Radar point cloud-based target track velocity estimation method and control unit

Abstract

This application provides a target trajectory velocity estimation method and control unit based on radar point clouds. The method includes: acquiring radar point cloud data of the target, including the position coordinates and observed radial velocity of each reflection point on the target, where the reflection points include reflection points of rotating components and reflection points of non-rotating components; fitting a sinusoidal model using the observed radial velocities of the reflection points of rotating components; determining the radial velocity compensation amount for each reflection point of rotating components based on the sinusoidal model; compensating the observed radial velocity of the reflection points of rotating components according to the radial velocity compensation amount to obtain the compensated radial velocity; and estimating the target's trajectory velocity based on the compensated radial velocity and the observed radial velocities of the reflection points of non-rotating components. This method effectively compensates for the interference of the rotational motion of the reflection points of rotating components on the target trajectory velocity estimation while fully utilizing all radar point cloud data, thus improving the accuracy of the target trajectory velocity estimation.

Description

Technical Field

[0001] This application relates to the fields of intelligent driving perception and vehicle radar signal processing technology, and in particular to a target trajectory speed estimation method and control unit based on radar point clouds. Background Technology

[0002] With the development of intelligent driving technology, vehicle-mounted angle radar has become one of the indispensable key sensors in vehicle perception systems. By emitting and receiving electromagnetic waves, vehicle-mounted angle radar can accurately measure information such as the distance, angle, and radial velocity of targets, providing crucial data support for estimating the motion state of surrounding targets.

[0003] Regarding target trajectory velocity estimation, the following two main technical solutions exist:

[0004] The first method involves acquiring the target's vehicle-mounted radar point cloud data, which includes reflection points from the vehicle body and wheels. Based on the least squares algorithm, the target's trajectory speed is estimated using this radar point cloud data. However, because the radial velocity of the wheel reflection points is affected by rotational motion and deviates from the overall vehicle motion trend, the accuracy of the target trajectory speed estimation is poor.

[0005] The second approach involves acquiring the vehicle-mounted radar point cloud data of the target, firstly filtering out wheel reflection points from the data based on wheel detection and point cloud segmentation algorithms; then, estimating the target's trajectory speed using a least-squares algorithm based on the vehicle body reflection points. This approach performs well when there are a sufficient number of vehicle body reflection points, but when these points are sparse, the robustness of the trajectory speed estimation drops sharply.

[0006] Therefore, there is an urgent need for an effective target trajectory velocity estimation scheme to meet the high precision and high reliability requirements of intelligent driving. Summary of the Invention

[0007] This application provides a target trajectory velocity estimation method and control unit based on radar point cloud, which can effectively compensate for the interference of the rotational motion of the reflection point of the rotating component on the target trajectory velocity estimation while making full use of all radar point cloud data, thereby improving the accuracy of target trajectory velocity estimation.

[0008] Firstly, this application provides a target trajectory velocity estimation method based on radar point clouds, including:

[0009] Acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating parts and reflection points of non-rotating parts.

[0010] By using the observed radial velocities at multiple reflection points of rotating components, a sinusoidal model is obtained through fitting. The sinusoidal model characterizes the mapping relationship between the observed radial velocities and the rotation angles of the reflection points of the rotating components.

[0011] Based on the sinusoidal law model, the radial velocity compensation amount at the reflection point of each rotating component is determined.

[0012] Based on the radial velocity compensation amount, the observed radial velocity at the reflection point of the rotating component is compensated to obtain the compensated radial velocity.

[0013] The target's trajectory speed is estimated based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

[0014] In one possible implementation, the sinusoidal law model satisfies the following relationship: the observed radial velocity of the reflection point of the rotating component is equal to the sum of the reference radial velocity of the non-rotating component and the sinusoidal fluctuation term. The reference radial velocity represents the projected velocity of the non-rotating component in the radar line-of-sight direction, and the sinusoidal fluctuation term represents the periodic fluctuation of the radial velocity caused by the rotation of the rotating component.

[0015] The instantaneous phase of the sinusoidal wave term is equal to the sum of the rotation angle of the reflection point of the rotating component relative to the center of the rotating component and the phase offset of the sinusoidal wave.

[0016] In one possible implementation, the radial velocity compensation amount at the reflection point of each rotating component is determined based on a sinusoidal law model, including:

[0017] The rotation angle of the reflection point of the rotating component is obtained, and the rotation angle is determined based on the position coordinates of the reflection point of the rotating component in the radar coordinate system.

[0018] The radial velocity compensation amount of the reflection point of the rotating component is determined based on the rotation angle, the amplitude of the sinusoidal wave term, and the phase offset of the sinusoidal wave.

[0019] In one possible implementation, the reflection point of the rotating component is determined in the following way:

[0020] Based on the least squares method, the initial trajectory velocity of the target is estimated from radar point cloud data;

[0021] For each reflection point on the target, the radial velocity residual corresponding to the reflection point is determined based on the observed radial velocity and the predicted radial velocity of the reflection point. The predicted radial velocity is determined based on the initial track velocity and the unit direction vector corresponding to the reflection point.

[0022] If the absolute value of the radial velocity residual is greater than the residual threshold, then the reflection point is determined as the reflection point of the rotating component.

[0023] In one possible implementation, the residual threshold is determined as follows:

[0024] The radial velocity residuals of all reflection points on the target are sorted to obtain a residual sequence; the change characteristic value between adjacent residuals in the residual sequence is calculated, and the residual threshold is determined based on the radial velocity residual corresponding to the position where the change characteristic value jumps. The change characteristic value includes difference value or gradient value.

[0025] Alternatively, based on the statistical characteristics of radar point cloud data, a residual threshold can be determined. These statistical characteristics include at least one of the mean, median, standard deviation, and quartiles of the radial velocity residuals.

[0026] In one possible implementation, the target trajectory velocity estimation method based on radar point clouds further includes:

[0027] Acquire visual image data that is collected synchronously with radar point cloud data;

[0028] The visual image data is processed to identify texture features or rotation speed markings on the rotating parts in order to estimate the real-time angular velocity of the rotating parts.

[0029] The amplitude of the sinusoidal wave term and the phase offset of the sinusoidal wave are dynamically corrected based on the real-time angular velocity.

[0030] Secondly, this application provides a target trajectory velocity estimation device based on radar point clouds, comprising:

[0031] The acquisition module is used to acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating parts and reflection points of non-rotating parts.

[0032] The fitting module is used to fit a sinusoidal model using the observed radial velocities of multiple rotating component reflection points. The sinusoidal model represents the mapping relationship between the observed radial velocities and the rotation angles of the rotating component reflection points.

[0033] The determination module is used to determine the radial velocity compensation amount of each rotating component's reflection point based on a sinusoidal law model.

[0034] The compensation module is used to compensate the observed radial velocity of the reflection point of the rotating component according to the radial velocity compensation amount, so as to obtain the compensated radial velocity.

[0035] The estimation module is used to estimate the target's trajectory velocity based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

[0036] In one possible implementation, the sinusoidal law model satisfies the following relationship:

[0037] The observed radial velocity at the reflection point of the rotating component is equal to the sum of the reference radial velocity of the non-rotating component and the sinusoidal fluctuation term. The reference radial velocity characterizes the projected velocity of the non-rotating component in the radar line-of-sight direction, and the sinusoidal fluctuation term characterizes the periodic fluctuation of the radial velocity caused by the rotation of the rotating component.

[0038] The instantaneous phase of the sinusoidal wave term is equal to the sum of the rotation angle of the reflection point of the rotating component relative to the center of the rotating component and the phase offset of the sinusoidal wave.

[0039] In one possible implementation, the determining module is specifically used for:

[0040] The rotation angle of the reflection point of the rotating component is obtained, and the rotation angle is determined based on the position coordinates of the reflection point of the rotating component in the radar coordinate system.

[0041] The radial velocity compensation amount of the reflection point of the rotating component is determined based on the rotation angle, the amplitude of the sinusoidal wave term, and the phase offset of the sinusoidal wave.

[0042] In one possible implementation, the reflection point of the rotating component is determined in the following way:

[0043] Based on the least squares method, the initial trajectory velocity of the target is estimated from radar point cloud data;

[0044] For each reflection point on the target, the radial velocity residual corresponding to the reflection point is determined based on the observed radial velocity and the predicted radial velocity of the reflection point. The predicted radial velocity is determined based on the initial track velocity and the unit direction vector corresponding to the reflection point.

[0045] If the absolute value of the radial velocity residual is greater than the residual threshold, then the reflection point is determined as the reflection point of the rotating component.

[0046] In one possible implementation, the residual threshold is determined as follows:

[0047] The radial velocity residuals of all reflection points on the target are sorted to obtain a residual sequence; the change characteristic value between adjacent residuals in the residual sequence is calculated, and the residual threshold is determined based on the radial velocity residual corresponding to the position where the change characteristic value jumps. The change characteristic value includes difference value or gradient value.

[0048] Alternatively, based on the statistical characteristics of radar point cloud data, a residual threshold can be determined. These statistical characteristics include at least one of the mean, median, standard deviation, and quartiles of the radial velocity residuals.

[0049] In one possible implementation, the target trajectory velocity estimation device based on radar point clouds further includes a correction module for:

[0050] Acquire visual image data that is collected synchronously with radar point cloud data;

[0051] The visual image data is processed to identify texture features or rotation speed markings on the rotating parts in order to estimate the real-time angular velocity of the rotating parts.

