Multi-radar splicing method based on lane geometric constraint and RCS adaptive gate
By adopting a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates, the problems of association failure and poor target adaptability in multi-radar stitching in curved scenarios are solved, achieving high-precision multi-radar track stitching, improving the continuity and integrity of target recognition, and making it suitable for edge computing devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU INSTITUTE OF TECHNOLOY XIDIAN UNIVERSITY
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-19
AI Technical Summary
Existing multi-radar stitching solutions are prone to problems such as failure of curve track association and poor adaptability to multiple types of targets on curves or S-shaped road sections. Traditional association algorithms are difficult to match targets correctly, resulting in duplicate targets or frequent ID switching. Furthermore, the gate size design cannot take into account different types of traffic participants.
A multi-radar stitching method based on lane geometry constraints and RCS adaptive gates is adopted. In the offline calibration stage, the offline track and latitude and longitude information of the vehicle target are obtained by RTK-GNSS positioning equipment. Rigid body transformation matrix and road tangent angle fitting function are constructed. In the online stitching stage, the size and direction of the associated gate are dynamically adjusted, and the target type is distinguished by RCS value to achieve adaptive association.
It improves the success rate of target stitching in curve and ramp scenarios, ensures the continuity and uniqueness of target IDs, adapts to the measurement accuracy of different types of vehicle targets, reduces computing power consumption and latency, and is suitable for edge computing devices.
Smart Images
Figure CN122063580A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent transportation systems and millimeter-wave radar fusion perception technology, specifically involving a multi-radar stitching method based on lane geometry constraints and RCS adaptive gate. Background Technology
[0002] With the development of intelligent transportation and vehicle-road cooperative technologies, roadside sensing units need to provide continuous, blind-spot-free real-time traffic flow data across the entire road surface. Millimeter-wave radar, due to its all-weather and high-precision characteristics, has become a core sensor. In practical deployments, multiple radars are typically cascaded to cover long-distance road sections.
[0003] In current edge computing architectures, a single radar typically possesses strong edge processing capabilities, enabling it to independently perform point cloud clustering, target tracking, and Kalman filtering, outputting relatively smooth target tracks (position, velocity, ID, etc.). Therefore, the core task of multi-radar stitching is no longer low-level point cloud processing, but rather how to correctly correlate targets output by different radars in overlapping areas and unify their IDs. Existing multi-radar stitching schemes mostly employ a centralized filtering architecture, which aggregates the point clouds from all radars and performs unified tracking filtering. Alternatively, during track-level fusion, nearest neighbor matching is performed using simple Euclidean distance or a fixed elliptical gate (direction determined by target velocity).
[0004] However, existing multi-radar stitching solutions still face the following challenges during the stitching process: 1. Failure of Track Association on Curves: Although a single radar has applied filtering, on curves or S-shaped sections, due to the inertia of the filter, the output velocity direction often diverges along the outer edge of the tangent. When two radars observe the same target from different angles, their output position and velocity vectors may have significant deviations, easily missing true matching targets located in the direction of the lane tangent angle. Traditional association algorithms (based on distance or velocity direction gates) easily identify them as two different targets, resulting in the trajectory not closing, generating duplicate targets or frequent ID switching.
[0005] 2. Poor adaptability to various targets: The gate size in existing technologies is usually fixed or only varies linearly with the covariance matrix, and is not sensitive to target size. This gate size design cannot take into account different traffic participants, that is, it cannot simultaneously adapt to small vehicles with weak RCS and long trucks, resulting in large targets being easily fragmented and small targets being easily lost. Summary of the Invention
[0006] To address the aforementioned problems in the existing technology, this invention provides a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates, wherein several radars are deployed around the lane; the corresponding multi-radar stitching method includes: Offline calibration stage: The vehicle target traveling in the lane is equipped with an RTK-GNSS positioning device; offline trajectory information of the vehicle target in the radar coordinate system is acquired using various radars, and offline latitude and longitude information of the vehicle target in the global latitude and longitude coordinate system is acquired using the RTK-GNSS positioning device; the offline trajectory information is corrected based on the offline trajectory information and the offline latitude and longitude information; a rigid body transformation matrix model is constructed based on the offline latitude and longitude information and the corrected offline trajectory information, and the parameters of the rigid body transformation matrix model are solved to obtain the calibration rigid body transformation matrix; a road tangent angle fitting function is constructed based on the offline latitude and longitude information. Online stitching operation phase: Online trajectory information of vehicle targets in the radar coordinate system is acquired using any adjacent radars A and B; based on the calibration rigid body transformation matrix, the corresponding online latitude and longitude information is calculated according to the online trajectory information; with the position of radar A as the center, the process includes: calculating the road tangent angle based on the road tangent angle fitting function and the online latitude and longitude information of radar A; constructing a gate rotation matrix based on the road tangent angle; dynamically adjusting the size of the associated gate according to the RCS value in the online trajectory information of radar A, and constructing a shape and orientation constraint matrix based on the gate rotation matrix and the adjusted size of the associated gate; based on the shape and orientation constraint matrix and the online trajectory information of radars A and B, determining whether the vehicle target acquired by radar B falls within the associated gate of radar A; if it does, the online trajectory information of radars A and B is fused.
