Method and system for improving detection capability by fusing radar frame data
By integrating radar frame data, the cost and volume increase problems caused by increasing radar wiring harness are solved, efficient radar detection capabilities are achieved, equipment cost and volume are reduced, and detection accuracy and robustness are improved.
Patent Information
- Application Number
- CN202411264249.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-11-26
AI Technical Summary
The prior art improves radar detection capabilities by increasing radar wiring harness, resulting in increased costs, larger equipment volume and increased damage probability, and inconvenient installation and use.
By fusing radar frame data, the pose relationship between historical radar frame data and current radar frame data is evaluated, and the pose alignment and data fusion is carried out to generate the fused radar frame data to improve detection capabilities. The method includes data acquisition, pose correction and data fusion steps, and uses state vectors, Kalman filtering, particle filtering and deep learning models to process.
Use low-cost radar to achieve detection effects similar to high-cost radar, improve the number of point clouds and detection capabilities of radar data, reduce equipment cost and volume, and improve detection accuracy and robustness.
Smart Images

Figure CN119024301A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of robotics and autonomous driving technology, and more specifically, to a method and system for improving detection capability by fusing radar frame data. Background Art
[0002] In the field of robotics and autonomous driving, radar plays a vital role as a key environmental perception sensor. It can provide three-dimensional information of the surrounding environment, including distance, speed, angle, etc., and is an important basis for decision-making and path planning of autonomous driving systems. However, with the continuous development of radar technology, especially the improvement in resolution, detection range and accuracy, the complexity and cost of radar systems have also increased. In particular, the increase in wiring harnesses (or the number of channels and antennas) has become a key factor driving up costs. In most scenarios, the demand for radar is necessary, whether it is single-beam radar or multi-beam radar, lidar or millimeter-wave radar, solid-state radar or mechanical radar, the cost is usually more expensive.
[0003] The current solution extracts a key step of real-time positioning and mapping from SLAM (Simultaneous Localization and Mapping) technology and combines it with radar detection algorithms. In order to improve the recognition ability of radar data, one of the traditional methods is to increase the radar beam. The increase in radar beams means that more laser beams can be emitted and received at the same time, thereby obtaining denser point cloud data.
[0004] Improving the recognition capability of radar data by adding radar harnesses will have the following problems: as the number of harnesses increases, the cost of the radar also increases rapidly, which will lead to an increase in the final overall cost; generally, the more harnesses a radar has, the larger its size is, which is inconvenient both during installation and subsequent use. In addition, due to its large size and the large number of radar transmitters and receivers, the probability of damage is also increased. Summary of the invention
[0005] An object of the present invention is to provide a method and system for improving detection capability by fusing radar frame data, so as to solve the aforementioned problem that as the number of wiring harnesses increases, the final overall cost increases, and the radar with more wiring harnesses has a larger volume, which is inconvenient for installation and subsequent use, and the probability of damage increases.
[0006] According to a first aspect of the present invention, there is provided a method for improving detection capability by fusing radar frame data, comprising the following steps:
[0007] S100, data acquisition step: acquiring one or more frames of historical radar frame data, wherein the historical radar frame data includes image information from a radar sensor at different time points in the past;
[0008] S200, posture correction step: evaluating the posture relationship of the historical radar frame data relative to the current radar frame data, and adjusting the posture of the historical radar frame data to align with the current radar frame data;
[0009] S300, data fusion step: fusing the historical radar frame data that has been aligned with the posture with the current radar frame data to generate fused radar frame data.
[0010] Optionally, the historical radar frame data also includes radar frame data that has been previously fused and processed.
[0011] Optionally, in the posture correction step S200, the posture relationship of the historical radar frame data relative to the current radar frame data is evaluated, including but not limited to calculating the position coordinates, orientation and speed of the object.
