Roadside traffic radar installation angle automatic calibration method and system

Abstract

The invention provides a roadside traffic radar installation angle automatic calibration method and system, which are applied to roadside radar installation angle automatic calibration, and the method comprises the steps: obtaining original static point cloud data; performing straight line fitting on the static point cloud by adopting an RANSAC (Random Sample Consensus) algorithm based on dynamic baseline constraint to obtain a static target straight line; verifying the slope and intercept fluctuation of the current straight line with the slope and intercept of each historical straight line through sliding window check, and when each fluctuation is smaller than a corresponding threshold value, passing the verification, and outputting a final straight line; and obtaining a radar installation angle according to the straight slope fitted by the static target. According to the method, the stationary target straight line is fitted through the improved RANSAC algorithm, the stability of the fitted straight line is verified through the sliding window and the historical straight line, the final straight line is output, the radar installation angle is obtained, and the radar deployment efficiency is improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of intelligent transportation technology, and in particular to a method, system, storage medium, and computer program product for automatically calibrating the installation angle of a roadside traffic radar. Background Art

[0002] In intelligent transportation systems, the installation angle of roadside radar is crucial to the accuracy of vehicle detection and tracking. Traditional manual calibration methods require closed roads. According to a 2023 study published in the China Journal of Highway and Transport, calibration of a single device takes an average of approximately 2.5 hours, and the lateral positioning deviation of a vehicle at 200 meters exceeds 5 meters. This results in low efficiency and large errors. While vision-based automated calibration solutions can significantly improve calibration speed and accuracy, they still suffer from issues such as failure in low-light conditions and poor adaptability to complex roads.

[0003] Vision-based automated calibration solutions typically rely on joint camera-radar calibration. The camera captures specific calibration targets in the road scene, such as checkerboard patterns, ArUco codes, or natural features like lane markings and traffic signs. Feature extraction is achieved by extracting key points using computer vision algorithms such as SIFT, ORB, and deep learning keypoint detection. Coordinate mapping is achieved by establishing a transformation relationship between the camera and radar coordinate systems. The radar's mounting angle is optimized using the PnP (Perspective-n-Point) algorithm or iterative nearest ICP registration to achieve angular optimization, such as pitch, yaw, and roll. In low-light conditions at night or in tunnels, the camera struggles to extract stable features. Strong light interference (such as headlight glare) can lead to feature mismatching. Curves, ramps, and non-standard intersections can cause deviations in the estimated pose of the calibration target. Dynamic vehicle interference (such as trucks obstructing the calibration target) can also occur. Vision-based calibration solutions have a high probability of failure in low-light environments, and the calibration failure rate can reach over 30% under complex road geometry. An automated, high-precision calibration solution is urgently needed. Summary of the Invention

[0004] The purpose of the embodiments of the present disclosure is to provide a method, system, storage medium and computer program product for automatic calibration of the installation angle of a roadside traffic radar, thereby solving the aforementioned problems existing in the prior art.

[0005] In order to achieve the above objectives, the technical solutions adopted in the embodiments of the present disclosure are as follows:

[0006] In one aspect, an embodiment of the present disclosure provides a method for automatically calibrating the installation angle of a roadside traffic radar, which is applied to automatically calibrating the installation angle of a roadside radar. The method includes:

[0007] Acquire original stationary point cloud data, wherein the original stationary point cloud data includes: point cloud data of a stationary target on the road;

[0008] The RANSAC algorithm based on dynamic baseline constraints is used to perform straight line fitting on the stationary point cloud to obtain the stationary target line;

[0009] The slope and intercept of the currently fitted stationary target line are checked and verified by a sliding window to see if they fluctuate with the slope and intercept of each historically fitted stationary target line. When each fluctuation is less than the corresponding threshold, the verification passes and the final line is output.

[0010] The radar installation angle is obtained based on the slope of the straight line fitted to the stationary target.

[0011] Optionally, before adopting the RANSAC algorithm based on dynamic baseline constraints to perform straight line fitting on the stationary point cloud, the method further includes: determining that the number of stationary point clouds in a single frame is greater than N0, and if the number of stationary point clouds in a single frame is greater than N0, adopting the RANSAC algorithm based on dynamic baseline constraints to perform straight line fitting on the stationary point cloud in the single frame;

[0012] If the number of static point clouds in a single frame is less than or equal to N0, multiple frames of static point cloud data are accumulated until the number of point clouds is greater than N0, and then the RANSAC algorithm based on dynamic baseline constraints is used to perform linear fitting on the static point clouds accumulated in multiple frames.

[0013] Optionally, the RANSAC algorithm based on dynamic baseline constraints is used to perform straight line fitting on the stationary point cloud to obtain a stationary target line, including:

[0014] Using dynamic baseline sampling, any two point pairs with a distance greater than the preset distance M are selected from the static point cloud as the line sample point set;

[0015] Calculate the straight line model based on the straight line sample point set;

[0016] Calculate the distance from each data point in the original point cloud data except the line sample point set to the line. If the distance from the point to the line is less than the distance error threshold, the corresponding point is an inline point.

