Related method, electronic device and storage medium for three-dimensional target speed perception

Abstract

The application provides a related method for three-dimensional target speed perception, an electronic device and a storage medium. The method for three-dimensional target speed perception comprises the following steps: in response to target ID data not satisfying a preset ID abnormal condition, generating a target trajectory of a three-dimensional target based on the target ID data; in response to any one of target scale data, target orientation data and target interval data not satisfying a preset position abnormal condition, generating a target initial speed of the three-dimensional target based on the target trajectory; and in response to the target initial speed not satisfying a preset speed abnormal condition, performing a smoothing operation on the target initial speed to obtain a target true value speed of the three-dimensional target. The application robustly generates the target true value speed of the three-dimensional target through abnormal judgment on target ID data, target scale data, target orientation data and target interval data of the three-dimensional target, realizes speed perception of the three-dimensional target, and thus performs abnormal judgment on the three-dimensional target according to the speed of the three-dimensional target, thereby improving labeling efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, in particular to a three-dimensional target speed perception method, an electronic device and a storage medium. BACKGROUND

[0002] At present, three-dimensional target perception based on deep learning has become a mainstream technical method, but such a method needs a large amount of labeled data for training and iteration of the deep learning model. The position, pose, scale, category and other information of the three-dimensional target can be determined using sensor data, and then the corresponding labeled data can be obtained by manual labeling.

[0003] Among them, the erroneous or unstable estimation of the speed of the three-dimensional target will have an adverse effect, but the speed of the three-dimensional target needs to be combined with adjacent multiple frames of sensor data to be calculated, and it is difficult to obtain by manual labeling. SUMMARY

[0004] The present application provides a three-dimensional target speed perception method, an electronic device and a storage medium to solve the above problems.

[0005] The first aspect of the present application provides a three-dimensional target speed perception method, comprising: acquiring multiple frames of labeled data in a preset time period; wherein the labeled data is obtained by labeling corresponding sensor data, and each frame of the labeled data includes target ID data, target scale data, target orientation data and target center point data of a three-dimensional target; in response to the target ID data not satisfying a preset ID abnormal condition, generating a target trajectory of the three-dimensional target based on the target ID data; in response to any one of the target scale data, the target orientation data and the target interval data not satisfying a preset position abnormal condition, generating a target initial speed of the three-dimensional target based on the target trajectory; wherein the target interval data is obtained according to the target center point data; in response to the target initial speed not satisfying a preset speed abnormal condition, performing a smoothing operation on the target initial speed to obtain a target true value speed of the three-dimensional target, thereby realizing the speed perception of the three-dimensional target.

[0006] In some embodiments, the labeled data further includes a timestamp; generating a target trajectory based on the target ID data comprises: acquiring at least one frame of the labeled data corresponding to the target ID data representing the same ID; and sorting at least one frame of the labeled data according to the timestamp of each frame to obtain the target trajectory.

[0007] In some embodiments, the labeled data further comprises point cloud data; the target initial speed is generated based on the target trajectory, comprising: traversing the timestamps of the target center point data corresponding to each frame in the target trajectory; starting from the timestamp of the target center point data in any frame, traversing forward to obtain first neighbor target data, and traversing backward to obtain second neighbor target data; performing calculation and processing on the first neighbor target data and the second neighbor target data to obtain position deviation data, and calculating the target intermediate speed using the position deviation data; performing filtering processing operation on the target intermediate speed to obtain the target initial speed corresponding to any frame in the target trajectory.

[0008] In some embodiments, the first neighbor target data comprises first neighbor point cloud data and first neighbor position data, and the second neighbor target data comprises second neighbor point cloud data and second neighbor position data; wherein the first neighbor position data and the second neighbor position data are characterized by the target center point data; the calculation and processing on the first neighbor target data and the second neighbor target data to obtain position deviation data comprises: sorting the point cloud data corresponding to multiple frames of the target trajectory according to the number of point cloud points from more to less to obtain a point cloud sequence, and generating a template point cloud using the first preset number of frames of point cloud data in the point cloud sequence; in response to the number of point cloud points in the template point cloud, the first neighbor point cloud data and the second neighbor point cloud data being greater than a preset point cloud threshold, performing matching operation processing on the first neighbor point cloud data and the template point cloud to obtain first deviation data, and performing matching operation processing on the second neighbor point cloud data and the template point cloud to obtain second deviation data; performing calculation and processing on the first deviation data and the second deviation data to obtain the position deviation data; or, in response to the number of point cloud points in any one of the template point cloud, the first neighbor point cloud data and the second neighbor point cloud data being less than a preset point cloud threshold, performing calculation and processing on the first neighbor position data and the second neighbor position data to obtain the position deviation data.

