Target optimization method and device, equipment and storage medium

Abstract

The application belongs to the technical field of target optimization, and provides a target optimization method, device, equipment and storage medium, the target optimization method comprises the following steps: S1, an equation model corresponding to target state data is established and initialized, wherein the target state data comprises position data, speed data and acceleration data; S2, target state data corresponding to point clouds is acquired, and the target state data is used as an input quantity of the equation model, wherein the number of the point clouds is multiple, different point clouds are acquired within different time thresholds, and different point clouds correspond to different target state data; S3, the position data and the speed data in the target state data are constrained, and a position constraint error corresponding to the position data is calculated. According to the application, all point clouds of a target in a period are acquired, then the optimal estimation of the target state data is calculated based on the point clouds and by using an error constraint and a cost function minimization method, so that the result of the target is more accurate.

Description

Technical Field

[0001] This invention belongs to the field of target optimization technology, specifically relating to target optimization methods, apparatus, equipment, and storage media. Background Technology

[0002] In recent years, research on autonomous driving has not only been a hot topic in the scientific community, but Level 3 autonomous driving has also become a key project for technology companies and traditional automakers to implement their technologies. Multi-sensor fusion is a commonly used solution in the industry today. It can leverage the perceptual characteristics of each sensor, with each sensor complementing and redundant with the others, to achieve accurate and reliable perception of environmental targets from all angles. Millimeter-wave radar has become a standard component in multi-sensor perception systems due to its low cost, high speed measurement accuracy, and minimal impact from weather conditions.

[0003] To improve the accuracy of target detection results obtained by sensors, multi-sensor target fusion software is used to fuse target data from multiple sensors. Currently, multi-sensor target fusion software often employs Kalman filtering to fuse data from various sensors, such as the perception fusion method for an autonomous driving system disclosed in application number CN202111250422.0, and the target fusion method, apparatus, storage medium, and electronic device disclosed in application number CN202010925757.7.

[0004] In Kalman filter fusion methods, typically only radar-sensed target data is fused. However, because this target data lacks associated point clouds, it may suffer from some information loss, resulting in inaccurate target data itself. Consequently, the fused results from this target data are neither stable nor accurate. Summary of the Invention

[0005] The purpose of this invention is to provide a target optimization method, apparatus, device, and storage medium to solve the problem mentioned in the background art, where target data may suffer some information loss due to the lack of associated point clouds, resulting in inaccurate target data.

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

[0007] Firstly, a target optimization method is provided, the target optimization method comprising:

[0008] Step S1: Establish and initialize the equation model corresponding to the target state data, wherein the target state data includes position data, velocity data and acceleration data;

[0009] Step S2: Obtain the target state data corresponding to the point cloud, and use the target state data as the input of the equation model. There are multiple point clouds, different point clouds are obtained within different time thresholds, and different point clouds correspond to different target state data.

[0010] Step S3: Constrain the position data and velocity data in the target state data, and calculate the position constraint error corresponding to the position data and the velocity constraint error corresponding to the velocity data;

[0011] Step S4: Based on the position constraint error and the velocity constraint error, and minimizing the cost function, calculate and output the optimal target state data.

[0012] In conjunction with the first aspect, in some optional implementations, in the equation model in step S1, the target state data is separated into horizontal and vertical directions, and assuming that the position data is a third-order curve equation that varies with time, then the velocity data is the derivative of the position data, and the acceleration data is the derivative of the velocity data.

[0013] In conjunction with the first aspect, in some optional implementations, in step S2, the method for obtaining the target state data corresponding to the point cloud includes:

[0014] Step S21: Obtain target detection data associated with the point cloud through sensors, wherein the target detection data includes the target relative position, target absolute velocity, and target absolute acceleration;

[0015] Step S22: Obtain the vehicle state data within the same time threshold as the point cloud, wherein the vehicle state data includes the vehicle position, the vehicle absolute speed, and the vehicle absolute acceleration;

[0016] Step S23: Calculate the target state data based on the target detection data and the vehicle state data, wherein the position data is equal to the relative position of the target, the speed data is equal to the difference between the absolute speed of the target and the absolute speed of the vehicle, and the acceleration data is the difference between the absolute acceleration of the target and the absolute acceleration of the vehicle.

