave radar dynamic point cloud data from a point attribute data set, and projects the potential millimeter wave radar dynamic point cloud data and non-potential millimeter wave radar dynamic point cloud data to a point cloud interface to obtain point cloud interface data, converts the point cloud position data set and the potential millimeter wave radar dynamic point cloud data to a camera coordinate system, and projects the potential millimeter wave radar dynamic point cloud data and the target camera image data set to an image interface after fusion to obtain image interface data, and adopts an interframe switching function to determine each potential millimeter wave radar dynamic point in which a position changes in the point cloud interface data and the image interface data for labeling. The application utilizes the principle that the time sequence change of multiple frames of point cloud data and the appearance characteristics of dynamic targets can be recognized by images, and realizes accurate calibration of dynamic targets.

Description

Technical Field

[0001] This invention relates to the field of data annotation technology, and more specifically, to a method and related apparatus for annotating dynamic targets. Background Technology

[0002] In the field of autonomous driving, millimeter-wave radar is widely used for target detection and tracking due to its excellent anti-interference capabilities and all-weather operation. Existing millimeter-wave radars primarily rely on single-frame point cloud data for target annotation. However, the resolution of single-frame point cloud data generated by current millimeter-wave radars is relatively low. Affected by noise and missing information, traditional single-frame point cloud data easily overlooks target motion trajectories, making it difficult to annotate dynamic targets in dynamic scenes. This impacts the intelligent perception performance of millimeter-wave radar in applications such as autonomous driving. Summary of the Invention

[0003] In view of this, the present invention discloses a method and related apparatus for labeling dynamic targets, so as to realize the labeling of dynamic targets in dynamic scenes and improve the intelligent perception effect of millimeter-wave radar in applications such as autonomous driving.

[0004] A method for labeling dynamic targets, comprising:

[0005] The millimeter-wave radar dataset and camera image dataset acquired within a preset time period are timestamped to obtain the target millimeter-wave radar dataset and the target camera image dataset. The target millimeter-wave radar dataset includes: a point cloud location dataset and a point attribute dataset.

[0006] Dynamic point cloud data for potential millimeter-wave radar are filtered from the point attribute dataset;

[0007] The potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset are both projected onto the point cloud interface to obtain point cloud interface data.

[0008] The point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system, and then fused with the target camera image dataset and projected onto the image interface to obtain the image interface data.

[0009] The inter-frame switching function is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data, and each potential millimeter-wave radar dynamic point is labeled as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point labeling.

[0010] Optionally, the step of aligning the millimeter-wave radar dataset and camera image dataset acquired within a preset time period with timestamps to obtain the target millimeter-wave radar dataset and target camera image dataset includes:

[0011] Within the preset time period, each frame of millimeter-wave radar data with a timestamp is acquired from the millimeter-wave radar and sequentially placed into the millimeter-wave radar data queue to obtain the millimeter-wave radar dataset.

[0012] Within the preset time period, each frame of camera image data with a timestamp is acquired from the camera sensor and sequentially placed into the camera image data queue to obtain the camera image dataset;

[0013] Read each frame of millimeter-wave radar data from the millimeter-wave radar dataset and each frame of camera image data from the camera image dataset in sequence;

[0014] For each frame of millimeter-wave radar data, a linear interpolation method is used to determine the target frame camera image data with the corresponding timestamp from all the camera image data of each frame, so as to obtain the target millimeter-wave radar dataset and the target camera image dataset with timestamp alignment.

[0015] Optionally, the step of filtering potential millimeter-wave radar dynamic point cloud data from the point attribute dataset includes:

[0016] Read the point attribute data of each frame in the point attribute dataset sequentially;

[0017] Determine the Euclidean distance and velocity of the attribute data for each frame point;

[0018] Filter out all target frame point attribute data from the point attribute dataset whose Euclidean distance is not greater than a distance threshold and whose velocity is not less than a velocity threshold;

[0019] The millimeter-wave radar points corresponding to the attribute data of each target frame point are used to form the potential millimeter-wave radar dynamic point cloud data.

[0020] Optionally, the step of transforming both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, fusing them with the target camera image dataset, and then projecting them onto the image interface to obtain image interface data includes:

[0021] Based on the camera intrinsic parameter matrix and the extrinsic parameter matrix from the millimeter-wave radar coordinate system to the camera coordinate system, the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system to obtain the target point cloud location dataset and the target potential millimeter-wave radar dynamic point cloud data.

[0022] In the camera coordinate system, the target point cloud location dataset, the target potential millimeter-wave radar dynamic point cloud data, and the target camera image dataset are fused to obtain fused data;

[0023] The fused data is projected onto the image interface to obtain the image interface data.

[0024] Optionally, the step of using inter-frame switching to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data, and labeling each potential millimeter-wave radar dynamic point as a target millimeter-wave radar dynamic point to obtain target millimeter-wave radar point cloud data with dynamic point labeling, includes:

[0025] The point cloud interface data of each frame in the point cloud interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame point cloud interface data, and to obtain the first millimeter-wave radar dynamic point set.

[0026] The image interface data of each frame in the image interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame image interface data, and to obtain the second millimeter-wave radar dynamic point set.

[0027] Identify each identical potential millimeter-wave radar dynamic point in the first millimeter-wave radar dynamic point set and the second millimeter-wave radar dynamic point set.

[0028] Each of the same potential millimeter-wave radar dynamic points is labeled as a target millimeter-wave radar dynamic point, resulting in the target millimeter-wave radar point cloud data after dynamic point labeling.