[0009] In some embodiments, the smoothing operation processing on the target initial speed comprises: performing time domain modeling processing on the target initial speed corresponding to each frame in the target trajectory, and performing interpolation processing operation by a preset interpolation method to obtain the target true value speed of each frame.

[0010] In some embodiments, the method further comprises: in response to at least one of the target ID data, the target scale data, the target orientation data, the target interval data and the target initial speed being abnormal, performing labeling and generating trajectory abnormality labeling data.

[0011] In some embodiments, the method further comprises: generating and outputting final data of the three-dimensional target in each frame of the labeled data within the preset time period; wherein the final data comprises labeled data, target true value speed and the trajectory anomaly label data.

[0012] The second aspect of the present application provides a model evaluation method, comprising: obtaining a target true value speed, wherein the target true value speed is obtained by using the three-dimensional target speed perception method in the first aspect; and performing evaluation operation processing on a preset model by using the target true value speed to obtain an evaluation result, so as to evaluate the model.

[0013] The third aspect of the present application provides an electronic device, comprising a memory and a processor coupled to each other, wherein the processor is configured to execute program instructions stored in the memory to implement the three-dimensional target speed perception method in the first aspect or the model evaluation method in the second aspect.

[0014] The fourth aspect of the present application provides a non-volatile computer readable storage medium, wherein the computer readable storage medium is configured to store program instructions, and the program instructions are configured to implement the three-dimensional target speed perception method in the first aspect or the model evaluation method in the second aspect when executed by a processor.

[0015] The above scheme comprises: obtaining multiple frames of labeled data within a preset time period, each frame of labeled data comprising target ID data, target scale data, target orientation data and target center point data of a three-dimensional target; in response to the target ID data not satisfying a preset ID anomaly condition, generating a target trajectory of the three-dimensional target based on the target ID data; in response to any one of the target scale data, the target orientation data and the target interval data not satisfying a preset position anomaly condition, generating a target initial speed of the three-dimensional target based on the target trajectory; and in response to the target initial speed not satisfying a preset speed anomaly condition, performing smoothing operation processing on the target initial speed to obtain a target true value speed of the three-dimensional target. The scheme of the present application realizes robust generation of the target true value speed of the three-dimensional target by combining the anomaly judgment of the target ID data, the target scale data, the target orientation data and the target interval data of the three-dimensional target, realizes speed perception of the three-dimensional target, and thus realizes anomaly judgment of the three-dimensional target according to the speed of the three-dimensional target, which is convenient to use and improves labeling efficiency.

[0016] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the present application. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0018] Figure 1 is a flowchart of a method for sensing a three-dimensional target speed in an embodiment of the present application;

[0019] Figure 2 is another flowchart of a method for sensing a three-dimensional target speed in an embodiment of the present application;

[0020] Figure 3 is a flowchart of a method for model evaluation in an embodiment of the present application;

[0021] Figure 4 is a structural diagram of an electronic device in an embodiment of the present application;

[0022] Figure 5 is a structural diagram of a non-volatile computer readable storage medium in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The present application will be further described by embodiments with reference to the drawings. It is to be expressly understood that the following embodiments are only used to illustrate the present application and should not be construed to limit the scope of the present application. Similarly, the following embodiments are only part of the embodiments of the present application, and all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0024] In the present application, the term "embodiment" means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase appears in various places in the specification does not necessarily refer to the same embodiment, nor is it mutually exclusive or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0025] The term "and / or" herein is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can represent the existence of A alone, the existence of A and B together, and the existence of B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship. In addition, "multiple" in this paper means two or more than two. In addition, the term "at least one" in this paper means any one of the plurality or any combination of at least two of the plurality, for example, including at least one of A, B and C, which can mean including any one or more elements selected from the set consisting of A, B and C. In addition, the terms "first", "second", "third" in this application are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

[0026] As described above, the position, pose, scale, category and other information of the three-dimensional target can be determined using sensor data, and then the corresponding labeled data is obtained by manual labeling. The erroneous or unstable estimation of the speed of the three-dimensional target will have an adverse effect, but the speed of the three-dimensional target needs to be calculated in combination with multiple adjacent frames of sensor data, and it is difficult to obtain by manual labeling.

[0027] To this end, the present application provides a related method for three-dimensional target speed perception, an electronic device and a storage medium to solve the above problems.

[0028] Please refer to Figure 1 and Figure 2 , Figure 1 is a flowchart of a method for perceiving the speed of a three-dimensional target in an embodiment of the present application, Figure 2 is another flowchart of a method for perceiving the speed of a three-dimensional target in an embodiment of the present application. It should be noted that the method of the present application is not limited to the order of the flowchart shown in Figure 1 . The method can be applied to an electronic device with computing functions. The electronic device can execute the method by receiving information collected by a sensor device. The electronic device of the present application can be a server or a system cooperated by a server and a terminal device. Further, the server can be hardware or software, which is not limited here.

