A vehicle-mounted radar position self-identification system and method

Abstract

This invention discloses a vehicle-mounted radar position self-identification system and method, belonging to the field of vehicle-mounted radar identification technology. It includes multiple anonymous radars installed at different locations on the vehicle, a vehicle bus interface, and a central processing unit. The central processing unit includes a data acquisition unit, a hierarchical logic classification unit, and an iterative vehicle motion refinement and classification result confirmation unit. It receives data from radar sensors and the vehicle bus interface, performs hierarchical logic classification and iterative vehicle motion refinement and classification result confirmation, and obtains the identification result. This invention reduces hardware and manufacturing costs, has high robustness and high accuracy, and can reliably identify various position and attitude errors, including front-to-back, left-to-right, and roll errors. It also features a closed-loop self-verification mechanism, and improves system reliability through the iterative vehicle motion refinement and classification result confirmation process.

Description

Technical Field

[0001] This invention relates to the field of vehicle radar identification technology, specifically to a vehicle radar location self-identification system and method. Background Technology

[0002] With the development of Advanced Driver Assistance Systems (ADAS) and autonomous driving technologies, the application of automotive millimeter-wave radar is becoming increasingly widespread. To reduce overall vehicle costs and system complexity, automotive sensor architecture is evolving from a few high-performance, centrally integrated radars to multiple, relatively simplified, and lower-cost distributed satellite radars. In this emerging architecture, satellite radars typically connect to the central processing unit only via a high-speed data bus (such as LVDS or automotive Ethernet) and employ Power over Ethernet (PoE) technology.

[0003] To further reduce costs, manufacturers tend to remove the hardware location identification pins from satellite radars. In traditional solutions, these pins hard-code the radar's installation location on the vehicle body (e.g., left side of the front bumper, right side of the rear bumper, etc.) through different combinations of voltage levels, open-circuit, or short-circuit states compared to the wiring harness. Removing this hardware makes all radars "anonymous" devices to the central system, preventing the system from directly knowing their physical installation location. This poses a significant obstacle to subsequent sensor data fusion and environmental perception.

[0004] The existing solutions mainly include the following, but all of them have obvious drawbacks:

[0005] (1) Manual configuration: During vehicle production or maintenance, the physical location information of each radar is manually written into the vehicle system using diagnostic tools. This method is inefficient, prone to errors, and cannot meet the needs of automatic identification after sensor replacement.

[0006] (2) End-of-line (EOL) calibration: Before the vehicle leaves the production line, the position and orientation of each sensor are calibrated using specialized and expensive equipment such as lasers and calibration boards. This method increases the investment cost of the production line and the manufacturing cost per vehicle, and cannot provide support throughout the entire life cycle of the vehicle (such as after sensor repair or replacement).

[0007] (3) Online calibration of installation angle: Some purely software-based online calibration methods exist in the prior art, such as those in US Patent US-11782126-B2 11 and the research of Kellner et al. They can estimate and correct small angular deviations (such as 1° or 2° azimuth deviation) in radar installation by using Doppler information of stationary targets detected by radar while the vehicle is in motion. However, the core of these methods is to perform continuous angle value estimation, aiming to solve the "alignment" problem rather than to perform discrete position identification. They cannot fundamentally distinguish between a radar installed at the front of the vehicle and a radar installed at the rear of the vehicle (their azimuth angles differ by about 180°), nor can they distinguish between left and right corner radars. When a "gross error" occurs, such as a radar being incorrectly installed in an undesigned position (e.g., a radar that should be installed on the front bumper is incorrectly installed on the rear bumper), these angle calibration algorithms cannot effectively identify and correct it.

[0008] Based on this, the present invention designs a low-cost, fully automatic, and highly robust vehicle-mounted radar position self-identification system and method to solve the above problems. Summary of the Invention

[0009] To address the aforementioned shortcomings of existing technologies, this invention provides a vehicle-mounted radar position self-identification system and method. Without relying on any specific hardware identification pins or external calibration equipment, it utilizes only the radar's own detection data and the vehicle's conventional motion information. Through a hierarchical, progressively refined identification logic, it automatically and accurately identifies the installation position of each radar during normal vehicle operation.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A vehicle-mounted radar position self-identification system includes:

[0012] Multiple radar sensors (101) are anonymous radars installed at different locations on the vehicle; some of these radars are 4D radars capable of measuring target height information.

[0013] The vehicle bus interface (102) is used to obtain vehicle motion information from the vehicle bus;

[0014] The central processing unit (104) includes a data acquisition unit, a hierarchical logic classification unit, and an iterative vehicle motion refinement and classification result confirmation unit. It is used to receive data from the radar sensor (101) and the vehicle bus interface (102), run hierarchical logic classification and iterative vehicle motion refinement and classification result confirmation, and obtain the recognition result.

[0015] The hierarchical logical classification unit is a logical classifier, which performs the following steps:

[0016] Step 2a: Longitudinal position classification, based on the statistical differences between the traffic environment observed by the forward radar and the rear radar during the vehicle's stable forward driving, distinguish between the forward radar and the rear radar.

[0017] Steps 2a-bis: Forward radar subdivision, based on the geometric differences in the detection field of view and target distribution caused by different installation positions, further distinguish between the front center radar installed directly in front of the vehicle and the front corner radar installed on both sides;

[0018] Step 2b: Lateral position classification. Based on traffic rules, target trajectories of traffic flow, and road boundary features, lateral position classification is performed on the front-angle radar group and the rear-angle radar group respectively.

[0019] Step 2c: Rotation installation error verification. Based on the height information provided by the 4D radar, detect installation errors such as the radar being rotated 180° along the axis of its emission direction.

[0020] The iterative vehicle motion refinement and classification result confirmation unit uses the preliminary vehicle motion information obtained by the data acquisition unit to perform a complete hierarchical logical classification to obtain a preliminary position recognition hypothesis; it uses the stationary target information detected by radar to calculate the corrected and more accurate vehicle motion state in reverse; it uses the corrected vehicle motion information to re-execute a complete hierarchical logical classification process; it judges whether the result of the secondary classification is consistent with the preliminary hypothesis, and makes decisions and locks based on the results.