[0052] The amplitude of the sinusoidal wave term and the phase offset of the sinusoidal wave are dynamically corrected based on the real-time angular velocity.

[0053] Thirdly, this application provides a control unit, including: a memory and a processor;

[0054] The memory stores instructions that the computer executes;

[0055] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0056] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible embodiments of the first aspect.

[0057] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0058] This application provides a target trajectory velocity estimation method and control unit based on radar point clouds. The method includes: acquiring radar point cloud data of the target, the radar point cloud data including the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system, the reflection points including reflection points of rotating components and reflection points of non-rotating components; using the observed radial velocities of multiple reflection points of rotating components, fitting a sinusoidal law model, the sinusoidal law model characterizing the mapping relationship between the observed radial velocity and the rotation angle of the reflection points of rotating components; determining the radial velocity compensation amount of each reflection point of rotating components based on the sinusoidal law model; compensating the observed radial velocity of the reflection points of rotating components according to the radial velocity compensation amount to obtain the compensated radial velocity; and estimating the target trajectory velocity based on the compensated radial velocity and the observed radial velocity of the reflection points of non-rotating components. This application acquires radar point cloud data of the target, establishes the law of its change with the rotation state for the observed radial velocity of the reflection point of the rotating component, determines the corresponding radial velocity compensation amount, compensates the observed radial velocity of the reflection point of the rotating component, and then estimates the target's trajectory velocity by combining the observed radial velocity of the reflection point of the non-rotating component. It can suppress the radial velocity deviation introduced by the rotational motion without directly eliminating the reflection point of the rotating component, thereby improving the accuracy, stability and robustness of the target trajectory velocity estimation while retaining more effective point cloud data. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] Figure 1 A schematic diagram of a scenario for the target trajectory velocity estimation method based on radar point clouds provided in the embodiments of this application;

[0061] Figure 2 A flowchart illustrating the target trajectory velocity estimation method based on radar point clouds provided in this application embodiment. Figure 1 ;

[0062] Figure 3 A flowchart illustrating the target trajectory velocity estimation method based on radar point clouds provided in this application embodiment. Figure 2 ;

[0063] Figure 4 A schematic diagram illustrating the sinusoidal law of the radial velocity of the wheel point provided in an embodiment of this application;

[0064] Figure 5 A comparison chart showing the change of target trajectory velocity estimation error over time for embodiments of this application;

[0065] Figure 6 A comparison chart of average errors in target trajectory velocity estimation provided for embodiments of this application;

[0066] Figure 7 Schematic diagram of the target trajectory velocity estimation device based on radar point cloud provided in the embodiments of this application Figure 1 ;

[0067] Figure 8 Schematic diagram of the target trajectory velocity estimation device based on radar point cloud provided in the embodiments of this application Figure 2 ;

[0068] Figure 9 This is a schematic diagram of the control unit provided in an embodiment of this application.

[0069] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0070] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0071] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0072] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0073] In intelligent driving systems, vehicle-mounted millimeter-wave radar is typically deployed at the front, sides, or rear of the vehicle to continuously acquire spatial position and radial velocity information of surrounding targets in complex traffic environments. This information provides speed decision-making support for scenarios such as meeting oncoming traffic, changing lanes, overtaking, following other vehicles at low speeds, and slow-moving traffic. Due to its advantages such as all-weather operation, long range, and strong resistance to obstruction, the point cloud data of vehicle-mounted millimeter-wave radar often becomes a crucial input for estimating the speed of targets (such as other vehicles). However, in close-range dynamic interaction scenarios, the echoes from vehicle-mounted millimeter-wave radar not only originate from non-rotating components such as the vehicle body but also clearly include reflection points from rotating components such as wheels and hubs. These reflection points simultaneously exhibit position coordinates and observed radial velocity information in the radar coordinate system, and their velocity components change periodically with rotation.

[0074] Therefore, how to utilize the velocity information of each reflection point in the radar point cloud to stably estimate the trajectory speed of the target vehicle without relying on high-cost perception links has become a very typical and practical application scenario in intelligent driving.

[0075] Current vehicle-mounted radar trajectory velocity estimation typically relies on point cloud clustering or target-level velocity fitting. This involves jointly processing multiple reflection points on the target to obtain the overall motion trend. Specifically, after millimeter-wave radar completes target detection, the system obtains radar point cloud data containing the three-dimensional positions of the reflection points and the observed radial velocities, and uses this data as input for trajectory velocity estimation. Common methods include weighted averaging of all radar point clouds or using the least squares method to fit the target trajectory velocity in the radar coordinate system.

[0076] The above method can achieve relatively acceptable estimation results when the target is mainly composed of a stable plane or vehicle body structure and the point cloud distribution is relatively uniform. However, in scenarios such as close-range oncoming traffic, overtaking and lane changing, and low-speed following, the proportion of reflection points of rotating parts in the radar point cloud will increase significantly. The observed radial velocity does not represent the actual translational motion of the vehicle body, but rather the additional velocity component caused by wheel rotation. If these reflection points of rotating parts are used for fitting together with the reflection points of non-rotating parts, the periodic fluctuations of the reflection points of rotating parts will directly contaminate the fitting results, causing the estimated values ​​to deviate, fluctuate, or even have systematic errors. If suspected reflection points of rotating parts are directly filtered out through thresholds or rules, although the influence of abnormal speeds can be reduced, the effective observation samples will be reduced simultaneously. Especially when the distance changes significantly, the point cloud is already sparse, or there is heavy local occlusion, the remaining radar point cloud number is insufficient to support stable fitting, which can easily introduce larger random fluctuations.

[0077] It is evident that existing technologies, when processing reflection points of rotating components, generally face the dilemma of "large errors when retaining all radar point clouds and insufficient data when filtering out abnormal point clouds." This makes it difficult to balance estimation accuracy, stability, and data utilization, and also fails to meet the real-time and safety requirements of intelligent driving.

[0078] Therefore, how to reduce or even eliminate the adverse effects of the reflection points of rotating components on track speed estimation while preserving effective reflection information in the target radar point cloud has become an urgent technical problem to be solved.

[0079] To address the aforementioned technical challenges, the target trajectory velocity estimation scheme based on radar point clouds provided in this application does not simply rely on eliminating reflection points from rotating components. Instead, it models and compensates for the velocity variation patterns of these reflection points, bringing them back to a velocity state consistent with the target's translational motion. This improves the accuracy and stability of target velocity estimation without sacrificing the number of effective point clouds. For close-range dynamic scenarios such as meeting oncoming traffic, lane changes, overtaking, and low-speed following, this method—based on radar point cloud observation velocity, using a sinusoidal model to achieve radial velocity compensation, and then estimating the trajectory velocity from the compensated velocity—can more effectively suppress periodic interference caused by rotating components, providing a more reliable velocity input for subsequent driving decisions.

[0080] Next, we will first explain the application scenarios of this application.

[0081] Figure 1 This is a schematic diagram illustrating a scenario for the target trajectory velocity estimation method based on radar point clouds provided in an embodiment of this application. Figure 1 As shown, the scenario includes an onboard millimeter-wave radar and at least one target within its detection range. Specifically, the onboard millimeter-wave radar is an onboard corner radar, mounted on the vehicle (e.g., on either side of the front or rear bumper), used to collect radar point cloud data of the surrounding environment. During vehicle operation, the radar applies the target trajectory velocity estimation method based on radar point clouds provided in this application to estimate the trajectory velocity of each detected target, thereby assisting in the implementation of intelligent driving functions such as adaptive cruise control, automatic emergency braking, blind spot monitoring, or lane change assist.

[0082] It should be noted that the execution entity of the target trajectory velocity estimation method based on radar point clouds provided in this application can be an on-board millimeter-wave radar or a vehicle domain controller, such as a body domain controller or an autonomous driving domain controller. When the execution entity is a domain controller, the on-board millimeter-wave radar is used to collect and transmit radar point cloud data, and the domain controller receives the radar point cloud data and performs target trajectory velocity estimation. This application does not limit the execution entity.

[0083] It should also be noted that the target trajectory velocity estimation method based on radar point clouds provided in this application is not limited to vehicle-mounted millimeter-wave radar perception scenarios, but can also be extended to other types of radar or sensor systems. Applicable moving targets include, but are not limited to: moving objects with rotating parts (such as wheels, rotors, rollers, etc.), such as automobiles, motorcycles, bicycles, robots, and drones.

[0084] To further understand the technical solution of this application, the following explanation is based on typical driving conditions. In close-range oncoming vehicle scenarios, the distance between the vehicle and the oncoming vehicle is small (e.g., 1 to 5 meters). The angle radar can detect detailed point clouds of the oncoming vehicle's wheels. In this case, the radial velocity modulation caused by wheel rotation significantly affects the speed estimation accuracy. This application can effectively improve the accuracy of trajectory speed estimation in oncoming vehicle scenarios by compensating for the wheel rotation component. In overtaking and lane-changing scenarios, the side and rear wheels of the target vehicle are within the main detection range of the angle radar. This application can improve the accuracy of the trajectory speed estimation of the vehicle in front, providing reliable data support for overtaking timing selection and lane-changing distance judgment. In low-speed following and congested scenarios, the distance between vehicles is small, and the radar can stably detect the wheel points of the vehicle in front. This application can provide a smoother and more stable speed tracking input for the adaptive cruise control system.