[0017] In conjunction with the first aspect, in some optional implementations, in step S3, based on the assumption that all point clouds are within the target bounding box, the nearest distance from the point cloud outside the target bounding box to the target bounding box is used as the error constraint, and the calculation equation for the position constraint error is as follows:

[0018] Ep_n=Pr-sprt(Pr_x*Pr_x+Pr_y*Pr_y),

[0019] Where Ep_n is the position constraint error of the nth point cloud, Pr is the radial distance of the point cloud, and Pr_x and Pr_y are the horizontal and vertical coordinates of the intersection point of the radial direction of the point cloud and the target, respectively.

[0020] In conjunction with the first aspect, in some optional implementations, in step S3, the lateral and longitudinal velocities of the target are decomposed into the angular directions of the point cloud, and constraints are established on the velocity data based on the radial velocity of the point cloud, wherein the velocity constraint error is as follows:

[0021] Ev_n=Vr-Vx*sin(θ)+Vy*cos(θ),

[0022] Where Ev_n is the velocity constraint error of the nth point cloud, Vr is the radial velocity of the point cloud, Vx is the x-axis velocity of the point cloud, Vy is the y-axis velocity of the point cloud, and θ is the horizontal angle of the point cloud.

[0023] In conjunction with the first aspect, in some optional implementations, in step S5, based on the position constraint error and the velocity constraint error, and minimizing the cost function, the optimal target state data is calculated and output, including:

[0024] Step S51: Calculate the total loss cost of the target state data based on the position constraint error and the velocity constraint error;

[0025] Step S52: Determine whether the total loss cost is less than a preset deviation threshold. If yes, take the target state data corresponding to the total loss cost as the optimal state data of the target. If no, repeat step S2 until the total loss cost is less than the deviation threshold.

[0026] Step S53: Output the optimal state data of the target.

[0027] Secondly, a target optimization apparatus is provided, the target optimization apparatus comprising:

[0028] An initialization module is used to establish and initialize an equation model corresponding to the target state data, wherein the target state data includes position data, velocity data, and acceleration data;

[0029] The acquisition module is used to acquire the target state data corresponding to the point cloud and use the target state data as the input of the equation model. There are multiple point clouds, different point clouds are acquired within different time thresholds, and different point clouds correspond to different target state data.

[0030] The constraint module is used to constrain the position data and velocity data in the target state data, and calculate the position constraint error corresponding to the position data and the velocity constraint error corresponding to the velocity data.

[0031] The calculation module is used to calculate and output the optimal target state data based on the position constraint error and the velocity constraint error, and based on minimizing the cost function.

[0032] Thirdly, a target optimization device is provided, the target optimization device including a controller and a memory, the memory storing a computer program, which, when executed by the controller, causes the optimization device to perform the target optimization method as described in any one of the first aspects.

[0033] Thirdly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is run on a computer, the computer causes the computer to perform the target optimization method as described in any one of the first aspects.

[0034] The invention employing the above technical solution has the following advantages:

[0035] This implementation method acquires all point clouds of the target within a certain time period, and then uses error constraint and cost function minimization to calculate the optimal estimate of the target state data. Compared with the existing technology, the target state data estimated by this method is more accurate, and the result obtained by fusing target state data from multiple sensors will also be more accurate and stable. Attached Figure Description

[0036] The present invention can be further illustrated by the non-limiting embodiments given in the accompanying drawings;

[0037] Figure 1 This is a flowchart illustrating the target optimization method in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the establishment of constraints on the position data of point clouds in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram illustrating the establishment of constraints on the velocity data of point clouds in an embodiment of the present invention. Detailed Implementation

[0040] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts are referred to by the same reference numerals in the drawings or description. Implementations not shown or described in the drawings are forms known to those skilled in the art. In addition, directional terms mentioned in the embodiments, such as "up," "down," "top," "bottom," "left," "right," "front," and "back," are only for reference to the directions in the drawings and are not intended to limit the scope of protection of the present invention.