[0021] Furthermore, it also includes a positioning module (103) for obtaining the vehicle's geographical location information and acquiring high-precision map or similar high-precision map data.

[0022] Furthermore, the data acquisition unit is used to collect raw point cloud data from each anonymous radar sensor (101), vehicle motion information received by the vehicle bus interface (102), traffic rules of the current driving area, and high-precision map or near-high-precision map data from the optional positioning module (103).

[0023] Furthermore, the raw point cloud data of the radar sensor (101) includes the distance, azimuth angle, radial relative velocity of each detected target, and the height information of the 4D radar; the vehicle motion information mainly includes the longitudinal velocity and yaw rate of the vehicle; the high-precision map or near-high-precision map data of the positioning module (103) includes the topological structure information of the road boundary.

[0024] Furthermore, the specific method for step 2a is as follows:

[0025] Step (1) Use the vehicle speed information to filter out the time window in which the vehicle is in a stable straight-line forward motion;

[0026] Step (2) Within this time window, the radial relative velocities of all target points acquired by each radar are statistically analyzed;

[0027] Step (3) Calculate the statistical mean of the radial velocity distribution of the target;

[0028] Step (4) Perform logical judgment:

[0029] For forward-facing radar, most targets in its field of view are either approaching or being overtaken, so its radial velocity distribution will be significantly biased towards negative values.

[0030] For rearward radar, most targets in its field of view are either moving away or being pushed back, so its radial velocity distribution will be significantly biased towards positive values.

[0031] Step (5) classifies the radar into “forward group” or “backward group” by setting a threshold.

[0032] Furthermore, the specific methods for steps 2a-bis are as follows:

[0033] Step (1) Analyze the spatial distribution characteristics of stationary targets detected by each radar in the "forward group" during stable straight-line movement;

[0034] Step (2) Perform logical judgment:

[0035] Front-to-center radar: The target point cloud detected by it, after a certain period of accumulation, will show a relatively symmetrical distribution range in the azimuth angle with 0 degrees as the center.

[0036] Forward-angle radar: The distribution of the target point cloud detected by it in the azimuth angle is completely asymmetrical;

[0037] Step (3) subdivides the “forward group” into the “front-center radar group” and the “front-angle radar group”.

[0038] Furthermore, the specific method for step 2b is as follows:

[0039] Step (1) First, call the traffic rules of the current driving area obtained by the data acquisition unit;

[0040] Step (2) uses the following features to identify left and right radars:

[0041] Feature 1: The target trajectory of oncoming traffic;

[0042] For the front corner radar group: when an oncoming target is tracked by two front corner radars at the same time, the left and right radars are distinguished by comparing the longitudinal position of the target when it disappears from the field of view of each radar.

[0043] For the rear corner radar group: when an oncoming target passes by the side of the vehicle, the initial generated position of the target is compared to distinguish the left and right rear corner radars;

[0044] Feature 2: Road boundary features;

[0045] A radar installed closer to the curb can detect road boundaries more stably and at a closer distance than a radar installed on the other side.

[0046] Step (3) By integrating the judgment results of one or more of the above features, “left” and “right” are distinguished.

[0047] Furthermore, step 2c specifically includes:

[0048] Step (1) The radar that is installed normally detects targets mainly in the vertical direction at an elevation angle of 0°.

[0049] Step (2) The radar installed by 180° roll has its coordinate system completely flipped, and has two detectable features: pitch angle reversal and azimuth angle reversal;

[0050] Step (3) Determine if the radar has a 180° roll installation error.

[0051] Furthermore, the iterative vehicle motion refinement and classification result confirmation unit specifically includes the following steps:

[0052] Step 601, Preliminary Classification: First, using the preliminary vehicle motion information from the vehicle bus, perform a complete hierarchical logical classification to obtain a preliminary, low-confidence position identification hypothesis;

[0053] Step 602: Refine the hypothetical vehicle motion: Adopt this initial hypothesis; Based on the geometric relationship that the radar is located at the "rear right corner", the system uses the stationary target information detected by the radar to calculate a corrected and more accurate vehicle motion state in reverse.

[0054] Step 603, Secondary Classification and Verification: Using this more accurate vehicle motion information, re-execute the complete hierarchical logical classification process;

[0055] Step 604, Decision and Locking:

[0056] If the result of the secondary classification is consistent with the initial hypothesis, then the hypothesis is proven to be correct, and the location identification result of the radar is locked.

[0057] If the result of the secondary classification is inconsistent with the initial hypothesis, it means that the initial hypothesis is wrong. The current hypothesis will be discarded, and a new hypothesis can be established based on the new classification result. The iteration will be repeated until the result converges.

[0058] To better achieve the objectives of this invention, this invention also provides a method for self-identification of vehicle radar position, which utilizes the aforementioned vehicle radar position self-identification system and includes the following steps:

[0059] Step 1: The system begins collecting data after the vehicle starts and starts moving;

[0060] Step Two: Hierarchical logical classification, including the following steps:

[0061] Step 2a: Longitudinal position classification, based on the statistical differences between the traffic environment observed by the forward radar and the rear radar during the vehicle's stable forward driving, distinguish between the forward radar and the rear radar.

[0062] Steps 2a-bis: Forward radar subdivision, based on the geometric differences in the detection field of view and target distribution caused by different installation positions, further distinguish between the front center radar installed directly in front of the vehicle and the front corner radar installed on both sides;

[0063] Step 2b: Lateral position classification. Based on traffic rules, target trajectories of traffic flow, and road boundary features, lateral position classification is performed on the front-angle radar group and the rear-angle radar group respectively.

[0064] Step 2c: Rotation installation error verification. Based on the height information provided by the 4D radar, detect installation errors such as the radar being rotated 180° along the axis of its emission direction.

[0065] Step 3: Iterative refinement and classification results of the vehicle motion, including the following steps:

[0066] Using the initial vehicle motion information acquired by the data acquisition unit, a complete hierarchical logical classification is performed to obtain an initial position identification hypothesis; using the stationary target information detected by radar, the corrected and more accurate vehicle motion state is calculated in reverse; using the corrected vehicle motion information, a complete hierarchical logical classification process is re-executed; it is determined whether the result of the secondary classification is consistent with the initial hypothesis, and a decision and lock are made based on the result.