[0029] The electronic device can execute the method by receiving information collected by a sensor device. The sensor device can be a millimeter wave radar, a laser radar, an inertial measurement unit (IMU) device, a global navigation satellite system (GNSS) device equipped in an autonomous vehicle. The sensor device perceives the relevant data of the autonomous vehicle and the dynamic scene of the real physical world around the vehicle during the driving of the autonomous vehicle. The scene can have vehicles, pedestrians, buildings and other objects on the road. In some possible implementation manners, the method for perceiving the speed of a three-dimensional target in an embodiment of the present application can be realized by a processor calling computer readable instructions stored in a memory. As shown in Figure 1 and Figure 2 , the method for perceiving the speed of a three-dimensional target includes the following steps:

[0030] S11, obtaining multiple frames of labeled data in a preset time period; wherein the labeled data is obtained by labeling corresponding sensor data, and each frame of labeled data includes target ID data, target scale data, target orientation data and target center point data of the three-dimensional target.

[0031] The sensor data is collected by a sensor device, which can be a radar sensor, an IMU device, a GNSS device, or an image sensor. The radar sensor can be used to collect point cloud data, the IMU device can be used to collect IMU data, and the GNSS device can be used to collect GNSS data. For example, the radar sensor, the IMU device, and the GNSS device are installed on a movable device, such as a robot or a model training vehicle.

[0032] In some embodiments, the radar sensor can be a laser radar sensor, such as a mechanical laser radar, a semi-solid laser radar, or a solid laser radar. In one embodiment, the radar sensor can be any radar device that can provide point cloud data and meet the requirements of perception accuracy for model training.

[0033] In an application scenario, an autonomous vehicle is driving on a road, and the radar sensor, the IMU device, and the GNSS device installed on the autonomous vehicle are used to collect point cloud data, IMU data, and GNSS data of the autonomous vehicle to obtain a plurality of frames of sensor data within a preset time period.

[0034] In the preset time period, each frame of sensor data can include sensor data of one or more three-dimensional targets, without specific limitation.

[0035] The collected plurality of frames of sensor data within the preset time period are labeled using a preset labeling method to obtain a plurality of frames of labeled data within the preset time period. The preset labeling method can be labeling the sensor data using a preset labeling model or labeling the sensor data manually, without specific limitation.

[0036] Each frame of labeled data within the preset time period includes target ID data, target scale data, target orientation data, and target center point data of a three-dimensional target. It can be understood that each three-dimensional target corresponds to a target ID data, and the same three-dimensional target appearing in different frames of labeled data has the same target ID data. The target scale data can be the three-dimensional scale of the three-dimensional target, such as length, width, and height. The target orientation data can be the orientation angle in the three-dimensional pose of the three-dimensional target, wherein the roll and pitch angles in the three-dimensional pose can be assumed to be zero. The target center point data can be the three-dimensional position coordinates of the center point of the three-dimensional target.

[0037] S12, in response to the target ID data not satisfying the preset ID abnormal condition, generating a target trajectory of the three-dimensional target based on the target ID data.

[0038] The target ID data of the three-dimensional target is judged for ID abnormality by the preset ID abnormality condition. It can be understood that different three-dimensional targets are marked with the same target ID data, that is, the ID abnormality is determined.

[0039] For example, the preset ID abnormality condition can be that two or more same target ID data appear in the same frame of labeled data. It can be understood that the same three-dimensional target can only appear once at a time, and when two or more same target ID data appear in the same frame, it can be determined that the target ID data is abnormal, that is, the preset ID abnormality condition is met.

[0040] For example, the preset ID abnormality condition can be that two or more same target ID data appear in the same frame of labeled data. It can be understood that the same three-dimensional target can only appear once at a time, and when two or more same target ID data appear in the same frame, it can be determined that the target ID data is abnormal, that is, the preset ID abnormality condition is met.

[0041] In other embodiments, the preset ID abnormality condition can also be represented by other judgment methods, which can determine that the target ID data of the three-dimensional target is abnormal. It can be selected and used according to actual use requirements, and is not limited.

[0042] In response to the target ID data not meeting the preset ID condition, that is, the target ID data of the three-dimensional target does not appear abnormal, the target trajectory of the three-dimensional target is generated based on the corresponding target ID data.

[0043] S13, in response to any one of the target scale data, the target orientation data, and the target interval data not meeting the preset position abnormality condition, generating a target initial speed of the three-dimensional target based on the target trajectory; wherein the target interval data is obtained according to the target center point data.

[0044] The target scale data, the target orientation data, and the target interval data are judged for data abnormality by the preset position abnormality condition.