[0029] Optionally, the step of dynamically switching the point cloud interface data of each frame in the point cloud interface data using the inter-frame switching function includes:

[0030] When a forward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the previous frame point cloud interface data.

[0031] When a backward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the next frame point cloud interface data.

[0032] Optionally, the step of dynamically switching the image interface data of each frame in the image interface data using the inter-frame switching function includes:

[0033] When a forward visualization data signal is received, the current frame image interface data in the image interface data is switched to the previous frame image interface data;

[0034] When a backward visualization data signal is received, the current frame image interface data in the image interface data is switched to the next frame image interface data.

[0035] A dynamic target annotation device, comprising:

[0036] The timestamp alignment unit is used to align the millimeter-wave radar dataset and camera image dataset acquired within a preset time period to obtain the target millimeter-wave radar dataset and the target camera image dataset. The target millimeter-wave radar dataset includes: a point cloud location dataset and a point attribute dataset.

[0037] A filtering unit is used to filter out potential millimeter-wave radar dynamic point cloud data from the point attribute dataset;

[0038] The first projection unit is used to project both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset onto the point cloud interface to obtain point cloud interface data.

[0039] The second projection unit is used to convert the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, and then project them onto the image interface after fusing them with the target camera image dataset to obtain image interface data.

[0040] The dynamic point annotation unit is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data by using the inter-frame switching function, and to annotate each potential millimeter-wave radar dynamic point as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point annotation.

[0041] A computer storage medium storing at least one instruction that, when executed by a processor, implements the dynamic target annotation method described above.

[0042] An electronic device, the electronic device comprising: a memory and a processor;

[0043] The memory is used to store at least one instruction;

[0044] The processor is used to execute the at least one instruction to implement the dynamic target annotation method described above.

[0045] As can be seen from the above technical solution, the present invention discloses a method and related apparatus for labeling dynamic targets. It aligns the timestamps of millimeter-wave radar datasets and camera image datasets acquired within a preset time period to obtain target millimeter-wave radar datasets and target camera image datasets. Potential millimeter-wave radar dynamic point cloud data is selected from the point attribute dataset. Both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset are projected onto a point cloud interface to obtain point cloud interface data. Both the point cloud position dataset and the potential millimeter-wave radar dynamic point cloud data are transformed into the camera coordinate system and fused with the target camera image dataset before being projected onto an image interface to obtain image interface data. An inter-frame switching function is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and image interface data. The target millimeter-wave radar dynamic points are then identified and labeled, thereby obtaining the target millimeter-wave radar point cloud data with dynamic point labeling. This application processes millimeter-wave radar datasets and camera image datasets as a whole. It utilizes the principle that multi-frame point cloud data in the millimeter-wave radar dataset can capture the time-series changes reflected by the dynamic changes of dynamic targets in space, and that images in the camera image dataset can identify the appearance features of dynamic targets. This allows for the simultaneous processing of fine features of static targets and motion trajectories of dynamic targets. It ensures accurate calibration of dynamic targets even when the point cloud is sparse, resulting in missing information or noise. This improves the intelligent perception effect of millimeter-wave radar in applications such as autonomous driving and better adapts to complex real-world scenarios. Attached Figure Description

[0046] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the published drawings without creative effort.

[0047] Figure 1 This is a flowchart of a dynamic target annotation method disclosed in an embodiment of the present invention;

[0048] Figure 2 This is a flowchart of a method for aligning timestamps of a millimeter-wave radar dataset and a camera image dataset, as disclosed in an embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram of an image interface and a point cloud interface before annotation, as disclosed in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of an annotated image interface and a point cloud interface disclosed in an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of the structure of a dynamic target annotation device disclosed in an embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] This invention discloses a method and related apparatus for labeling dynamic targets. It processes millimeter-wave radar datasets and camera image datasets as a whole. It utilizes the principle that multi-frame point cloud data in the millimeter-wave radar dataset can capture the time-series changes reflected by the dynamic changes of dynamic targets in space, and that images in the camera image dataset can identify the appearance features of dynamic targets. It can simultaneously process the fine features of static targets and the motion trajectory of dynamic targets, ensuring accurate labeling of dynamic targets even when the point cloud is sparse, resulting in missing information or noise. This improves the intelligent perception effect of millimeter-wave radar in applications such as autonomous driving and better adapts to complex real-world scenarios.

[0055] See Figure 1 The present invention discloses a flowchart of a dynamic target annotation method, which includes:

[0056] Step S101: Timestamp alignment is performed on the millimeter-wave radar dataset and camera image dataset acquired within a preset time period to obtain the target millimeter-wave radar dataset and the target camera image dataset.

[0057] It should be noted that the millimeter-wave radar dataset and camera image dataset obtained in this application are data acquired within the same preset time period. The value of the preset time period depends on actual needs, such as 1 hour, and this application does not limit it here.

[0058] In practical applications, millimeter-wave radar datasets are acquired through millimeter-wave radar installed on the vehicle, and camera image datasets are acquired through camera sensors installed on the same vehicle.

[0059] Because millimeter-wave radar and camera sensors typically operate on different principles and have different sampling frequencies, the data they acquire may differ at different points in time. This application aims to ensure that the millimeter-wave radar dataset and the camera image dataset are synchronized in time, enabling more accurate association and fusion of the two datasets. This is achieved by aligning the timestamps of the millimeter-wave radar dataset and the camera image dataset, adjusting the data from these different points in time to the same time reference, thereby ensuring data synchronization.

[0060] The target millimeter-wave radar dataset includes: point cloud location dataset and point attribute dataset.

