Vehicle-mounted millimeter wave radar target size enhancement estimation method based on multipath compensation

Abstract

The application discloses a kind of based on multi-path compensation's vehicle-mounted millimeter wave radar target size enhancement estimation method.The method includes whether current radar detection environment is multi-path significant scene according to the received radar point cloud data, if current radar detection environment is multi-path significant scene, the reflecting surface of multi-path point cloud data is obtained, target point cloud and multi-path point cloud data are sequentially extracted from radar point cloud data, according to the geometric relationship of multi-path point cloud data and reflecting surface propagation model is established, and according to the propagation model the coordinates of multi-path point are mapped to target body location, to form compensation point cloud, the target point cloud is fused with compensation point cloud, to form enhanced target point cloud set, and the size of target is estimated based on enhanced target point cloud set.The application makes up the deficiency of direct detection point cloud, can convert multi-path interference into beneficial information, improve system robustness, and is suitable for existing vehicle-mounted millimeter wave radar system.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and specifically to a target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation. Background Technology

[0002] With the widespread application of automotive millimeter-wave radar in intelligent driving systems, target size estimation has become a key task. In existing technologies, target size estimation mainly relies on attributes such as the target point cloud distribution and radar cross-section (RCS) detected by radar. Specifically, the target outline is delineated using the point cloud, and the target size is determined by combining the RCS value. For example, in adjacent lane scenarios, radar can detect the front (or rear) and sides of the target, forming an "L"-shaped outline, making it relatively easy to estimate the target size.

[0003] However, in lane-following scenarios, such as when a front radar detects a vehicle ahead in its lane, it can typically only acquire point clouds of the target vehicle's rear due to the radar's limited field of view, failing to obtain side information. Furthermore, limited by radar resolution and point cloud sparsity, relying solely on the rear point cloud makes it difficult to accurately delineate the target's outline, leading to inaccurate size estimation.

[0004] Furthermore, multipath reflection is a common phenomenon in millimeter-wave radar. For example, in environments such as tunnels and metal fences, radar will receive multipath points formed by reflections from reflective surfaces. Existing processing methods typically treat multipath points as interference and eliminate them by filtering or deleting them at the target level to avoid the generation of false targets.

[0005] The existing technology has the following drawbacks:

[0006] 1. Incomplete point cloud information: In the same lane scenario, the radar can only detect one side of the target (such as the rear of the vehicle) and cannot obtain the side point cloud, resulting in incomplete outline and large size estimation deviation.

[0007] 2. Multipath points are not effectively utilized: Existing technologies treat multipath points as negative interference and eliminate them, ignoring the additional useful information contained in multipath points (such as target side structure information), resulting in information waste.

[0008] 3. Poor scene adaptability: In environments with significant multipath effects, such as tunnels and metal guardrails, existing methods do not utilize the additional information brought by multipath reflection, which limits the accuracy and robustness of target size estimation. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by providing a target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation.

[0010] To achieve the above objectives, this invention provides a target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation, comprising:

[0011] Step 1: Identify whether the current radar detection environment is a multipath saliency scene based on the received radar point cloud data. If the current radar detection environment is a multipath saliency scene, obtain the reflection surface of the multipath point cloud data.

[0012] Step 2: Extract the target point cloud and multipath point cloud data sequentially from the radar point cloud data;

[0013] Step 3: Establish a propagation model based on the geometric relationship between the multipath point cloud data and the reflecting surface, and map the coordinates of the multipath points to the target body position according to the propagation model to form a compensation point cloud;

[0014] Step 4: Fuse the target point cloud with the compensation point cloud to form an enhanced target point cloud set;

[0015] Step 5: Estimate the size of the target based on the enhanced target point cloud set.

[0016] Furthermore, the multipath salient scenes include tunnel scenes and metal guardrail edge scenes.

[0017] Furthermore, the method for identifying the tunnel scene is as follows:

[0018] If the radar point cloud data shows a number of high-altitude point clouds in front that exceed a set threshold, and the radar point cloud data has relatively clear boundary lines on both sides, then the current scene is identified as a tunnel scene.

[0019] Furthermore, the method for recognizing the scene along the metal guardrail is as follows:

[0020] If there is one or more static point cloud data that is parallel to the vehicle's direction of travel, has a reflection intensity higher than a set intensity threshold, and is linearly distributed, then the current scene is identified as a scene near a metal guardrail.

[0021] Furthermore, the scene recognition results from other sensors are used as external inputs and fused with the recognition results based on radar point cloud data to obtain the final multipath salient scene recognition results.