[0045] For example, the preset position abnormal condition can include comparing the target scale data with preset scale verification data to determine whether the target scale data is abnormal. The target scale data of the three-dimensional target with the same category data can be counted according to the category data of the three-dimensional target in the preset number of frames of label data to obtain the preset scale verification data corresponding to the category data, for example, the preset scale verification data can be scale prior distribution data. Then, the target scale data of the current frame is compared with the corresponding preset scale verification data according to the preset criterion to determine whether the target scale data is abnormal; for example, if the target scale data of the current frame does not meet the preset criterion in the preset scale verification data, it is determined that the target scale data of the current frame is abnormal, that is, the preset position abnormal condition is met. The preset criterion can be 3σ criterion or other criteria that can be implemented, which can be selected according to actual use requirements and is not limited.

[0046] For example, the preset position abnormal condition can also include comparing the target orientation data of the current frame with the target orientation data of the adjacent frame to determine whether the target orientation data of the current frame is abnormal. For example, the corresponding angular velocity can be calculated using the target orientation data of the current frame and the target orientation data of the adjacent frame, and it is determined whether the angular velocity is greater than a preset angular velocity threshold, for example, if the angular velocity is greater than the preset angular velocity threshold, the target orientation data of the current frame is abnormal, that is, the preset position abnormal condition is met.

[0047] The target interval data is obtained according to the target center point data, for example, the target center point data can be the three-dimensional position coordinates of the center point of the three-dimensional target and the corresponding time stamp, and then the target interval data of different three-dimensional targets can be obtained according to the three-dimensional position coordinates of the center points of different three-dimensional targets and the corresponding time stamps. It can be understood that the target interval data includes time interval and space interval, the corresponding time interval can be obtained according to the corresponding time stamps of different three-dimensional targets, and the corresponding space interval can be obtained according to the three-dimensional position coordinates of the center points of different three-dimensional targets.

[0048] For example, the preset position abnormal condition can further include comparing a time difference between a timestamp of the current frame and a timestamp of the adjacent frame with a preset time threshold, and comparing a position distance between the three-dimensional coordinate of the center point of the three-dimensional target of the current frame and the three-dimensional coordinate of the center point of the three-dimensional target of the adjacent frame with a preset distance threshold, to determine whether the target interval data of the current frame is abnormal. For example, the time difference between the timestamp of the current frame and the timestamp of the adjacent frame is greater than the preset time threshold, that is, the time interval between the timestamp of the current frame and the timestamp of the adjacent frame is greater than the preset time threshold, and / or the position distance between the three-dimensional coordinate of the center point of the three-dimensional target of the current frame and the three-dimensional coordinate of the center point of the three-dimensional target of the adjacent frame is greater than the preset distance threshold, that is, the spatial interval between the three-dimensional coordinate of the center point of the three-dimensional target of the current frame and the three-dimensional coordinate of the center point of the three-dimensional target of the adjacent frame is greater than the preset distance threshold, then the target interval data of the three-dimensional target of the current frame is abnormal, that is, the preset position abnormal condition is met.

[0049] In other embodiments, the preset position abnormal condition can also be represented by other judgment methods, which can determine that the target scale data, the target orientation data, and the target interval data of the three-dimensional target are abnormal. The actual use requirements can be selected for use, and are not limited specifically.

[0050] In response to any one of the target scale data, the target orientation data, and the target interval data not meeting the preset position abnormal condition, that is, any one of the target scale data, the target orientation data, and the target interval data not being abnormal, the target initial speed of the three-dimensional target is generated based on the target trajectory.

[0051] S14, in response to the target initial speed not meeting the preset speed abnormal condition, performing a smoothing operation on the target initial speed to obtain a target true value speed of the three-dimensional target, to realize speed perception of the three-dimensional target.

[0052] The preset speed abnormal condition is used to judge the speed abnormality of the target initial speed of the three-dimensional target.

[0053] For example, the preset speed abnormal condition can be comparing the target initial speed of the three-dimensional target with a preset speed threshold, to determine whether the target initial speed is abnormal. For example, the target initial speed is greater than the preset speed threshold, then the target initial speed is abnormal, that is, the preset speed abnormal condition is met.

[0054] For example, the preset speed abnormal condition can also be that the target initial speed of the current frame is calculated and processed with the target initial speed of the adjacent frame to determine whether the target initial speed of the current frame is abnormal. For example, a speed vector difference is calculated from the target initial speed of the current frame and the target initial speed of the adjacent frame, that is, the size change and direction change of the speed are considered at the same time, and the calculated speed vector difference is divided by time to obtain a corresponding quotient; it is determined whether the quotient is greater than a preset vector threshold, for example, if the quotient is greater than the preset vector threshold, the target initial speed of the current frame is abnormal, that is, the preset speed abnormal condition is met.

[0055] In other embodiments, the preset speed abnormal condition can also be represented by other judgment methods, which can determine that the target initial speed of the three-dimensional target is abnormal. The actual use requirements can be selected and used, and are not limited.