[0041] This application provides a target optimization device, which is mainly installed on a vehicle. The target optimization device includes a controller and a memory. The memory stores a computer program. When the computer program is executed by the controller, the optimization device performs the target optimization method described below.

[0042] In addition, the target optimization device includes multiple sensors, primarily millimeter-wave radar, which is mainly used to detect the point cloud of the target.

[0043] It should be noted that in this embodiment, "target" can refer to both the physical object detected by the sensor, such as a person, vehicle, or obstacle, and the sensor's detection result. For example, it can be a data model describing the physical object. In a computer, this model can exist as a structure with multiple attributes (corresponding to members in the structure), which describe the target's state. In the following text, unless there is ambiguity, the two meanings of the term "target" will not be specifically distinguished; they can be understood in the context of the specific scenario.

[0044] In addition, the detected point cloud contains associated information about the target, including but not limited to the target's coordinates, orientation angle, velocity, acceleration, and size.

[0045] like Figure 1 As shown in the embodiments of this application, a target optimization method is also provided, wherein the target optimization method may include the following steps:

[0046] Step S1: Establish and initialize the equation model corresponding to the target state data, wherein the target state data includes position data, velocity data and acceleration data;

[0047] The method for fusing multi-sensor targets mentioned in the background technology mainly involves fusing the position data, velocity data, and acceleration data of the target detected by multiple sensors separately. Step S1 mainly involves establishing an equation model for these data.

[0048] Step S2: Obtain the target state data corresponding to the point cloud, and use the target state data as the input of the equation model. There are multiple point clouds, different point clouds are obtained within different time thresholds, and different point clouds correspond to different target state data.

[0049] After the equation model is established and initialized in step S1, the target state data corresponding to the point cloud is acquired. By inputting the target state data corresponding to each point cloud into the equation model, the specific equation corresponding to that target state data can be calculated. However, the number of point clouds acquired by the same sensor is often more than one, and different point clouds can correspond to different target state data. Therefore, in this embodiment, a time threshold is set. One point cloud is acquired within each time threshold, and the target state data within that time threshold can be calculated based on each point cloud. For example, if the time threshold is set to 1 second, one point cloud is acquired every second from the 1st second to the nth second, and n seconds is considered a time period, meaning only all point clouds within n seconds are acquired. Then, the optimal target state data is calculated using the methods in steps 3 and 4.

[0050] Step S3: Constrain the position data and velocity data in the target state data, and calculate the position constraint error corresponding to the position data and the velocity constraint error corresponding to the velocity data;

[0051] It is understandable that the vehicle is generally in a dynamic movement process, and the target is also generally in a dynamic movement process. The target's state data mainly depends on the target's position data and speed data. Therefore, there will inevitably be some error in these two data. Constraining these two data makes it easier to estimate the optimal target state data in step 4.

[0052] Step S4: Based on the position constraint error and the velocity constraint error, and minimizing the cost function, calculate and output the optimal target state data.

[0053] Understandably, this step uses the cost function minimization method, taking the position constraint error and velocity constraint error as related quantities of the cost function. This allows the calculation of the loss cost corresponding to the point cloud, and the target state data with the smaller loss cost is taken as the optimal target state data, making the detected target results more accurate.

[0054] In summary, this implementation method acquires all point clouds of the target within a certain time period, and then calculates the optimal estimate of the target state data by using error constraints and cost function minimization. Compared with the existing technology that directly uses target data, the target state data estimated by this method is more accurate, and the result obtained by fusing target state data from multiple sensors will also be more accurate and stable.