[0007] In one embodiment of the present invention, the offline track information is corrected based on the offline track information and the offline latitude and longitude information, including: Calculate the vehicle target's speed information in the global latitude and longitude coordinate system based on the offline latitude and longitude information; Based on the vehicle target speed information in the radar coordinate system and the vehicle target speed information in the global latitude and longitude coordinate system from the offline track information, a cross-correlation function of the speed sequence is constructed; The cross-correlation peak value of the cross-correlation function is determined, and the cross-correlation peak value is used as the time deviation; The offline track information is corrected based on the time deviation.
[0008] In one embodiment of the present invention, the cross-correlation function of the constructed velocity sequence is expressed as follows: ; in, Indicates clock skew. The cross-correlation function representing the velocity sequence, express The vehicle target's speed information in the radar coordinate system at all times. express The speed information of the vehicle target in the global latitude and longitude coordinate system at all times.
[0009] In one embodiment of the present invention, the constructed rigid body transformation matrix model is expressed by the following formula: ; in, Indicates offline latitude and longitude information. This indicates the location information of the vehicle target in the offline track information. , These are all parameters of the rigid body transformation matrix model, representing translation and rotation angles, respectively.
[0010] In one embodiment of the present invention, the constructed gate rotation matrix is expressed by the formula: ; in, Represents the gate rotation matrix. Indicates the tangent angle of the road.
[0011] In one embodiment of the present invention, dynamically adjusting the size of the associated gate based on the online track information of radar A includes: The associated gate is elliptical or rectangular; the dynamic adjustment process for the dimensions of the long side and the short side of the associated gate is the same; dynamically adjusting the dimensions of the long side of the associated gate according to the online track information of radar A includes: setting a reference associated gate, a gate threshold, and a gate adjustment size; determining the relationship between the RCS value in the online track information of radar A and the gate threshold, and determining the associated gate size correction coefficient based on the determination result and the gate adjustment size; calculating the adjusted dimensions of the long side of the associated gate according to the associated gate size correction coefficient and the reference associated gate.
[0012] In one embodiment of the present invention, the determined associated gate size correction coefficient is expressed by the formula: ; in, This represents the correlation gate size correction factor. This represents the RCS value in the online track information of radar A. This represents the lower limit of the gate threshold. This indicates the upper limit of the gate threshold. This indicates the minimum value of the gate adjustment size. This indicates the maximum value of the gate adjustment size.
[0013] In one embodiment of the present invention, the constructed shape and orientation constraint matrix is expressed by the formula: ; in, Represents the shape and orientation constraint matrix. Represents the gate rotation matrix. express The transpose operation, This indicates the size of the longer side in the adjusted associated gate. This indicates the size of the short side in the adjusted associated gate.
[0014] In one embodiment of the present invention, the formula for determining whether the vehicle target acquired by radar B falls within the associated gate of radar A is as follows: ; in, This indicates the position information of the vehicle target in the online track information acquired by radar B. This indicates the position information of the vehicle target in the online track information acquired by radar A. Represents the shape and orientation constraint matrix. express The transpose operation.
[0015] In one embodiment of the present invention, the corresponding multi-radar stitching method further includes: If the vehicle target acquired by radar B falls within the associated gate of radar A, then the ID information in the online track information of radar A and radar B will be unified.