[0067] Compared with the prior art, the beneficial effects of this invention are: 1) Reduced hardware and manufacturing costs: It completely eliminates the need for hardware position identification pins, special wiring harness design and expensive end-of-line (EOL) calibration equipment, which significantly reduces the material cost of the sensor and the manufacturing cost of the whole vehicle.

[0068] 2) High robustness and high accuracy: It adopts a hierarchical, multimodal logical classification method, which integrates kinematic statistical features, traffic rules, environmental features, and optional high-precision map information. Compared with existing technologies that rely solely on basic physical models for angle estimation, it is more adaptable to noise and various working conditions, and can reliably identify various position and posture errors, including forward and backward, left and right, and roll.

[0069] 3) It has a closed-loop self-verification mechanism: Through an iterative process of refining and classifying the vehicle motion results, the mutual promotion and verification of position recognition and vehicle motion estimation are realized, which greatly improves the final performance and reliability of the system. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0071] Figure 1 This is a block diagram of the architecture of a vehicle-mounted radar position self-identification system according to the present invention.

[0072] Figure 2 This is a flowchart of a vehicle-mounted radar position self-identification method according to the present invention.

[0073] Figure 3 A schematic diagram illustrating the principle of vertical position classification (front / back).

[0074] Figure 4 A schematic diagram illustrating the principle of horizontal position classification (left / right).

[0075] Figure 5 A schematic diagram illustrating the error detection principle for 180° roll-up installation.

[0076] Figure 6 A flowchart for confirming the results of iterative vehicle motion refinement and classification. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0078] Example 1: Please refer to the accompanying drawings in the instruction manual. Figure 1 A vehicle-mounted radar position self-identification system, comprising:

[0079] There are multiple radar sensors 101 (referred to as radars), which are anonymous radars installed in different positions on the vehicle (such as the front center, front left corner, front right corner, rear left corner, rear right corner, etc.); some of these radars are 4D radars, which have the ability to measure target height information.

[0080] The vehicle bus interface 102 is used to obtain vehicle motion information from the vehicle controller area network (CAN) or a similar bus.

[0081] Optional positioning module 103, such as Global Navigation Satellite System (GNSS), is used to acquire the vehicle's geographic location information and high-precision map or near-high-precision map data, particularly the topology of the road's outer edge.

[0082] The central processing unit 104 includes a data acquisition unit, a hierarchical logic classification unit, and an iterative vehicle motion refinement and classification result confirmation unit. It is used to receive data from the radar sensor 101, the vehicle bus interface 102, and the positioning module 103, run the position self-identification algorithm, and output the identification result 105 to the upper-layer application (such as data fusion, ADAS function, etc.).

[0083] The data acquisition unit is used to collect raw point cloud data from each anonymous radar sensor 101, vehicle motion information received by the vehicle bus interface 102, traffic rules of the current driving area (left-hand drive / right-hand drive, etc.), and high-precision map or similar high-precision map data from the positioning module 103 (optional).

[0084] The raw point cloud data from radar sensor 101 includes the distance, azimuth, radial relative velocity, and (for 4D radar) altitude information for each detected target. Vehicle motion information primarily includes the vehicle's longitudinal velocity and yaw rate. The high-precision map or near-high-precision map data from positioning module 103 includes topological information about road boundaries.

[0085] The traffic rules for the current driving area can be obtained through any one or a combination of the following methods:

[0086] a) Based on the positioning module: If the vehicle is equipped with an optional positioning module 103, it can query the database or obtain information through network services based on the geographical location.

[0087] b) Self-learning: The system learns and makes judgments autonomously by statistically analyzing oncoming traffic detected by radar over a period of time. For example, in a right-hand drive driving area (where vehicles drive on the left), the probability of an oncoming vehicle (a target with a very high negative relative speed) appearing on the right side of the vehicle will be significantly greater than that on the left, and vice versa.

[0088] c) Multi-sensor fusion: This information is obtained from the perception results of other vehicle sensors (such as cameras).

[0089] The hierarchical logical classification unit is a hierarchical logical classifier based on physical and environmental features. This classifier sequentially resolves positional ambiguities in the vertical, horizontal, and rotational dimensions. It includes the following steps:

[0090] 2a: Vertical position classification (distinguishing between "front" and "back"). Please refer to the instruction manual appendix. Figure 3 This step aims to distinguish between radars installed in the front half of the vehicle (forward radar) and radars installed in the rear half (rear radar). The principle is based on the statistically significant differences in the traffic environment observed by the front and rear radars during stable forward driving. The specific method is as follows:

[0091] (1) The system uses the vehicle speed information to filter out the time window when the vehicle is in a stable straight-line forward state.

[0092] (2) Within this time window, the radial relative velocity of all target points collected by each radar is statistically analyzed.

[0093] (3) Calculate the statistical mean of the radial velocity distribution of the target.

[0094] (4) Perform logical judgments:

[0095] For the forward-facing radar 301, most targets in its field of view (oncoming vehicles, slower vehicles in the same direction ahead, stationary objects) are either approaching or being overtaken, so its radial velocity distribution 302 will be significantly biased towards negative values.

[0096] For the rearward radar 303, most targets in its field of view (vehicles overtaking from behind, vehicles traveling in the same direction from behind) are either moving away or being pushed back, so its radial velocity distribution 304 will be significantly biased towards positive values.

[0097] (5) By setting a threshold (e.g., 0), the system can classify radars into “forward group” or “backward group” with high confidence.

[0098] 2a-bis: Forward Radar Subdivision (Distinguishing between "Front Center" and "Front Corner"). After identifying the "Forward Group" radars, this step aims to further distinguish between the "Front Center Radar" mounted directly in front of the vehicle and the "Front Corner Radars" mounted on either side. This is based on the geometric differences in the detection field of view and target distribution caused by different mounting positions. The specific method is as follows:

[0099] (1) System analysis of the spatial distribution characteristics of stationary targets (such as road boundaries, guardrails, streetlights, etc.) detected by each radar in the "forward group" when traveling steadily in a straight line.