[0085] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0086] It should be noted that, for ease of understanding of the technical solutions of this application, the following embodiments are all illustrated using a vehicle as the target and a wheel as a rotating component. However, those skilled in the art should understand that the methods provided in this application are also applicable to other types of moving targets (such as motorcycles, bicycles, robots, drones, etc.) and other types of rotating components (such as rotors, rollers, gears, etc.).

[0087] Figure 2 A flowchart illustrating the target trajectory velocity estimation method based on radar point clouds provided in this application embodiment. Figure 1 .like Figure 2 As shown, the target trajectory velocity estimation method based on radar point clouds includes:

[0088] S201. Acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating parts and reflection points of non-rotating parts.

[0089] Radar point cloud data refers to a dataset consisting of multiple discrete reflection points. Each reflection point is associated with at least spatial position coordinates and radial velocity observations. The position coordinates characterize the relative position of the reflection point in the radar coordinate system, and the observed radial velocity characterizes the relative motion component of the reflection point along the radar's line-of-sight direction. The radar coordinate system can be a three-dimensional Cartesian coordinate system with the radar installation center as the origin, where the forward direction is the longitudinal axis, the lateral direction is the lateral axis, and the vertical direction is the height axis. Alternatively, an equivalent coordinate definition can be used depending on the vehicle installation method, as long as it can uniformly describe the positional relationship of the reflection point relative to the radar.

[0090] Reflection points from rotating components include those originating from the target vehicle's wheels, hubs, or structures rotating synchronously with the wheels. The radial velocity of these reflection points includes both translational and rotational components. Reflection points from non-rotating components include those originating from doors, bumpers, side panels, rearview mirror housings, or other structural surfaces that do not rotate with the wheel system. The radial velocity of these reflection points primarily characterizes the overall translational motion of the target.

[0091] For example, an onboard millimeter-wave radar can continuously output detection results at a fixed cycle, such as outputting a frame of point cloud data with a refresh cycle of 20 milliseconds to 100 milliseconds. Each frame of point cloud data is clustered or associated at the target level by the target detection module to obtain a set of reflection points belonging to the same target vehicle. To ensure that subsequent speed compensation is performed for the same target, in this embodiment, target tracking processing can be performed first, matching the target detection results in consecutive frames according to spatial proximity, speed continuity, and heading consistency, and assigning a unique identifier to the target. Subsequently, the position coordinates and observed radial velocity of each reflection point are extracted from the point cloud corresponding to the target identifier, forming the original radar point cloud set of the target to be processed. In one possible embodiment, a single frame of point cloud can be processed directly, thereby reducing latency and meeting real-time control requirements. In another possible embodiment, adjacent multiple frames of point clouds can be time-aligned and spatially corrected before forming a fused point cloud to alleviate the problem of insufficient point count and large noise fluctuations in a single frame. Time alignment can be achieved through vehicle self-motion compensation, and spatial correction can be achieved by unifying coordinates at different times to the same reference time using radar extrinsic parameters and installation attitude parameters.

[0092] Regarding the classification of reflection points, in this embodiment of the application, the spatial distribution characteristics of the target vehicle point cloud, the radial velocity residual characteristics, and the prior geometric position can be used to classify the reflection points.

[0093] For example, in one possible implementation, the main body region of the vehicle body and the possible region of the wheel system can be estimated based on the target bounding box or the point cloud boundary, and points located in the neighborhood of the wheel and whose speed change is significantly higher than that of the main body point can be marked as candidate rotating component reflection points.

[0094] In another possible implementation, the initial translational velocity vector of the target can be estimated using the least squares method based on all radar point cloud data. Then, for each reflection point, the predicted radial velocity is calculated based on the initial translational velocity vector, as well as the residual of the observed radial velocity of each reflection point relative to the initial translational velocity vector. Combining the residual threshold, the reflection points are divided into reflection points of rotating components and reflection points of non-rotating components. For details, please refer to the embodiments below, which will not be elaborated here.

[0095] In another possible implementation, the radial velocity of a reflection point on a rotating component fluctuates periodically with time or rotation angle, while the radial velocity of a reflection point on a non-rotating component remains relatively stable. Therefore, point cloud data of the target vehicle can be acquired across multiple consecutive frames, and each reflection point can be tracked to obtain its radial velocity time series. Spectral analysis (such as Fast Fourier Transform) can be performed on the radial velocity time series of each reflection point to extract its frequency components. If a significant periodic component corresponding to the wheel rotation frequency exists in the radial velocity time series of a reflection point, that point is identified as a reflection point of a rotating component; otherwise, it is identified as a reflection point of a non-rotating component.

[0096] Through the above method, the system obtained target vehicle point cloud data that simultaneously includes reflection points of rotating and non-rotating components, providing a data foundation for the subsequent establishment of a rotation speed model.

[0097] It should be understood that S201 not only acquires target-level radar observation data but also performs point-level classification directly related to velocity compensation, ensuring that subsequent processing does not treat all reflection points as homogeneous observations. Based on the above analysis, if reflection points of rotating and non-rotating components are not distinguished before trajectory velocity estimation, the additional radial velocity introduced by wheel rotation will directly enter the overall fitting process, leading to offsets and jitter in the estimation results. However, this embodiment establishes point-level structural attributes simultaneously during the data acquisition stage, enabling targeted modeling and compensation of reflection points of rotating components, thus creating conditions for improving the accuracy and stability of trajectory velocity estimation from the source.

[0098] S202. By using the observed radial velocities at multiple reflection points of rotating components, a sinusoidal law model is obtained through fitting. The sinusoidal law model characterizes the mapping relationship between the observed radial velocity and the rotation angle of the reflection points of the rotating components.

[0099] In this embodiment, a sinusoidal model is used to describe the periodic radial velocity change of the reflecting point of the rotating component due to the wheel's rotation around its axle. The rotation angle refers to the angular position of the reflecting point of the rotating component relative to the center of rotation in the wheel plane, reflecting the velocity projection relationship corresponding to different orientations of the reflecting point on the wheel rim. Since the instantaneous velocity direction of any fixed scattering point on the wheel changes periodically with the angle during rotation, its projection in the radar line-of-sight direction usually changes in a sinusoidal form. Therefore, a sinusoidal function can be used to model the additional rotational velocity.

[0100] For example, the sinusoidal law model can be expressed as: , This represents the observed radial velocity of the reflection point of the rotating component at a certain rotation angle; This represents the reference radial velocity related to the translation of the target vehicle. This indicates the amplitude of the velocity fluctuation caused by rotation; This represents phase parameters related to radar observation angle, gear train reference zero position, installation geometry, etc. This represents the current rotation angle corresponding to the reflection point of the rotating component. In the above model, the reference velocity, amplitude, and phase together determine the law of change of the velocity of the reflection point of the rotating component with the angle. Among them, the reference velocity reflects the overall translation trend of the target, the amplitude reflects the radial projection intensity of the rotational velocity of the gear train, and the phase reflects the spatial observation geometric difference.

[0101] In the specific fitting process, the rotation angles corresponding to the reflection points of multiple rotating components can be determined first. The rotation angles can be obtained by estimating the wheel's rotation center and calculating the geometric angles between the reflection points and the rotation center. For example, a cluster of points approximating the tire region can be identified in the radar point cloud of the target. The local center of this cluster in the vehicle's lateral and vertical directions can be used as the approximate location of the rotation center, and the geometric angles can be calculated using the relative coordinates of the reflection points and the rotation center. If the radar is mainly deployed in a forward or lateral position, the calculation of the rotation angles can be corrected by combining the wheel plane normal, vehicle attitude, and radar line-of-sight direction to ensure that the obtained angles are consistent with the radial velocity projection relationship. In another possible implementation, when the distribution of rotating component reflection points within a single frame is insufficient, rotating component reflection points already associated with the same wheel region in several adjacent frames can be introduced. Angle and velocity samples can be accumulated through a sliding window to improve the fitting stability.

[0102] After obtaining the correspondence between the rotation angle and the observed radial velocity for multiple sample points, the parameters of the sinusoidal law model can be solved using a nonlinear least squares algorithm. Specifically, an objective function can be constructed to minimize the sum of squared errors between the observed radial velocities of all rotating component reflection points and the output velocities of the sinusoidal law model. To reduce the impact of outliers on the fitting results, robust weights can also be introduced into the objective function in this embodiment, reducing the weights for samples with low signal-to-noise ratios, unstable angle estimations, or large velocity mutations.

[0103] In one possible embodiment, the overall translational velocity of the target is first roughly estimated based on the reflection point of the non-rotating component, and this rough velocity is used as a parameter. The initial values ​​are then obtained through iterative optimization to obtain the parameters. and parameters This accelerates the convergence speed.

[0104] In another possible embodiment, parameters can also be simultaneously... ,parameter and parameters A joint solution is performed, and the fitting results of historical frames are used as priors. The model parameters are updated in time using Kalman filtering or exponential smoothing to adapt to model drift caused by the acceleration and deceleration of the target vehicle, attitude changes and radar observation angle changes.