[0012] Optionally, the posture correction step S200 specifically includes:
[0013] S201 defines the state vector: the state vector includes position, velocity, attitude and the error terms of these states. According to the initial inertial measurement unit (IMU), real-time kinematic carrier phase differential technology (RTK) or global positioning system (GPS) data, the initial state of the system is estimated and the covariance matrix of the state estimation is set;
[0014] S202 state prediction: Use the IMU acceleration and angular velocity data and dynamic model to predict the state and covariance matrix at the next moment;
[0015] S203 Observation Update: Use error state Kalman filtering to update the error terms and covariance matrix in the state estimation. The error terms include position error, velocity error, and attitude error.
[0016] S204 calculates the pose difference between radar frames: obtains radar point cloud data of the current frame and the previous frame and their timestamps, and calculates the pose difference of the previous radar frame relative to the current frame using the state estimation updated by the error state Kalman filter (ESKF);
[0017] S205 Radar point cloud pose correction: Based on the calculated pose difference, each point in the radar point cloud of the previous frame is transformed so that it is spatially aligned with the point cloud of the current frame. The transformation operation includes translation and / or rotation.
[0018] Optionally, in the step of calculating the pose difference between radar frames in S204, the pose estimation of the historical radar frame relative to the current radar frame may be performed by using Kalman filtering or particle filtering.
[0019] Optionally, in the posture correction step S200, the posture is evaluated and aligned by comparing feature points, edge information or other identifiable features in the historical radar frame data and the current radar frame data using a transformation matrix, an interpolation algorithm or an optimization algorithm.
[0020] Optionally, in the data fusion step S300, any one or more of weighted averaging method, Kalman filtering, particle filtering, and deep learning model are used to fuse the historical radar frame data and the current radar frame data.
[0021] Optionally, the method for improving detection capability by fusing radar frame data further includes:
[0022] S400, data post-processing step: optimizing the fused radar frame data, including but not limited to target recognition, target tracking and target classification.
[0023] According to a first aspect of the present invention, there is provided a system for improving detection capability by fusing radar frame data, comprising:
[0024] Data acquisition module: used to acquire one or more frames of historical radar frame data, wherein the historical radar frame data includes image information from the radar sensor at different time points in the past;
[0025] A posture correction module: used to evaluate the posture relationship of the historical radar frame data relative to the current radar frame data, and adjust the posture of the historical radar frame data to align with the current radar frame data;
[0026] Data fusion module: used to fuse the historical radar frame data that has been aligned with the current radar frame data to generate fused radar frame data;
[0027] Data post-processing module: used to optimize the processing of fused radar frame data, including but not limited to target recognition, target tracking and target classification.
[0028] Optionally, the system for improving detection capability by fusing radar frame data further includes:
[0029] Output execution module: including but not limited to obstacle detection device and path planning device, used for path planning or obstacle detection according to the optimized and fused radar frame data.
[0030] The method and system for improving detection capability by fusing radar frame data disclosed in the present invention have the following technical effects:
[0031] By evaluating the pose of the historical radar frame relative to the current radar frame, aligning the pose of the historical radar frame data to the current radar frame data, and fusing the historical radar frame data with the current radar frame data as the latest radar frame data, low-cost radars can be used to achieve effects similar to high-cost radars, which can increase the number of point clouds in radar data and improve detection capabilities.
[0032] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
[0034] Figure 1 A flow chart of a method for improving detection capability by fusing radar frame data provided by an embodiment of the present invention;
[0035] Figure 2 A schematic diagram of the logic implementation of a method for improving detection capability by fusing radar frame data provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0036] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.
[0037] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
[0038] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.
[0039] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0040] In the prior art, in order to improve the recognition ability of the radar, it is necessary to increase the radar beam and cost. When identifying targets in radar data, it is often impossible to accurately identify them because the number of points is not enough, resulting in low accuracy and recall rates. This application evaluates the pose of the historical radar frame relative to the current radar frame, aligns the pose of the historical radar frame data to the current radar frame data, and fuses the historical radar frame and the current radar frame data as the latest radar frame data. Low-cost radars can be used to achieve effects similar to high-cost radars, increase the number of point clouds in radar data, and improve detection capabilities.