[0017] Record the number or proportion of inliers in the current straight line model;

[0018] After multiple iterations, multiple candidate straight line models are generated;

[0019] Select the straight line model with the largest number of inliers or the largest proportion of inliers as the optimal straight line for this fitting;

[0020] According to the interior point set of the optimal straight line model, the straight line parameters are re-fitted by the least squares method.

[0021] Optionally, the sliding window is used to check and verify whether the slope and intercept of the currently fitted stationary target straight line fluctuate with the slope and intercept of each historically fitted stationary target straight line. When each fluctuation is less than a corresponding threshold, the verification is passed and the final straight line is output, including:

[0022] Maintain a sliding window of length N to store the most recent N straight-line fitting results. Each fitting result includes the slope and intercept.

[0023] Calculate the slope error ratio and intercept error ratio between the current line and each historical line in the sliding window, or the slope error and intercept error;

[0024] If the slope error ratios of the current line and each line in the sliding window are all less than the slope ratio threshold, and the intercept error ratios are all less than the intercept ratio threshold, or if the slope error is less than the slope threshold and the intercept error is less than the intercept threshold, then the current straight line fitting result is determined to be stable and the final straight line equation is output.

[0025] Optionally, obtaining the radar installation angle according to the slope of the straight line fitted to the stationary target includes:

[0026] Calculate the original angle θ based on the slope kˊ of the target fitting line line =arctan(kˊ)*180 / π, inversely calculate the actual installation angle; where kˊ is the slope of the line and arctan(kˊ) is the returned radian value;

[0027] If the slope of the fitted line is less than 0, the radar is facing the direction of the car, the road is on the right side of the radar, and the radar installation angle θ install The calculation method is θ install =θ line +90°;

[0028] If the slope of the fitted line is > 0, the radar is facing the direction of the car, the road is on the left side of the radar, and the radar installation angle θ install The calculation method is θ install =θ line -90°.

[0029] Optionally, after obtaining the radar installation angle, the method further includes: calculating the slope kˊ of the radar line and the preset slope k preset The slope difference of the road is determined according to the slope difference, road width and length, and the installation deflection compensation coefficient λ is determined, λ=f(Δθ, W, L), and the radar installation angle is corrected; when the radar installation angle θ install =θ line When ±90°, the radar installation angle correction formula is θ install =θ line ±90°+f(Δθ,W,L)μ, where, θ lineis the angle corresponding to the slope of the fitting line, μ is the proportional constant, λ is the installation deflection compensation coefficient, Δθ is the installation angle error, W is the road width, and L is the calibration distance.

[0030] Optionally, determining the installation deflection compensation coefficient λ according to the slope difference, road width, and road length includes:

[0031] According to the slope kˊ of the radar line and the preset slope k preset The slope difference Δk is used to calculate the installation angle error Δθ=arctan(kˊ)-arctan(k preset ), and then determine the installation deflection compensation coefficient λ through λ=f(Δθ, W, L)=α·Δθ+β·W / L+γ, where α, β, and γ represent parameters.

[0032] Another aspect of the present disclosure provides a system for automatically calibrating the installation angle of a roadside traffic radar, the system comprising:

[0033] An acquisition module is used to acquire original stationary point cloud data; wherein the original stationary point cloud data includes: point cloud data of stationary targets on the road;

[0034] The straight line fitting module is used to fit the static point cloud using the RANSAC algorithm based on dynamic baseline constraints to obtain the static target straight line;

[0035] The verification module is used to check and verify the fluctuation of the slope and intercept of the currently fitted stationary target line with the slope and intercept of each historically fitted stationary target line through a sliding window. When each fluctuation is less than the corresponding threshold, the verification is passed and the final line is output;

[0036] The determination module is used to obtain the radar installation angle according to the slope of the straight line fitted by the stationary target.

[0037] Another aspect of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-described method when executed by a processor.

[0038] Another aspect of the present disclosure provides a computer program product, including a computer program, which implements the steps of the above method when executed by a processor.