[0022] Furthermore, the extraction method for the multipath point cloud data is as follows:

[0023] Based on the geometric information of the reflector and the spatial distribution of the target point cloud, the mirror region of the target point cloud region with respect to the reflector is calculated. For each point in the radar point cloud data, it is determined whether it falls within the mirror region. Points falling within the mirror region are output as a set of candidate multipath points after position feature filtering.

[0024] A motion model of the target point cloud is established using a point cloud tracking algorithm. The expected position of each point in the target point cloud in the current frame is predicted, and the expected motion trajectory of each point in the target point cloud is obtained. For each point in the candidate multipath point set after position feature filtering, the corresponding theoretical mirror motion trajectory is calculated according to the geometric relationship of the reflecting surface. The degree of agreement between the expected motion trajectory of the point and the theoretical mirror motion trajectory is compared. If the degree of agreement is higher than a preset threshold, the point is confirmed as a multipath point.

[0025] Furthermore, the reflective surface is obtained through navigation maps or by filtering and fitting radar point cloud data, and it is represented by a reflection line with the vehicle-mounted millimeter-wave radar as the origin. The reflection line is represented as follows: ,in, , The x and y coordinates are on the reflection line. , All are coefficients. This is a constant term.

[0026] Furthermore, each compensation point in the compensation point cloud Represented as:

[0027] ;

[0028] ;

[0029] ;

[0030] in, , Compensation points x and y coordinates , They are respectively the compensation points Corresponding multipath points x and y coordinates Multipath point The directional distance to the reflected ray.

[0031] Furthermore, if there are more than two reflecting surfaces in the current radar detection environment, the compensation point cloud is calculated separately for each reflecting surface and then fused with the target point cloud.

[0032] Furthermore, an uncertainty factor is introduced into the propagation model to assess the confidence level of the calculated compensation point location, and the calculated confidence level is used as the basis for the target size estimation weight.

[0033] Beneficial effects: 1. Improved size estimation accuracy: By utilizing the target side information contained in multipath points, the shortcomings of direct detection of point clouds are made up for, especially in the same lane scenario, the size estimation effect is significantly improved.

[0034] 2. Enhanced scenario adaptability: In environments with significant multipath interference, such as tunnels and metal guardrails, this method can transform multipath interference into useful information, thereby improving system robustness.

[0035] 3. No additional hardware costs required: This method is entirely based on algorithm implementation and is applicable to existing vehicle-mounted millimeter-wave radar systems. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation, according to an embodiment of the present invention.

[0037] Figure 2 It is a schematic diagram of mapping the coordinates of multipath points to the target body position according to the propagation model. Detailed Implementation

[0038] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0039] like Figure 1 and Figure 2 As shown, this embodiment of the invention provides a target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation, including:

[0040] Step 1: Based on the received radar point cloud data, identify whether the current radar detection environment is a salient multipath scene. If the current radar detection environment is a salient multipath scene, obtain the reflecting surface of the multipath point cloud data. Specifically, the aforementioned salient multipath scenes include tunnel scenes and metal guardrail edge scenes, and the identification methods are as follows:

[0041] Tunnel Scene Recognition: In a tunnel environment, radar detects reflection points at the tunnel's side boundaries and ceiling. The point cloud is densely distributed at lower altitudes at the tunnel's side boundaries and at higher altitudes in the middle (detecting overhead traffic signs, streetlights, etc.). Therefore, if the radar point cloud data shows more high-altitude point clouds than a set threshold in front, and the radar point cloud data has relatively clear boundary lines on both sides, the current scene is identified as a tunnel scene.

[0042] Metal guardrail scene recognition: In scenes where there are metal guardrails on both sides of the road, the guardrails, as strong reflective surfaces, will produce clear mirror point clouds. Recognition features include the point cloud appearing as one or more linear distributions parallel to the vehicle's direction of travel, with clear boundaries. These linear point clouds typically have high reflection intensity and are static points. Therefore, if the radar point cloud data contains more than one static point cloud data point parallel to the vehicle's direction of travel with a reflection intensity higher than a set intensity threshold and exhibiting a linear distribution, the current scene is identified as a metal guardrail edge scene.

[0043] Furthermore, scene recognition results from other sensors (vehicle cameras, LiDAR, etc.) can be used as external inputs and fused with recognition results based on radar point cloud data to obtain the final multipath salient scene recognition result, thereby improving the reliability of scene recognition. For example, cameras can directly determine whether the current scene is a "tunnel" or a "road with guardrails" through image recognition technology; LiDAR can provide high-precision 3D point clouds, more clearly depicting the geometry of tunnels or guardrails. Filtrifying this information with the radar's own recognition results at the decision-level (such as weighted voting) can more accurately confirm whether the current scene is a multipath salient scene.