[0105] Step S202 is crucial for resolving the velocity contamination problem of rotating components. Its core lies not in simply identifying the reflection points of the rotating components, but in explicitly establishing a functional mapping relationship between the velocity of these reflection points and the rotation angle. Since the deviations caused by the reflection points of the rotating components exhibit significant periodicity, using only threshold filtering would result in a significant reduction in effective samples. However, by fitting a sinusoidal model, the velocity changes, originally considered interference, can be transformed into interpretable, predictable, and compensable structured information. Based on the above analysis, this embodiment of the application, by introducing a sinusoidal model, transforms the periodic radial velocity changes caused by wheel rotation from a random noise problem into a parametric modeling problem, thereby providing a calculable basis for subsequent point-by-point compensation. While preserving the effective spatial distribution information of the reflection points of the rotating components, it improves the overall point cloud utilization and the stability of trajectory velocity estimation.

[0106] S203. Based on the sinusoidal law model, determine the radial velocity compensation amount of the reflection point of each rotating component.

[0107] In this embodiment, the radial velocity compensation amount refers to the correction value used to eliminate the additional radial velocity component introduced by the rotational motion at the reflection point of the rotating component. Since a sinusoidal law model characterizing the mapping relationship between the observed radial velocity and the rotation angle has been obtained in S202, the velocity deviation caused by rotation at each reflection point of the rotating component can be calculated from the sinusoidal law model based on the rotation angle corresponding to its current position.

[0108] To improve the stability of the calculation results, reasonable constraints can be set on the radial velocity compensation amount in this embodiment. For example, the absolute value of the radial velocity compensation amount should not exceed the theoretical upper limit derived from the target wheel diameter estimation, the target's current translational speed, and the physical relationship between wheel speed, thereby avoiding excessive radial velocity compensation amount due to individual outliers, angular errors, or fitting distortion. Smoothing can also be performed based on the radial velocity compensation amount change trend in the neighborhood of the same reflection point in adjacent frames, making the radial velocity compensation amount change continuously over time and preventing single-frame model jitter from propagating to the final velocity estimation result. In another possible implementation, point-level confidence can be introduced, applying only a portion of the radial velocity compensation amount to the reflection points of rotating components that are far away, have low signal-to-noise ratios, or poor angular geometry, to balance correction effect and robustness.

[0109] Step S203 applies the fitted global periodicity pattern to each reflection point of the rotating component, achieving point-level, sample-by-sample quantification of velocity deviation. Based on the above analysis, if only the overall periodic deviation of the rotating component's reflection points is recognized, but corresponding correction values ​​cannot be calculated for each specific reflection point, then the translational velocity state represented by each point cannot be accurately recovered. However, this embodiment calculates a one-to-one radial velocity compensation amount based on the current rotation angle, improving the error correction of the rotating component's reflection points from coarse-grained suppression to fine-grained compensation, thereby effectively reducing the estimation error caused by inconsistent velocity deviations at different angular positions.

[0110] S204. Based on the radial velocity compensation amount, compensate the observed radial velocity at the reflection point of the rotating component to obtain the compensated radial velocity.

[0111] In this embodiment, the compensation process can be understood as correcting the original observed value (i.e., the observed radial velocity) to transform the velocity of the rotating component's reflection point from a mixed state of "vehicle translational velocity component plus rotational additional component" to a state closer to "only representing the vehicle translational velocity component". If the radial velocity compensation amount obtained in S203 is defined as the rotational additional velocity, then the compensated radial velocity of the current rotating component's reflection point can be obtained by "subtracting the radial velocity compensation amount from the observed radial velocity".

[0112] For example, iterate through all the identified rotating component reflection points in the current frame, read the observed radial velocity and the radial velocity compensation amount calculated in S203 for each rotating component reflection point, perform point-level subtraction operation to generate the compensated radial velocity, and store the result together with the original spatial coordinates of the rotating component reflection point into the updated point cloud data structure.

[0113] To avoid extreme outliers affecting subsequent overall trajectory velocity estimation, this embodiment can add a rationality check after compensation. For example, if the radial velocity of a rotating component reflection point after compensation still deviates too much from the main velocity distribution of non-rotating component reflection points in the same frame, it is determined that the rotating component reflection point may have an angle estimation error, misclassification, or abnormal measurement noise, and further truncation or elimination processing can be performed on the point. In another possible implementation, the continuity of velocity change before and after compensation can also be constrained. If the radial velocity compensation result of a rotating component reflection point has a jump variable exceeding a threshold in adjacent frames, the compensation result of the previous frame and the sinusoidal law model output of the current frame are fused to improve temporal smoothness.

[0114] Compared to directly filtering out reflection points from rotating components, the embodiments of this application can maintain a sufficient number of effective samples even when the point cloud is sparse, there is significant local occlusion, and the target distance varies greatly. Therefore, it is beneficial to reduce random fluctuations caused by insufficient samples. Based on the above analysis, by compensating for rather than eliminating reflection points from rotating components, the periodic velocity contamination caused by wheel rotation is weakened on the one hand, and the number and geometric distribution integrity of reflection points are preserved on the other hand, so that subsequent trajectory velocity estimation can simultaneously take into account both accuracy and stability.

[0115] S205. Estimate the target's trajectory speed based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

[0116] The trajectory velocity can be understood as the speed of the target along its actual trajectory. Its output can be a scalar velocity value (i.e., rate) or a velocity vector component in the radar coordinate system, vehicle coordinate system, or ground coordinate system. Since the compensated velocities of the rotating component reflection points have largely eliminated the additional effects of wheel rotation, these points, together with the reflection points of non-rotating components, constitute a more complete and consistent set of target motion observations. The radial velocities of all reflection points in this set primarily reflect the overall translational motion of the target, thus providing high-quality input data for subsequent trajectory velocity estimation. When using this set for trajectory velocity estimation, least squares fitting, weighted least squares fitting, robust regression, or a state estimation method based on Bayesian filtering can be employed.

[0117] For example, for each reflection point The line-of-sight vector corresponding to its spatial location With the target velocity vector to be estimated Establish the observation equation:

[0118]

[0119] In the formula, For the first The observed radial velocity at each reflection point is as follows: for reflection points of rotating components, the compensated radial velocity is used; for reflection points of non-rotating components, the original observed radial velocity is used. To point from the radar to the first Unit direction vector of each reflection point; Let V be the target velocity vector to be estimated. For the first Measurement error at each reflection point.

[0120] for A number of reflection points (including compensated rotating component points and uncompensated non-rotating component points) can be used to construct... The observation equations (i.e., the overdetermined equations) are solved using the least squares method to obtain an estimate of the target velocity vector.

[0121]

[0122] Optionally, the current frame estimation result and the historical frame velocity estimation result can be input into a Kalman filter to obtain the final trajectory velocity output after time-smoothed processing, thereby meeting the requirements of the intelligent driving system for continuous and stable velocity input.

[0123] In practical applications, the obtained target trajectory speed can be further provided to modules such as oncoming vehicle decision-making, lane-change risk assessment, following vehicle control, collision warning, or target trajectory prediction. Because this embodiment retains the compensated effective speed information of the rotating component reflection points, it can maintain high speed calculation stability even in scenarios with a small number of target reflection points, insufficient main reflection surfaces, and a high proportion of wheel system points in close-range observations. Compared to related technologies that include all rotating component reflection points in the fitting process, leading to increased systematic errors, or that filter out all rotating component reflection points, resulting in insufficient samples, this embodiment eliminates rotational speed interference while preserving point cloud utilization. Based on the above analysis, the trajectory speed estimation step, by jointly using the compensated rotating component reflection points and the non-rotating component reflection points unaffected by rotation, enables the target trajectory speed estimation algorithm to simultaneously possess a sufficient number of samples and high observation consistency, thereby significantly improving the estimation accuracy, robustness, and real-time applicability in close-range dynamic interaction scenarios.

[0124] In this embodiment, radar point cloud data of the target is acquired, and the observed radial velocity of the rotating component reflection point is established to determine the law of its change with the rotation state and the corresponding radial velocity compensation amount is determined. The observed radial velocity of the rotating component reflection point is compensated, and the target's trajectory velocity is estimated by combining the observed radial velocity of the non-rotating component reflection point. This can suppress the radial velocity deviation introduced by the rotational motion without directly eliminating the rotating component reflection point, thereby improving the accuracy, stability and robustness of the target trajectory velocity estimation while retaining more effective point cloud data.

[0125] In some embodiments, the sinusoidal law model satisfies the following relationship: the observed radial velocity of the reflection point of the rotating component is equal to the sum of the reference radial velocity of the non-rotating component and the sinusoidal wave term. The reference radial velocity represents the projected velocity of the non-rotating component in the radar line-of-sight direction, and the sinusoidal wave term represents the periodic fluctuation of the radial velocity caused by the rotation of the rotating component. The instantaneous phase of the sinusoidal wave term is equal to the sum of the rotation angle of the reflection point of the rotating component relative to the center of the rotating component and the sinusoidal wave phase offset.

[0126] For example, this sinusoidal law model can be expressed mathematically as follows:

[0127]

[0128] in, This represents the observed radial velocity of the reflection point of the rotating component at a certain rotation angle; This represents the baseline radial velocity related to the translation of the target vehicle. It characterizes the projected velocity of non-rotating components (such as the vehicle body and frame) on the target in the radar line-of-sight direction. For the target vehicle, the non-rotating components undergo overall translational motion with the target. It reflects the projection component of the target's translational velocity vector along the line connecting the radar and the target; This is a sinusoidal fluctuation term, which exhibits periodic fluctuation characteristics; The amplitude of the speed fluctuation reflects the combined effect of the wheel spokes, tire radius, and rotational speed of the rotating parts on the radial speed change. The larger the amplitude, the more obvious the modulation of the observed radial speed by the rotating parts. The sinusoidal wave phase offset represents the inherent phase offset of the system and is related to the radar installation location, the wheel position relative to the vehicle body, etc. It represents the instantaneous phase of the sinusoidal wave term.