[0056] In response to the target initial speed not meeting the preset speed abnormal condition, that is, the target initial speed is not abnormal, the target initial speed is smoothed to obtain the target true value speed of the three-dimensional target, so as to realize the speed perception of the three-dimensional target.

[0057] The above scheme obtains a plurality of frames of labeling data in a preset time period, each frame of labeling data includes target ID data, target scale data, target orientation data and target center point data of a three-dimensional target; in response to the target ID data not meeting a preset ID abnormal condition, a target trajectory of the three-dimensional target is generated based on the target ID data; in response to any one of the target scale data, the target orientation data and the target interval data not meeting a preset position abnormal condition, a target initial speed of the three-dimensional target is generated based on the target trajectory; and in response to the target initial speed not meeting a preset speed abnormal condition, the target initial speed is smoothed to obtain a target true value speed of the three-dimensional target. The scheme of the present application combines the abnormal judgment of the target ID data, the target scale data, the target orientation data and the target interval data of the three-dimensional target, and then realizes the robust generation of the target true value speed of the three-dimensional target, realizes the speed perception of the three-dimensional target, and thus realizes the abnormal judgment of the three-dimensional target according to the speed of the three-dimensional target, which is convenient to use and improves the labeling efficiency.

[0058] As described above, in response to the target ID data not meeting the preset ID abnormal condition, the target trajectory of the three-dimensional target is generated based on the target ID data. In an embodiment of the present application, the labeling data further includes a timestamp; the target trajectory is generated based on the target ID data, which includes: obtaining at least one frame of labeling data corresponding to the target ID data representing the same ID; and sorting the at least one frame of labeling data according to the timestamp corresponding to each frame to obtain the target trajectory.

[0059] Each frame of annotation data corresponds to a timestamp, and the timestamps of different frame annotation data are different. Each three-dimensional target has a target ID data, and since the same three-dimensional target can be continuously moving over time, there can be multiple frame annotation data including corresponding data of the same three-dimensional target. At least one frame of annotation data corresponding to the target ID data representing the same ID is obtained, and according to the timestamp corresponding to each frame of annotation data, at least one frame of annotation data can be sorted in chronological order, thereby obtaining the target trajectory of the three-dimensional target.

[0060] For example, the target trajectory T = {o t |t∈[0,L)}, wherein t represents a timestamp, which can be represented by an integer; L represents the total number of frames of at least one frame of annotation data; o t represents annotation data. The timestamps in the target trajectory are not necessarily continuous, and there can be a case of missing frames.

[0061] As described above, in response to any one of the target scale data, the target orientation data, and the target interval data not satisfying the preset position abnormal condition, the target initial speed of the three-dimensional target is generated based on the target trajectory. In an embodiment of the present application, the annotation data further includes point cloud data; the target initial speed is generated based on the target trajectory, including: traversing the timestamp of the target center point data corresponding to each frame in the target trajectory; starting from the timestamp corresponding to the target center point data in any frame, traversing forward to obtain first neighbor target data, and traversing backward to obtain second neighbor target data; calculating and processing the first neighbor target data and the second neighbor target data to obtain position deviation data, and calculating the target intermediate speed using the position deviation data; performing filtering processing operation on the target intermediate speed to obtain the target initial speed corresponding to any frame in the target trajectory.

[0062] According to the timestamp of each frame of annotation data, the annotation data is sorted in chronological order to obtain the target trajectory of the three-dimensional target. The timestamp corresponding to the target center point data in the annotation data of each frame in the target trajectory is traversed, starting from the timestamp corresponding to the target center point data in any frame, traversing forward along the target trajectory to obtain first neighbor target data, and traversing backward along the target trajectory to obtain second neighbor target data.

[0063] For example, first, starting from the timestamp of the target center point data of the first frame in the target trajectory, forward traversal is performed, and at a time interval from the first frame reaching a preset neighbor threshold, first neighbor target data corresponding to the first frame is determined; starting from the timestamp of the target center point data of the first frame in the target trajectory, backward traversal is performed, and at a time interval from the first frame reaching a preset neighbor threshold, second neighbor target data corresponding to the first frame is determined. Then, starting from the timestamp of the target center point data of the second frame in the target trajectory, forward traversal is performed to obtain first neighbor target data corresponding to the second frame, and backward traversal is performed to obtain second neighbor target data corresponding to the second frame. Similarly, first neighbor target data and second neighbor target data corresponding to the target center point data of each frame in the target trajectory can be obtained. The preset neighbor threshold can be 3 frames, 5 frames, or other values that can be implemented, and is set according to actual use requirements and is not specifically limited.

[0064] The first neighbor target data and the second neighbor target data are calculated and processed, and position deviation data corresponding to each frame of annotation data can be obtained, and the target intermediate speed can be calculated using the position deviation data. For example, the target intermediate speed of the current frame is calculated using the formula Δp / (t fut-t his), wherein Δp represents the position deviation data, t fut represents the timestamp corresponding to the second neighbor target data, and t his represents the timestamp corresponding to the first neighbor target data.