[0061] Point cloud location datasets refer to the location information of target objects detected by millimeter-wave radar in three-dimensional space.

[0062] A point attribute dataset refers to the specific attributes and parameters associated with each detection point (or target point, reflection point) detected and acquired by a millimeter-wave radar system. Examples include ID number, lateral distance, longitudinal distance, lateral velocity, longitudinal velocity, label, angular velocity, length, and width.

[0063] Step S102: Filter out potential millimeter-wave radar dynamic point cloud data from the point attribute dataset.

[0064] In this application, potential millimeter-wave radar dynamic point cloud data refers to data that has a high probability of becoming a target millimeter-wave radar dynamic point.

[0065] Step S103: Project both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset onto the point cloud interface to obtain the point cloud interface data.

[0066] After filtering out the potential millimeter-wave radar dynamic point cloud data from the point attribute dataset, the remaining data is the non-potential millimeter-wave radar dynamic point cloud data, which is actually static point cloud data.

[0067] Step S104: Convert both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, fuse them with the target camera image dataset, and project them onto the image interface to obtain image interface data.

[0068] In this application, both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are in the millimeter-wave radar coordinate system, while the target camera image dataset is in the camera coordinate system. Therefore, before data fusion, the data needs to be transformed into the same coordinate system. In this embodiment, both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are transformed into the camera coordinate system, and then fused with the target camera image dataset in the camera coordinate system to obtain fused data. The fused data is then projected onto the image interface to obtain the image interface data.

[0069] The point cloud data displayed in this application contains positional information in three-dimensional space. Therefore, the point cloud interface provides a display and interaction of three-dimensional information. The shape, structure and positional information of an object can be seen intuitively through the point cloud interface.

[0070] The image data displayed in the image interface is two-dimensional. Therefore, the image interface provides the display and interaction of two-dimensional information. Through the image interface, you can view the color, texture, shape and other features in the image.

[0071] Step S105: Use the inter-frame switching function to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data, and label each potential millimeter-wave radar dynamic point as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point labeling.

[0072] In practical applications, this application employs an inter-frame switching function, utilizing the time-series changes reflected in multiple frames of point cloud data in the point cloud interface data, as well as the high-resolution details such as image textures provided in the image interface data, to achieve accurate annotation of dynamic points of the target millimeter-wave radar. Therefore, by fusing multi-frame point cloud data and image information, this application can simultaneously process the fine features of static objects and the motion trajectory of dynamic targets, improving the point cloud annotation effect of millimeter-wave radar in dynamic environments and enhancing the robustness and reliability of the system in dynamic target detection and tracking. Furthermore, it can effectively annotate whether the target is stationary or moving, thereby improving the detection and annotation capabilities of millimeter-wave radar point clouds in complex dynamic scenes.

[0073] In summary, this invention discloses a method for labeling dynamic targets. It involves aligning the timestamps of millimeter-wave radar datasets and camera image datasets acquired within a preset time period to obtain target millimeter-wave radar datasets and target camera image datasets. Potential millimeter-wave radar dynamic point cloud data is selected from the point attribute dataset. Both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data from the point attribute dataset are projected onto a point cloud interface to obtain point cloud interface data. Both the point cloud position dataset and the potential millimeter-wave radar dynamic point cloud data are transformed into the camera coordinate system and fused with the target camera image dataset before being projected onto an image interface to obtain image interface data. An inter-frame switching function is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and image interface data. The target millimeter-wave radar dynamic points are then identified and labeled, thereby obtaining the dynamically labeled target millimeter-wave radar point cloud data. This application processes millimeter-wave radar datasets and camera image datasets as a whole. It utilizes the principle that multi-frame point cloud data in the millimeter-wave radar dataset can capture the time-series changes reflected by the dynamic changes of dynamic targets in space, and that images in the camera image dataset can identify the appearance features of dynamic targets. This allows for the simultaneous processing of fine features of static targets and motion trajectories of dynamic targets. It ensures accurate calibration of dynamic targets even when the point cloud is sparse, resulting in missing information or noise. This improves the intelligent perception effect of millimeter-wave radar in applications such as autonomous driving and better adapts to complex real-world scenarios.

[0074] Furthermore, unlike traditional methods that rely solely on single-frame or point cloud data, this application can better handle moving targets in dynamic scenes, improving annotation accuracy and system robustness. This combined approach is not only suitable for dynamic target detection but also enhances the system's ability to handle complex scenes such as fast-moving objects and occlusion, significantly improving the perception performance of millimeter-wave radar point clouds in applications such as autonomous driving.

[0075] In one embodiment, see Figure 2 The flowchart of a method for timestamp alignment of millimeter-wave radar datasets and camera image datasets disclosed in this embodiment of the invention, specifically step S101 may include:

[0076] Step S201: Within a preset time period, acquire each frame of millimeter-wave radar data with a timestamp from the millimeter-wave radar and put it into the millimeter-wave radar data queue in sequence to obtain the millimeter-wave radar dataset.

[0077] In practical applications, an empty millimeter-wave radar data queue and an empty camera image data queue can be created. The length of the queue depends on the actual needs, such as 20, but this application does not limit it.

[0078] By putting each frame of millimeter-wave radar data with a timestamp obtained from the millimeter-wave radar into the created millimeter-wave radar data queue, a millimeter-wave radar data set can be obtained.

[0079] Step S202: Within a preset time period, obtain each frame of camera image data with a timestamp from the camera sensor and sequentially put it into the camera image data queue to obtain a camera image data set.

[0080] By putting each frame of camera image data with a timestamp obtained from the camera sensor into the camera image data queue, a camera image data set can be obtained.