[0129] In practical implementation, a fitting dataset can be constructed first based on the time-series observations of radial velocity samples at the reflection points of the rotating component, and then the baseline radial velocity can be obtained through nonlinear least squares estimation. Speed ​​fluctuation amplitude Phase offset of sinusoidal wave This allows the sinusoidal law model to accurately reflect the speed fluctuation characteristics of vehicles under conditions such as meeting oncoming traffic, changing lanes, or following other vehicles at low speeds.

[0130] This sinusoidal model shows that the observed radial velocity at the reflection point of the rotating component is composed of two superpositions: a reference radial component that translates with the overall target. and the periodic fluctuation component caused by the rotation of the rotating component. .

[0131] In this embodiment, the observed radial velocity of the rotating component's reflection point is decomposed into a reference radial velocity and a periodic fluctuation, allowing the additional velocity caused by rotation to be quantitatively characterized. Therefore, during velocity compensation processing, the velocity of the rotating component's reflection point can be corrected to match the vehicle's translational motion, and then used together with the reflection point of the non-rotating component to participate in trajectory velocity estimation, thereby reducing the interference of tire rotation on the overall velocity fitting.

[0132] In some embodiments, the radial velocity compensation amount of each rotating component reflection point is determined based on a sinusoidal law model, including: obtaining the rotation angle of the rotating component reflection point, the rotation angle being determined based on the position coordinates of the rotating component reflection point in the radar coordinate system; and determining the radial velocity compensation amount of the rotating component reflection point based on the rotation angle, the amplitude of the sinusoidal wave term, and the phase offset of the sinusoidal wave.

[0133] For example, the rotation angle can be calculated from the geometric relationship between the coordinates of the reflection point in the radar coordinate system and the coordinates of the rotation center. First, obtain the position coordinates of the reflection point of the rotating component in the radar coordinate system, and calculate the rotation angle of the reflection point relative to the center of the rotating component based on these position coordinates and the center position of the rotating component (which can be obtained through target geometric prior or point cloud fitting); then, substitute the rotation angle, the amplitude of the sinusoidal fluctuation term, and the phase offset of the sinusoidal fluctuation into the radial velocity compensation formula: Calculated This is the radial velocity compensation amount at the reflection point of the rotating component.

[0134] As can be seen from the above analysis, since the radial velocity compensation varies with the rotation angle, it is possible to implement differentiated corrections for reflection points at different positions such as the rim and spokes, thereby avoiding the direct inclusion of the periodic fluctuations of rotating components into the trajectory velocity estimation process.

[0135] By employing the aforementioned method for calculating radial velocity compensation, the reflection points of rotating components are no longer simply eliminated. Instead, they are regressed to the vehicle's translational velocity range through angle-related velocity compensation, thereby improving point cloud utilization and reducing velocity fitting bias. Simultaneously, this radial velocity compensation method reduces the impact of tire speed variations and vehicle attitude changes on velocity estimation results, making the target trajectory velocity output more stable and continuous. This, in turn, enhances the accuracy of velocity perception for vehicles ahead, oncoming vehicles, and lane-changing vehicles in intelligent driving scenarios.

[0136] In some embodiments, the reflection point of the rotating component is determined in the following manner:

[0137] Step 1.1: Based on the least squares method, estimate the target's initial trajectory speed according to radar point cloud data.

[0138] This step is used to obtain a preliminary estimate of the target trajectory velocity, which will serve as the benchmark for subsequent residual calculations and reflection point classification.

[0139] For example, this step can employ an estimation method similar to that of step S205. The difference is that step S205 (final track velocity estimation) uses the compensated radial velocity (for the reflection point of the rotating component) and the original observed radial velocity (for the reflection point of the non-rotating component); while this step (initial track velocity estimation) uses the original observed radial velocity of all reflection points, that is, no compensation is made for the radial velocity of any reflection point.

[0140] In this embodiment, the least squares method can be ordinary least squares, weighted least squares, or robust least squares. This application does not limit the specific least squares method. Ordinary least squares solves for the target velocity by minimizing the sum of squares between the observed radial velocity and the predicted radial velocity. Weighted least squares assigns different weights based on the signal-to-noise ratio or distance of the reflection point to improve the robustness of the fit. Robust least squares reduces the impact of outliers on the fitting results by introducing a robust weight function.

[0141] Step 1.2: For each reflection point on the target, determine the radial velocity residual corresponding to the reflection point based on the observed radial velocity and the predicted radial velocity. The predicted radial velocity is determined based on the initial track velocity and the unit direction vector corresponding to the reflection point.

[0142] For example, the unit direction vector of each reflection point With initial trajectory speed By projecting the image, the predicted radial velocity of the reflection point under the assumption of overall target translation is obtained. Then, the predicted radial velocity is subtracted from the observed radial velocity directly measured by the radar to form the radial velocity residual.

[0143] Step 1.3: If the absolute value of the radial velocity residual is greater than the residual threshold, then the reflection point is determined as the reflection point of the rotating component.

[0144] The residual threshold can be preset based on the radar measurement noise level, the sparsity of the point cloud, and the target speed range, or it can be adaptively adjusted according to the scene statistical characteristics to improve the robustness of identifying reflection points of rotating parts such as hubs and wheels.

[0145] It should be understood that the radial velocity residual is used to reflect the degree of deviation between the observed radial velocity and the radial velocity predicted based on the overall motion of the target. When the degree of deviation exceeds the residual threshold, it indicates that the velocity component of the reflection point is greatly affected by the rotational motion and is determined to be a reflection point of a rotating component; otherwise, it is determined to be a reflection point of a non-rotating component.

[0146] The working principle of this embodiment is to first establish a prediction benchmark, and then identify outliers by using the deviation between the observed radial velocity and the predicted radial velocity. Since the reflection points of rotating components are superimposed with the additional radial velocity caused by wheel rotation, their residuals are usually significantly larger than those of the static structure reflection points of the vehicle body. Therefore, they can be effectively separated from the target by threshold determination. In this way, subsequent compensation for the reflection points of rotating components and trajectory velocity estimation can be based on more accurate classification, thereby avoiding mistaking points such as tires and wheel hubs as samples of the overall vehicle translational velocity.

[0147] This implementation method improves the accuracy of rotating component reflection point identification while preserving effective point cloud information, reduces the interference of abnormal speeds on trajectory speed estimation, and enhances the stability and consistency of target trajectory speed estimation in scenarios such as meeting oncoming traffic, lane changing, overtaking, and low-speed following. Furthermore, identifying rotating component reflection points does not require complex point cloud segmentation and wheel detection algorithms; it is accomplished solely based on radar point cloud and least-squares estimation results, reducing computational complexity.

[0148] In some embodiments, the residual threshold is determined by: sorting the radial velocity residuals of all reflection points on the target to obtain a residual sequence; calculating the change characteristic value between adjacent residuals in the residual sequence, and determining the residual threshold based on the radial velocity residual corresponding to the position where the change characteristic value jumps, wherein the change characteristic value includes difference value or gradient value; or, determining the residual threshold based on the statistical characteristics of radar point cloud data, wherein the statistical characteristics include at least one of the mean, median, standard deviation and quartiles of the radial velocity residuals.

[0149] In one implementation, the radial velocity residuals of all reflection points on the target are sorted to obtain a residual sequence; the change characteristic value between adjacent residuals in the residual sequence is calculated, and the residual threshold is determined based on the radial velocity residual corresponding to the position where the change characteristic value jumps. The change characteristic value includes difference value or gradient value.

[0150] The residual sequence is used to arrange all reflection points in order of residual magnitude, thereby making it clearer to distinguish the reflection points of non-rotating parts (such as the vehicle body) from those of rotating parts (such as the wheels) in terms of residual value distribution. The position where the difference value jumps usually corresponds to the boundary where the residual transitions from a relatively concentrated area to an outlier area. The position where the gradient value jumps can also characterize the inflection point of this boundary. Therefore, the residual threshold adapted to the current radar point cloud distribution can be determined accordingly.

[0151] For example, the method for identifying the location (i.e., the jump point) where the changing feature value changes includes at least one of the following: determining the location with the largest difference value as the jump point; calculating the difference of the differences (i.e., the second gradient), and determining the location corresponding to the peak of the second gradient as the jump point. Then, the radial velocity residual corresponding to the jump point is determined as the residual threshold, or the average of the two radial velocity residuals before and after the jump point is used as the residual threshold.

[0152] In another implementation, the residual threshold is determined based on the statistical characteristics of radar point cloud data. The statistical characteristics include at least one of the mean, median, standard deviation, and quartiles of the radial velocity residuals.

[0153] Among them, statistical properties are used to characterize the overall distribution of residuals. When the number of radar point clouds is small or the local distribution is discontinuous, the residual threshold can be set directly based on the mean, median, standard deviation or quartiles to improve the stability of residual threshold determination.

[0154] For example, calculate the statistical characteristics of the radial velocity residuals at all reflection points. These statistical characteristics include the mean, median, standard deviation, or quartiles of the radial velocity residuals. Based on these statistical characteristics and empirical values, determine the residual threshold. For example, using the mean multiple method ( , The mean, Standard deviation, (Preset coefficients, with values ​​ranging from 1.5 to 3), median multiple method, standard deviation multiple method ( ), interquartile range method ( , This is a preset coefficient, with a value range of, for example, 1 to 2. It is the third quartile. (The first quartile).