[0044] The aforementioned reflective surface can be obtained through high-precision map or point cloud fitting, and is usually simplified to represent by reflection lines, as shown in the following specific scheme:

[0045] Based on high-precision maps: If the vehicle is equipped with a high-precision map, the map already contains precise geometric information (location, orientation) of reflective surfaces such as tunnel walls and guardrails. The radar, through its own positioning (e.g., combined with GPS / IMU) and map matching, can directly obtain the reflection line equation of the reflective surface with the vehicle-mounted millimeter-wave radar as the origin. ,in, , The x and y coordinates are on the reflection line. , All are coefficients. This is a constant term.

[0046] Real-time point cloud fitting: In the absence of high-precision maps or when maps are not updated in a timely manner, radar point cloud data can be used to fit the reflective surface in real time. The specific steps are as follows:

[0047] 1. Filter out points from the point cloud that may belong to static reflective surfaces (such as tunnel walls and guardrails). These points usually have high reflectivity and zero velocity.

[0048] 2. Apply the least squares method or RANSAC algorithm to the selected point set to perform line fitting, and obtain the reflection line equation with the vehicle-mounted millimeter-wave radar as the origin. Parameters ( , , ).

[0049] 3. Combine the vehicle's heading angle to determine whether the fitted reflection line corresponds to a possible reflecting surface (such as a wall that is parallel to the vehicle's direction of travel or at a specific angle).

[0050] Step 2: Extract the directly detected target point cloud and multipath point cloud data sequentially from the radar point cloud data. The target point cloud is extracted through data processing and tracking. The multipath point cloud data is extracted from the radar point cloud data based on the point cloud's positional features and motion consistency features. Details are as follows:

[0051] 1. Location-based filtering (preliminary filtering)

[0052] Input: Current frame radar point cloud data, target point cloud data, and reflection line equation of the reflector surface.

[0053] Screening process:

[0054] Region mapping: Based on the geometric information of the reflector and the spatial distribution of the target point cloud, calculate the mirror region of the target point cloud with respect to the reflector. This does not involve calculating a single mirror point, but rather mapping the entire target point cloud contour to the other side of the reflector, forming a "virtual target contour region." For example, if the directly detected point cloud is an approximately rectangular region at the rear of a vehicle, then its mirror region is another rectangular region symmetrical about the reflector.

[0055] Region matching: For each point in the radar point cloud data, determine whether it falls within the aforementioned mirror region. The determination can be based on the Euclidean distance between the point and the bounding box of the mirror region, or whether the distance between the point and the centroid of the mirror region is within a certain range (this range is related to the target size and radar resolution).

[0056] Logical explanation: The principle behind this step is that multipath points usually originate from the "projection" of the target body on the other side of the reflecting surface. Therefore, they are more likely to appear near the target's mirror area, rather than being randomly distributed. This avoids misjudgments caused by calculation errors of a single point and adapts to the sparsity of point clouds.

[0057] Output: A set of candidate multipath points after location feature filtering.

[0058] 2. Screening based on motion consistency characteristics (final confirmation)

[0059] Input: A set of candidate multipath points after location feature filtering, and the motion trajectory of the target point cloud (obtained through multi-frame tracking).

[0060] Screening process:

[0061] A motion model of the target point cloud is established using point cloud tracking algorithms (such as Kalman filtering or nearest neighbor association), predicting the expected position of each point in the target point cloud in the current frame, and obtaining the predicted motion trajectory of each point in the target point cloud.

[0062] For each point in the candidate multipath point set after location feature screening, calculate its corresponding theoretical mirror motion trajectory based on the geometric relationship of the reflecting surface (it should maintain a fixed geometric transformation relationship with the motion trajectory of the directly detected target, such as mirror translation).

[0063] Consistency assessment: Compare the predicted trajectory of the point with the theoretical mirror trajectory (e.g., calculate the angle between their motion vectors and the difference in their velocities). If the consistency is higher than a preset threshold, the point is confirmed as a multipath point.

[0064] Output: Final extracted multipath point cloud data.

[0065] Step 3: Establish a propagation model based on the geometric relationship between the multipath point cloud data and the reflecting surface, and map the coordinates of the multipath points to the target body position according to the propagation model to form a compensation point cloud. Specifically, each compensation point in the compensation point cloud... Represented as:

[0066] ;

[0067] ;

[0068] ;

[0069] in, , Compensation points x and y coordinates , They are respectively the compensation points Corresponding multipath points x and y coordinates Multipath point The directional distance to the reflected ray.