[0016] The beneficial effects of this invention are: This invention proposes a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates, comprising an offline calibration stage and an online stitching operation stage. In the offline calibration stage, using vehicle targets equipped with RTK-GNSS positioning devices, two tasks are completed simultaneously: radar spatial registration and geometric fitting of road tangent angles, to assist in association during the online stitching operation stage. In the online stitching operation stage, a dual adaptive association gate technique is proposed: the fitted road tangent angle is used to forcibly constrain the association gate angle, and the size of the association gate is dynamically adjusted using the RCS value of the vehicle target. This achieves high-precision full-domain stitching of multi-radar tracks in a latitude and longitude coordinate system, significantly improving the stitching success rate for curves, ramps, and different types of vehicle targets. By forcibly constraining the association gate angle with road tangent angles, this invention effectively corrects the association deviation caused by single radar at curves due to viewing angle differences and filtering lag, thereby decoupling the association process from dependence on the instantaneous velocity vector of the vehicle target. This solves the problem of missed association leading to false vehicle targets in curve scenarios, ensuring the continuity and uniqueness of target IDs in curve scenarios. This invention is adaptable to various vehicle targets. Based on a size-adaptive mechanism using associated gates with RCS values, it ensures measurement accuracy for small vehicle targets while preventing trajectory fragmentation for large vehicle targets, thus improving the integrity of global perception and achieving precise stitching. This invention significantly reduces computational power consumption and latency. By abandoning the secondary filtering of fusion and directly utilizing mature tracks from a single radar for stitching, the algorithm has low complexity and does not require maintaining a complex global covariance matrix, making it suitable for edge computing devices and possessing extremely high application value.
[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the road tangent angle fitting principle provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of dynamically adjusting the size of the associated gate provided in an embodiment of the present invention. Detailed Implementation
[0019] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0020] Please see Figure 1 This invention provides a multi-radar stitching method based on lane geometry constraints and RCS adaptive gates, wherein several radars are deployed around the lane; the corresponding multi-radar stitching method includes: S10, Offline Calibration Stage: The vehicle target traveling in the lane is equipped with an RTK-GNSS positioning device; offline trajectory information of the vehicle target in the radar coordinate system is acquired using various radars, and offline latitude and longitude information of the vehicle target in the global latitude and longitude coordinate system is acquired using the RTK-GNSS positioning device; the offline trajectory information is corrected based on the offline trajectory information and offline latitude and longitude information; a rigid body transformation matrix model is constructed based on the offline latitude and longitude information and the corrected offline trajectory information, and the parameters of the rigid body transformation matrix model are solved to obtain the calibration rigid body transformation matrix; a road tangent angle fitting function is constructed based on the offline latitude and longitude information.
[0021] In this embodiment of the invention, an RTK (Real-time Kinematic)-GNSS (Global Navigation Satellite System) positioning device is introduced during the offline calibration stage. The vehicle target is equipped with this RTK-GNSS positioning device and travels along its lane. The offline trajectory information of the vehicle target in the radar coordinate system is obtained using radar deployed on the roadside, denoted as... , This represents the timestamps corresponding to each point on the vehicle target. The ID representing the vehicle target, ( ) represents the location information of the vehicle target, ( ) indicates that the vehicle target is in ( The velocity component in the direction of ) The RCS value, representing the vehicle target, reflects the target's ability to reflect radar waves and is used to distinguish target type, size, etc. Simultaneously, offline latitude and longitude information of the vehicle target in the global latitude and longitude coordinate system is obtained using RTK-GNSS positioning equipment, denoted as... , Indicates offline longitude information. This indicates offline latitude information.
[0022] Furthermore, although the radar performs hardware-level NTP (Network Time Protocol) time synchronization with the RTK-GNSS positioning device, a millisecond-level time discrepancy still exists between the two due to factors such as network fluctuations. Therefore, it is assumed that there is a clock deviation between the radar and the RTK-GNSS positioning device. Then, by calculating the cross-correlation function of the velocity sequences of the two... To determine the optimal alignment time and correct the track information, this invention proposes correcting offline track information based on offline track information and offline latitude and longitude information, including: calculating the vehicle target speed information in the global latitude and longitude coordinate system based on the offline latitude and longitude information; constructing a cross-correlation function of the speed sequence based on the vehicle target speed information in the radar coordinate system and the vehicle target speed information in the global latitude and longitude coordinate system from the offline track information; solving for the cross-correlation peak value of the cross-correlation function and using the cross-correlation peak value as the time deviation; and correcting the offline track information based on the time deviation. More specifically: This invention calculates the vehicle target's speed information in the global latitude and longitude coordinate system based on prior and subsequent offline latitude and longitude information. Assuming... The offline latitude and longitude information at that moment is , The offline latitude and longitude information at that moment is Since this is a distance calculation based on a local area of the Earth, the influence of the Earth's curvature is ignored. Therefore, the Euclidean distance is used directly to calculate the distance between two points: (1); (2); The cross-correlation function of the velocity sequence constructed in this embodiment of the invention is expressed by the following formula: (3); in, Indicates clock skew. The cross-correlation function representing the velocity sequence, express The vehicle target's speed information in the radar coordinate system at all times. express The vehicle target's velocity information at any given moment in the global latitude and longitude coordinate system. Calculate the time deviation corresponding to the peak cross-correlation value. and utilize Correct the offline track information. Here, 't' can be obtained from the offline track information. To determine, corresponding time, Time can be , -1 hour.