[0081] Step S203: Sequentially read each frame of millimeter-wave radar data in the millimeter-wave radar data set and each frame of camera image data in the camera image data set.

[0082] Among them, sequentially reading each frame of millimeter-wave radar data in the millimeter-wave radar data set is actually sequentially reading each frame of millimeter-wave radar data in the millimeter-wave radar data queue.

[0083] Similarly, sequentially reading each frame of camera image data in the camera image data set is actually sequentially reading each frame of camera image data in the camera image data queue.

[0084] Step S204: For the timestamp of each frame of millimeter-wave radar data, use the linear interpolation method to determine the target frame of camera image data with the corresponding timestamp from all the frames of camera image data, and obtain the target millimeter-wave radar data set and the target camera image data set with aligned timestamps.

[0085] Specifically, sequentially read each frame of millimeter-wave radar data in the millimeter-wave radar data queue, and record the timestamp of the current frame of millimeter-wave radar data as Tr. Then sequentially read each frame of camera image data in the camera image data queue, record the timestamp of the current frame of camera image data as Tc1, and record the timestamp of the next frame of camera image data as Tc2.

[0086] The process of using the linear interpolation method is as follows:

[0087] When Tc1 < Tr < Tc2, calculate the absolute value of the difference between Tc1 - Tr as a1, and calculate the absolute value of the difference between Tc2 - Tr as a2.

[0088] When a1 < a2, change the value of Tc1 to Tr;

[0089] When a1 > a2, change the value of Tc2 to Tr.

[0090] Obtain the camera image data corresponding to the millimeter-wave radar data with the timestamp Tr.

[0091] The millimeter-wave radar data and camera image data are all stored with timestamps of Tr.

[0092] Following the timestamp alignment method described above, timestamp-aligned target millimeter-wave radar datasets and target camera image datasets are obtained. The target millimeter-wave radar dataset is a collection of multiple timestamp-aligned millimeter-wave radar data, and the target camera image dataset is a collection of multiple timestamp-aligned camera image data.

[0093] It should be noted that each millimeter-wave radar data in the target millimeter-wave radar dataset includes: point cloud location data and point attribute data. The timestamps of the point cloud location data and point attribute data are the same as the timestamps of the corresponding millimeter-wave radar data. For example, the x-axis and y-axis data of all points in the millimeter-wave radar data with timestamp Tr constitute the point cloud location data of the millimeter-wave radar with timestamp Tr.

[0094] In one embodiment, step S102 may specifically include:

[0095] (1) Read the point attribute data of each frame in the point attribute dataset in sequence.

[0096] (2) Determine the Euclidean distance and velocity of the point attribute data for each frame.

[0097] (3) Filter out all target frame point attribute data whose Euclidean distance is not greater than the distance threshold and whose velocity is not less than the velocity threshold from the point attribute dataset;

[0098] (4) The millimeter-wave radar points corresponding to the attribute data of each target frame point are used to form the potential millimeter-wave radar dynamic point cloud data.

[0099] The value of the distance threshold is determined according to actual needs, and this invention does not limit it.

[0100] Specifically, each frame of point attribute data Pi in the point attribute dataset is read sequentially, and the x-axis position data of each frame of point attribute data Pi is defined as Pix and the y-axis position data as Piy.

[0101] The purpose of distance threshold filtering is to remove noise points that are far from the millimeter-wave radar and reduce data redundancy.

[0102] Set distance thresholds thre_x for the x-axis and thre_y for the y-axis. If the x-axis position data Pix of point attribute data Pi exceeds the distance threshold thre_x, and / or the y-axis position data Piy of point attribute data Pi exceeds the distance threshold thre_y, discard the millimeter-wave radar points corresponding to these point attribute data that exceed the distance thresholds thre_x for the x-axis and / or thre_y for the y-axis.

[0103] The principle of velocity threshold filtering is as follows: points that may be dynamic are selected based on velocity attributes and designated as potential dynamic points for millimeter-wave radar. The velocity value Vp of the millimeter-wave radar point is calculated based on its x-axis and y-axis velocities and compared with the velocity Vb of the previous frame's millimeter-wave radar point cloud, which serves as the velocity threshold. If Vp exceeds Vb, the millimeter-wave radar point is considered a potential dynamic point for millimeter-wave radar.

[0104] In one embodiment, step S104 may specifically include:

[0105] (1) Based on the camera intrinsic parameter matrix and the extrinsic parameter matrix from the millimeter-wave radar coordinate system to the camera coordinate system, the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system to obtain the target point cloud location dataset and the target potential millimeter-wave radar dynamic point cloud data.

[0106] (2) In the camera coordinate system, the target point cloud location dataset, the target potential millimeter-wave radar dynamic point cloud data and the target camera image dataset are fused to obtain fused data.

[0107] (3) Project the fused data onto the image interface to obtain the image interface data.

[0108] Specifically, the system sequentially reads each frame of millimeter-wave radar point cloud location data from the point cloud location dataset, each frame of potential millimeter-wave radar dynamic points from the potential millimeter-wave radar dynamic point cloud data, and each frame of target camera image data from the target camera image dataset.

[0109] Let Ci be the current frame millimeter-wave radar point cloud location data, Fdi be the current frame potential millimeter-wave radar dynamic point, and Ii be the current frame camera image data.

[0110] Based on the camera intrinsic parameter matrix Ri and the extrinsic parameter matrix Te from the millimeter-wave radar coordinate system to the camera coordinate system, Ci and Fdi are transformed into the camera coordinate system to obtain the target point cloud location dataset and the target potential millimeter-wave radar dynamic point cloud data.