[0041] The present invention proposes an embodiment of a method for improving detection capability by fusing radar frame data. Specifically, Figure 1 As shown, the following steps are included:
[0042] S100, data acquisition step: acquiring one or more frames of historical radar frame data, the historical radar frame data including image information from radar sensors at different time points in the past; wherein the radar sensor can continuously scan the surrounding environment and generate images containing information such as object position, speed, shape, etc. The historical radar frame data contains reflection information of all objects scanned by the radar, such as distance, angle, speed, etc., which exists in the form of data points and is converted into images or point clouds through specific algorithms for easy processing.
[0043] S200, posture correction step: evaluating the posture relationship of the historical radar frame data relative to the current radar frame data, and adjusting the posture of the historical radar frame data to align with the current radar frame data;
[0044] S300, data fusion step: fusing the historical radar frame data after posture alignment with the current radar frame data to generate fused radar frame data. The fused data can more comprehensively reflect the environmental status and improve the detection capability.
[0045] In the embodiment of the present invention, the historical radar frame data also includes radar frame data that has been fused and processed before. After the initial fusion process, the fused radar frame data obtained can also be regarded as a kind of "historical data". In the process of data acquisition and processing, not only the original frame data directly obtained from the radar sensor is considered, but also the frame data that has been processed by a certain fusion algorithm is considered. This processing method further expands the diversity and richness of the data, which helps to improve the accuracy and robustness of subsequent analysis or detection.
[0046] In the embodiment of the present invention, in the posture correction step S200, the posture relationship of the historical radar frame data relative to the current radar frame data is evaluated, including but not limited to calculating the position coordinates, orientation and speed of the object. Specifically, the specific position of the object in the three-dimensional space is determined, and the orientation of the object in the space is determined; the orientation can be represented by the angle between the main axis of the object (such as the forward and sideways direction of the vehicle) and the axis of the coordinate system; the speed of the object moving in the space is determined, including the size and direction.
[0047] In the embodiment of the present invention, the posture correction step S200 specifically includes:
[0048] S201 defines the state vector: the state vector includes position (such as x, y, z coordinates in three-dimensional space), velocity (such as vx, vy, vz velocity in three-dimensional space), attitude (such as rotation represented by quaternion q), and error terms of these states (such as position error δx, δy, δz, velocity error δvx, δvy, δvz, attitude error δθ, δφ, δψ, or use a more compact error representation, such as rotation vector error). Estimate the initial state of the system based on the initial inertial measurement unit (IMU), real-time kinematic carrier phase difference technology (RTK) or global positioning system (GPS) data, and set the initial uncertainty covariance matrix of the state estimate;
[0049] S202 state prediction: Use the acceleration and angular velocity data of the IMU and a dynamic model (such as a rigid body kinematic model) to predict the state (including position and attitude) and covariance matrix at the next moment;
[0050] S203 Observation Update: Use the error state Kalman filter to update the error terms and covariance matrix in the state estimation, where the error terms include position error, velocity error, and attitude error; optimize the state estimation by minimizing the observation residual;
[0051] S204 calculates the pose difference between radar frames: obtains radar point cloud data of the current frame and the previous frame and their timestamps, and uses the state estimation updated by the error state Kalman filter (ESKF) to calculate the pose difference of the previous radar frame relative to the current frame. Specifically, Figure 2 As shown;
[0052] S205 Radar point cloud pose correction: Based on the calculated pose difference, each point in the radar point cloud of the previous frame is transformed so that it is spatially aligned with the point cloud of the current frame. The transformation operation includes translation and / or rotation.