[0100] (2) Perform logical judgments:

[0101] Front-center radar: After a certain period of accumulation, the target point cloud detected by it will show a relatively symmetrical distribution range in the azimuth angle centered on 0 degrees. Although the target point density directly in front of the vehicle may not be the highest, the static environmental points detected by it (such as road boundaries, guardrails, etc.) are basically symmetrical in the angular range on the left and right sides (there may be a slight deviation of less than 10 degrees due to installation errors).

[0102] Front-angle radar: The distribution of the target point cloud detected by it is completely asymmetrical in azimuth. For example, a radar installed at the left front corner (such as at 45 degrees) will have the center of its entire point cloud distribution significantly biased towards the positive angle region, rather than being symmetrical with 0 degrees as the center.

[0103] (3) By analyzing the symmetry of the target point cloud distribution or the position of the centroid, the system can subdivide the "forward group" into the "front-middle radar group" and the "front-angle radar group".

[0104] 2b: Horizontal position classification (distinguishing between "left" and "right"). Please refer to the instruction manual appendix. Figure 4 After completing the forward radar segmentation, the system performs lateral position classification for the forward-facing radar group and the rearward radar group (if multiple rear radars exist). This step integrates traffic rules and environmental characteristics, exhibiting extremely high robustness. The specific method is as follows:

[0105] (1) The system first calls the traffic rules of the current driving area obtained by the data acquisition unit.

[0106] (2) The following features can be used for left and right radar identification:

[0107] Feature 1: Target trajectory of oncoming traffic flow 401. This feature distinguishes left and right sides by comparing the detection trajectory of the same oncoming target with that of the contrast-angle radar. Oncoming target identification can be based on its extremely high negative relative velocity and can be cross-verified by other sensors such as cameras.

[0108] For the front corner radar array: When an oncoming target (e.g., in the right lane of a country where driving is on the left) is tracked simultaneously by two front corner radars, the system compares the longitudinal position (X-coordinate, positive for the vehicle's direction of travel) of the target when it disappears from the field of view of each radar. Due to the limitation of the field of view (FOV), the front corner radar on the oncoming side (right) will track the target for a longer time than the radar on the other side, causing it to disappear at a later position (e.g., X = -10m). The front corner radar on the non-oncoming side (left) will lose the target earlier (e.g., X = 20m). The left and right radars can be distinguished by comparing the X-coordinate of the target's disappearance point.

[0109] For the rear corner radar group: the logic is similar to that of the front corner radar, but the comparison is made based on the initial generation position of the target. When an oncoming target passes by the side of the vehicle, the rear corner radar on the oncoming side (right side) will detect the target earlier (e.g., generating a trajectory at X=1m), while the rear corner radar on the non-oncoming side (left side), due to the limitations of FOV and installation angle, may not detect the target until it has passed far beyond the center of the vehicle's rear axle (e.g., generating a trajectory at X=-20m), or may not detect it at all. By comparing the X coordinates of the target generation point, the left and right rear corner radars can be distinguished.

[0110] Feature 2: Road Boundary Features. Road boundaries (such as curbs / guardrails, etc.) 403 typically appear as a continuous, stationary target line with a radial velocity close to the projection of the vehicle's velocity. The radar installed closer to the curb (left radar 404) will detect this boundary line more stably and at a closer distance than the radar on the other side (right radar 402). The algorithm analyzes the spatial distribution of stationary target points, fits the road boundary line, and uses this to identify the radar on the curb side.

[0111] (Optional) Feature 3: Road boundaries on the map. If the system has acquired high-precision map data, it can match the road boundary features detected by radar with the topological structure of the road edges on the map. Radars that successfully match the left road boundary are identified as left-side radars, and vice versa. This method can be used as a high-confidence verification method or as the main basis for judgment when other features are not obvious.

[0112] (3) By integrating the judgment results of one or more of the above features, the system can distinguish "left" and "right" with high confidence.

[0113] 2c: Rotational installation error check (180° roll detection). Please refer to the instruction manual appendix. Figure 5 This step is used to detect a serious installation error: the radar has been rotated 180° along its emission axis. This detection relies on altitude information provided by the 4D radar. Specifically, it includes:

[0114] (1) The radar 501, which is normally installed (roll angle = 0°), has a certain installation height (e.g., 30cm to 100cm). The vertical distribution 502 of the target points it detects is mainly concentrated in the 0° pitch angle (corresponding to other vehicles and infrastructure) and negative pitch angle area (corresponding to ground reflection).

[0115] (2) The radar 503, mounted with a 180° roll, has its coordinate system completely flipped. This results in two significant, detectable features:

[0116] Pitch angle reversal: The portion of the antenna that was originally pointing towards the ground is now pointing towards the sky. Therefore, the ground reflection point that would normally appear at a negative pitch angle will now appear in a large positive pitch angle region 504. At the same time, the beam that was originally horizontally pointed towards other vehicles will now point towards the ground, resulting in a significant decrease in target detection capability near a 0° pitch angle.

[0117] Horizontal angle (azimuth) reversal: The radar's left and right directions are also reversed. A target physically located to the left of the vehicle will be incorrectly reported as being to the right (i.e., its azimuth angle is multiplied by -1).

[0118] (3) By jointly analyzing the density distribution of the detection points in the elevation angle (whether there is a large positive angle dense area) and the symmetry in the azimuth angle (compared with adjacent, confirmed radars, whether there is a consistent left and right reversal), the algorithm can determine with high confidence that the radar has a 180° roll installation error.

[0119] Iterative autonomous vehicle motion refinement and classification result confirmation unit.