[0111] In the camera coordinate system, the target point cloud location dataset, the target potential millimeter-wave radar dynamic point cloud data, and the target camera image dataset are fused to obtain fused data, and the fused data is projected onto the image interface to obtain image interface data.

[0112] The dynamic point cloud data of potential millimeter-wave radar and the dynamic point cloud data of non-potential millimeter-wave radar in the point attribute dataset are both projected onto the point cloud interface to obtain the point cloud interface data.

[0113] See Figure 3 The diagram shows the image interface and point cloud interface. In the image interface data displayed, black represents static points and green represents the selected potential millimeter-wave radar dynamic points.

[0114] Similarly, in the point cloud interface data displayed, black represents static points, and green represents the selected potential dynamic points for millimeter-wave radar.

[0115] In practical applications, a rectangular box can be drawn on the graphical interface for each potential millimeter-wave radar dynamic point to visually identify these potential millimeter-wave radar dynamic points.

[0116] In one embodiment, step S105 may specifically include:

[0117] (1) Use the inter-frame switching function to dynamically switch the point cloud interface data of each frame in the point cloud interface data, determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame point cloud interface data, and obtain the first millimeter-wave radar dynamic point set.

[0118] (2) Use the inter-frame switching function to dynamically switch the image interface data of each frame in the image interface data, determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame image interface data, and obtain the second millimeter-wave radar dynamic point set.

[0119] (3) Determine each identical potential millimeter-wave radar dynamic point in the first millimeter-wave radar dynamic point set and the second millimeter-wave radar dynamic point set.

[0120] (4) Each of the same potential millimeter-wave radar dynamic points is labeled as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point labeling.

[0121] The inter-frame switching function refers to the switching between adjacent frames, such as switching from the current frame to the previous frame, or switching from the current frame to the next frame.

[0122] This application significantly improves the accuracy of dynamic target calibration by comparing a first set of dynamic millimeter-wave radar points determined based on point cloud interface data with a second set of dynamic millimeter-wave radar points determined based on image interface data, and using the same potential dynamic millimeter-wave radar points from both sets as the final target dynamic millimeter-wave radar points. This enhances the intelligent perception capabilities of millimeter-wave radar in applications such as autonomous driving, enabling it to better adapt to complex real-world scenarios.

[0123] In one embodiment, the process of dynamically switching between frames of point cloud interface data using the inter-frame switching function may include:

[0124] When a forward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the previous frame point cloud interface data.

[0125] When a backward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the next frame point cloud interface data.

[0126] Suppose that the user generates a forward visualization data signal by pressing the "s" key on the touch keyboard. After the user generates the forward visualization data signal by pressing the "s" key on the touch keyboard, the current frame point cloud interface data in the point cloud interface data can be switched to the previous frame point cloud interface data.

[0127] By generating a backward visualization data signal using the "d" key on the touch keyboard, the current frame point cloud interface data in the point cloud interface data can be switched to the next frame point cloud interface data.

[0128] In practical applications, the "s" and "d" keys on the keyboard can be used to switch between the previous and next frame of point cloud interface data, enabling switching between multiple frames of point cloud interface data. This allows for the identification of potential millimeter-wave radar dynamic points whose positions change within adjacent frames of point cloud interface data, thus obtaining the first set of millimeter-wave radar dynamic points.

[0129] In one embodiment, the process of dynamically switching between frames of image interface data using inter-frame switching functionality may include:

[0130] When a forward visualization data signal is received, the current frame image interface data in the image interface data is switched to the previous frame image interface data;

[0131] When a backward visualization data signal is received, the current frame image interface data in the image interface data is switched to the next frame image interface data.

[0132] Similarly, assuming that the user generates a forward visualization data signal by pressing the "s" key on the touch keyboard, the current frame image interface data in the image interface data can be switched to the previous frame image interface data after the "s" key on the touch keyboard generates the forward visualization data signal.

[0133] By generating a backward visualization data signal using the "d" key on the touch keyboard, the current frame image interface data in the image interface data can be switched to the next frame image interface data.

[0134] In practical applications, the "s" and "d" keys on the keyboard can be used to switch between the previous and next frame of image interface data, enabling switching between multiple frames of image interface data. This allows for the identification of potential millimeter-wave radar dynamic points whose positions change within adjacent frames of image interface data, thus obtaining a second set of millimeter-wave radar dynamic points.

[0135] To facilitate user viewing, each identical potential millimeter-wave radar dynamic point in the first and second millimeter-wave radar dynamic point sets can be labeled.

[0136] See Figure 4 As shown, the annotation process can be as follows:

[0137] Point selection and annotation: Select a point in the point cloud interface or image interface by using "Shift + left mouse button". The selected point is displayed in red, indicating that it is marked as a target millimeter-wave radar dynamic point.

[0138] To unlabel a point: Click the selected red dot with "Shift + left mouse button". The dot will turn black, indicating that the label has been removed.

[0139] After all target millimeter-wave radar dynamic points are labeled, the position data of all points in the point cloud interface is read. Combined with the millimeter-wave radar point attribute data of the current frame, the attribute data of the target millimeter-wave radar dynamic points and the attribute data of the unlabeled target millimeter-wave radar dynamic points are saved as a file. The class attribute of each point in the file indicates whether it is labeled as a dynamic point.

[0140] Explanation of the class attribute:

[0141] class=1: This point is marked as a target millimeter-wave radar dynamic point.

[0142] class=0: This point is not labeled, indicating that it is a static point.

[0143] This completes the labeling of dynamic points on the millimeter-wave radar.