[0053] In an embodiment of the present invention, in the step of calculating the pose difference between radar frames in S204, the pose estimation of the historical radar frame relative to the current radar frame can also be performed by Kalman filtering or particle filtering. Kalman filtering is an efficient recursive filter (autoregressive filter) that can estimate the state of a dynamic system from a series of measurements containing statistical noise. In radar data processing, the position and speed of objects in the radar frame can be used as system states, and the optimal values of these states can be estimated by Kalman filtering. Particle filtering is a recursive Bayesian filter based on the Monte Carlo method, which is used to estimate the state of nonlinear non-Gaussian dynamic systems. In radar pose estimation, particles can represent the position, orientation and other states of objects. Observation data is used to evaluate the degree of match between each particle and the object detected by the radar, and the weight of the particle is updated accordingly. For the pose estimation of the historical radar frame relative to the current radar frame, the current radar frame can be first regarded as a reference frame, and then Kalman filtering or particle filtering can be applied to estimate the pose of the object in the historical radar frame.
[0054] In an embodiment of the present invention, in the posture correction step S200, by comparing the feature points, edge information or other identifiable features in the historical radar frame data and the current radar frame data, the posture is evaluated and aligned using a transformation matrix, an interpolation algorithm or an optimization algorithm. Among them, feature points refer to those points in the radar data that have significant characteristics and are easy to be identified and tracked. In the radar data, these points correspond to strong reflection points, corner points or edge intersections of objects. The edge is where the attributes such as brightness and reflection intensity in the image or radar data change significantly, usually corresponding to the boundary of the object. In the radar data, the edge information can be extracted by gradient calculation, threshold processing and other methods. In addition to feature points and edge information, the radar data may also contain other identifiable features, such as texture features, speed features, etc., which can also be used for comparison and matching between frames. The calculation of the transformation matrix can be implemented by a variety of methods, such as the least squares method, the iterative closest point (ICP) algorithm, etc. These methods use the geometric relationship between corresponding points to solve the transformation matrix. The interpolation algorithm can estimate the value of the unknown point based on the information of the surrounding known points, thereby filling the blank areas in the data. In radar data processing, the interpolation algorithm can be used to improve the density and completeness of the data.
[0055] In an embodiment of the present invention, in the step of data fusion S300, any one or more of weighted average method, Kalman filter, particle filter, and deep learning model are used to fuse the historical radar frame data and the current radar frame data. Among them, the weighted average method is a simple data fusion method. In radar data fusion, weights are assigned according to the timeliness of the historical radar frame data and the real-time of the current radar frame data, and the corresponding features (such as position, speed, etc.) in the historical radar frame data and the current radar frame data are weighted averaged to obtain the fused feature value. Kalman filtering is an efficient recursive filter that uses the dynamic model and observation data of the system to estimate the state of the system. In radar data fusion, the historical radar frame data can be regarded as the historical state estimation of the system, and the current radar frame data can be regarded as the new observation data. Through the prediction step of Kalman filtering, the current state is predicted according to the dynamic model of the system; in the update step, the current radar frame data is used to update the state estimation. Particle filtering is a recursive Bayesian filter based on the Monte Carlo method. It approximates the posterior probability distribution of the state through a set of random samples (particles). In radar data fusion, the weights and positions of the particles are updated by simulating the dynamic process and observation process of the system. Then, resampling is performed according to the weights of the particles to generate a new set of particles that better reflects the current state of the system. Through particle filtering, historical radar frame data and current radar frame data can be fused to obtain a more accurate state estimate. Deep learning models, such as convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants (such as LSTM, GRU), can automatically learn feature representation and pattern recognition from data. In radar data fusion, deep learning models can be used to extract high-level features from radar frame data and learn the complex relationship between historical radar frame data and current radar frame data. By training the deep learning model, it can be enabled to predict or infer the fused state information.