[0065] The obtained target intermediate speed is filtered by a preset filtering algorithm, and the target initial speed corresponding to any frame in the target trajectory is obtained. For example, the preset filtering algorithm can be a Kalman filtering algorithm, or other algorithms that can be implemented, and is not specifically limited.

[0066] As described above, starting from the timestamp corresponding to the target center point data in any frame of the target trajectory, the first neighbor target data is obtained by forward traversal, and the second neighbor target data is obtained by backward traversal. In an embodiment of the present application, the first neighbor target data includes first neighbor point cloud data and first neighbor position data, and the second neighbor target data includes second neighbor point cloud data and second neighbor position data; wherein the first neighbor position data and the second neighbor position data are characterized by the target center point data; the first neighbor target data and the second neighbor target data are calculated and processed to obtain position deviation data, including: sorting the multiple frames of point cloud data corresponding to the target trajectory according to the number of point cloud points from more to less to obtain a point cloud sequence, and generating a template point cloud using the first preset number of frames of point cloud data in the point cloud sequence; in response to the number of point cloud points in the template point cloud, the first neighbor point cloud data and the second neighbor point cloud data being greater than a preset point cloud threshold, the first neighbor point cloud data is matched with the template point cloud to obtain first deviation data; and the second neighbor point cloud data is matched with the template point cloud to obtain second deviation data; the first deviation data and the second deviation data are calculated and processed to obtain the position deviation data; or, in response to the number of point cloud points in any one of the template point cloud, the first neighbor point cloud data and the second neighbor point cloud data being less than the preset point cloud threshold, the first neighbor position data and the second neighbor position data are calculated and processed to obtain the position deviation data.

[0067] The annotation data includes point cloud data, and the multiple frames of point cloud data corresponding to the target trajectory are sorted according to the number of point cloud points from more to less to obtain a point cloud sequence. For example, the target trajectory includes three frames of annotation data A, B and C, the number of point cloud data a in the annotation data A is 500, the number of point cloud data b in the annotation data B is 800, and the number of point cloud data c in the annotation data C is 300; the point cloud data is sorted according to the number of point cloud points to obtain a point cloud sequence, that is, point cloud data b, point cloud data a and point cloud data c.

[0068] After obtaining the point cloud sequence, a template point cloud is generated using the first preset number of frames of point cloud data in the point cloud sequence. For example, the point cloud sequence includes 1000 frames of point cloud data, and a template point cloud is generated using the point cloud data of the first 200 frames. The preset number can also be other numbers that can be implemented, which can be set according to actual use requirements and is not limited in particular.

[0069] In response to the number of point cloud points in the template point cloud, the first near-neighbor point cloud data, and the second near-neighbor point cloud data being greater than a preset point cloud threshold, the first near-neighbor point cloud data is subjected to point cloud matching operation processing with the template point cloud to obtain first deviation data, and the second near-neighbor point cloud data is subjected to point cloud matching operation processing with the template point cloud to obtain second deviation data. After obtaining the first deviation data and the second deviation data, the first deviation data and the second deviation data are subjected to calculation processing to obtain position deviation data. For example, the position deviation data is obtained by subtracting the first deviation data from the second deviation data.

[0070] In response to the number of point cloud points in any one of the template point cloud, the first near-neighbor point cloud data, and the second near-neighbor point cloud data being less than a preset point cloud threshold, the first near-neighbor position data and the second near-neighbor position data are subjected to calculation processing to obtain position deviation data. The first near-neighbor position data and the second near-neighbor position data are represented by target center point data. For example, the target center point data can be three-dimensional coordinate data of a center point of a three-dimensional target. The position deviation data is obtained by subtracting the three-dimensional coordinate corresponding to the first near-neighbor position data from the three-dimensional coordinate corresponding to the second near-neighbor position data.

[0071] As described above, in response to the target initial speed not satisfying the preset speed abnormality condition, the target initial speed is subjected to smoothing operation processing. In an embodiment of the present application, the smoothing operation processing on the target initial speed includes: time-domain modeling processing on the target initial speed corresponding to each frame in the target trajectory, and interpolation processing operation by a preset interpolation method to obtain the target true value speed of each frame.

[0072] The target initial speed (t, v x , v y , v z ) corresponding to each frame in the target trajectory is subjected to time-domain modeling processing, where t represents a time stamp, v x , v y , v z represent the target initial speed, and interpolation processing is performed by a preset interpolation method to obtain the target true value speed t represents a time stamp, and v represent the target true value speed. The preset interpolation method can be a time-domain interpolation method or any other implementable method, and is not limited in particular.