[0144] Corresponding to the above method embodiments, the present invention also discloses a dynamic target annotation device.

[0145] See Figure 5 The present invention discloses a structural schematic diagram of a dynamic target annotation device, which may include:

[0146] The timestamp alignment unit 301 is used to perform timestamp alignment on the millimeter-wave radar dataset and camera image dataset acquired within a preset time period to obtain the target millimeter-wave radar dataset and the target camera image dataset.

[0147] It should be noted that the millimeter-wave radar dataset and camera image dataset obtained in this application are data acquired within the same preset time period. The value of the preset time period depends on actual needs, such as 1 hour, and this application does not limit it here.

[0148] In practical applications, millimeter-wave radar datasets are acquired through millimeter-wave radar installed on the vehicle, and camera image datasets are acquired through camera sensors installed on the same vehicle.

[0149] Because millimeter-wave radar and camera sensors typically operate on different principles and have different sampling frequencies, the data they acquire may differ at different points in time. This application aims to ensure that the millimeter-wave radar dataset and the camera image dataset are synchronized in time, enabling more accurate association and fusion of the two datasets. This is achieved by aligning the timestamps of the millimeter-wave radar dataset and the camera image dataset, adjusting the data from these different points in time to the same time reference, thereby ensuring data synchronization.

[0150] The target millimeter-wave radar dataset includes: point cloud location dataset and point attribute dataset.

[0151] Point cloud location datasets refer to the location information of target objects detected by millimeter-wave radar in three-dimensional space.

[0152] A point attribute dataset refers to the specific attributes and parameters associated with each detection point (or target point, reflection point) detected and acquired by a millimeter-wave radar system. Examples include ID number, lateral distance, longitudinal distance, lateral velocity, longitudinal velocity, label, angular velocity, length, and width.

[0153] The filtering unit 302 is used to filter out potential millimeter-wave radar dynamic point cloud data from the point attribute dataset.

[0154] In this application, potential millimeter-wave radar dynamic point cloud data refers to data that has a high probability of becoming a target millimeter-wave radar dynamic point.

[0155] The first projection unit 303 is used to project both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset onto the point cloud interface to obtain point cloud interface data.

[0156] After filtering out the potential millimeter-wave radar dynamic point cloud data from the point attribute dataset, the remaining data is the non-potential millimeter-wave radar dynamic point cloud data, which is actually static point cloud data.

[0157] The second projection unit 304 is used to convert the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, and then project them onto the image interface after fusing them with the target camera image dataset to obtain image interface data.

[0158] In this application, both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are in the millimeter-wave radar coordinate system, while the target camera image dataset is in the camera coordinate system. Therefore, before data fusion, the data needs to be transformed into the same coordinate system. In this embodiment, both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are transformed into the camera coordinate system, and then fused with the target camera image dataset in the camera coordinate system to obtain fused data. The fused data is then projected onto the image interface to obtain the image interface data.

[0159] The point cloud data displayed in this application contains positional information in three-dimensional space. Therefore, the point cloud interface provides a display and interaction of three-dimensional information. The shape, structure and positional information of an object can be seen intuitively through the point cloud interface.

[0160] The image data displayed in the image interface is two-dimensional. Therefore, the image interface provides the display and interaction of two-dimensional information. Through the image interface, you can view the color, texture, shape and other features in the image.

[0161] The dynamic point annotation unit 305 is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data by using the inter-frame switching function, and to annotate each potential millimeter-wave radar dynamic point as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point annotation.

[0162] In practical applications, this application employs an inter-frame switching function, utilizing the time-series changes reflected in multiple frames of point cloud data in the point cloud interface data, as well as the high-resolution details such as image textures provided in the image interface data, to achieve accurate annotation of dynamic points of the target millimeter-wave radar. Therefore, by fusing multi-frame point cloud data and image information, this application can simultaneously process the fine features of static objects and the motion trajectory of dynamic targets, improving the point cloud annotation effect of millimeter-wave radar in dynamic environments and enhancing the robustness and reliability of the system in dynamic target detection and tracking. Furthermore, it can effectively annotate whether the target is stationary or moving, thereby improving the detection and annotation capabilities of millimeter-wave radar point clouds in complex dynamic scenes.

[0163] In summary, this invention discloses a dynamic target annotation device. It aligns the timestamps of millimeter-wave radar datasets and camera image datasets acquired within a preset time period to obtain target millimeter-wave radar datasets and target camera image datasets. Potential millimeter-wave radar dynamic point cloud data is selected from the point attribute dataset. Both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data from the point attribute dataset are projected onto a point cloud interface to obtain point cloud interface data. Both the point cloud position dataset and the potential millimeter-wave radar dynamic point cloud data are transformed into the camera coordinate system and fused with the target camera image dataset before being projected onto an image interface to obtain image interface data. An inter-frame switching function is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and image interface data. The target millimeter-wave radar dynamic points are then identified and annotated, thereby obtaining the dynamically annotated target millimeter-wave radar point cloud data. This application processes millimeter-wave radar datasets and camera image datasets as a whole. It utilizes the principle that multi-frame point cloud data in the millimeter-wave radar dataset can capture the time-series changes reflected by the dynamic changes of dynamic targets in space, and that images in the camera image dataset can identify the appearance features of dynamic targets. This allows for the simultaneous processing of fine features of static targets and motion trajectories of dynamic targets. It ensures accurate calibration of dynamic targets even when the point cloud is sparse, resulting in missing information or noise. This improves the intelligent perception effect of millimeter-wave radar in applications such as autonomous driving and better adapts to complex real-world scenarios.