[0120] Using the initial vehicle motion information acquired by the data acquisition unit, a complete hierarchical logical classification is performed to obtain a preliminary, low-confidence position identification hypothesis. Then, using stationary target information detected by radar, a revised, more accurate vehicle motion state is calculated. This more accurate vehicle motion information is then used to re-execute the complete hierarchical logical classification process. The result of the second classification is compared to the initial hypothesis, and a decision and lock are made based on the result. This method can resolve classification errors that may be caused by inaccurate initial vehicle motion information and greatly improves the confidence of the final identification result. Please refer to the instruction manual appendix. Figure 6 Specifically, it includes the following steps:

[0121] 601. Preliminary Classification: The system first uses the preliminary vehicle motion information from the vehicle bus, which may be biased, to perform a complete hierarchical logical classification (step two) to obtain a preliminary, low-confidence position identification hypothesis (for example, assuming that a certain radar is the "rear right corner" radar).

[0122] 602. Refinement of the Hypothetical-Driven Vehicle Motion: The system adopts this initial assumption. Based on the geometric relationship of the radar being located at the "rear right corner," the system uses the stationary target information detected by the radar to calculate a corrected and more accurate vehicle motion state (speed and yaw rate). This step can draw on the principle of using radar data to calibrate vehicle motion in existing technologies, but its application is premised on the initial classification assumption of this invention.

[0123] 603. Secondary Classification and Verification: The system uses this more accurate vehicle motion information to re-execute a complete hierarchical logical classification process.

[0124] 604. Decision-making and Lock-in:

[0125] If the result of the secondary classification is consistent with the initial hypothesis, then the hypothesis is proven to be correct, and the system locks the radar's location identification result with high confidence.

[0126] If the result of the secondary classification is inconsistent with the initial hypothesis, it indicates that the initial hypothesis is incorrect (most likely due to a bias in the initial vehicle motion information). The system will discard the current hypothesis and may choose to build a new hypothesis based on the new classification result, repeating the iterative process until the result converges.

[0127] Through this iterative verification closed loop, the present invention can not only identify the radar position, but also simultaneously optimize the accuracy of the vehicle motion estimation, forming a virtuous cycle of mutual reinforcement.

[0128] Example 2: Figure 2 As shown, a method for self-identification of vehicle-mounted radar location includes the following steps:

[0129] Step 1: The system begins collecting data after the vehicle starts and starts moving. The data acquisition unit collects raw point cloud data from each anonymous radar sensor 101, vehicle motion information received from the vehicle bus interface 102, traffic rules of the current driving area (left-hand drive / right-hand drive, etc.), and high-precision map or near-high-precision map data from the positioning module 103 (optional).

[0130] The raw point cloud data from radar sensor 101 includes the distance, azimuth, radial relative velocity, and (for 4D radar) altitude information for each detected target. Vehicle motion information primarily includes the vehicle's longitudinal velocity and yaw rate. The high-precision map or near-high-precision map data from positioning module 103 includes topological information about road boundaries.

[0131] The traffic rules for the current driving area can be obtained through any one or a combination of the following methods:

[0132] a) Based on the positioning module: If the vehicle is equipped with an optional positioning module 103, it can query the database or obtain information through network services based on the geographical location.

[0133] b) Self-learning: The system learns and makes judgments autonomously by statistically analyzing oncoming traffic detected by radar over a period of time. For example, in a right-hand drive driving area (where vehicles drive on the left), the probability of an oncoming vehicle (a target with a very high negative relative speed) appearing on the right side of the vehicle will be significantly greater than that on the left, and vice versa.

[0134] c) Multi-sensor fusion: This information is obtained from the perception results of other vehicle sensors (such as cameras).

[0135] Step Two: Hierarchical logical classification, specifically including the following steps:

[0136] 2a: Vertical position classification (distinguishing between "front" and "back"). Please refer to the instruction manual appendix. Figure 3 This step aims to distinguish between radars installed in the front half of the vehicle (forward radar) and radars installed in the rear half (rear radar). The principle is based on the statistically significant differences in the traffic environment observed by the front and rear radars during stable forward driving. The specific method is as follows:

[0137] (1) The system uses the vehicle speed information to filter out the time window when the vehicle is in a stable straight-line forward state.

[0138] (2) Within this time window, the radial relative velocity of all target points collected by each radar is statistically analyzed.

[0139] (3) Calculate the statistical mean of the radial velocity distribution of the target.

[0140] (4) Perform logical judgments:

[0141] For the forward-facing radar 301, most targets in its field of view (oncoming vehicles, slower vehicles in the same direction ahead, stationary objects) are either approaching or being overtaken, so its radial velocity distribution 302 will be significantly biased towards negative values.

[0142] For the rearward radar 303, most targets in its field of view (vehicles overtaking from behind, vehicles traveling in the same direction from behind) are either moving away or being pushed back, so its radial velocity distribution 304 will be significantly biased towards positive values.

[0143] (5) By setting a threshold (e.g., 0), the system can classify radars into “forward group” or “backward group” with high confidence.

[0144] 2a-bis: Forward Radar Subdivision (Distinguishing between "Front Center" and "Front Corner"). After identifying the "Forward Group" radars, this step aims to further distinguish between the "Front Center Radar" mounted directly in front of the vehicle and the "Front Corner Radars" mounted on either side. This is based on the geometric differences in the detection field of view and target distribution caused by different mounting positions. The specific method is as follows:

[0145] (1) System analysis of the spatial distribution characteristics of stationary targets (such as road boundaries, guardrails, streetlights, etc.) detected by each radar in the "forward group" when traveling steadily in a straight line.

[0146] (2) Perform logical judgments:

[0147] Front-center radar: After a certain period of accumulation, the target point cloud detected by it will show a relatively symmetrical distribution range in the azimuth angle centered on 0 degrees. Although the target point density directly in front of the vehicle may not be the highest, the static environmental points detected by it (such as road boundaries, guardrails, etc.) are basically symmetrical in the angular range on the left and right sides (there may be a slight deviation of less than 10 degrees due to installation errors).

[0148] Front-angle radar: The distribution of the target point cloud detected by it is completely asymmetrical in azimuth. For example, a radar installed at the left front corner (such as at 45 degrees) will have the center of its entire point cloud distribution significantly biased towards the positive angle region, rather than being symmetrical with 0 degrees as the center.

[0149] (3) By analyzing the symmetry of the target point cloud distribution or the position of the centroid, the system can subdivide the "forward group" into the "front-middle radar group" and the "front-angle radar group".