[0023] Furthermore, in this embodiment of the invention, a rigid body transformation matrix (rotation and translation matrix) model is constructed from the radar coordinate system to a global latitude and longitude coordinate system, such as the WGS84 global coordinate system. This is the projection matrix from each radar coordinate system to the global latitude and longitude coordinate system, expressed as: (4); in, This represents the offline latitude and longitude information of the vehicle target in the global latitude and longitude coordinate system. Represents the rotation matrix. = , This indicates the position information of the vehicle target in the offline track information in the radar coordinate system. This represents the translation vector.
[0024] Expanded into matrix form, the final rigid body transformation matrix model is expressed by the following formula: (5); in, Indicates offline latitude and longitude information. This indicates the location information of the vehicle target in the offline track information. , These are parameters in the rigid body transformation matrix model, representing translation and rotation angles, respectively. The rotation angle is solved using the least squares method. Translational displacement This step eliminates installation errors between multiple radars, achieving absolute spatial registration.
[0025] Furthermore, such as Figure 2 As shown, this embodiment of the invention fits a function about the road tangent angle based on the offline latitude and longitude information obtained by the RTK-GNSS positioning device. This function represents the change of the road tangent angle with position, thus directly obtaining the road tangent angle at the current location based on the latitude and longitude information. For details on how to fit the corresponding function based on known point information, please refer to existing technologies. Here, the road tangent angle fitting function is defined as follows: Its input is latitude and longitude information, and its output is the road tangent angle in the global latitude and longitude coordinate system: (6); in, This represents the road tangent angle. It can be seen that this embodiment of the invention establishes a strong "position-direction" prior constraint, which is used for associated gate correction during the subsequent online splicing operation phase.
[0026] S20. Online stitching operation phase: Utilizing any adjacent radars A and B, acquire online trajectory information of vehicle targets in the radar coordinate system; based on the calibration rigid body transformation matrix, calculate the corresponding online latitude and longitude information according to the online trajectory information; with the position of radar A as the center, the execution process includes: calculating the road tangent angle based on the road tangent angle fitting function and the online latitude and longitude information of radar A; constructing a gate rotation matrix based on the road tangent angle; dynamically adjusting the size of the associated gate according to the RCS value in the online trajectory information of radar A, and constructing a shape and orientation constraint matrix based on the gate rotation matrix and the adjusted size of the associated gate; based on the shape and orientation constraint matrix and the online trajectory information of radars A and B, determine whether the vehicle target acquired by radar B falls within the associated gate of radar A; if so, fuse the online trajectory information of radars A and B.
[0027] During system operation, taking any two adjacent radars A and B as an example: Radar A and Radar B acquire online trajectory information of vehicle targets in the radar coordinate system, similar to offline trajectory information. Using the rigid body transformation matrix model of formula (5), the position information of the vehicle target in the online track information output by radar A and radar B is obtained. Real-time projection onto the global latitude and longitude coordinate system yields online latitude and longitude information. Then utilize online latitude and longitude information Construct an adaptive correlation gate. Specifically: Determine the position information of the vehicle target in the online track information acquired by radar B. Does it belong to the vehicle target's position information in the online track information acquired by radar A? , that is Figure 3 As shown, when determining whether target A acquired by radar A and target B acquired by radar B are the same vehicle target, the position information of the vehicle target in the online track information acquired by radar A is used. A correlated gate is constructed around the center. It is evident that this invention abandons the traditional approach that relies solely on the velocity vector output of a single radar. Specifically, the following strategy is employed to achieve adaptive dynamic adjustment of the correlated gate's size: This invention embodiment will use online latitude and longitude information Substituting into formula (6), we obtain the corresponding road tangent angle. According to the tangent angle of the road Construct the gate rotation matrix. The constructed gate rotation matrix is expressed by the following formula: (7); in, Represents the gate rotation matrix. This represents the road tangent angle. It can be seen that regardless of how the vehicle target speed and direction fluctuate from the output of a single radar, the major axis of the associated gate (ellipse or rectangle) is forcibly rotated to an angle with the road tangent based on lane geometry information. Consistent. The specific principle is as follows: when stitching together curves, the vehicle target mainly travels along the lane. Even if the track after single radar filtering has a tendency to tangentially deviate, the real associated vehicle target will inevitably appear in the area extending along the lane. Locking the lane direction can capture the real vehicle target to the greatest extent.