[0055] As an optional implementation, in the equation model in step S1, the target state data is separated into horizontal and vertical directions, and it is assumed that the position data is a third-order curve equation that varies with time. Then, the velocity data is the derivative of the position data, and the acceleration data is the derivative of the velocity data.

[0056] The specific equations for calculating the location data are as follows:

[0057] Px=a0+a1*t+a2 / 2*t*t+a3 / 6*t*t*t;

[0058] Py=b0+b1*t+b2 / 2*t*t+b3 / 6*t*t*t;

[0059] Based on the calculation equation for position data, the specific calculation equation for velocity data is as follows:

[0060] Vx = a1 + a2*t + a3 / 2*t*t;

[0061] Vy = b1 + b2*t + b3 / 2*t*t;

[0062] Based on the equations for calculating velocity data, the specific equations for calculating acceleration data are as follows:

[0063] Ax = a2 + a3 / t;

[0064] Ay = b2 + b3 / t;

[0065] Based on the three sets of equations above, it can be determined that the parameters that need to be determined in this equation model include eight parameters: a0, a1, a2, a3 and b0, b1, b2, b3. Since the first frame time within each time threshold can be considered as 0, the expression equation for each parameter can be obtained from the three sets of equations above. The specific first frame initialization equation for each parameter is as follows:

[0066] a0=Px, a1=Vx, a2=Ax, a3=0;

[0067] b0=Py, b1=Vy, b2=Ay, b3=0;

[0068] In the above equation model, P represents position, V represents velocity, A represents acceleration, x represents the horizontal axis, y represents the vertical axis, and t represents the time threshold.

[0069] As an optional implementation, in step S2, the method for obtaining the target state data corresponding to the point cloud includes:

[0070] Step S21: Obtain target detection data associated with the point cloud through sensors, wherein the target detection data includes the target relative position, target absolute velocity, and target absolute acceleration;

[0071] Step S22: Obtain the vehicle state data within the same time threshold as the point cloud, wherein the vehicle state data includes the vehicle position, the vehicle absolute speed, and the vehicle absolute acceleration;

[0072] Step S23: Calculate the target state data based on the target detection data and the vehicle state data, wherein the position data is equal to the relative position of the target, the speed data is equal to the difference between the absolute speed of the target and the absolute speed of the vehicle, and the acceleration data is the difference between the absolute acceleration of the target and the absolute acceleration of the vehicle.

[0073] It is understandable that both the target and the vehicle may be in a dynamic process. By adopting the above implementation method, the vehicle can obtain relative state data of the target relative to the vehicle, which makes it easier for the vehicle to analyze the state of the target.

[0074] As an optional implementation method, such as Figure 2 As shown, in step S3, based on the assumption that all point clouds are within the target's bounding box, the shortest distance from the point cloud outside the target's bounding box to the target bounding box is used as the error constraint. The calculation equation for the position constraint error is as follows:

[0075] Ep_n=Pr-sprt(Pr_x*Pr_x+Pr_y*Pr_y),

[0076] Where Ep_n is the position constraint error of the nth point cloud, Pr is the radial distance of the point cloud, and Pr_x and Pr_y are the horizontal and vertical coordinates of the intersection point of the radial direction of the point cloud and the target, respectively.

[0077] As an optional implementation method, such as Figure 3 As shown, in step S3, the horizontal and vertical velocities of the target are decomposed into the angular directions of the point cloud, and constraints are established on the velocity data based on the radial velocity of the point cloud. The velocity constraint error is as follows:

[0078] Ev_n=Vr-Vx*sin(θ)+Vy*cos(θ),

[0079] Where Ev_n is the velocity constraint error of the nth point cloud, Vr is the radial velocity of the point cloud, Vx is the x-axis velocity of the point cloud, Vy is the y-axis velocity of the point cloud, and θ is the horizontal angle of the point cloud.