[0070] Step 4: Fuse the target point cloud with the compensation point cloud to form an enhanced target point cloud set, that is, combine the compensation point cloud with the directly detected target point cloud for use.

[0071] Step 5: Estimate the target size based on the enhanced target point cloud set. Specifically, methods such as contour fitting, minimum bounding rectangle, or point cloud density analysis can be used to estimate the target size. The estimated target size can be used for subsequent target tracking, classification, or decision control.

[0072] In addition, in complex detection environments (such as tunnels with two walls), there may be more than two reflecting surfaces. In this case, the compensation point cloud is calculated separately for each reflecting surface and then fused with the target point cloud.

[0073] In practical applications, the reflective surface may not be perfectly smooth or may be deformed. An uncertainty factor can be introduced into the propagation model to assess the confidence level of the calculated compensation point location, and the calculated confidence level can be used as the basis for estimating the target size.

[0074] The above description is merely a preferred embodiment of the present invention. It should be noted that for those skilled in the art, other parts not specifically described are existing technology or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation, characterized in that, include: Step 1: Identify whether the current radar detection environment is a multipath saliency scene based on the received radar point cloud data. If the current radar detection environment is a multipath saliency scene, obtain the reflection surface of the multipath point cloud data. Step 2: Extract the target point cloud and multipath point cloud data sequentially from the radar point cloud data; Step 3: Establish a propagation model based on the geometric relationship between the multipath point cloud data and the reflecting surface, and map the coordinates of the multipath points to the target body position according to the propagation model to form a compensation point cloud; Step 4: Fuse the target point cloud with the compensation point cloud to form an enhanced target point cloud set; Step 5: Estimate the size of the target based on the enhanced target point cloud set; The extraction method for the multipath point cloud data is as follows: Based on the geometric information of the reflector and the spatial distribution of the target point cloud, the mirror region of the target point cloud region with respect to the reflector is calculated. For each point in the radar point cloud data, it is determined whether it falls within the mirror region. Points falling within the mirror region are output as a set of candidate multipath points after position feature filtering. A motion model of the target point cloud is established using a point cloud tracking algorithm. The expected position of each point in the target point cloud in the current frame is predicted, and the expected motion trajectory of each point in the target point cloud is obtained. For each point in the candidate multipath point set after position feature filtering, the corresponding theoretical mirror motion trajectory is calculated according to the geometric relationship of the reflecting surface. The degree of agreement between the expected motion trajectory of the point and the theoretical mirror motion trajectory is compared. If the degree of agreement is higher than a preset threshold, the point is confirmed as a multipath point.

2. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 1, characterized in that, The multipath salient scenes include tunnel scenes and metal guardrail edge scenes.

3. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 2, characterized in that, The tunnel scene is identified as follows: If the radar point cloud data shows a number of high-altitude point clouds in front that exceed a set threshold, and the radar point cloud data has relatively clear boundary lines on both sides, then the current scene is identified as a tunnel scene.

4. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation as described in claim 3, characterized in that, The method for recognizing the scene along the metal railing is as follows: If there is one or more static point cloud data that is parallel to the vehicle's direction of travel, has a reflection intensity higher than a set intensity threshold, and is linearly distributed, then the current scene is identified as a scene near a metal guardrail.

5. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 1, characterized in that, The scene recognition results from other sensors are used as external inputs and fused with the recognition results based on radar point cloud data to obtain the final multipath salient scene recognition result.

6. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 1, characterized in that, The reflective surface is obtained through navigation maps or by filtering and fitting radar point cloud data, and it is represented by a reflection line with the vehicle-mounted millimeter-wave radar as the origin. The reflection line is represented as follows: ,in, , The x and y coordinates are on the reflection line. , All are coefficients. This is a constant term.

7. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation as described in claim 6, characterized in that, Each compensation point in the compensation point cloud Represented as: ； in, , Compensation points x and y coordinates , They are respectively the compensation points Corresponding multipath points x and y coordinates Multipath point The directional distance to the reflected ray.

8. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 1, characterized in that, If there are more than two reflecting surfaces in the current radar detection environment, the compensation point cloud is calculated for each reflecting surface and then fused with the target point cloud.

9. The target size enhancement estimation method for vehicle-mounted millimeter-wave radar based on multipath compensation according to claim 1, characterized in that, An uncertainty factor is introduced into the propagation model to assess the confidence level of the calculated compensation point location, and the calculated confidence level is used as the basis for the target size estimation weight.

Images