[0028] Typically, based on the RCS (Radar Cross Section) value of a vehicle target, it's possible to distinguish between small RCS targets (such as cars / pedestrians) and large RCS targets (such as trucks / buses). For small RCS targets, the radar reflection points are concentrated, resulting in high single-radar positioning accuracy. Shrinking the gate can avoid misassociating with roadside debris; therefore, the correlation gate needs to be appropriately reduced. For large RCS targets, which are extended targets, two radars may detect the front and rear of the vehicle separately, leading to positional discrepancies of several meters. Therefore, the correlation gate needs to be significantly expanded. Enlarging the gate ensures that the observations from both radars fall within the same correlation area, preventing them from being identified as two separate vehicles. Based on this analysis, the embodiments of the present invention dynamically adjust the size of the associated gate according to the online track information of radar A, including: the associated gate is elliptical or rectangular; the dynamic adjustment process of the size of the long side and the size of the short side of the associated gate is the same; dynamically adjusting the size of the long side of the associated gate according to the online track information of radar A includes: setting a reference associated gate, a gate threshold, and a gate adjustment size; determining the relationship between the RCS value and the gate threshold in the online track information of radar A, and determining the associated gate size correction coefficient according to the determination result and the gate adjustment size; calculating the size of the long side of the adjusted associated gate according to the associated gate size correction coefficient and the reference associated gate. Wherein, The correlation gate size correction coefficient determined in this embodiment of the invention is expressed by the following formula: (8); in, This represents the correlation gate size correction factor. This represents the RCS value in the online track information of radar A. This represents the lower limit of the gate threshold. This indicates the upper limit of the gate threshold. This indicates the minimum value of the gate adjustment size. This indicates the maximum value of the gate adjustment size.
[0029] The adjusted size of the associated gate is expressed by the formula: (9); in, This indicates the adjusted size of the associated gate. This represents the dimensions of the reference correlated gate. The dimensions of the long side and the short side of the adjusted correlated gate can be calculated using formula (9), and are denoted as follows: , . , The calculation process is the same, only the reference associated gate, gate threshold, and gate adjustment size are different.
[0030] Furthermore, in this embodiment of the invention, a shape and orientation constraint matrix is constructed based on the gate rotation matrix and the adjusted dimensions of the associated gate. The constructed shape and orientation constraint matrix is expressed by the following formula: (10); in, Represents the shape and orientation constraint matrix. Represents the gate rotation matrix. express The transpose operation, This indicates the size of the longer side in the adjusted associated gate. This indicates the size of the short side in the adjusted associated gate. The rotation direction of the associated gate is defined, and the diagonal matrix is defined. The size of the associated gate is defined.
[0031] Next, the position information of the vehicle target in the online track information acquired by radar B is determined. Whether the target falls within the associated gate of radar A after the aforementioned rotation and scaling. The formula for determining whether the vehicle target acquired by radar B falls within the associated gate of radar A is as follows: (11); in, This indicates the position information of the vehicle target in the online track information acquired by radar B. This indicates the position information of the vehicle target in the online track information acquired by radar A. Represents the shape and orientation constraint matrix. express The transpose operation.
[0032] If the vehicle target acquired by radar B falls within the associated gate of radar A, the fused position information is directly output, for example, a weighted average is taken as the final result, and Kalman filtering smoothing is not performed; it is directly broadcast downstream. If the vehicle target acquired by radar B does not fall within the associated gate of radar A, it is determined to be a new vehicle target, and its individual ID is retained and output independently.
[0033] Furthermore, the multi-radar stitching method based on lane geometry constraints and RCS adaptive gates provided in this embodiment of the invention also includes: If the vehicle target acquired by radar B falls within the associated gate of radar A, then the ID information in the online track information of radar A and radar B will be unified. The ID information unification can adopt the least-ID principle.