[0150] 2b: Horizontal position classification (distinguishing between "left" and "right"). Please refer to the instruction manual appendix. Figure 4 After completing the forward radar segmentation, the system performs lateral position classification for the forward-facing radar group and the rearward radar group (if multiple rear radars exist). This step integrates traffic rules and environmental characteristics, exhibiting extremely high robustness. The specific method is as follows:

[0151] (1) The system first calls the traffic rules of the current driving area obtained by the data acquisition unit.

[0152] (2) The following features can be used for left and right radar identification:

[0153] Feature 1: Target trajectory of oncoming traffic flow 401. This feature distinguishes left and right sides by comparing the detection trajectory of the same oncoming target with that of the contrast-angle radar. Oncoming target identification can be based on its extremely high negative relative velocity and can be cross-verified by other sensors such as cameras.

[0154] For the front corner radar array: When an oncoming target (e.g., in the right lane of a country where driving is on the left) is tracked simultaneously by two front corner radars, the system compares the longitudinal position (X-coordinate, positive for the vehicle's direction of travel) of the target when it disappears from the field of view of each radar. Due to the limitation of the field of view (FOV), the front corner radar on the oncoming side (right) will track the target for a longer time than the radar on the other side, causing it to disappear at a later position (e.g., X = -10m). The front corner radar on the non-oncoming side (left) will lose the target earlier (e.g., X = 20m). The left and right radars can be distinguished by comparing the X-coordinate of the target's disappearance point.

[0155] For the rear corner radar group: the logic is similar to that of the front corner radar, but the comparison is made based on the initial generation position of the target. When an oncoming target passes by the side of the vehicle, the rear corner radar on the oncoming side (right side) will detect the target earlier (e.g., generating a trajectory at X=1m), while the rear corner radar on the non-oncoming side (left side), due to the limitations of FOV and installation angle, may not detect the target until it has passed far beyond the center of the vehicle's rear axle (e.g., generating a trajectory at X=-20m), or may not detect it at all. By comparing the X coordinates of the target generation point, the left and right rear corner radars can be distinguished.

[0156] Feature 2: Road Boundary Features. Road boundaries (such as curbs / guardrails, etc.) 403 typically appear as a continuous, stationary target line with a radial velocity close to the projection of the vehicle's velocity. The radar installed closer to the curb (left radar 404) will detect this boundary line more stably and at a closer distance than the radar on the other side (right radar 402). The algorithm analyzes the spatial distribution of stationary target points, fits the road boundary line, and uses this to identify the radar on the curb side.

[0157] (Optional) Feature 3: Road boundaries on the map. If the system has acquired high-precision map data, it can match the road boundary features detected by radar with the topological structure of the road edges on the map. Radars that successfully match the left road boundary are identified as left-side radars, and vice versa. This method can be used as a high-confidence verification method or as the main basis for judgment when other features are not obvious.

[0158] (3) By integrating the judgment results of one or more of the above features, the system can distinguish "left" and "right" with high confidence.

[0159] 2c: Rotational installation error check (180° roll detection). Please refer to the instruction manual appendix. Figure 5 This step is used to detect a serious installation error: the radar has been rotated 180° along its emission axis. This detection relies on altitude information provided by the 4D radar. Specifically, it includes:

[0160] (1) The radar 501, which is normally installed (roll angle = 0°), has a certain installation height (e.g., 30cm to 100cm). The vertical distribution 502 of the target points it detects is mainly concentrated in the 0° pitch angle (corresponding to other vehicles and infrastructure) and negative pitch angle area (corresponding to ground reflection).

[0161] (2) The radar 503, mounted with a 180° roll, has its coordinate system completely flipped. This results in two significant, detectable features:

[0162] Pitch angle reversal: The portion of the antenna that was originally pointing towards the ground is now pointing towards the sky. Therefore, the ground reflection point that would normally appear at a negative pitch angle will now appear in a large positive pitch angle region 504. At the same time, the beam that was originally horizontally pointed towards other vehicles will now point towards the ground, resulting in a significant decrease in target detection capability near a 0° pitch angle.

[0163] Horizontal angle (azimuth) reversal: The radar's left and right directions are also reversed. A target physically located to the left of the vehicle will be incorrectly reported as being to the right (i.e., its azimuth angle is multiplied by -1).

[0164] (3) By jointly analyzing the density distribution of the detection points in the elevation angle (whether there is a large positive angle dense area) and the symmetry in the azimuth angle (compared with adjacent, confirmed radars, whether there is a consistent left and right reversal), the algorithm can determine with high confidence that the radar has a 180° roll installation error.

[0165] Step 3: Iterative Refinement and Classification Result Confirmation of Vehicle Motion. Using the preliminary vehicle motion information acquired by the data acquisition unit, a complete hierarchical logical classification is performed to obtain a preliminary, low-confidence position identification hypothesis. Using the stationary target information detected by radar, a corrected and more accurate vehicle motion state is calculated. Using this more accurate vehicle motion information, a complete hierarchical logical classification process is executed again. The result of the second classification is compared with the preliminary hypothesis, and a decision and locking mechanism are made based on the result. This method can resolve classification errors that may be caused by inaccurate preliminary vehicle motion information and greatly improves the confidence of the final recognition result. Please refer to the instruction manual appendix. Figure 6 Specifically, it includes the following steps:

[0166] 601. Preliminary Classification: The system first uses the preliminary vehicle motion information from the vehicle bus, which may be biased, to perform a complete hierarchical logical classification (step two) to obtain a preliminary, low-confidence position identification hypothesis (for example, assuming that a certain radar is the "rear right corner" radar).

[0167] 602. Refinement of the Hypothetical-Driven Vehicle Motion: The system adopts this initial assumption. Based on the geometric relationship of the radar being located at the "rear right corner," the system uses the stationary target information detected by the radar to calculate a corrected and more accurate vehicle motion state (speed and yaw rate). This step can draw on the principle of using radar data to calibrate vehicle motion in existing technologies, but its application is premised on the initial classification assumption of this invention.

[0168] 603. Secondary Classification and Verification: The system uses this more accurate vehicle motion information to re-execute a complete hierarchical logical classification process.