[0164] Furthermore, unlike traditional methods that rely solely on single-frame or point cloud data, this application can better handle moving targets in dynamic scenes, improving annotation accuracy and system robustness. This combined approach is not only suitable for dynamic target detection but also enhances the system's ability to handle complex scenes such as fast-moving objects and occlusion, significantly improving the perception performance of millimeter-wave radar point clouds in applications such as autonomous driving.

[0165] In one embodiment, the timestamp alignment unit 301 can be specifically used for:

[0166] Within the preset time period, each frame of millimeter-wave radar data with a timestamp is acquired from the millimeter-wave radar and sequentially placed into the millimeter-wave radar data queue to obtain the millimeter-wave radar dataset.

[0167] Within the preset time period, each frame of camera image data with a timestamp is acquired from the camera sensor and sequentially placed into the camera image data queue to obtain the camera image dataset;

[0168] Read each frame of millimeter-wave radar data from the millimeter-wave radar dataset and each frame of camera image data from the camera image dataset in sequence;

[0169] For each frame of millimeter-wave radar data, a linear interpolation method is used to determine the target frame camera image data with the corresponding timestamp from all the camera image data of each frame, so as to obtain the target millimeter-wave radar dataset and the target camera image dataset with timestamp alignment.

[0170] In one embodiment, the filtering unit 302 can be specifically used for:

[0171] Read the point attribute data of each frame in the point attribute dataset sequentially;

[0172] Determine the Euclidean distance and velocity of the attribute data for each frame point;

[0173] Filter out all target frame point attribute data from the point attribute dataset whose Euclidean distance is not greater than a distance threshold and whose velocity is not less than a velocity threshold;

[0174] The millimeter-wave radar points corresponding to the attribute data of each target frame point are used to form the potential millimeter-wave radar dynamic point cloud data.

[0175] In one embodiment, the second projection unit 304 can be specifically used for:

[0176] Based on the camera intrinsic parameter matrix and the extrinsic parameter matrix from the millimeter-wave radar coordinate system to the camera coordinate system, the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system to obtain the target point cloud location dataset and the target potential millimeter-wave radar dynamic point cloud data.

[0177] In the camera coordinate system, the target point cloud location dataset, the target potential millimeter-wave radar dynamic point cloud data, and the target camera image dataset are fused to obtain fused data;

[0178] The fused data is projected onto the image interface to obtain the image interface data.

[0179] In one embodiment, the dynamic point annotation unit 305 can be specifically used for:

[0180] The point cloud interface data of each frame in the point cloud interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame point cloud interface data, and to obtain the first millimeter-wave radar dynamic point set.

[0181] The image interface data of each frame in the image interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame image interface data, and to obtain the second millimeter-wave radar dynamic point set.

[0182] Identify each identical potential millimeter-wave radar dynamic point in the first millimeter-wave radar dynamic point set and the second millimeter-wave radar dynamic point set;

[0183] Each of the same potential millimeter-wave radar dynamic points is labeled as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point labeling.

[0184] In one embodiment, the dynamic point annotation unit 305 can be specifically used for:

[0185] When a forward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the previous frame point cloud interface data.

[0186] When a backward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the next frame point cloud interface data.

[0187] In one embodiment, the dynamic point annotation unit 305 can be specifically used for:

[0188] When a forward visualization data signal is received, the current frame image interface data in the image interface data is switched to the previous frame image interface data;

[0189] When a backward visualization data signal is received, the current frame image interface data in the image interface data is switched to the next frame image interface data.

[0190] It should be noted that for the specific working principles of each component in the device embodiment, please refer to the corresponding section of the method embodiment, which will not be repeated here.

[0191] Corresponding to the above embodiments, the present invention also discloses a computer storage medium that stores at least one instruction, which, when executed by a processor, implements the above-described method for labeling dynamic targets.

[0192] Computer storage media can be tangible media that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. Computer storage media can be machine-readable signal media or machine-readable storage media. Computer storage media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0193] Corresponding to the above embodiments, such as Figure 6 As shown, the present invention also provides an electronic device, which may include: a processor 1 and a memory 2;

[0194] The processor 1 and memory 2 communicate with each other via communication bus 3.

[0195] Processor 1, for executing at least one instruction;

[0196] Memory 2 is used to store at least one instruction;

[0197] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0198] Memory 2 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0199] The processor executes at least one instruction to implement the steps shown in the embodiment of the dynamic target annotation method.

[0200] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0201] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0202] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for labeling dynamic targets, characterized in that, include: The millimeter-wave radar dataset and camera image dataset acquired within a preset time period are timestamped to obtain the target millimeter-wave radar dataset and the target camera image dataset. The target millimeter-wave radar dataset includes: a point cloud location dataset and a point attribute dataset. Dynamic point cloud data for potential millimeter-wave radar are filtered from the point attribute dataset; The potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset are both projected onto the point cloud interface to obtain point cloud interface data. The point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system, and then fused with the target camera image dataset and projected onto the image interface to obtain the image interface data. The point cloud interface data of each frame in the point cloud interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame point cloud interface data, and to obtain the first millimeter-wave radar dynamic point set. The multi-frame point cloud data in the point cloud interface data can capture the time series changes reflected by the dynamic changes of dynamic targets in space. The image interface data of each frame in the image interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame image interface data, and to obtain the second set of millimeter-wave radar dynamic points. The image interface data can identify the appearance features of dynamic targets. Identify each identical potential millimeter-wave radar dynamic point in the first millimeter-wave radar dynamic point set and the second millimeter-wave radar dynamic point set. Each of the same potential millimeter-wave radar dynamic points is labeled as a target millimeter-wave radar dynamic point, resulting in target millimeter-wave radar point cloud data after dynamic point labeling.