[0073] In an embodiment of the present application, the perception method of the three-dimensional target speed further includes: in response to at least one of the target ID data, the target scale data, the target orientation data, the target interval data, and the target initial speed being abnormal, performing marking and generating trajectory abnormality marking data.

[0074] In response to an abnormality existing in at least one of the target ID data, the target scale data, the target orientation data, the target interval data, and the target initial speed, i.e., the target ID data satisfies a preset ID abnormality condition, the target scale data satisfies a preset position abnormality condition, the target orientation data satisfies a preset position abnormality condition, the target interval data satisfies a preset position abnormality condition, and / or the target initial speed satisfies a preset speed abnormality condition, the abnormal data is marked and trajectory abnormality marking data is generated.

[0075] For example, an error_flag field can be added to mark the abnormal data, and an error_info field can be added, in which specific abnormal information corresponding to the data is noted.

[0076] As described above, in response to an abnormality existing in at least one of the target ID data, the target scale data, the target orientation data, the target interval data, and the target initial speed, the abnormal data is marked and trajectory abnormality marking data is generated. In an embodiment of the present application, the three-dimensional target speed perception method further includes: generating and outputting final data of the three-dimensional target in each frame of labeling data within a preset time period; wherein the final data includes the labeling data, the target true value speed, and the trajectory abnormality marking data.

[0077] The final data of the three-dimensional target in each frame of labeling data within a preset time period is generated and outputted, wherein there can be multiple three-dimensional targets in any frame of labeling data, and then the final data of each three-dimensional target is generated and outputted. The final data includes the labeling data, the target true value speed, and the trajectory abnormality marking data.

[0078] Please refer to Figure 3 , Figure 3 is a flowchart of a model evaluation method in an embodiment of the present application. In some possible implementation manners, the model evaluation method of the embodiment of the present application can be implemented by a processor calling computer readable instructions stored in a memory. As shown in Figure 3 , the model evaluation method includes the following steps:

[0079] S21, obtaining a target true value speed, wherein the target true value speed is obtained by using the three-dimensional target speed perception method in the above embodiment.

[0080] Obtaining a target true value speed, for example, obtaining a plurality of frames of labeled data in a preset time period; wherein the labeled data is obtained by labeling corresponding sensor data, and each frame of labeled data includes target ID data, target scale data, target orientation data and target center point data of a three-dimensional target; in response to the target ID data not satisfying a preset ID abnormal condition, generating a target trajectory of the three-dimensional target based on the target ID data; in response to any one of the target scale data, the target orientation data and target interval data not satisfying a preset position abnormal condition, generating a target initial speed of the three-dimensional target based on the target trajectory; wherein the target interval data is obtained according to the target center point data; in response to the target initial speed not satisfying a preset speed abnormal condition, performing a smoothing operation on the target initial speed to obtain the target true value speed of the three-dimensional target.

[0081] S22, performing evaluation operation processing on the preset model by using the target true value speed to obtain an evaluation result, so as to realize evaluation on the model.

[0082] The corresponding sensor data is processed by using the preset model to obtain a model speed of the three-dimensional target. For the same sensor data, the target true value speed is obtained by using the three-dimensional target speed perception method in the above embodiment, and then the target true value speed is compared with the model speed to evaluate the preset model, so as to obtain an evaluation result, thereby realizing evaluation on the model.

[0083] Those skilled in the art can understand that, in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined according to its function and possible internal logic.

[0084] Please refer to Figure 4 , Figure 4 is a structural schematic diagram of an electronic device in the embodiment of the present application. The electronic device 400 includes a memory 401 and a processor 402 coupled with each other. The processor 402 is configured to execute program instructions stored in the memory 401 to implement the steps in the three-dimensional target speed perception method embodiment or the steps in the model evaluation method embodiment described above. In one specific implementation scenario, the electronic device 400 can include but is not limited to a microcomputer, a server, without limitation.

[0085] Specifically, the processor 402 is configured to control itself and the memory 401 to implement the steps in the three-dimensional target speed perception method embodiment or the steps in the model evaluation method embodiment described above. The processor 402 can also be referred to as a CPU (Central Processing Unit). The processor 402 can be an integrated circuit chip having a processing capability. The processor 402 can also be a general processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The general processor can be a microprocessor or the processor can also be any conventional processor. In addition, the processor 402 can be implemented by an integrated circuit chip together.

[0086] Please refer to Figure 5 , Figure 5 is a structural diagram of a non-volatile computer readable storage medium in the embodiments of the present application. The computer readable storage medium 500 is configured to store program instructions 501, and the program instructions 501, when executed by the processor 402, are configured to implement the steps in the three-dimensional target speed perception method embodiment or the steps in the model evaluation method embodiment described above.

[0087] The above description of the various embodiments tends to emphasize the differences between the various embodiments, and the same or similar parts can be referred to each other, and for brevity, the details are not repeated herein.