[0039] The beneficial effects of the embodiments of the present disclosure are:

[0040] The method of the disclosed embodiment is based on the RANSAC algorithm with dynamic baseline constraints. It performs straight line fitting on the acquired stationary target point cloud, checks the fluctuation of the slope and intercept of the fitted line with the slope and intercept of multiple historical fitted lines through a sliding window, outputs the final line parameters after verification, and automatically calculates the radar installation angle, thereby improving calibration efficiency and accuracy, avoiding the inefficiency of manual calibration, and allowing calibration at any time without being restricted by weather, thereby improving radar deployment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a method for automatically calibrating the installation angle of a roadside traffic radar proposed in an embodiment of the present disclosure;

[0042] Figure 2 This is a schematic diagram of the structure of an automatic calibration system for the installation angle of a roadside traffic radar proposed in an embodiment of the present disclosure;

[0043] Figure 3 This is a schematic diagram of a process for fitting a straight line using a RANSAC algorithm with dynamic baseline constraints in an automatic calibration method for the installation angle of a roadside traffic radar proposed in an embodiment of the present disclosure;

[0044] Figure 4 This is a schematic diagram of the results of stationary target detection in a method for automatic calibration of the installation angle of a roadside traffic radar proposed in an embodiment of the present disclosure;

[0045] Figure 5 This is a schematic diagram of the structure of a roadside traffic radar after installation in a method for automatically calibrating the installation angle of the radar proposed in an embodiment of the present disclosure;

[0046] Figure 6 This is a structural diagram of radar data processing in a roadside traffic radar installation angle automatic calibration system proposed in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure more clear, the embodiments of the present disclosure are further described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the embodiments of the present disclosure and are not used to limit the embodiments of the present disclosure. The embodiments of the present disclosure relate to a method and system for automatically calibrating the installation angle of a roadside millimeter-wave radar based on radar barrier detection data, which is suitable for the automated calibration and rapid deployment of intelligent transportation infrastructure.

[0048] Example 1, as Figure 1 As shown, an embodiment of the present disclosure provides a method for automatically calibrating the installation angle of a roadside traffic radar, which is applied to automatically calibrating the installation angle of a roadside radar. The method includes:

[0049] Step S100: Acquire original stationary point cloud data, wherein the original stationary point cloud data includes: point cloud data of stationary targets on the road.

[0050] Before the radar acquires point cloud data, the radar to be calibrated is fixedly installed at the corresponding position on the roadside, such as Figure 5 As shown, the radar is mounted horizontally at a certain angle, facing the road. The height is set between 1.5m and 9m. The optimal horizontal mounting angle is between 10° and 45°. A too small angle will result in blind spots near the road, while a too large angle will result in blind spots far away, affecting subsequent target detection. After the radar to be calibrated is installed and powered on, it continuously collects point cloud data of the surrounding environment. This point cloud data includes point cloud data of stationary targets on the road. Stationary targets can be fences in the middle of the road or roadside fences. Stationary targets are straight or nearly straight lines. Point cloud data for stationary targets is collected from stationary objects on the road within approximately 50 meters of the radar normal. High-speed or expressway roads generally have relatively small curvatures over short distances, meaning the curvature radius is typically large. This is suitable for roads with a curvature radius ≥ R, which is close to a straight line. For curved roads, the number of radars is generally increased, and the deployment is more dense to ensure lane coverage. R is set based on measured data and can be 1000 meters.

[0051] The radar installation angle automatic calibration method of the disclosed embodiments is applicable to millimeter-wave radars. While laser radars generally lack Doppler, the method of the disclosed embodiments can be used to calibrate the radar installation angle when the point cloud collected by special laser radars or other radars contains information such as Doppler velocity, which is sufficient to distinguish between moving and stationary point clouds. The acquired raw radar point cloud data undergoes preprocessing, filtering, ghosting removal, and data accumulation. For millimeter-wave radars that simultaneously collect moving and stationary point clouds, the moving and stationary point clouds must be separated from the collected raw point cloud data. The extracted stationary point cloud is then used to track moving targets. A modified RANSAC (random sampling consensus) algorithm can be applied to the preprocessed stationary points for line fitting, line stability verification, and angle calculation. The stationary point cloud includes stationary target point cloud data, which can include road fence point cloud data. Each radar point cloud includes attribute information such as range, azimuth, Doppler velocity, signal-to-noise ratio (SNR), and radar cross section (RCS). The method of distinguishing between moving and static point clouds is as follows: moving point clouds are points with a doppler value other than 0, and static point clouds are points with a doppler value of 0. In other words, static point clouds with a doppler value of 0 are extracted.

[0052] After acquiring the static point cloud data, determine whether the number of point clouds is greater than the preset threshold N0. If the number of static point clouds in the current frame is greater than N0, perform linear fitting. Otherwise, skip linear fitting and end the operation. Alternatively, accumulate multiple frames of point cloud data until the number of static point clouds is greater than N0. If the number of point clouds is too small, no linear fitting is performed.

[0053] In step S100 of the embodiment of the present disclosure, point cloud data can be acquired frame by frame. When the number of point clouds is greater than a preset threshold value N0, the following steps are performed on the corresponding single-frame static point cloud.

[0054] Step S200: Using the RANSAC algorithm with dynamic baseline constraints, a straight line is fitted to the stationary point cloud to obtain a stationary target straight line.

