Laser radar target edge feature point detection method based on waveform decomposition

Abstract

The present invention discloses a laser radar target edge feature point detection method based on waveform decomposition. The edge detection of the target in the scene is realized through the laser radar echo waveform decomposition method, and the 3D feature point position of the target edge is detected based on the edge judgment result. The present invention innovatively introduces laser echo waveform analysis to determine the target contour edge, realizes rapid detection of the target edge, and thus detects the 3D feature point position of the target. The method does not require full-target three-dimensional imaging detection, but only requires target edge waveform perception. The process is simple, the data volume is small, the calculation speed is fast, and the target edge detection accuracy is high. By introducing threshold adaptive judgment, the adaptive detection of the feature point position of the laser radar calibration plate is realized, the edge detection accuracy is improved, and the detection process of edge feature points is simplified.

Description

Technical Field

[0001] The present invention relates to the fields of laser radar signal processing, calculation and vision, and in particular to a laser radar target edge feature point detection method based on waveform decomposition. Technical Background

[0002] In recent years, with the development of autonomous driving, robots and other unmanned motion systems, the system's demand for three-dimensional scene information has increased day by day. Traditional visual imaging is a two-dimensional image with rich scene information such as color, texture and shape obtained by visual sensors such as CMOS and CCD. Usually, such sensors cannot directly obtain three-dimensional information of the scene and are easily affected by light. In order to solve this problem, laser radar is introduced to obtain the distance information of the scene, and multi-sensor fusion technology is used to fuse the obtained 3D data with the two-dimensional image data obtained by the visual sensor, so as to give full play to their respective advantages, thereby providing rich color image information of the scene and accurate location point cloud information. Feature point detection of 3D data is a necessary way to realize the fusion of laser radar and other sensors. In the feature point detection of 3D data, the edge detection process of the target object is particularly critical. In traditional laser radar information processing, the information contained in the laser radar echo has not been effectively utilized. The use of laser radar echo information can quickly and effectively determine whether the scanning position is the edge or plane of the object.

[0003] Since LiDAR is not sensitive to plane scanning, and studies have shown that the checkerboard pattern will affect the ranging accuracy of LiDAR and the final calibration results. Therefore, in order to minimize the error caused by the change in reflectivity, a white calibration board with a narrow black border on the edge is used (Park Y, Yun S, Won CS, et al. Calibration between color camera and 3D LIDAR instruments with a polygonal planar board [J]. Sensors, 2014, 14 (3): 5333-5353.). In the 3D edge detection process of the calibration board, it is generally necessary to first perform point cloud segmentation of the calibration target, and then perform feature point detection based on the point cloud segmentation results. In this type of method, the point cloud segmentation operation increases the complexity of the calibration work, and the segmentation results are usually not accurate, which will affect the accuracy of the final calibration. Summary of the invention

[0004] The purpose of the present invention is to propose a laser radar target edge feature point detection method based on waveform decomposition to simplify the detection workflow and improve the edge feature point detection accuracy.

[0005] The object of the present invention is achieved by at least one of the following technical solutions: A laser radar target edge feature point detection method based on waveform decomposition, comprising the following steps:

[0006] S1. Place a calibration plate at a fixed distance from the laser radar detection device, use the laser radar detection device to collect data on the scene, obtain the original echo waveform data of the scene and save it according to the position information;

[0007] S2, using a waveform denoising algorithm to denoise the original echo waveform data collected by the laser radar detection device to obtain filtered data;

[0008] S3, processing the denoised data by a laser radar waveform decomposition method based on exponential fitting, taking the position where double echoes exist in the waveform decomposition and the echo amplitude difference is within a certain threshold as the target edge position and saving the target edge position;

[0009] S4. According to the saved target edge position, a random sampling consistency algorithm is used to solve the straight line where the target edge is located, and the coordinate value of the intersection point of the target calibration plate edge straight line is calculated, that is, the coordinate value of the feature point in the laser radar coordinate system.

[0010] Furthermore, the calibration plate used for target imaging in step S1 is a white square plane plate with a known side length, and a black border is covered on the edge of the plane plate. The four vertices of the calibration plate are used as feature points of the target, and the calibration plate is placed vertically in front of the full-waveform lidar; the lidar detection device with echo collection function detects the scene according to a certain step size and saves the echo waveform information of the corresponding position.