[0080] As an optional implementation, in step S5, based on the position constraint error and the velocity constraint error, and minimizing the cost function, the optimal target state data is calculated and output, including:

[0081] Step S51: Calculate the total loss cost of the target state data based on the position constraint error and the velocity constraint error;

[0082] Step S52: Determine whether the total loss cost is less than a preset deviation threshold (the deviation threshold is generally calibrated through actual engineering experience). If yes, the target state data corresponding to the total loss cost is taken as the optimal state data of the target. If no, repeat step S2 until the total loss cost is less than the deviation threshold.

[0083] Step S53: Output the optimal state data of the target.

[0084] The formulas for calculating the total loss and making the judgment are as follows:

[0085] Loss=Ev_n*Sv_n*Ev_n+Ep_n*Sp_n*Ep_n;

[0086] [a0,a1,a2,a3,b0,b1,b2,b3]=argmin(Loss);

[0087] Where S is the information weight coefficient of the error term, which is generally calibrated through practical engineering experience.

[0088] In the implementation of steps S51 to S53, the idea of ​​iteration is adopted to calculate the optimal target state data, which not only facilitates the calculation but also simplifies the workload and produces results relatively quickly.

[0089] This application also provides a target optimization device, which includes an initialization module, an acquisition module, a constraint module, and a calculation module, wherein the functions of each module are as follows:

[0090] An initialization module is used to establish and initialize an equation model corresponding to the target state data, wherein the target state data includes position data, velocity data, and acceleration data;

[0091] The acquisition module is used to acquire the target state data corresponding to the point cloud and use the target state data as the input of the equation model. There are multiple point clouds, different point clouds are acquired within different time thresholds, and different point clouds correspond to different target state data.

[0092] The constraint module is used to constrain the position data and velocity data in the target state data, and calculate the position constraint error corresponding to the position data and the velocity constraint error corresponding to the velocity data.

[0093] The calculation module is used to calculate and output the optimal target state data based on the position constraint error and the velocity constraint error, and based on minimizing the cost function.

[0094] Optionally, the acquisition module is also used for:

[0095] Target detection data associated with the point cloud is obtained through sensors, wherein the target detection data includes the target's relative position, target's absolute velocity, and target's absolute acceleration;

[0096] Obtain vehicle state data within the same time threshold as the point cloud, wherein the vehicle state data includes vehicle position, vehicle absolute speed, and vehicle absolute acceleration;

[0097] The target state data is calculated based on the target detection data and the vehicle state data, wherein the position data is equal to the target's relative position, the speed data is equal to the difference between the target's absolute speed and the vehicle's absolute speed, and the acceleration data is the difference between the target's absolute acceleration and the vehicle's absolute acceleration.

[0098] Optionally, the calculation module is also used for:

[0099] Calculate the total loss cost of the target state data based on the position constraint error and the velocity constraint error;

[0100] Determine whether the total loss cost is less than a preset deviation threshold. If so, take the target state data corresponding to the total loss cost as the optimal state data of the target. If not, repeat step S2 until the total loss cost is less than the deviation threshold.

[0101] Output the optimal state data of the target.

[0102] This application embodiment also provides a target optimization device, which includes a controller and a memory. The memory stores a computer program. When the computer program is executed by the controller, the optimization device performs the target optimization method described above.

[0103] This application also provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the target optimization method described above.

[0104] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, braking device, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0105] The above provides a detailed description of the target optimization method, apparatus, device, and storage medium provided by the present invention. The specific embodiments are described only to aid in understanding the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method of target optimization, characterized by, The target optimization method comprises: Step S1, establishing and initializing an equation model corresponding to target state data, wherein the target state data comprises position data, velocity data and acceleration data; Step S2, obtaining the target state data corresponding to the point cloud, and taking the target state data as the input quantity of the equation model, wherein the number of the point cloud is multiple, different point clouds are obtained within different time thresholds, and different point clouds correspond to different target state data; Step S3, constraining the position data and the velocity data in the target state data, calculating the position constraint error corresponding to the position data, and the velocity constraint error corresponding to the velocity data; Step S4, calculating the optimal target state data according to the position constraint error and the velocity constraint error, and outputting based on the cost function minimization.