[0055] Step S200 of the embodiment of the present disclosure adopts an improved RANSAC algorithm. Specifically, through dynamic baseline sampling, optimal line selection and least squares optimization, robust straight line fitting of fence points is achieved, and abnormal lines with too large or too small angles are eliminated.

[0056] like Figure 3 As shown in the figure, for some radars, if the number of stationary point clouds in the current frame is greater than N0, a straight line fitting is performed on the stationary point cloud data. If the number of stationary point clouds in the current frame is less than or equal to N0, the straight line fitting is skipped and the operation is terminated.

[0057] like Figure 3 As shown, the RANSAC algorithm with dynamic baseline constraints is used to perform straight line fitting on the stationary point cloud to obtain a stationary target line, including:

[0058] Step S210 : Dynamic baseline sampling is used to randomly select point pairs with a distance greater than a preset distance M from the static point cloud multiple times as line sample point sets for generating candidate line models.

[0059] It should be noted that the dynamic baseline sampling is to preferentially select any adjacent point pairs with a spacing greater than a preset distance M from the static point cloud of the current frame as the static target straight line fitting data, so as to improve the fitting reliability in complex scenes and suppress local noise interference. Among them, M is greater than 3m, and M can be set to 5m, 10m, etc. M is set according to the actual situation, and the straight line fitting uses static point cloud data. Dynamic baseline sampling optimizes the point selection strategy for straight line fitting, excludes point pairs that are too close, and avoids small-sized objects from generating erroneous straight lines, which not only reduces unnecessary calculations but also increases the accuracy of the fitting results. The dynamic baseline constraint limits the sampling point spacing by presetting the distance threshold M (M≥5 meters), eliminating erroneous straight line fitting caused by close-range noise points (such as local deformation of the fence, isolated obstacles), and improves the fitting robustness of the long strip fence structure compared to the random sampling of the classic RANSAC.

[0060] Step S220: Generate a candidate line model each time based on the line sample point set.

[0061] Step S220 of the disclosed embodiment is a model assumption to generate straight line parameters. The general equation of the straight line is calculated as ax+by+c=0, where a, b, and c are the three coefficients of the straight line equation; the disclosed embodiment converts it into a slope-intercept formula y=kx+b″, where k is the slope, b″ is the intercept, and x refers to the horizontal coordinate of the point cloud, and y is the vertical coordinate. The center of the radar is taken as the origin, the y-axis is the radar detection direction, the vertical coordinate, and x is the horizontal coordinate. One iteration results in a candidate straight line model, and after multiple iterations, multiple candidate straight line models can be obtained.

[0062] Step S230: Calculate the distance d between each data point j and the line except the line sample point set in the original point cloud data. j , if d j < distance error threshold∈, then point j is determined to be an interior point of the line.

[0063] In step S230 of the embodiment of the present disclosure, it is determined whether the remaining data points j fall near the current straight line according to the distance error threshold ∈. If the distance d from point j to the straight line is j <∈, then the point j is determined to be a point inside the current line. The remaining data points are point cloud data in the original static point cloud data except for the line sample point set determined in step S210.

[0064] Calculate the inner point: Calculate the distance d from all other points to the straight line. The general equation ax+by+c=0 can be used to calculate the distance d from point j to the fitted stationary target straight line. j , If d j <∈, the point j is an interior point of the straight line.

[0065] Here, a, b, and c represent the parameters of the line equation. An inlier is a point whose distance from the line is less than a set threshold (e.g., M meters). Because a fence typically corresponds to multiple, relatively continuous point clouds, a sufficient number of points must fall near the line to ensure that the fitted line is most likely the fence line. A line with fewer inliers may have many discrete points fitted into it, resulting in many incorrect lines.

[0066] Step S240: Record the number or ratio of inliers in the line model of the current frame.

[0067] Among them, the inlier ratio = the number of inliers / the number of original stationary point clouds.

[0068] Step S250: After multiple iterations, multiple candidate line models are generated.

[0069] Repeat the above steps S210 to S240 N1 times, such as N1 = 100, and generate different candidate models L1, L2, ..., L each time. N The number of inliers or the ratio of inliers in the corresponding line model is recorded. The number of inliers is determined at each iteration. If the number of inliers in any candidate line is less than K, the line model for that iteration is invalid. If the number of inliers in any candidate line is greater than or equal to K, the line model for that iteration is valid.

[0070] Step S260 , optimal straight line selection, selecting the straight line model with the largest number of inliers or the largest proportion of inliers from the valid straight line models as the optimal straight line for this fitting.

[0071] It should be noted that from multiple valid candidate lines, the model L with the most inliers or the largest inlier ratio is selected. best As the optimal straight line. Optimal Model Selection (OMS) selects the best fitting straight line by evaluating indicators such as the number of inliers.

[0072] Step S270: refit the interior point set corresponding to the optimal straight line model by the least square method to determine the straight line parameters.

[0073] Least Squares Refinement is used to refine the interior point set of the candidate line and improve accuracy.