[0034] To verify the effectiveness of the multi-radar stitching method based on lane geometry constraints and RCS adaptive gates provided in this embodiment of the invention, the following experiments were conducted.
[0035] The roadside millimeter-wave radar uses a 350m traffic radar, and the edge computing server uses the EA-B310, whose CPU is a 6-core NVIDIA Carmel ARM processor. The system features a v8.2 64-bit CPU, 6MB L2 + 4MB L3 RAM, 8GB 128-bit LPDDR4x memory at 1600MHz (51.2GB / s), and 16GB eMMC 5.1 storage. The experimental scenario is a curved section of a highway, with two roadside millimeter-wave radars (Radar A and Radar B) deployed, the overlapping area being the curve.
[0036] During the offline calibration phase: An engineering vehicle equipped with an RTK-GNSS positioning device drives past at a speed of 60 km / h. The system records the offline track information of the two radars and the offline latitude and longitude information of the RTK-GNSS. The calibration rigid body transformation matrices of radar A and radar B are calculated, and the road tangent angle fitting function for the curve is generated.
[0037] During the online stitching operation phase: Within the curve area, radar A and radar B acquire online trajectory information of the vehicle target in the radar coordinate system. The system converts this online trajectory information into online latitude and longitude information in the latitude and longitude coordinate system. At this point, due to the different curve perspectives, the two points are 1.5 meters apart on the map. The system constructs a correlation gate based on radar A, queries the road tangent angle fitting function at the location of the vehicle target, and finds that the lane at this location is curved to the right by 20 degrees (…). Although the velocity direction displayed in the online track information of radar A is straight, the system forcibly rotates the gate 20 degrees clockwise. Based on the RCS value in the online track information of radar A, the size of the associated gate is confirmed, and adaptive adjustments are made to the associated gate in both direction and size dimensions. The position information in the online track information of radar B is checked. If a traditional velocity gate is used ( If radar B is located outside the associated gate, the association fails, and two vehicles appear on the screen. This invention uses the lane constraint gate (…). If radar B successfully falls into the associated gate, the association is successful, and a car will appear on the screen.
[0038] In summary, the multi-radar stitching method based on lane geometry constraints and RCS adaptive gate proposed in this invention includes an offline calibration stage and an online stitching operation stage. In the offline calibration stage, using vehicle targets equipped with RTK-GNSS positioning devices, two tasks are completed simultaneously: radar spatial registration and geometric fitting of road tangent angles, to assist in association during the online stitching operation stage. In the online stitching operation stage, a dual adaptive association gate technology is proposed: the fitted road tangent angle is used to forcibly constrain the angle of the association gate, and the size of the association gate is dynamically adjusted using the RCS value of the vehicle target. This achieves high-precision full-domain stitching of multi-radar tracks in the latitude and longitude coordinate system, significantly improving the stitching success rate for curves, ramps, and different types of vehicle targets. This invention, by forcibly constraining the angle of the association gate by the road tangent angle, effectively corrects the association deviation caused by single radar at curves due to viewing angle differences and filtering lag, thereby decoupling the association process from dependence on the instantaneous velocity vector of the vehicle target. This solves the problem of missed association leading to false vehicle targets in curve scenarios, ensuring the continuity and uniqueness of target IDs in curve scenarios. This invention is adaptable to various vehicle targets. Based on a size-adaptive mechanism using associated gates with RCS values, it ensures measurement accuracy for small vehicle targets while preventing trajectory fragmentation for large vehicle targets, thus improving the integrity of global perception and achieving precise stitching. This invention significantly reduces computational power consumption and latency. By abandoning the secondary filtering of fusion and directly utilizing mature tracks from a single radar for stitching, the algorithm has low complexity and does not require maintaining a complex global covariance matrix, making it suitable for edge computing devices and possessing extremely high application value.
[0039] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0040] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the specification and accompanying drawings, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.