[0169] 604. Decision-making and Lock-in:

[0170] If the result of the secondary classification is consistent with the initial hypothesis, then the hypothesis is proven to be correct, and the system locks the radar's location identification result with high confidence.

[0171] If the result of the secondary classification is inconsistent with the initial hypothesis, it indicates that the initial hypothesis is incorrect (most likely due to a bias in the initial vehicle motion information). The system will discard the current hypothesis and may choose to build a new hypothesis based on the new classification result, repeating the iterative process until the result converges.

[0172] Through this iterative verification closed loop, the present invention can not only identify the radar position, but also simultaneously optimize the accuracy of the vehicle motion estimation, forming a virtuous cycle of mutual reinforcement.

[0173] Experimental Example: Suppose a car equipped with five anonymous radars (front left corner, front right corner, front center, rear left corner, and rear right corner) is started and driven for the first time in a country where it drives on the left. The specific identification process using the system of this invention is as follows:

[0174] 1) Upon system startup, the central processing unit 104 begins receiving point cloud data from the five radar sensors 101 and acquires preliminary vehicle speed and yaw rate from the vehicle bus interface 102. By analyzing the statistical characteristics of oncoming traffic, the system autonomously learns and determines that it is currently in a left-hand driving zone.

[0175] 2) When the vehicle is When traveling steadily in a straight line at a speed of km / h on an urban road, the system enters the longitudinal position classification stage (distinguishing between "forward" and "rear"). The processing unit analysis reveals that the average radial velocity of targets from three radars is significantly negative, while the average velocity of the other two radars is significantly positive. Based on this, the system classifies the former as the "forward group" and the latter as the "rear group."

[0176] 3) Subsequently, the system further subdivides the three radars in the "forward-facing group" (distinguishing between "forward-center" and "forward-angle"). By analyzing the static target point cloud detected over a period of time, the system finds that the point cloud distribution of one radar is symmetrical about 0 degrees in azimuth (e.g., from -60 degrees to +60 degrees). The point cloud distribution of the other two radars is significantly biased to one side (e.g., one concentrated in the range of -90 degrees to 0 degrees, and the other concentrated in the range of 0 degrees to +90 degrees). Based on this, the system identifies the symmetrically distributed radar as the "forward-center" radar, and classifies the other two into the "forward-angle group".

[0177] 4) Next, the system classifies the two radars in the "front corner group" by their lateral position (distinguishing between "left" and "right"). The system identifies an oncoming target approaching from the right lane, which is simultaneously tracked by both front corner radars. Analysis reveals that one radar tracks the target until its longitudinal position X = -10m before losing track, while the other radar loses track at X = 20m. Based on the principle of "longer tracking time and later disappearance position," the system identifies the former as the "front right corner" radar and the latter as the "front left corner" radar. A similar logic is applied to the radars in the "rear corner group," identifying them as "rear left corner" and "rear right corner."

[0178] 5) The system performs rotational installation error checks (180° roll detection) on all five radars. Analysis of their 4D point cloud data distribution at the elevation angle reveals that target points for all radars are concentrated in the 0° and negative angle regions, with no errors observed. Figure 5 The image shows an abnormal distribution of dense point clouds in a large positive elevation angle region. Meanwhile, its azimuth detection results are consistent with other radars, and no left-right reversal phenomenon was found. Therefore, it is determined that the roll angles of all radars are normal.

[0179] 6) At this point, the system obtains a preliminary identification result: the five radars are for the front left corner, front right corner, front center, rear left corner, and rear right corner. The system then enters the iterative stage of refining and confirming the vehicle motion and classification results.

[0180] The system adopts the identification results of the "front right corner" radar and uses the stationary targets such as streetlights and buildings detected by the radar to calculate a more accurate vehicle speed of 50.2 km / h and a yaw rate of 0.01 rad / s.

[0181] The system uses this refined vehicle motion information to re-execute the classification logic of step 2 on all radars. It finds that all classification results are completely consistent with the initial results.

[0182] The system determines that the identification results are convergent and reliable, binds the IDs of the five radars to their physical locations (front left corner, front right corner, front center, rear left corner, and rear right corner), and outputs the results to the upper-layer application with a high confidence level. The entire self-identification process is then complete.

[0183] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle-mounted radar position self-identification system, characterized in that, include: Multiple radar sensors (101) are anonymous radars installed at different locations on the vehicle; some of these radars are 4D radars capable of measuring target height information. The vehicle bus interface (102) is used to obtain vehicle motion information from the vehicle bus; The central processing unit (104) includes a data acquisition unit, a hierarchical logic classification unit, and an iterative vehicle motion refinement and classification result confirmation unit. It is used to receive data from the radar sensor (101) and the vehicle bus interface (102), run hierarchical logic classification and iterative vehicle motion refinement and classification result confirmation, and obtain the recognition result. The hierarchical logical classification unit is a logical classifier, which performs the following steps: Step 2a: Longitudinal position classification, based on the statistical differences between the traffic environment observed by the forward radar and the rear radar during the vehicle's stable forward driving, distinguish between the forward radar and the rear radar. Steps 2a-bis: Forward radar subdivision, based on the geometric differences in the detection field of view and target distribution caused by different installation positions, further distinguish between the front center radar installed directly in front of the vehicle and the front corner radar installed on both sides; Step 2b: Lateral position classification. Based on traffic rules, target trajectories of traffic flow, and road boundary features, lateral position classification is performed on the front-angle radar group and the rear-angle radar group respectively. Step 2c: Rotation installation error verification. Based on the height information provided by the 4D radar, detect installation errors such as the radar being rotated 180° along the axis of its emission direction. The iterative vehicle motion refinement and classification result confirmation unit uses the preliminary vehicle motion information obtained by the data acquisition unit to perform a complete hierarchical logical classification to obtain a preliminary position identification hypothesis; it uses the stationary target information detected by radar to calculate the corrected and more accurate vehicle motion state in reverse; it uses the corrected vehicle motion information to re-execute a complete hierarchical logical classification process; it judges whether the result of the secondary classification is consistent with the preliminary hypothesis, and makes decisions and locks based on the results; The iterative vehicle motion refinement and classification result confirmation unit specifically includes the following steps: Step 601, Preliminary Classification: First, using the preliminary vehicle motion information from the vehicle bus, perform a complete hierarchical logical classification to obtain a preliminary, low-confidence position identification hypothesis; Step 602: Refine the hypothetical vehicle motion: Adopt this initial hypothesis; Based on the geometric relationship of the radar being located at the "rear right corner", use the stationary target information detected by the radar to calculate a corrected and more accurate vehicle motion state in reverse. Step 603, Secondary Classification and Verification: Using this more accurate vehicle motion information, re-execute the complete hierarchical logical classification process; Step 604, Decision and Locking: If the result of the secondary classification is consistent with the initial hypothesis, then the hypothesis is proven to be correct, and the location identification result of the radar is locked. If the result of the secondary classification is inconsistent with the initial hypothesis, it means that the initial hypothesis is wrong. The current hypothesis will be discarded, and a new hypothesis will be established based on the new classification result. This process will be repeated until the result converges.