[0074] The interior point set least squares optimization formula is as follows:

[0075] Let the interior point set be The equation of the line is y = k'x + b', and the goal is to minimize the sum of squared errors. m is the number of inliers.

[0076] Solution

[0077]

[0078] The improved RANSAC algorithm in the disclosed embodiment is a customized optimization solution for the special requirements of fence point cloud fitting, such as long strip structures, angle constraints, etc.

[0079] like Figure 4 As shown, the improved RANSAC algorithm is used to fit the straight line l formed by the fence points.

[0080] Step S300: Check and verify through a sliding window whether the slope and intercept of the current straight line fluctuate with the slope and intercept of each historically fitted static target straight line. When each fluctuation is less than the corresponding threshold, the verification passes and the final straight line is output.

[0081] Step S300 of the disclosed embodiment uses a multi-frame sliding window mechanism to verify the stability of the fitting result. The fitted line is verified by checking the fluctuation of the slope and intercept of the line through the sliding window. When the verification passes, the final line is output and stored in the window.

[0082] Maintain a fitting result queue of length N; the slope change rate threshold is set to P k %, the intercept change rate threshold is set to P b %; N consecutive verifications are completed by triggering calibration.

[0083] Linear stability verification: maintain a sliding window of length N to store the fitting results, check the slope and intercept consistency between the current result and the historical result, and check whether the slope fluctuates for N consecutive times < P k % and intercept N consecutive fluctuations <P b %, it indicates that the result is stable and the final straight line is confirmed.

[0084] Maintain a sliding window of length N to store historical fitting results, check the slope and intercept consistency between the current result and the historical results. When the current result is the straight line fitted by the current frame, assume that the straight line equation of the current frame is y = kˊ*x+bˊ, where kˊ is the slope of the current frame straight line, bˊ is the intercept of the current frame straight line, and the straight line equation of the historical N fitting results is y i =k i *x+b i , where k i is the slope of the straight line in the history frame i, b i is the intercept of the line in the history frame i, i = 1, 2...N, and the sliding window stores the k-th intercept of the corresponding line. i and b i , calculate the slope error ratio e between the current frame and each historical frame ki =|k i -kˊ| / kˊ and the intercept error ratio e between the current frame and the historical frame bi =|b i -bˊ| / bˊ, when e of N historical frames is satisfied ki Both are smaller than P k % and e bi Both are smaller than P b %, it is considered that a stable fence line has been found, and the current frame is determined to be a valid frame and added to the window, otherwise the frame is discarded. k % and P b % is preset, N, P k%, and P b % needs to be adjusted according to the actual data, and generally can be set to 5%, or the slope error and intercept error can be directly restricted. For example, e bi = |b i - bˊ|, e ki = |k i - kˊ|, |e bi | < e b _thr, |e ki | < ek_thr, where b i is the intercept of the i-th historical frame, bˊ is the intercept of the current frame, k i is the slope of the i-th historical frame, kˊ is the slope of the current frame, e bi is the intercept error, e ki is the slope error, eb_thr is the maximum intercept error, and ek_thr is the maximum slope error. The slope of the fence line usually changes slowly, so P k % can be set smaller, such as 1%. Alternatively, when the window is full, the mean value μ k of the slope and μ b of the intercept in the output window are used as the final slope and intercept, and the installation angle θ = arctan(μ k ). Alternatively, when the window is full or the predetermined calibration times or time are reached, the slope and intercept of the latest frame line, or the slope and intercept of any historical line in the window, or the median of the average of the historical line slope and intercept, etc. are used as the final slope and intercept. Through window verification, the line converges and tends to be stable. The improved RANSAC algorithm is adopted to improve the robustness of line fitting. This method is particularly suitable for the point cloud fitting of long strip structures such as fences and walls, and its advantages are as follows: the influence of short-distance noise is reduced through baseline constraints; the semantic rationality is improved by combining angle prior knowledge; the geometric accuracy of the final model is ensured by least squares optimization. Dynamic baseline sampling mechanism: Point pairs with a spacing exceeding M meters are preferentially selected for line fitting to improve the fitting reliability in complex scenarios. Multi-frame sliding window verification: It is confirmed to be valid when the slope fluctuates continuously N times < Pk% and the intercept fluctuates continuously N times < Pb% to ensure the stability of the calibration result.

[0085] Step S400: Obtain the actual installation angle of the radar according to the slope of the stationary target line finally fitted.

[0086] That is to say, the radar installation angle is calculated according to the stable line parameters. Specifically, according to the line fitted to the target, the angle corresponding to the line slope k is obtained, so as to obtain the actual installation angle of the radar. As Figure 5 shown, the installation angle is the included angle between the perpendicular line y-axis of the radar plane and the line where the road is located, that is, the azimuth angle, and the y-axis is the radar detection direction. Calculate the angle θ line=arctan(kˊ)*180 / π; Determine the installation angle (θ according to the angle corresponding to the slope of the straight line kˊ line ±90°), where θ line Right now Figure 4 Middle θ.