2. The objective optimization method of claim 1, wherein In the equation model in step S1, the longitudinal and lateral directions of the target state data are separated, and it is assumed that the position data is a three-order curve equation changing with time, the velocity data is the differential of the position data, and the acceleration data is the differential of the velocity data.

3. The objective optimization method of claim 1, wherein In step S2, the method for obtaining the target state data corresponding to the point cloud comprises: Step S21, obtaining target detection data associated with the point cloud through a sensor, wherein the target detection data comprises target relative position, target absolute velocity and target absolute acceleration; Step S22, obtaining vehicle state data within the same time threshold as the point cloud, wherein the vehicle state data comprises vehicle position, vehicle absolute velocity and vehicle absolute acceleration; Step S23, calculating the target state data according to the target detection data and the vehicle state data, wherein the position data is equal to the target relative position, the velocity data is equal to the difference between the target absolute velocity and the vehicle absolute velocity, and the acceleration data is the difference between the target absolute acceleration and the vehicle absolute acceleration.

4. The objective optimization method of claim 1, wherein In step S3, based on the assumption that all the point clouds are within the target frame, the nearest distance from the point cloud outside the target frame to the target frame is taken as the error to establish the constraint, and the calculation equation of the position constraint error is as follows: ， wherein, is the position constraint error for the nth point cloud, is the radial distance of the point cloud, and are the horizontal and vertical coordinates of the intersection of the radial of the point cloud and the target, respectively.

5. The objective optimization method of claim 4, wherein In step S3, the lateral and longitudinal velocities of the target are decomposed into the angle direction of the point cloud, and the velocity data is constrained based on the radial velocity of the point cloud, and the velocity constraint error is as follows: , wherein, is the velocity constraint error for the nth point cloud, is the radial velocity of the point cloud, is the x-velocity of the point cloud, is the y-velocity of the point cloud, is the horizontal angle of the point cloud.

6. The objective optimization method of claim 5, wherein, In step S5, according to the position constraint error and the velocity constraint error, and based on the cost function minimization, the optimal target state data is calculated and outputted, comprising: Step S51, calculating the total loss cost of the target state data according to the position constraint error and the velocity constraint error; Step S52, judging whether the total loss cost is less than a preset deviation threshold, if yes, taking the target state data corresponding to the total loss cost as the optimal state data of the target, if not, repeating step S2 until the total loss cost is less than the deviation threshold. Step S53, outputting the optimal state data of the target.

7. An object optimization apparatus characterized by comprising: The target optimization device comprises: An initialization module, configured to establish and initialize an equation model corresponding to target state data, wherein the target state data comprises position data, velocity data and acceleration data; An acquisition module, configured to acquire the target state data corresponding to point clouds, and take the target state data as an input of the equation model, wherein the point clouds are multiple, different point clouds are acquired within different time thresholds, and different point clouds correspond to different target state data; A constraint module, configured to constrain the position data and the velocity data in the target state data, calculate a position constraint error corresponding to the position data, and a velocity constraint error corresponding to the velocity data; A calculation module, configured to calculate optimal target state data according to the position constraint error and the velocity constraint error, and based on cost function minimization, and output the optimal target state data.

8. An object optimization apparatus characterized by comprising: The target optimization device comprises a controller and a memory, and the memory stores a computer program, when the computer program is executed by the controller, the optimization device executes the target optimization method as claimed in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, when the computer program is executed on a computer, the computer executes the target optimization method as claimed in any one of claims 1-6.

Images