[0011] Further, in step S1, the position information of the original echo waveform data W of the target scene is saved in the form of a spherical coordinate system (r, ω, θ) based on the laser radar detection device, where r is the laser radar ranging distance, ω is the zenith angle, and θ is the azimuth angle;

[0012] The laser radar detection device with echo collection function scans the scene at a known fixed angle interval, and the zenith angle and azimuth angle of the nth scanning position are calculated as follows:

[0013] ω n =nS c ,θ n =nS r

[0014] In the formula, S c and S r are the scanning step intervals in the vertical and horizontal directions respectively.

[0015] Furthermore, in step S2, the laser radar echo waveform denoising algorithm adopts a Savitzky-Golay filter of order k and window length n, where k and n are both natural numbers.

[0016] Furthermore, the specific method of analyzing the denoised data by the LiDAR waveform decomposition method based on exponential fitting is as follows:

[0017] The waveform preprocessing parameters θ = (N, c, μ) are determined by the L1 regularized deconvolution method;

[0018] The waveform decomposition process at the i-th position is as follows:

[0019]

[0020]

[0021] Where N represents the number of decomposed echoes, c = {c1,...,c N} represents the amplitude of the N decomposed echoes, μ={μ1,...,μ N} represents the peak time of the decomposed N echoes, W fi represents the echo waveform data after denoising at the i-th position, c n ,μ n are the amplitude and time of the decomposed nth echo, λ n is the scale factor of the system waveform, S(t) represents the laser radar system waveform fitted by exponential fitting, G k , γ k ,ω k , is a real-valued model parameter, k is the number of components;

[0022] The denoised data is processed by the laser radar waveform decomposition method based on exponential fitting to obtain waveform data composed of N echoes with different amplitudes. The edge of the target is judged based on these N echo waveform data. The filtering conditions are as follows:

[0023]

[0024] The constraints are as follows:

[0025] E={P i |I i <Δ};

[0026] Among them, I i represents the weight of whether the ith position is an edge, c1 and c2 represent the amplitudes of two echoes when the number of decomposed echoes N=2, Δ represents the threshold for judging whether the ith position is an edge, and E represents the set of detected edge positions.

[0027] Furthermore, in step S4, according to the set E of edge positions obtained in step S3, according to the spatial position of the calibration plate, the set E is divided into four categories with the horizontal and vertical coordinates of the center of the calibration plate as thresholds, and four edge position sets E1, E2, E3, and E4 are obtained to fit the target edge straight lines respectively.

[0028] Furthermore, the specific steps of fitting the target edge line using the random sampling consensus algorithm are as follows:

[0029] (1) Select two points from the set and fit a spatial line;

[0030] (2) Calculate the distance d from other points to the line. By setting the threshold Td, all points with d ≤ Td become inliers, and count the number of inliers.

[0031] (3) Repeat the above steps and select the result with the largest number of internal points as the optimal result;

[0032] Through the fitted four edge straight lines, the intersection of the four edge straight lines of the target calibration plate is calculated, and the intersection position is the position of the target edge feature point.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects: the present invention introduces a laser radar waveform decomposition method and makes full use of the echo waveform information of the laser radar, so that there is no need to segment the calibration target in the scene, and the position at the edge of the calibration plate can be effectively identified, which effectively reduces the difficulty of target edge detection, improves the accuracy of edge detection, simplifies the workflow of feature point detection, and is a new laser radar target edge feature point detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the target calibration plate design.

[0035] Figure 2 Schematic diagram of a lidar data acquisition system with echo acquisition function.

[0036] Figure 3 The figure is a flowchart of a method for detecting edge feature points of laser radar targets based on waveform decomposition. DETAILED DESCRIPTION

[0037] The specific implementation modes of the present invention are further described below in conjunction with the accompanying drawings, but the implementation modes of the present invention are not limited thereto.

[0038] Example:

[0039] A laser radar target edge feature point detection method based on waveform decomposition, such as Figure 3 As shown, the following steps are included:

[0040] S1. Place a calibration plate at a fixed distance and use a laser radar detection device with echo collection function to collect data on the scene (such as Figure 2 As shown), the data containing the original echo waveform W of the scene is obtained and saved according to the position information:

[0041] In this embodiment, the calibration plate is as follows Figure 1 As shown in the figure, the calibration plate is a square white plane plate with a length of 70 cm. There is a black border with a width of 1 cm on the edge of the plane plate. The four vertices of the calibration plate are used as calibration feature points. The calibration plate is rotated 45 degrees clockwise and placed vertically in front of the lidar system.