[0087] Angle calculation: Calculate the original angle θ based on the slope kˊ of the target fitting line line =arctan(kˊ)*180 / π to calculate the actual installation angle. kˊ is the slope of the line, and arctan(kˊ) is the returned radian value.

[0088] If the slope of the fitting line kˊ<0, that is, θ line <0, the radar is facing the direction of the car, the road is on the right side of the radar, and the radar installation angle is calculated as θ install =θ line +90°.

[0089] If the slope of the fitting line kˊ>0, that is, θ line >0, the radar is facing the direction of the car, the road is on the left side of the radar, and the radar installation angle is calculated as θ install =θ line -90°.

[0090] The angle constraint is the absolute value of the inversely calculated radar installation angle |θ install |<θ max (e.g. 60°), θ install is the radar installation angle, θ max is the set maximum angle. Otherwise, recalibrate and execute step S100 again.

[0091] Step S500: Automatically set the detection area boundary, emergency lane position, and calculate coordinate conversion parameters according to the installation angle.

[0092] like Figure 2 As shown, area configuration: automatically configure the detection area and calculate the coordinate conversion parameters. Automatically set the detection area to improve the accuracy of subsequent vehicle detection. The method of the embodiment of the present disclosure reduces the installation and debugging costs and improves the efficiency of system deployment. Figure 5 As shown in Figure 2, the radar detection results after calibrating the installation angle.

[0093] According to actual tests, the embodiments of the present disclosure have significant advantages over traditional methods:

[0094] index Traditional vision-based methods Methods of the embodiments of the present disclosure Calibration time 150 minutes <1 minute Environmental dependence Sunny / daytime All-weather

[0095] 1. Fully automatic calibration of roadside radar installation angles without manual intervention; 2. Utilizes an improved RANSAC algorithm to improve the robustness of straight-line fitting; 3. Ensures the stability of calibration results through multi-frame verification; 4. Automatically sets detection areas to improve the accuracy of subsequent vehicle detection; 5. Reduces installation and debugging costs and improves system deployment efficiency.

[0096] Table 2 shows the radar detection results after calibrating the radar installation angle

[0097] Serial number Actual installation angle (°) Radar calibration angle (°) Calibration error (°) 1 5 5.3 0.3 2 10 10.2 0.2 3 15 15.3 0.3 4 20 19.7 -0.3 5 25 25.2 0.2 6 30 29.6 -0.4

[0098] The disclosed embodiment is a method for automatically calibrating the installation angle of a roadside radar based on linear fitting of fence points detected by the radar. By fitting the fence point data detected by the radar, the radar installation angle is automatically calculated, thereby improving the calibration efficiency and accuracy.

[0099] like Figure 6 As shown, another embodiment of the present disclosure provides a roadside traffic radar installation angle automatic calibration system, which is applied to the automatic calibration of the roadside radar installation angle. The system includes:

[0100] The acquisition module 100 is used to acquire original stationary point cloud data; wherein the original stationary point cloud data includes: point cloud data of stationary targets on the road.

[0101] The straight line fitting module 200 is used to perform straight line fitting on the static point cloud using the RANSAC algorithm based on dynamic baseline constraints to obtain a static target straight line.

[0102] The verification module 300 is used to check and verify the fluctuation of the slope and intercept of the currently fitted stationary target straight line with the slope and intercept of each historically fitted stationary target straight line through a sliding window. When each fluctuation is less than the corresponding threshold, the verification is passed and the final straight line is output.

[0103] The determination module 400 is configured to obtain the radar installation angle according to the slope of the straight line fitted to the stationary target.

[0104] like Figure 2As shown, the system of the embodiment of the present disclosure also includes: an area configuration module, which is used to automatically set the detection area boundary, the emergency lane position and calculate the coordinate conversion parameters according to the installation angle. Millimeter wave RF front-end module: used to transmit and receive specific millimeter wave signal waveforms and perform analog-to-digital conversion. Radar signal processing module: detects the original signal and outputs the original radar point cloud data, including moving and stationary point cloud data. Radar data processing module: pre-processes the millimeter wave original radar point cloud (filtering, ghost removal, data accumulation, etc.), tracks moving targets, and performs an improved RANSAC (random sampling consistency) algorithm on the stationary points to perform straight line fitting, straight line stability verification and angle solution. The radar data processing module includes: an acquisition module, a pre-processing module, a straight line fitting module, a verification module and a determination module. The pre-processing module pre-processes the millimeter wave original radar point cloud (filtering, ghost removal, data accumulation, etc.). Data output module: sets the detection area and performs coordinate conversion on the data according to the angle calibration results, and outputs target information and traffic event information.