2. The dynamic target annotation method according to claim 1, characterized in that, The process of aligning the timestamps of the millimeter-wave radar dataset and camera image dataset acquired within a preset time period to obtain the target millimeter-wave radar dataset and target camera image dataset includes: Within the preset time period, each frame of millimeter-wave radar data with a timestamp is acquired from the millimeter-wave radar and sequentially placed into the millimeter-wave radar data queue to obtain the millimeter-wave radar dataset. Within the preset time period, each frame of camera image data with a timestamp is acquired from the camera sensor and sequentially placed into the camera image data queue to obtain the camera image dataset; Read each frame of millimeter-wave radar data from the millimeter-wave radar dataset and each frame of camera image data from the camera image dataset in sequence; For each frame of millimeter-wave radar data, a linear interpolation method is used to determine the target frame camera image data with the corresponding timestamp from all the camera image data of each frame, so as to obtain the target millimeter-wave radar dataset and the target camera image dataset with timestamp alignment.

3. The method for labeling dynamic targets according to claim 1 or 2, characterized in that, The step of filtering potential millimeter-wave radar dynamic point cloud data from the point attribute dataset includes: Read the point attribute data of each frame in the point attribute dataset sequentially; Determine the Euclidean distance and velocity of the attribute data for each frame point; Filter out all target frame point attribute data from the point attribute dataset whose Euclidean distance is not greater than a distance threshold and whose velocity is not less than a velocity threshold; The millimeter-wave radar points corresponding to the attribute data of each target frame point are used to construct the potential millimeter-wave radar dynamic point cloud data.

4. The method for labeling dynamic targets according to claim 1 or 2, characterized in that, The process of transforming both the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, fusing them with the target camera image dataset, and projecting them onto the image interface to obtain image interface data includes: Based on the camera intrinsic parameter matrix and the extrinsic parameter matrix from the millimeter-wave radar coordinate system to the camera coordinate system, the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data are both transformed into the camera coordinate system to obtain the target point cloud location dataset and the target potential millimeter-wave radar dynamic point cloud data. In the camera coordinate system, the target point cloud location dataset, the target potential millimeter-wave radar dynamic point cloud data, and the target camera image dataset are fused to obtain fused data; The fused data is projected onto the image interface to obtain the image interface data.

5. The method for labeling dynamic targets according to claim 1, characterized in that, The method of dynamically switching point cloud interface data frame by frame switching includes: When a forward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the previous frame point cloud interface data. When a backward visualization data signal is received, the current frame point cloud interface data in the point cloud interface data is switched to the next frame point cloud interface data.

6. The method for labeling dynamic targets according to claim 5, characterized in that, The method of dynamically switching the image interface data of each frame in the image interface data using the inter-frame switching function includes: When a forward visualization data signal is received, the current frame image interface data in the image interface data is switched to the previous frame image interface data; When a backward visualization data signal is received, the current frame image interface data in the image interface data is switched to the next frame image interface data.

7. A dynamic target labeling device, characterized in that, include: The timestamp alignment unit is used to align the millimeter-wave radar dataset and camera image dataset acquired within a preset time period to obtain the target millimeter-wave radar dataset and the target camera image dataset. The target millimeter-wave radar dataset includes: a point cloud location dataset and a point attribute dataset. A filtering unit is used to filter out potential millimeter-wave radar dynamic point cloud data from the point attribute dataset; The first projection unit is used to project both the potential millimeter-wave radar dynamic point cloud data and the non-potential millimeter-wave radar dynamic point cloud data in the point attribute dataset onto the point cloud interface to obtain point cloud interface data. The second projection unit is used to convert the point cloud location dataset and the potential millimeter-wave radar dynamic point cloud data into the camera coordinate system, and then project them onto the image interface after fusing them with the target camera image dataset to obtain image interface data. The dynamic point annotation unit is used to determine each potential millimeter-wave radar dynamic point whose position changes in the point cloud interface data and the image interface data by using the inter-frame switching function, and to annotate each potential millimeter-wave radar dynamic point as a target millimeter-wave radar dynamic point to obtain the target millimeter-wave radar point cloud data after dynamic point annotation. The dynamic point annotation unit is specifically used for: The point cloud interface data of each frame in the point cloud interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame point cloud interface data, and to obtain the first millimeter-wave radar dynamic point set. The multi-frame point cloud data in the point cloud interface data can capture the time series changes reflected by the dynamic changes of dynamic targets in space. The image interface data of each frame in the image interface data is dynamically switched by the inter-frame switching function to determine the potential millimeter-wave radar dynamic points whose positions change in adjacent frame image interface data, and to obtain the second set of millimeter-wave radar dynamic points. The image interface data can identify the appearance features of dynamic targets. Identify each identical potential millimeter-wave radar dynamic point in the first millimeter-wave radar dynamic point set and the second millimeter-wave radar dynamic point set. Each of the same potential millimeter-wave radar dynamic points is labeled as a target millimeter-wave radar dynamic point, resulting in target millimeter-wave radar point cloud data after dynamic point labeling.

8. A computer storage medium, characterized in that, The computer storage medium stores at least one instruction, which, when executed by a processor, implements the dynamic target annotation method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, The electronic device includes: a memory and a processor; The memory is used to store at least one instruction; The processor is used to execute the at least one instruction to implement the dynamic target annotation method as described in any one of claims 1 to 6.

Images