[0155] In this embodiment, by sorting the radial velocity residual distribution of reflection points and identifying abrupt change points, the residual threshold is made independent of fixed empirical values ​​and adaptively changes with the actual distribution of the current target radar point cloud. This allows for a more accurate distinction between vehicle body stability reflection points and reflection points from rotating components such as tires and wheel hubs. Because the residual threshold can be dynamically adjusted with scene changes, it maintains good recognition stability even in situations with sparse point clouds, strong occlusion, or significant target attitude changes, and provides reliable input for subsequent radial velocity compensation and trajectory velocity estimation.

[0156] In some embodiments, the target trajectory velocity estimation method based on radar point cloud further includes: acquiring visual image data synchronously collected with radar point cloud data; processing the visual image data to identify texture features or rotation speed indicators on the rotating component in order to estimate the real-time angular velocity of the rotating component; and dynamically correcting the amplitude of the sinusoidal wave term and the phase offset of the sinusoidal wave term based on the real-time angular velocity.

[0157] It should be understood that, in dynamic scenarios such as the target vehicle accelerating, decelerating, or turning, the amplitude in the sinusoidal model... Phase offset of sinusoidal wave It changes over time and relies on frame-by-frame or sliding window fitting of the radar point cloud for updating. and When the point cloud is sparse or noisy, there may be update delays or estimation errors. This embodiment introduces a visual sensor to directly observe the motion state of the rotating component, thereby achieving [the desired effect]. and Real-time dynamic correction.

[0158] For example, while the vehicle-mounted millimeter-wave radar collects point cloud data, visual image data of the target is simultaneously acquired by camera devices deployed at the front, sides, or rear of the vehicle. The vehicle-mounted millimeter-wave radar and the camera devices have completed time synchronization and spatial calibration to ensure that the image data and point cloud data are aligned in time and space. The visual image data can be RGB images, grayscale images, or images after distortion correction. The camera devices can be global shutter industrial cameras, vehicle-mounted surround-view cameras, or front-view cameras. In practical applications, other models of camera devices can also be selected, and this application embodiment does not limit this.

[0159] For example, when processing visual image data, the target vehicle can be detected and segmented first, and then texture features can be extracted from the areas near the tires, rims, or axles. Texture features can include tire treads, rim stripes, reflective markings, coded patterns, or corner features. The displacement changes of the texture in consecutive frames can be calculated by combining optical flow estimation, feature matching, or deep learning recognition models. The pixel displacement can be converted into the actual rotation angle by combining the camera projection model and the wheel geometry parameters (radius), and then divided by the inter-frame time interval to obtain the real-time angular velocity of the rotating component.

[0160] Taking angular velocity estimation based on wheel hub stripe feature point tracking as an example, assuming the target vehicle's wheel radius is 0.3 meters, and the camera on the vehicle is positioned at 30 frames per second (i.e., the inter-frame time interval)... Images are acquired in seconds. The target vehicle's wheel area is located in the image using a target detection algorithm, and a clear spoke edge on the wheel hub is extracted as a feature point. In the second... In the frame, the image coordinates of this feature point are ( ) pixels, in the In the frame, the coordinates of the matched feature points are ( If the pixel is 0, then the pixel displacement is 0. , The total pixel displacement is Pixel, actual arc length displacement , Distance to the target vehicle. Let be the camera focal length; then the change in rotation angle... , Given the wheel radius, the real-time angular velocity is further obtained. .

[0161] If a rotation speed indicator is provided on the rotating component, the wheel speed can be calculated and the real-time angular velocity can be obtained by recognizing changes in the angular position of the rotation speed indicator, changes in the direction of the barcode, or the time interval between repeated appearances of the mark.

[0162] After obtaining the real-time angular velocity, one implementation method combines the prior geometric parameters of the rotating component to dynamically correct the amplitude parameters. ,For example, , The radius of the rotating component (which can be obtained through vehicle geometry priors or visual measurements). For real-time angular velocity, This is the angle between the radar line-of-sight direction and the rotation plane of the rotating component (calculated using the relative geometry of the radar and the target). In another implementation, if the radius of the rotating component... and included angle If the information is unknown or contains errors, it can also be corrected in the following ways: , The preset scaling factor can be obtained through offline calibration or online learning.

[0163] sinusoidal wave phase offset Frame-by-frame matching and correction can be performed by visually tracking specific reference points on rotating components (such as valve stems or wheel spoke clearances). For example, the real-time angular position of this reference point can be determined. According to the sinusoidal law model, the radial velocity at this reference point should be: , combined The corrected sinusoidal wave phase offset can be obtained by inverse solution. .

[0164] The correction can be performed in each radar sampling cycle, or it can be triggered to update when the angular velocity change exceeds a preset threshold.

[0165] In this embodiment, by introducing visual image data acquired synchronously with the radar, the system can still obtain external constraints on the wheel speed state when the point cloud of the rotating component is sparse, the radar echo is blocked, or the wheel speed changes rapidly. Based on this, the sinusoidal law model is dynamically corrected, thereby improving the accuracy of radial speed compensation of the rotating component, reducing the estimation deviation caused by changes in wheel speed, viewing angle, installation attitude, or radar error, and thus improving the stability and robustness of target trajectory speed estimation.

[0166] Based on the above embodiments, by Figure 3 A detailed explanation of the target trajectory velocity estimation method based on radar point clouds is provided.

[0167] Figure 3 A flowchart illustrating the target trajectory velocity estimation method based on radar point clouds provided in this application embodiment. Figure 2 ,like Figure 3 As shown, the target trajectory velocity estimation method based on radar point clouds includes:

[0168] S301. Acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating parts and reflection points of non-rotating parts.

[0169] For example, an onboard millimeter-wave radar can continuously output detection results at a fixed period, such as outputting a frame of point cloud data with a refresh period of 20 milliseconds to 100 milliseconds. Each frame of point cloud data is clustered or associated at the target level by the target detection module to obtain a set of reflection points belonging to the same target vehicle.

[0170] For example, consider the point cloud of the right rear wheel of the target vehicle (i.e., another vehicle) in the left front corner of the vehicle's onboard angle radar. The target vehicle is traveling at a speed of 15.0 m / s, and the onboard angle radar is located at the left front corner of the vehicle. It collects a mixed point cloud containing 50 body points (reflection points of non-rotating parts) and 50 wheel points (reflection points of rotating parts). The wheel points are randomly distributed on the circumference of the wheel.

[0171] S302. Based on radar point cloud data, estimate the target's initial trajectory velocity and analyze the radial velocity residuals corresponding to all reflection points to identify the reflection points of rotating components.

[0172] For example, this step may specifically include: estimating the initial trajectory velocity of the target based on radar point cloud data using the least squares method; determining the radial velocity residual corresponding to each reflection point on the target based on the observed radial velocity and the predicted radial velocity of the reflection point; and determining the reflection point as the reflection point of the rotating component if the absolute value of the radial velocity residual is greater than the residual threshold.

[0173] For details on the specific implementation principle, please refer to steps 1.1 to 1.3, which will not be repeated here.

[0174] For example, using all 100 point cloud data points, the target vehicle's three-dimensional velocity vector, i.e., the initial trajectory velocity, is initially estimated using the least squares method. Then, the radial velocity residual (i.e., the observed radial velocity minus the predicted radial velocity) is calculated for each reflection point, and the standard deviation of the radial velocity residual is statistically analyzed. Finally, reflection points with radial velocity residuals greater than 1.5 times the standard deviation are marked as wheel points.

[0175] S303. By using the observed radial velocities at multiple reflection points of rotating components, a sinusoidal law model is obtained through fitting. The sinusoidal law model characterizes the mapping relationship between the observed radial velocity and the rotation angle of the reflection points of the rotating components.

[0176] For example, the sinusoidal law model can be expressed mathematically as follows:

[0177]

[0178] In the formula, This represents the observed radial velocity of the reflection point of the rotating component at a certain rotation angle; This represents the reference radial velocity related to the translation of the target vehicle. This refers to the amplitude of the velocity fluctuation caused by rotation; This represents the phase offset of the sinusoidal wave. This indicates the current rotation angle corresponding to the reflection point of the rotating component.

[0179] Specifically, based on the point cloud data of the aforementioned 50 wheel points, the relationship between the radial velocity and rotation angle of the wheel points is analyzed. A sine curve is fitted using the nonlinear least squares method to obtain the parameters: -14.565 m / s, 1.828 m / s 1.590 rad, the fitted sinusoidal law model is: .

[0180] For example, Figure 4 This is a schematic diagram illustrating the sinusoidal law of the radial velocity of the wheel point provided in an embodiment of this application, as shown below. Figure 4 As shown, the horizontal axis represents the rotation angle of the wheel point relative to the wheel center. (Unit: rad), the vertical axis represents the radial velocity of the wheel point (unit: m / s). Figure 4 The graph includes discrete dots and a solid curve. The discrete dots represent measured data samples of multiple reflection points on the target vehicle's wheels, with each dot corresponding to a wheel point. The horizontal axis represents the rotation angle of that point, and the vertical axis represents the observed radial velocity. The solid curve represents the fitted curve obtained by sinusoidally fitting the discrete dots. Figure 4 It can be seen that the radial velocity of the wheel point exhibits obvious periodic fluctuation characteristics with the rotation angle.