[0105] The system in this disclosed embodiment collects radar point cloud data from road barriers and uses an improved random sampling consensus algorithm to perform straight-line fitting. After verifying the linear stability using a multi-frame sliding window, it calculates the installation angle and ultimately automatically configures the detection area. Compared to traditional methods, this method offers high calibration efficiency and is unaffected by ambient lighting conditions. This disclosed embodiment is suitable for the rapid deployment of roadside sensing devices in intelligent transportation systems.

[0106] Another aspect of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-described method when executed by a processor.

[0107] Another aspect of the present disclosure provides a computer program product, including a computer program, which implements the steps of the above method when executed by a processor.

[0108] In the second embodiment, for some other radars, when the single-frame point cloud data is too small, multiple frames of point cloud data can be superimposed and accumulated according to the actual situation. After obtaining the static point cloud data in step S100 of the first embodiment, the number of single-frame point clouds is determined. If the number of static point clouds in a single frame is less than or equal to N0, multiple frames of static point cloud data can be accumulated until the number of point clouds is greater than N0. Then, the RANSAC algorithm based on dynamic baseline constraints in step S200 of the first embodiment is used to perform straight line fitting on the static point clouds accumulated in multiple frames. And the subsequent steps of the first embodiment are executed. In step S300 of the first embodiment, the corresponding sliding window stores the straight line results of the historical multi-frame static point cloud fitting, which is convenient for verifying that the slope and intercept of the straight line fitted by the current multi-frame point cloud data fluctuate with the slope and intercept of each straight line fitted by the historical multi-frame static point cloud in the sliding window. It will not be described in detail here.

[0109] In the third embodiment, after obtaining the radar installation angle in step S400 of the first or second embodiment, the method further comprises: calculating the slope k′ of the stationary target line scanned by the radar and the preset slope k preset The difference is calculated, and the parameter λ is set according to the difference, road width and length to correct the radar installation angle.

[0110] θ install =θ line ±90°+λμ

[0111] Among them, θ line The angle corresponding to the slope of the fitted line is calculated. μ is the proportional constant (unit conversion or scaling factor), which is calibrated experimentally. λ is the installation deflection compensation factor, which is obtained by looking up the difference between the measured error and the preset slope (pre-calibrated) or using a corresponding fitting formula. The radar installation angle refers to the angle between the radar normal and true north.

[0112] Table 1 records the calibration results under various known combinations of installation angle error Δθ, road width W, and calibration distance L according to the experimental design.

[0113] Δθ(°) W(m) L(m) Measured λ Fitting λ 1.5 3.75 50 0.6 0.62 －2.0 4.50 30 －0.8 －0.79

[0114] Fitting results in λ = f (Δθ, W, L) = α·Δθ+β·W / L+γ

[0115] Among them, α, β, and γ are calibration constants obtained through fitting in previous experiments. W / L reflects the impact of the road width-to-length ratio on compensation. Applicable scenarios: small range error |Δθ| < 5° and regular road geometry. Preset slope k preset It can be obtained through high-precision maps or manual measurement.

[0116] Preset slope k preset The a priori slope of the radar installation angle scanning stationary target should be estimated. preset =0, Δk=k′-k preset , Δθ≈arctan(Δk)≈Δk, Δθ≈arctan(k′)≈k′ (small angle approximation)

[0117] At this time, Δk directly reflects the angle error and can be used to calculate λ.

[0118] If k preset ≠0, Δθ=arctan(k′)-arctan(k preset )

[0119] The slope difference must first be converted to an angle difference Δθ through inverse tangent conversion, and then substituted into λ = f(Δθ, W, L) to obtain:

[0120] θinstall =θ line ±90°+f(Δθ, W, L)·μ.

[0121] Among them, Δθ is the angular deviation between the measured stationary target line and the preset line, which directly reflects the radar installation angle error. install Is the absolute value of the radar installation angle within a reasonable range? install |<θ max , if yes, then output; otherwise recalibrate.

[0122] The above is only a preferred implementation of the embodiment of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the embodiment of the present disclosure. These improvements and modifications should also be considered within the scope of protection of the embodiment of the present disclosure.

Claims

1. A method for automatically calibrating the installation angle of a roadside traffic radar, which is applied to the automatic calibration of the installation angle of a roadside radar, and is characterized in that: The method comprises: Acquire original stationary point cloud data; wherein the original stationary point cloud data includes: point cloud data of stationary targets on the road; The RANSAC algorithm based on dynamic baseline constraints is used to perform straight line fitting on the stationary point cloud to obtain the stationary target line; The slope and intercept of the currently fitted stationary target line are checked and verified by a sliding window to see if they fluctuate with the slope and intercept of each historically fitted stationary target line. When each fluctuation is less than the corresponding threshold, the verification passes and the final line is output. The radar installation angle is obtained based on the slope of the straight line fitted to the stationary target.