[0042] The position information of the scene echo waveform data W containing the target is stored in the form of a spherical coordinate system (r, ω, θ) based on the laser radar detection device, where r is the laser radar ranging distance, ω is the zenith angle, and θ is the azimuth angle;

[0043] The laser radar detection device with echo collection function scans the scene at a known fixed angle interval, and the zenith angle and azimuth angle of the nth scanning position are calculated as follows:

[0044] ω n =nS c ,θ n =nS r

[0045] By converting the position information based on the spherical coordinate system (r, ω, θ) into the rectangular coordinate system (X L ,Y L ,Z L ), the specific conversion is as follows:

[0046] x Ln =rcos(ω n )cos(θ n ),

[0047] y Ln =rsin(ω n ),

[0048] z Ln =rcos(ω n )sin(θ n );

[0049] Among them, S c and S r are the scanning step intervals in the vertical and horizontal directions respectively. The laser radar three-dimensional coordinate system takes the laser radar position as the coordinate origin O L , along the positive forward direction is Z LAxis, along the positive right direction is X L Axis, vertically downward direction is Y L axis;

[0050] S2. Use the Savitzky-Golay time series filtering algorithm to denoise the original echo waveform data W collected by the laser radar three-dimensional imaging system to obtain the filtered data W f ;

[0051] Set the number of single-side points m to be fitted, the order of the polynomial n, and the echo waveform data W of the i-th point cloud to be fitted i , according to the Savitzky-Golay smoothing formula:

[0052]

[0053] in, is the smoothing coefficient, obtained by fitting the polynomial using the least squares method. Represents the data value of the kth point in the echo waveform data sequence of the i-th point cloud.

[0054] In this example, a 4th-order Savitzky-Golay filter with a window length of 31 is used to filter out noise while ensuring that the shape and width of the signal remain unchanged. The denoised signal is W f .

[0055] S3, the denoised data W is decomposed by the laser radar waveform decomposition method based on exponential fitting f After analysis, the position where double echoes exist in the waveform decomposition and the difference in echo amplitude is within a certain threshold is the detected target edge position, and this position information is saved;

[0056] The laser radar waveform decomposition method based on exponential fitting is to use the exponential function to decompose the denoised laser radar echo waveform W. f For analysis, we first determine the waveform preprocessing parameters θ = (N, c, μ) by using the L1 regularized deconvolution method. After the parameters are determined, the waveform decomposition process at the i-th position is as follows:

[0057]

[0058]

[0059] Where N represents the number of decomposed echoes, c = {c1,…,c N} represents the amplitude of the N decomposed echoes, μ={μ1,…,μ N} represents the peak time of the decomposed N echoes, represents the echo waveform data after denoising at the i-th position, c n,μ n are the amplitude and time of the decomposed nth echo, λ n is the scale factor of the system waveform, S(t) represents the laser radar system waveform fitted by exponential fitting, G k , γ k ,ω k , is a real-valued model parameter and k is the number of components

[0060] The data processed by the laser radar waveform decomposition method based on exponential fitting will obtain waveform data composed of N echoes with different amplitudes. The edge of the target is judged based on these N echo waveform data. The filtering conditions are as follows:

[0061]

[0062] The constraints are as follows:

[0063] E={P i |I i <Δ};

[0064] Among them, I i represents the weight of whether the ith position is an edge, c1 and c2 represent the amplitudes of two echoes when the number of decomposed echoes N=2, Δ represents the threshold for judging whether the ith position is an edge, and E represents the set of detected edge positions.

[0065] S4. Using the saved target edge position, the random sampling consensus algorithm (RANSAC) is used to solve the straight line where the target edge is located, and the coordinate value of the intersection point of the target calibration plate edge straight line is calculated, that is, the coordinate value of the feature point in the laser radar coordinate system:

[0066] In this example, according to the set E of edge positions obtained in step S3, the set E is divided into four categories according to the spatial position of the calibration plate and the horizontal and vertical coordinates of the center of the calibration plate as thresholds. Suppose the horizontal and vertical coordinate thresholds are S C , S R , the coordinates of the position to be classified are (x, y):

[0067]

[0068] The four edge position sets E1, E2, E3, and E4 are obtained to fit the target edge straight lines respectively. The straight line fitting method adopts a more robust random sampling consistency algorithm for fitting:

[0069] (1) Select two points from the set and fit a spatial line;

[0070] (2) Calculate the distance d from other points to the line. By setting the threshold Td, all points with d ≤ Td become inliers, and count the number of inliers.

[0071] (3) Repeat the above steps and select the result with the largest number of internal points as the optimal result.

[0072] By fitting the four edge lines, the intersection of the four edge lines of the target calibration plate is calculated. The intersection position is the target edge feature point position: (X 1L ,Y 1L ,Z 1L ), (X 2L ,Y 2L ,Z 2L ), (X 3L ,Y 3L ,Z 3L ) and (X 4L ,Y 4L ,Z 4L ), where 1, 2, 3, and 4 represent the four vertex numbers of the calibration plate.