[0056] In an embodiment of the present invention, the method for improving detection capability by fusing radar frame data further includes:
[0057] S400, data post-processing step: optimize the fused radar frame data, including but not limited to target recognition, target tracking and target classification. In the fused radar frame data, target recognition aims to accurately distinguish different types of targets, such as vehicles, pedestrians, obstacles, etc. Target tracking refers to the continuous estimation and prediction of state parameters such as position, velocity, acceleration, etc. of a specific target in a continuous data sequence. In a radar system, target tracking is an important means to achieve dynamic monitoring of targets. Target classification refers to classifying the identified targets according to their attributes or characteristics. In-depth analysis is performed on the extracted target features, such as shape features, motion features, reflection intensity features, etc. By analyzing these features, the intrinsic attributes and category information of the target can be revealed.
[0058] In addition, in the data post-processing step, the fused radar frame data needs to be optimized to improve the quality and availability of the data. The optimization process includes but is not limited to the following aspects:
[0059] Noise suppression: Use filtering techniques (such as median filtering, Gaussian filtering, etc.) to denoise radar data to reduce the interference of noise on target identification, tracking and classification;
[0060] Data compression: compress the fused radar frame data to reduce the burden of data storage and transmission. At the same time, data compression can also retain the main features and information of the data to ensure the availability and accuracy of the data;
[0061] Performance evaluation: Perform performance evaluation on the results of target recognition, tracking and classification, such as calculating indicators such as recognition rate, tracking accuracy, and classification accuracy. Through performance evaluation, we can understand the effects and problems of the data post-processing steps, and provide a basis for subsequent optimization and improvement.
[0062] The present invention also provides an embodiment of a method for improving detection capability by fusing radar frame data, comprising the following steps:
[0063] Data acquisition step: acquiring one or more frames of historical radar frame data, where the historical radar frame data includes image information from a radar sensor at different time points in the past;
[0064] Data fusion step: directly fuse the historical radar frame data with the current radar frame data to generate fused radar frame data.
[0065] In an embodiment of the present invention, in the data fusion step, the historical radar frame data and the current radar frame data are directly fused and processed by specifically using any one of machine learning, deep learning, and reinforcement learning algorithms.
[0066] It should be noted that in radar data fusion, machine learning algorithms can learn the relationship between historical radar frames and current radar frames, extract key features, and use them for fusion processing. Deep learning algorithms are a branch of machine learning that learn complex patterns in data by building multi-layer neural networks. In radar data fusion, deep learning algorithms can automatically extract high-level feature representations from radar frames and achieve more accurate fusion processing. Reinforcement learning algorithms allow models to continuously trial and error and learn in the environment, with the goal of maximizing cumulative rewards. In radar data fusion, reinforcement learning algorithms simulate the process of radar data fusion and optimize the fusion strategy through interaction with the environment (i.e., radar data).
[0067] The present invention also provides an embodiment of a system for improving detection capability by fusing radar frame data, comprising:
[0068] Data acquisition module: used to acquire one or more frames of historical radar frame data, where the historical radar frame data includes image information from the radar sensor at different time points in the past;
[0069] Posture correction module: used to evaluate the posture relationship between the historical radar frame data and the current radar frame data, and adjust the posture of the historical radar frame data to align with the current radar frame data;
[0070] Data fusion module: used to fuse the historical radar frame data that has been aligned with the current radar frame data to generate fused radar frame data.
[0071] Data post-processing module: used to optimize the processing of fused radar frame data, including but not limited to target recognition, target tracking and target classification.
[0072] In an embodiment of the present invention, the system for improving detection capability by fusing radar frame data also includes an output execution module, which includes but is not limited to an obstacle detection device and a path planning device, and is used to perform path planning or obstacle detection based on the optimized and fused radar frame data.
[0073] In addition, the algorithm concept of the present invention can be expanded to the time series state analysis of multiple continuous frames, and multiple frames can be fused to improve the density and richness of the point cloud, thereby improving the overall algorithm accuracy; low-beam, such as multiple frames of 16-line lidar, can be fused into one frame, thereby achieving to a certain extent (during motion) the effect of a lidar comparable to that of 32-line, 64-line or even more-beam lidar.