[0041] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A multi-radar stitching method based on lane geometry constraints and RCS adaptive gate, characterized in that, Several radars are deployed around the lane; corresponding multi-radar stitching methods include: Offline calibration stage: The vehicle target traveling in the lane is equipped with an RTK-GNSS positioning device; offline trajectory information of the vehicle target in the radar coordinate system is acquired using various radars, and offline latitude and longitude information of the vehicle target in the global latitude and longitude coordinate system is acquired using the RTK-GNSS positioning device; the offline trajectory information is corrected based on the offline trajectory information and the offline latitude and longitude information; a rigid body transformation matrix model is constructed based on the offline latitude and longitude information and the corrected offline trajectory information, and the parameters of the rigid body transformation matrix model are solved to obtain the calibration rigid body transformation matrix; a road tangent angle fitting function is constructed based on the offline latitude and longitude information. Online stitching operation phase: Online trajectory information of vehicle targets in the radar coordinate system is acquired using any adjacent radars A and B; based on the calibration rigid body transformation matrix, the corresponding online latitude and longitude information is calculated according to the online trajectory information; with the position of radar A as the center, the process includes: calculating the road tangent angle based on the road tangent angle fitting function and the online latitude and longitude information of radar A; constructing a gate rotation matrix based on the road tangent angle; dynamically adjusting the size of the associated gate according to the RCS value in the online trajectory information of radar A, and constructing a shape and orientation constraint matrix based on the gate rotation matrix and the adjusted size of the associated gate; based on the shape and orientation constraint matrix and the online trajectory information of radars A and B, determining whether the vehicle target acquired by radar B falls within the associated gate of radar A; if it does, the online trajectory information of radars A and B is fused.
2. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, Correcting the offline track information based on the offline track information and the offline latitude and longitude information includes: Calculate the vehicle target's speed information in the global latitude and longitude coordinate system based on the offline latitude and longitude information; Based on the vehicle target speed information in the radar coordinate system and the vehicle target speed information in the global latitude and longitude coordinate system from the offline track information, a cross-correlation function of the speed sequence is constructed; The cross-correlation peak value of the cross-correlation function is determined, and the cross-correlation peak value is used as the time deviation; The offline track information is corrected based on the time deviation.
3. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 2, characterized in that, The cross-correlation function of the constructed velocity sequences is expressed as follows: ； in, Indicates clock skew. The cross-correlation function representing the velocity sequence, express The vehicle target's speed information in the radar coordinate system at all times. express The speed information of the vehicle target in the global latitude and longitude coordinate system at all times.
4. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, The constructed rigid body transformation matrix model is expressed by the following formula: ； in, Indicates offline latitude and longitude information. This indicates the location information of the vehicle target in the offline track information. , These are all parameters of the rigid body transformation matrix model, representing translation and rotation angles, respectively.
5. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, The constructed gate rotation matrix is expressed by the formula: ； in, Represents the gate rotation matrix. Indicates the tangent angle of the road.
6. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, The size of the associated gate is dynamically adjusted based on the online track information of radar A, including: The associated gate is elliptical or rectangular; the dynamic adjustment process for the dimensions of the long side and the short side of the associated gate is the same; dynamically adjusting the dimensions of the long side of the associated gate according to the online track information of radar A includes: setting a reference associated gate, a gate threshold, and a gate adjustment size; determining the relationship between the RCS value in the online track information of radar A and the gate threshold, and determining the associated gate size correction coefficient based on the determination result and the gate adjustment size; calculating the adjusted dimensions of the long side of the associated gate according to the associated gate size correction coefficient and the reference associated gate.
7. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 6, characterized in that, The determined correlation gate size correction factor is expressed by the formula: ； in, This represents the correlation gate size correction factor. This represents the RCS value in the online track information of radar A. This represents the lower limit of the gate threshold. This indicates the upper limit of the gate threshold. This indicates the minimum value of the gate adjustment size. This indicates the maximum value of the gate adjustment size.
8. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 6, characterized in that, The constructed shape and orientation constraint matrix is expressed by the following formula: ； in, Represents the shape and orientation constraint matrix. Represents the gate rotation matrix. express The transpose operation, This indicates the size of the longer side in the adjusted associated gate. This indicates the size of the short side in the adjusted associated gate.
9. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, The formula for determining whether a vehicle target acquired by radar B falls within the associated gate of radar A is as follows: ； in, This indicates the position information of the vehicle target in the online track information acquired by radar B. This indicates the position information of the vehicle target in the online track information acquired by radar A. Represents the shape and orientation constraint matrix. express The transpose operation.
10. The multi-radar stitching method based on lane geometry constraints and RCS adaptive gates according to claim 1, characterized in that, Corresponding multi-radar stitching methods also include: If the vehicle target acquired by radar B falls within the associated gate of radar A, then the ID information in the online track information of radar A and radar B will be unified.