2. The vehicle-mounted radar position self-identification system according to claim 1, characterized in that, It also includes a positioning module (103) for obtaining the vehicle's geographical location information and obtaining high-precision map or similar high-precision map data.

3. The vehicle-mounted radar position self-identification system according to claim 2, characterized in that, The data acquisition unit is used to collect raw point cloud data from each anonymous radar sensor (101), vehicle motion information received by the vehicle bus interface (102), traffic rules of the current driving area, and high-precision map or similar high-precision map data from the positioning module (103).

4. The vehicle-mounted radar position self-identification system according to claim 3, characterized in that, The raw point cloud data of the radar sensor (101) includes the distance, azimuth angle, radial relative velocity of each detected target, and the height information of the 4D radar; the vehicle motion information includes the longitudinal velocity and yaw rate of the vehicle; the high-precision map or near-high-precision map data of the positioning module (103) includes the topological structure information of the road boundary.

5. The vehicle-mounted radar position self-identification system according to claim 1, characterized in that, The specific method for step 2a is as follows: Step (1) Use the vehicle speed information to filter out the time window in which the vehicle is in a stable straight-line forward motion; Step (2) Within this time window, the radial relative velocities of all target points acquired by each radar are statistically analyzed; Step (3) Calculate the statistical mean of the radial velocity distribution of the target; Step (4) Perform logical judgment: For forward-facing radar, most targets in its field of view are either approaching or being overtaken, so its radial velocity distribution will be significantly biased towards negative values. For rearward radar, most targets in its field of view are either moving away or being pushed back, so its radial velocity distribution will be significantly biased towards positive values. Step (5) classifies the radar into "forward group" or "backward group" by setting a threshold.

6. The vehicle-mounted radar position self-identification system according to claim 5, characterized in that, The specific methods for steps 2a-bis are as follows: Step (1) Analyze the spatial distribution characteristics of stationary targets detected by each radar in the "forward group" during stable straight-line movement; Step (2) Perform logical judgment: Front-to-center radar: The target point cloud detected by it, after a certain period of accumulation, will show a relatively symmetrical distribution range in the azimuth angle with 0 degrees as the center. Forward-angle radar: The distribution of the target point cloud detected by it in the azimuth angle is completely asymmetrical; Step (3) subdivide the "forward group" into the "front-center radar group" and the "front-angle radar group".

7. The vehicle-mounted radar position self-identification system according to claim 1, characterized in that, The specific method for step 2b is as follows: Step (1) First, call the traffic rules of the current driving area obtained by the data acquisition unit; Step (2) uses the following features to identify left and right radars: Feature 1: The target trajectory of oncoming traffic; For the front corner radar group: when an oncoming target is tracked by two front corner radars at the same time, the left and right radars are distinguished by comparing the longitudinal position of the target when it disappears from the field of view of each radar. For the rear corner radar group: when an oncoming target passes by the side of the vehicle, the initial generated position of the target is compared to distinguish the left and right rear corner radars; Feature 2: Road boundary features; A radar installed closer to the curb can detect road boundaries more stably and at a closer distance than a radar installed on the other side. Step (3) By integrating the judgment results of one or more of the above features, "left" and "right" are distinguished.

8. The vehicle-mounted radar position self-identification system according to claim 1, characterized in that, Step 2c specifically includes: Step (1) The radar that is installed normally will detect target points in a vertical direction that are concentrated at a 0° elevation angle; Step (2) The radar installed by 180° roll has its coordinate system completely flipped, and has two detectable features: pitch angle reversal and azimuth angle reversal; Step (3) Determine if the radar has a 180° roll installation error.

9. A method for self-identifying the position of a vehicle-mounted radar, utilizing the vehicle-mounted radar position self-identification system according to any one of claims 1 to 8, characterized in that, Includes the following steps: Step 1: Start collecting data after the vehicle is started and moving; Step Two: Hierarchical logical classification, including the following steps: Step 2a: Longitudinal position classification, based on the statistical differences between the traffic environment observed by the forward radar and the rear radar during the vehicle's stable forward driving, distinguish between the forward radar and the rear radar. Steps 2a-bis: Forward radar subdivision, based on the geometric differences in the detection field of view and target distribution caused by different installation positions, further distinguish between the front center radar installed directly in front of the vehicle and the front corner radar installed on both sides; Step 2b: Lateral position classification. Based on traffic rules, target trajectories of traffic flow, and road boundary features, lateral position classification is performed on the front-angle radar group and the rear-angle radar group respectively. Step 2c: Rotation installation error verification. Based on the height information provided by the 4D radar, detect installation errors such as the radar being rotated 180° along the axis of its emission direction. Step 3: Iterative refinement and classification results of the vehicle motion, including the following steps: Using the initial vehicle motion information acquired by the data acquisition unit, a complete hierarchical logical classification is performed to obtain an initial position identification hypothesis; using the stationary target information detected by radar, the corrected and more accurate vehicle motion state is calculated in reverse; using the corrected vehicle motion information, a complete hierarchical logical classification process is re-executed; it is determined whether the result of the secondary classification is consistent with the initial hypothesis, and a decision and lock are made based on the result.

Images