[0181] It should be noted that, Figure 4 The discrete point distribution and fitted curve shape shown are merely illustrative diagrams. In actual applications, due to differences in target vehicle type, driving status, radar installation location, environmental noise, and other factors, the distribution range of discrete points and the specific parameter values ​​of the fitted curve will vary. , , The magnitudes of the fitting residuals may vary, but the overall characteristic of the radial velocity exhibiting a sinusoidal pattern with the rotation angle remains unchanged.

[0182] S304. Based on the sinusoidal law model, determine the radial velocity compensation amount of the reflection point of each rotating component.

[0183] For example, for the wheel point, based on its angle on the wheel... The radial velocity compensation amount is calculated using the following formula: .

[0184] S305. Based on the radial velocity compensation amount, compensate for the observed radial velocity at the reflection point of the rotating component to obtain the compensated radial velocity.

[0185] For example, the compensated radial velocity is obtained by subtracting the corresponding radial velocity compensation amount from the observed radial velocity at the wheel point. :

[0186]

[0187] S306. Estimate the target's trajectory speed based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

[0188] For example, for each reflection point The line-of-sight vector corresponding to its spatial location With the target velocity vector to be estimated Establish the observation equation:

[0189]

[0190] In the formula, For the first The observed radial velocity at each reflection point is as follows: for reflection points of rotating components, the compensated radial velocity is used; for reflection points of non-rotating components, the original observed radial velocity is used. To point from the radar to the first Unit direction vector of each reflection point; Let V be the target velocity vector to be estimated. For the first Measurement error at each reflection point.

[0191] for A number of reflection points (including compensated rotating component points and uncompensated non-rotating component points) can be used to construct... The observation equations (i.e., the overdetermined equations) are solved using the least squares method to obtain an estimate of the target velocity vector.

[0192]

[0193] For example, Figure 5 A comparison chart showing the change of target trajectory velocity estimation error over time, provided in the embodiments of this application, is shown below. Figure 5 As shown, the horizontal axis represents time (in seconds), and the vertical axis represents the trajectory velocity estimation error (in m / s). Curve 1 shows the curve of the estimated target trajectory velocity error versus time without radial velocity compensation, curve 2 shows the curve of the estimated target trajectory velocity error versus time with radial velocity compensation, dashed line a represents the average error of the estimated target trajectory velocity without radial velocity compensation, and dashed line b represents the average error of the estimated target trajectory velocity with radial velocity compensation.

[0194] from Figure 5It can be seen that the fluctuation amplitude of curve 2 is significantly smaller than that of curve 1, indicating that the target trajectory velocity estimation method provided in this application embodiment can effectively suppress the error fluctuation of trajectory velocity estimation and improve the estimation stability; the position of dashed line b is significantly lower than that of dashed line a, indicating that after radial velocity compensation, the average error of trajectory velocity estimation is significantly reduced.

[0195] Figure 6 A comparison chart of the average error of target trajectory velocity estimation provided in the embodiments of this application, such as... Figure 6 As shown, the horizontal axis represents two estimation methods (uncompensated and compensated), the vertical axis represents the average error of the track speed estimation (unit: m / s), the left bar represents the average error of the target track speed without radial speed compensation, and the right bar represents the average error of the target track speed after radial speed compensation.

[0196] from Figure 6 As can be seen, after processing by the target trajectory velocity estimation method provided in this application embodiment, the average error of the target trajectory velocity is reduced from 3.050m / s to 1.7477m / s, and the average error is improved by about 42.7%, which proves the significant effect of the technical solution of this application in improving the accuracy of trajectory velocity estimation.

[0197] In summary, the technical solution provided in this application is particularly applicable to the following actual driving scenarios, in which the proportion of the reflection point (wheel point) of the rotating component in the radar point cloud increases significantly, and its impact on the accuracy of trajectory speed estimation is particularly prominent. This application can effectively improve the estimation accuracy and ensure driving safety by compensating for the speed fluctuations introduced by wheel rotation.

[0198] Scenario 1: Close-range vehicle encounter.

[0199] For example, when driving on narrow roads or two-way single-lane roads, if a vehicle encounters an oncoming vehicle at close range, the wheels of the oncoming vehicle closest to the vehicle's corner radar are within the radar's close-range detection area. Due to the small distance between vehicles (typically 1-5 meters), the radar can detect detailed structures of the oncoming vehicle's wheels (such as wheel rims and tire sidewalls). In this scenario, this application can effectively compensate for radial velocity estimation errors caused by wheel rotation, ensuring accurate speed judgment during oncoming traffic and thus improving driving safety.

[0200] Scenario 2: Overtaking and lane changing.

[0201] For example, when a vehicle overtakes another vehicle, or when another vehicle merges into the vehicle's lane, the rear side wheels of the target vehicle (such as the right rear wheel of the vehicle in front) are usually within the primary detection range of the vehicle's left front corner radar. Since accurately estimating the speed of the vehicle in front is crucial for choosing the overtaking opportunity and judging the lane-changing distance, this application improves the accuracy of estimating the speed of the vehicle in front by compensating for wheel rotation interference, providing reliable data support for overtaking decisions and lane-changing control.

[0202] Scenario 3: Low-speed following and traffic jams.

[0203] For example, in low-speed following or congested areas, where vehicle distances are small, radar can detect the wheel points of the vehicle in front. This application improves the accuracy of estimating the speed of the vehicle in front, providing a smoother and more stable speed tracking input for the adaptive cruise control system, thereby improving following comfort and safety.

[0204] Based on real-vehicle testing, this application achieves the best estimation results under the following conditions:

[0205] 1) When the distance between the target vehicle and the vehicle is within 1-5 meters, the radar can obtain sufficient reflection points of rotating parts (wheel points).

[0206] 2) Each wheel point cluster contains at least 8-10 effective reflection points to ensure the reliability of the sinusoidal law fitting.

[0207] 3) The target vehicle travels at a relatively stable speed to avoid rapid changes in the sine parameters caused by drastic acceleration or deceleration.

[0208] It should be noted that the above conditions are preferred conditions for this application, and not limitations on its scope of protection. In practical applications, even if all the above conditions are not met (such as the target distance being greater than 5 meters, the number of wheel points being small, or the target accelerating or decelerating drastically), this application can still adjust the algorithm parameters according to the actual point cloud quality (such as using weighted least squares, introducing time window accumulation, and combining visual sensor correction, etc.) to maintain the accuracy and robustness of the trajectory speed estimation to a certain extent.

[0209] In summary, this application has at least the following advantages:

[0210] 1. By acquiring radar point cloud data of the target, establishing the law of its change with the rotation state for the observed radial velocity of the rotating component reflection point, determining the corresponding radial velocity compensation amount, compensating for the observed radial velocity of the rotating component reflection point, and then combining the observed radial velocity of the non-rotating component reflection point to estimate the target's trajectory velocity, it is possible to suppress the radial velocity deviation introduced by the rotational motion without directly eliminating the rotating component reflection point, thereby improving the accuracy, stability and robustness of target trajectory velocity estimation while retaining more effective point cloud data.

[0211] Second, by decomposing the observed radial velocity of the rotating component's reflection point into a reference radial velocity and a periodic fluctuation, the additional velocity caused by rotation can be quantitatively characterized. Therefore, during velocity compensation processing, the velocity of the rotating component's reflection point can be corrected to match the vehicle's translational motion, and then used together with the reflection point of non-rotating components to participate in trajectory velocity estimation, thereby reducing the interference of tire rotation on the overall velocity fitting.

[0212] Third, identifying the reflection points of rotating parts does not require complex point cloud segmentation and wheel detection algorithms. It is accomplished solely based on radar point clouds, least squares estimation results, and residual statistics methods, thereby reducing computational complexity and improving computational efficiency.

[0213] Fourth, by sorting the radial velocity residual distribution of reflection points and identifying abrupt change points, the residual threshold is made independent of fixed empirical values, but rather adaptively changes with the actual distribution of the current target radar point cloud. This allows for a more accurate distinction between vehicle stability reflection points and reflection points from rotating components such as tires and wheel hubs. Because the residual threshold can be dynamically adjusted with scene changes, it can maintain good recognition stability even in situations with sparse point clouds, strong occlusion, or significant changes in target attitude, and provides reliable input for subsequent radial velocity compensation and trajectory velocity estimation.

[0214] Fifth, by introducing visual image data acquired synchronously with the radar, the system can still obtain external constraints on the wheel speed state when the point cloud of the rotating component is sparse, the radar echo is blocked, or the wheel speed changes rapidly. Based on this, the sinusoidal law model is dynamically corrected, thereby improving the accuracy of radial speed compensation of the rotating component, reducing the estimation deviation caused by changes in wheel speed, viewing angle, installation attitude, or radar error, and thus improving the stability and robustness of target trajectory speed estimation.

[0215] Figure 7 Schematic diagram of the target trajectory velocity estimation device based on radar point cloud provided in the embodiments of this application Figure 1 ,like Figure 7 As shown, the target trajectory velocity estimation device 70 based on radar point clouds provided in this embodiment includes: an acquisition module 71, a fitting module 72, a determination module 73, a compensation module 74, and an estimation module 75. Wherein:

[0216] The acquisition module 71 is used to acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating parts and reflection points of non-rotating parts.