[0073] The present invention proposes a laser radar target edge feature point detection method based on waveform decomposition from the perspective of laser radar echo waveform analysis, which is a new 3D feature point detection method.

Claims

1. A laser radar target edge feature point detection method based on waveform decomposition, characterized in that: The following steps are involved: S1. Place a calibration plate at a fixed distance from the laser radar detection device, use the laser radar detection device to collect data on the scene, obtain the original echo waveform data of the scene and save it according to the position information; S2, using a waveform denoising algorithm to denoise the original echo waveform data collected by the laser radar detection device to obtain filtered data; S3. Process the denoised data by a laser radar waveform decomposition method based on exponential fitting, and take the position where there are double echoes in the waveform decomposition and the echo amplitude difference is within a certain threshold as the target edge position and save the target edge position; the specific method is: The waveform preprocessing parameters θ = (N, c, μ) are determined by the L1 regularized deconvolution method; The waveform decomposition process at the i-th position is as follows: Where N represents the number of decomposed echoes, c = {c1,…,c N } represents the amplitude of the N decomposed echoes, μ={μ1,...,μ N } represents the peak time of the decomposed N echoes, represents the echo waveform data after denoising at the i-th position, c n ,μ n are the amplitude and time of the decomposed nth echo, λ n is the scale factor of the system waveform, S(t) represents the laser radar system waveform fitted by exponential fitting, G k , γ k ,ω k , is a real-valued model parameter, k is the number of components; The denoised data is processed by the laser radar waveform decomposition method based on exponential fitting to obtain waveform data composed of N echoes with different amplitudes. The edge of the target is judged based on these N echo waveform data. The filtering conditions are as follows: The constraints are as follows: E={P i |I i <D}; Among them, I i represents the weight of whether the ith position is an edge, c1 and c2 represent the amplitudes of two echoes when the number of decomposed echoes N=2, Δ represents the threshold for judging whether the ith position is an edge, and E represents the set of detected edge positions; S4. According to the saved target edge position, a random sampling consistency algorithm is used to solve the straight line where the target edge is located, and the coordinate value of the intersection point of the target calibration plate edge straight line is calculated, that is, the coordinate value of the feature point in the laser radar coordinate system.

2. The laser radar target edge feature point detection method based on waveform decomposition according to claim 1 is characterized in that: The calibration plate used for target imaging in step S1 is a white square plane plate with a known side length, and a black border is covered on the edge of the plane plate. The four vertices of the calibration plate are used as feature points of the target, and the calibration plate is placed vertically in front of the full waveform laser radar; The laser radar detection device with echo collection function detects the scene according to a certain step size and saves the echo waveform information of the corresponding position.

3. The laser radar target edge feature point detection method based on waveform decomposition according to claim 1 is characterized in that: In step S1, the position information of the original echo waveform data W of the target scene is saved in the form of a spherical coordinate system (r, ω, θ) based on the laser radar detection device, where r is the laser radar ranging distance, ω is the zenith angle, and θ is the azimuth angle; The laser radar detection device with echo collection function scans the scene at a known fixed angle interval, and the zenith angle and azimuth angle of the nth scanning position are calculated as follows: ω n =nS c ,θ n =nS r In the formula, S c and S r are the scanning step intervals in the vertical and horizontal directions respectively.

4. The laser radar target edge feature point detection method based on waveform decomposition according to claim 1 is characterized in that: In step S2, the laser radar echo waveform denoising algorithm adopts a Savitzky-Golay filter of order k and window length h, where k and h are both natural numbers.

5. The laser radar target edge feature point detection method based on waveform decomposition according to claim 1 is characterized in that: In step S4, according to the set E of edge positions obtained in step S3, according to the spatial position of the calibration plate, the set E is divided into four categories with the horizontal and vertical coordinates of the center of the calibration plate as thresholds, and four edge position sets E1, E2, E3, and E4 are obtained to fit the target edge lines respectively.

6. The laser radar target edge feature point detection method based on waveform decomposition according to claim 5 is characterized in that: The specific steps of fitting the target edge line using the random sampling consensus algorithm are as follows: (1) Select two points from the set and fit a spatial line; (2) Calculate the distance d from other points to the line. By setting the threshold Td, all points with d ≤ Td become inliers, and count the number of inliers. (3) Repeat the above steps and select the result with the largest number of internal points as the optimal result; Through the fitted four edge straight lines, the intersection of the four edge straight lines of the target calibration plate is calculated, and the intersection position is the position of the target edge feature point.

Images