2. The method according to claim 1, characterized in that Before performing straight line fitting on the stationary point cloud using the RANSAC algorithm based on dynamic baseline constraints, the method further includes: Determine whether the number of stationary point clouds in a single frame is greater than N0. If the number of stationary point clouds in a single frame is greater than N0, use the RANSAC algorithm based on dynamic baseline constraints to perform straight line fitting on the stationary point clouds in the single frame. If the number of static point clouds in a single frame is less than or equal to N0, multiple frames of static point cloud data are accumulated until the number of point clouds is greater than N0, and then the RANSAC algorithm based on dynamic baseline constraints is used to perform linear fitting on the static point clouds accumulated in multiple frames.

3. The method according to claim 1 or 2, characterized in that The RANSAC algorithm based on dynamic baseline constraints is used to perform straight line fitting on the stationary point cloud to obtain the stationary target line, including: Using dynamic baseline sampling, any two point pairs with a distance greater than the preset distance M are selected from the static point cloud as the line sample point set; Calculate the straight line model based on the straight line sample point set; Calculate the distance from each data point in the original point cloud data except the line sample point set to the line. If the distance from the point to the line is less than the distance error threshold, the corresponding point is an inline point. Record the number or proportion of inliers in the current straight line model; After multiple iterations, multiple candidate straight line models are generated; Select the straight line model with the largest number of inliers or the largest proportion of inliers as the optimal straight line for this fitting; According to the interior point set of the optimal straight line model, the straight line parameters are re-fitted by the least squares method.

4. The method according to any one of claims 1 to 3, characterized in that The sliding window is used to check and verify that the slope and intercept of the currently fitted stationary target straight line fluctuate with the slope and intercept of each historically fitted stationary target straight line. When each fluctuation is less than the corresponding threshold, the verification is passed and the final straight line is output, including: Maintain a sliding window of length N to store the most recent N straight-line fitting results. Each fitting result includes the slope and intercept. Calculate the slope error ratio and intercept error ratio of the current line and each historical line in the sliding window, or the slope error and intercept error; If the slope error ratios of the current line and each line in the sliding window are all less than the slope ratio threshold, and the intercept error ratios are all less than the intercept ratio threshold, or if the slope error is less than the slope threshold and the intercept error is less than the intercept threshold, then the current straight line fitting result is determined to be stable and the final straight line equation is output.

5. The method according to any one of claims 1 to 3, characterized in that Obtaining the radar installation angle according to the slope of the straight line fitted to the stationary target includes: Calculate the original angle θ based on the slope kˊ of the target fitting line line =arctan(kˊ)*180 / π, inversely calculate the actual installation angle; where kˊ is the slope of the line and arctan(kˊ) is the returned radian value; If the slope of the fitted line is less than 0, the radar is facing the direction of the car, the road is on the right side of the radar, and the radar installation angle θ install The calculation method is θ install =θ line +90°; If the slope of the fitted line is > 0, the radar is facing the direction of the car, the road is on the left side of the radar, and the radar installation angle θ install The calculation method is θ install =θ line -90°.

6. The method according to claim 5, characterized in that After obtaining the radar installation angle, the method further includes: calculating the slope kˊ of the radar line and the preset slope k preset The slope difference of the road is used to determine the installation deflection compensation coefficient λ according to the slope difference, road width and length, λ = f(Δθ, W, L), and correct the radar installation angle; When the radar is installed at an angle θ install =θ line When ±90°, the radar installation angle correction formula is θ install =θ line ±90°+f(Δθ,W,L)μ; where, θ line is the angle corresponding to the slope of the fitting line, μ is the proportional constant, Δθ is the installation angle error, W is the road width, and L is the calibration distance.

7. The method according to claim 6, characterized in that The step of determining the installation deflection compensation coefficient λ according to the slope difference, the road width, and the road length includes: According to the slope kˊ of the radar line and the preset slope k preset The slope difference Δk is used to calculate the installation angle error Δθ=arctan(kˊ)-arctan(k preset ), and then determine the installation deflection compensation coefficient λ through λ=f(Δθ, W, L)=α·Δθ+β·W / L+γ, where α, β, and γ represent parameters.

8. A roadside traffic radar installation angle automatic calibration system, used for roadside radar installation angle automatic calibration, characterized in that: The system comprises: An acquisition module is used to acquire original stationary point cloud data; wherein the original stationary point cloud data includes: point cloud data of stationary targets on the road; The straight line fitting module is used to fit the static point cloud using the RANSAC algorithm based on dynamic baseline constraints to obtain the static target straight line; The verification module is used to check and verify the fluctuation of the slope and intercept of the currently fitted stationary target line with the slope and intercept of each historically fitted stationary target line through a sliding window. When each fluctuation is less than the corresponding threshold, the verification is passed and the final line is output; The determination module is used to obtain the radar installation angle according to the slope of the straight line fitted by the stationary target.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images