[0074] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules are based on the same concept as the method embodiment of the present application, and are a system corresponding to the above-mentioned method of improving detection capability by fusing radar frame data. All implementation methods in the above-mentioned method embodiment are applicable to the embodiment of the device. Its specific functions and technical effects brought about can be found in the method embodiment part, which will not be repeated here.
[0075] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.
Claims
1. A method for improving detection capability by fusing radar frame data, characterized in that: The following steps are involved: S100, data acquisition step: acquiring one or more frames of historical radar frame data, wherein the historical radar frame data includes image information from a radar sensor at different time points in the past; S200, posture correction step: evaluating the posture relationship of the historical radar frame data relative to the current radar frame data, and adjusting the posture of the historical radar frame data to align with the current radar frame data; S300, data fusion step: fusing the historical radar frame data that has been aligned with the posture with the current radar frame data to generate fused radar frame data.
2. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: The historical radar frame data also includes radar frame data that has been fused and processed before.
3. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: In the posture correction step S200, the posture relationship of the historical radar frame data relative to the current radar frame data is evaluated, including but not limited to calculating the position coordinates, orientation and speed of the object.
4. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: The S200 posture correction steps specifically include: S201 defines the state vector: the state vector includes position, velocity, attitude and the error terms of these states. According to the initial inertial measurement unit (IMU), real-time kinematic carrier phase differential technology (RTK) or global positioning system (GPS) data, the initial state of the system is estimated and the covariance matrix of the state estimation is set; S202 state prediction: Use the IMU acceleration and angular velocity data and dynamic model to predict the state and covariance matrix at the next moment; S203 Observation Update: Use error state Kalman filtering to update the error terms and covariance matrix in the state estimation. The error terms include position error, velocity error, and attitude error. S204 calculates the pose difference between radar frames: obtains radar point cloud data of the current frame and the previous frame and their timestamps, and calculates the pose difference of the previous radar frame relative to the current frame using the state estimation updated by the error state Kalman filter (ESKF); S205 Radar point cloud pose correction: Based on the calculated pose difference, each point in the radar point cloud of the previous frame is transformed so that it is spatially aligned with the point cloud of the current frame. The transformation operation includes translation and / or rotation.
5. The method for improving detection capability by fusing radar frame data according to claim 4, characterized in that: In the step of calculating the pose difference between radar frames in S204, the pose estimation of the historical radar frame relative to the current radar frame may also be performed by using Kalman filtering or particle filtering.
6. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: In the posture correction step S200, by comparing feature points, edge information or other identifiable features in the historical radar frame data and the current radar frame data, the posture is evaluated and aligned using a transformation matrix, an interpolation algorithm or an optimization algorithm.
7. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: In the S300 data fusion step, any one or more of weighted average method, Kalman filter, particle filter, and deep learning model are used to fuse the historical radar frame data and the current radar frame data.
8. The method for improving detection capability by fusing radar frame data according to claim 1, characterized in that: Also includes: S400, data post-processing step: optimizing the fused radar frame data, including but not limited to target recognition, target tracking and target classification.
9. A system for improving detection capability by fusing radar frame data, characterized in that: include: Data acquisition module: used to acquire one or more frames of historical radar frame data, wherein the historical radar frame data includes image information from the radar sensor at different time points in the past; A posture correction module: used to evaluate the posture relationship of the historical radar frame data relative to the current radar frame data, and adjust the posture of the historical radar frame data to align with the current radar frame data; Data fusion module: used to fuse the historical radar frame data that has been aligned with the current radar frame data to generate fused radar frame data; Data post-processing module: used to optimize the processing of fused radar frame data, including but not limited to target recognition, target tracking and target classification.
10. The system for improving detection capability by fusing radar frame data according to claim 9, characterized in that: Also includes: Output execution module: including but not limited to obstacle detection device and path planning device, used for path planning or obstacle detection according to the optimized and fused radar frame data.