[0217] The fitting module 72 is used to fit a sinusoidal law model using the observed radial velocities of multiple rotating component reflection points. The sinusoidal law model characterizes the mapping relationship between the observed radial velocity and the rotation angle of the rotating component reflection points.

[0218] Module 73 is used to determine the radial velocity compensation amount of each rotating component's reflection point based on a sinusoidal law model.

[0219] The compensation module 74 is used to compensate the observed radial velocity of the reflection point of the rotating component according to the radial velocity compensation amount, so as to obtain the compensated radial velocity.

[0220] The estimation module 75 is used to estimate the target's trajectory velocity based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

[0221] In implementation, the acquisition module 71 first retains the point cloud observations of both rotating and non-rotating components on the target, enabling the fitting module 72 to establish a sinusoidal model for the periodic velocity changes of the reflection points of rotating components, thus moving beyond simply eliminating outliers (i.e., reflection points of rotating components). The determination module 73 provides the radial velocity compensation for each reflection point of rotating components based on this model. The compensation module 74 then corrects for the additional velocity components introduced by rotation, making the compensated radial velocity of the reflection points of rotating components closer to the target's true translational velocity. The estimation module 75 further combines the compensated radial velocity with the observed radial velocity of the reflection points of non-rotating components for trajectory velocity estimation. This retains more effective point cloud samples while suppressing periodic interference caused by wheel rotation, thereby improving the accuracy, stability, and real-time reliability of trajectory velocity estimation in close-range passing, lane changing, overtaking, and following scenarios.

[0222] In one possible implementation, the sinusoidal law model satisfies the following relationship: the observed radial velocity of the reflection point of the rotating component is equal to the sum of the reference radial velocity of the non-rotating component and the sinusoidal wave term. The reference radial velocity represents the projected velocity of the non-rotating component in the radar line-of-sight direction, and the sinusoidal wave term represents the periodic fluctuation of the radial velocity caused by the rotation of the rotating component. The instantaneous phase of the sinusoidal wave term is equal to the sum of the rotation angle of the reflection point of the rotating component relative to the center of the rotating component and the phase offset of the sinusoidal wave term.

[0223] In one possible implementation, the determining module 73 is specifically used to: obtain the rotation angle of the rotating component reflection point, the rotation angle being determined based on the position coordinates of the rotating component reflection point in the radar coordinate system; and determine the radial velocity compensation amount of the rotating component reflection point based on the rotation angle, the amplitude of the sinusoidal wave term, and the sinusoidal wave phase offset.

[0224] In one possible implementation, the reflection point of the rotating component is determined as follows: based on the least squares method, the initial trajectory velocity of the target is estimated from radar point cloud data; for each reflection point on the target, the radial velocity residual corresponding to the reflection point is determined based on the observed radial velocity and the predicted radial velocity of the reflection point, wherein the predicted radial velocity is determined based on the initial trajectory velocity and the unit direction vector corresponding to the reflection point; if the absolute value of the radial velocity residual is greater than the residual threshold, then the reflection point is determined as the reflection point of the rotating component.

[0225] In one possible implementation, the residual threshold is determined as follows:

[0226] Sort the radial velocity residuals of all reflection points on the target to obtain a residual sequence; calculate the change characteristic value between adjacent residuals in the residual sequence, and determine the residual threshold based on the radial velocity residual corresponding to the position where the change characteristic value jumps. The change characteristic value includes difference value or gradient value; or, determine the residual threshold based on the statistical characteristics of radar point cloud data. The statistical characteristics include at least one of the mean, median, standard deviation and quartiles of the radial velocity residuals.

[0227] like Figure 8 As shown, in one possible implementation, the target trajectory velocity estimation device 70 based on radar point cloud further includes a correction module 76, which is used to acquire visual image data synchronously acquired with radar point cloud data; process the visual image data to identify texture features or rotation speed markings on the rotating component to estimate the real-time angular velocity of the rotating component; and dynamically correct the amplitude of the sinusoidal wave term and the phase offset of the sinusoidal wave term according to the real-time angular velocity.

[0228] The target trajectory velocity estimation device based on radar point cloud provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0229] Figure 9 This is a schematic diagram of the control unit provided in an embodiment of this application. Figure 9 As shown, the control unit 90 provided in this embodiment includes at least one processor 901 and a memory 902. Optionally, the control unit 90 further includes a communication component 903. The processor 901, memory 902, and communication component 903 are connected via a bus 904.

[0230] In a specific implementation, at least one processor 901 executes computer execution instructions stored in memory 902, causing at least one processor 901 to perform the above-described method.

[0231] The specific implementation process of processor 901 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0232] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0233] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0234] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0235] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0236] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0237] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0238] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0239] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0240] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0241] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0242] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0243] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0244] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A radar point cloud based target track velocity estimation method, characterized in that, include: Acquire radar point cloud data of the target, the radar point cloud data including the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system, the reflection points including reflection points of rotating parts and reflection points of non-rotating parts; By using the observed radial velocities at multiple reflection points of the rotating components, a sinusoidal law model is obtained through fitting. The sinusoidal law model characterizes the mapping relationship between the observed radial velocities and the rotation angles of the reflection points of the rotating components. Based on the sinusoidal law model, the radial velocity compensation amount of each of the rotating components' reflection points is determined; Based on the radial velocity compensation amount, the observed radial velocity of the reflection point of the rotating component is compensated to obtain the compensated radial velocity. The target's trajectory speed is estimated based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

2. The radar point cloud based target track velocity estimation method of claim 1, wherein, The sinusoidal law model satisfies the following relationship: The observed radial velocity of the reflection point of the rotating component is equal to the sum of the reference radial velocity of the non-rotating component and the sinusoidal fluctuation term. The reference radial velocity represents the projected velocity of the non-rotating component in the radar line-of-sight direction, and the sinusoidal fluctuation term represents the periodic fluctuation of the radial velocity caused by the rotation of the rotating component. The instantaneous phase of the sinusoidal wave term is equal to the sum of the rotation angle of the reflection point of the rotating component relative to the center of the rotating component and the phase offset of the sinusoidal wave.

3. The target trajectory velocity estimation method based on radar point clouds according to claim 2, characterized in that, The determination of the radial velocity compensation amount at the reflection point of each rotating component based on the sinusoidal law model includes: The rotation angle of the reflection point of the rotating component is obtained, and the rotation angle is determined based on the position coordinates of the reflection point of the rotating component in the radar coordinate system; The radial velocity compensation amount of the reflection point of the rotating component is determined based on the rotation angle, the amplitude of the sinusoidal wave term, and the phase offset of the sinusoidal wave.

4. The target trajectory velocity estimation method based on radar point clouds according to any one of claims 1 to 3, characterized in that, The reflection point of the rotating component is determined in the following way: Based on the least squares method, the initial trajectory velocity of the target is estimated according to the radar point cloud data; For each reflection point on the target, the radial velocity residual corresponding to the reflection point is determined based on the observed radial velocity and the predicted radial velocity of the reflection point. The predicted radial velocity is determined based on the initial track velocity and the unit direction vector corresponding to the reflection point. If the absolute value of the radial velocity residual is greater than the residual threshold, then the reflection point is determined as the reflection point of the rotating component.

5. The target trajectory velocity estimation method based on radar point clouds according to claim 4, characterized in that, The residual threshold is determined in the following way: The radial velocity residuals of all reflection points on the target are sorted to obtain a residual sequence; the change characteristic value between adjacent residuals in the residual sequence is calculated, and the residual threshold is determined based on the radial velocity residual corresponding to the position where the change characteristic value jumps. The change characteristic value includes difference value or gradient value. Alternatively, the residual threshold can be determined based on the statistical characteristics of the radar point cloud data, wherein the statistical characteristics include at least one of the mean, median, standard deviation, and quartiles of the radial velocity residuals.

6. The target trajectory velocity estimation method based on radar point clouds according to claim 2 or 3, characterized in that, Also includes: Acquire visual image data that is synchronously collected with the radar point cloud data; The visual image data is processed to identify texture features or rotation speed indicators on the rotating component in order to estimate the real-time angular velocity of the rotating component. The amplitude of the sinusoidal wave term and the phase offset of the sinusoidal wave are dynamically corrected based on the real-time angular velocity.

7. A target trajectory velocity estimation device based on radar point clouds, characterized in that, include: The acquisition module is used to acquire radar point cloud data of the target. The radar point cloud data includes the position coordinates and observed radial velocity of each reflection point on the target in the radar coordinate system. The reflection points include reflection points of rotating components and reflection points of non-rotating components. The fitting module is used to fit a sinusoidal law model using the observed radial velocities of multiple reflection points of the rotating components. The sinusoidal law model characterizes the mapping relationship between the observed radial velocities and the rotation angles of the reflection points of the rotating components. The determination module is used to determine the radial velocity compensation amount of each of the rotating component reflection points based on the sinusoidal law model. The compensation module is used to compensate the observed radial velocity of the reflection point of the rotating component according to the radial velocity compensation amount, so as to obtain the compensated radial velocity. An estimation module is used to estimate the target's trajectory velocity based on the compensated radial velocity and the observed radial velocity at the reflection point of the non-rotating component.

8. A control unit, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the target trajectory velocity estimation method based on radar point clouds as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the target trajectory velocity estimation method based on radar point clouds as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes a computer program that, when executed, implements the target trajectory velocity estimation method based on radar point clouds as described in any one of claims 1 to 6.

Images