[0088] In several embodiments provided in the present application, it should be understood that the disclosed method and related equipment can be implemented by other ways. For example, the above-described related equipment implementation is only schematic, for example, the division of the module or unit is only a logical function division, and actual implementation can have another division manner, for example, the unit or component can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication disconnecting can be indirect coupling or communication disconnecting through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0089] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0090] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application, essentially or in part, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform all or part of the steps of the methods in the various embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various other media that can store program codes.

Claims

1. A method of perceiving a three-dimensional target velocity, characterized by, The method comprises: acquiring a plurality of frames of labeled data within a preset time period; wherein the labeled data is obtained by labeling corresponding sensor data, and each frame of the labeled data comprises target ID data, target scale data, target orientation data, and target center point data of a three-dimensional target; in response to the target ID data not satisfying a preset ID abnormal condition, generating a target trajectory of the three-dimensional target based on the target ID data; in response to any one of the target scale data, the target orientation data, and target interval data not satisfying a preset position abnormal condition, generating a target initial speed of the three-dimensional target based on the target trajectory, wherein the target interval data is obtained from the target center point data; in response to the target initial speed not satisfying a preset speed abnormal condition, performing a smoothing operation on the target initial speed to obtain a target true value speed of the three-dimensional target, thereby realizing speed perception of the three-dimensional target.

2. The method of claim 1, wherein, The labeled data further comprises a timestamp; generating a target trajectory based on the target ID data comprises: acquiring at least one frame of the labeled data corresponding to the target ID data representing the same ID; sorting the at least one frame of the labeled data according to the corresponding timestamps of each frame to obtain the target trajectory.

3. The method of claim 2, wherein, The labeled data further comprises point cloud data; generating a target initial speed based on the target trajectory comprises: traversing the timestamps of the target center point data corresponding to each frame in the target trajectory; starting from the timestamp corresponding to the target center point data in any frame, traversing forward to obtain first neighbor target data and traversing backward to obtain second neighbor target data; performing calculation processing on the first neighbor target data and the second neighbor target data to obtain position deviation data, and calculating a target intermediate speed using the position deviation data; performing filtering processing on the target intermediate speed to obtain the target initial speed corresponding to any frame in the target trajectory.

4. The method of claim 3, wherein, The first neighbor target data comprises first neighbor point cloud data and first neighbor position data, and the second neighbor target data comprises second neighbor point cloud data and second neighbor position data; wherein the first neighbor position data and the second neighbor position data are represented by the target center point data; performing calculation processing on the first neighbor target data and the second neighbor target data to obtain position deviation data comprises: sorting a plurality of frames of point cloud data corresponding to the target trajectory according to the number of point cloud points from more to less to obtain a point cloud sequence, and generating a template point cloud using the first preset number of frames of point cloud data in the point cloud sequence; in response to the number of point cloud points in the template point cloud, the first neighbor point cloud data, and the second neighbor point cloud data all being greater than a preset point cloud threshold, performing matching operation processing on the first neighbor point cloud data and the template point cloud to obtain first deviation data, and performing matching operation processing on the second neighbor point cloud data and the template point cloud to obtain second deviation data; performing calculation processing on the first deviation data and the second deviation data to obtain the position deviation data; or In response to the number of point cloud points in any one of the template point cloud, the first near-neighbor point cloud data, and the second near-neighbor point cloud data being less than a preset point cloud threshold, the first near-neighbor position data and the second near-neighbor position data are calculated to obtain the position deviation data.

5. The method of claim 3, wherein, The target initial speed is subjected to a smoothing operation, including: The target initial speed corresponding to each frame in the target trajectory is subjected to time-domain modeling processing, and is subjected to interpolation processing by a preset interpolation method to obtain a target true value speed of each frame.

6. The method of claim 1, wherein, The method further includes: In response to at least one of the target ID data, the target scale data, the target orientation data, the target interval data, and the target initial speed being abnormal, a label is generated and trajectory abnormality label data is generated.

7. The method of claim 6, wherein, The method further includes: Final data of the three-dimensional target in each frame of label data within the preset time period is generated and output; wherein the final data includes label data, a target true value speed, and the trajectory abnormality label data.

8. A model evaluation method characterized by, It includes: A target true value speed is obtained, wherein the target true value speed is obtained by using the three-dimensional target speed perception method of any one of claims 1-7; An evaluation result is obtained by evaluating a preset model by using the target true value speed, so as to evaluate the model.

9. An electronic device, comprising: The processor is configured to execute program instructions stored in the memory to implement the three-dimensional target speed perception method of any one of claims 1-7 or the model evaluation method of claim 8.

10. A non-transitory computer readable storage medium, comprising: The computer readable storage medium is configured to store program instructions, which, when executed by the processor, are configured to implement the three-dimensional target speed perception method of any one of claims 1-7 or the model evaluation method of claim 8.

Images