Array gm-apd lidar three-dimensional point cloud reconstruction method and system based on echo waveform decomposition

Abstract

The application discloses a method and system for three-dimensional point cloud reconstruction of array GM-APD laser radar based on echo waveform decomposition, and belongs to the technical field of three-dimensional point cloud reconstruction. The method solves the problems of low reconstruction density, poor uniformity and lack of representation ability for target structure characteristics of the existing three-dimensional point cloud reconstruction method for complex structure target point cloud. The method comprises the following steps: collecting echo waveforms of each pixel in each sampling surface of a target, decomposing the echo waveforms, obtaining a plurality of response distance values, and fitting a plurality of sub-waveforms; calculating distance difference values between the plurality of sub-waveforms, and calculating a rotation angle of the corresponding sampling surface around an x axis according to the distance difference values; calculating a minimum plane fitting error of each sampling surface, and calculating a rotation angle of the corresponding sampling surface around a z axis according to the minimum plane fitting error; calculating three-dimensional coordinates of a plurality of response points in each pixel through rotation and translation change, and reconstructing a three-dimensional point cloud. The application is suitable for automatic driving and vehicle navigation, urban planning and building surveying scenes.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of point cloud three-dimensional reconstruction, and particularly relates to a laser radar three-dimensional point cloud reconstruction technology. BACKGROUND

[0002] The GM-APD laser radar has high optical sensitivity and high time resolution, can quickly obtain the intensity image, depth image and three-dimensional point cloud of the detected target, and has been widely applied to long-distance target detection and identification.

[0003] In the array GM-APD laser radar detection process, since the divergence angle of the laser is fixed, when there are targets at different distances in the scene, the pixel field of view sizes corresponding to the targets at different distances are different. Specifically, the farther the distance of the detected target, the larger the field of view range of the pixel. At the same time, each pixel usually only restores one response distance value of the target in its field of view to represent the detected target surface in the field of view. Therefore, due to these factors, the collected three-dimensional point cloud will present the characteristics of sparsity and uneven density. This means that there will be a lot of incomplete information missing in the three-dimensional point cloud, which affects the characterization ability of the three-dimensional point cloud to the characteristics of the detected target, and further leads to the decline of the recognition ability of the detected target with missing shape or similar shape, and the detection accuracy of the small target at a long distance with less point cloud will be reduced. In order to solve these problems, it is necessary to reconstruct the three-dimensional point cloud to improve its characterization ability to the characteristics of the detected target and improve the accuracy of target detection and identification.

[0004] The key to reconstructing the three-dimensional point cloud is how to accurately calculate the three-dimensional coordinates of the target. The commonly used three-dimensional point cloud reconstruction method is to up-sample the three-dimensional data, add additional points based on the spatial position relationship between the original point clouds, so as to improve the density and accuracy of the three-dimensional point cloud. This kind of method can well deal with simple targets, but the reconstruction effect of point cloud of complex structure targets may not be good. Some deep learning-based methods have also been proposed, such as PU-Net, PU-GAN and PU-GCN, which can effectively learn the spatial structure of point cloud from data and realize up-sampling of three-dimensional point cloud. Although the deep learning method has good effect in dealing with non-smooth three-dimensional point cloud, it needs a large amount of training data and computing resources. Moreover, for unknown targets in actual detection, the reconstruction effect cannot be guaranteed. The waveform received by the laser radar is the superposition of the backscattered waveforms of each sampling surface in the field of view of the laser radar. Through waveform decomposition, the number and parameters of the sub-waveforms in the field of view of the laser radar can be estimated, and then the distance of each sampling surface in the field of view of the laser radar can be obtained. Although some waveform decomposition researches have successfully decomposed the sub-waveforms in the field of view of the laser radar, they cannot further reconstruct the three-dimensional point cloud of the target. SUMMARY

[0005] The application provides an array GM-APD laser radar three-dimensional point cloud reconstruction method and system based on echo waveform decomposition, and aims to solve the problems of low reconstruction density, poor uniformity and lack of characterization ability for target structure characteristics of the existing three-dimensional point cloud reconstruction method for complex structure target point cloud.

[0006] The application provides an array GM-APD laser radar three-dimensional point cloud reconstruction method based on echo waveform decomposition, and the method comprises the following steps:

[0007] S1: collecting echo waveforms of each pixel in each sampling surface of a to-be-detected target, and decomposing the echo waveforms to obtain a plurality of response distance values of each pixel, and fitting a plurality of sub-waveforms;

[0008] S2: calculating distance difference values between the plurality of sub-waveforms on each sampling surface respectively, and calculating a rotation angle of the corresponding sampling surface around an axis according to the distance difference values; x

[0009] S3: calculating a minimum plane fitting error of each sampling surface, and calculating a rotation angle of the corresponding sampling surface around an axis according to the minimum plane fitting error; z

[0010] S4: calculating three-dimensional coordinates of a plurality of response points in each pixel through rotation and translation changes, and reconstructing a three-dimensional point cloud.

[0011] Further, the preferred scheme is provided: the model of the echo waveform is composed of a Gaussian function and a generalized Gaussian function.

[0012] Further, the preferred scheme is provided: in S1, the echo waveform of each pixel is decomposed by using an empirical mode decomposition method, and the method comprises the following steps:

[0013] obtaining peak position of all IMF signals, and the peak position is represented as , wherein i is an IMF signal index, k is an index of the peak position in the IMF signal, i is a peak position of the echo waveform, and is a full width at half maximum of the transmitted signal. ;

[0014] According to and , the is screened, and the peak position of the IMF signal in the range of is reserved. ;

[0015] The peak position with the highest frequency of occurrence is selected as the initial position of the first signal component. ;​​

[0016] The distance difference between the remaining peak position and is calculated, and the peak position with the smallest distance difference and the highest frequency is selected as the initial position of the second signal component ; ;

[0017] Two sub-waveforms are obtained by fitting using the Levenberg-Marquardt algorithm.

[0018] Further, the preferred solution is provided: the S2 includes:

[0019] According to the obtained two sub-waveforms, two distance difference values of the corresponding sampling surfaces are calculated.

[0020] Let the center coordinates of the sampling surface where the i-th pixel is located be , and the sampling surface where the i-th pixel is located is between the two sampling surfaces corresponding to the two sub-waveforms. The coordinates of the two sampling surfaces corresponding to the two sub-waveforms are respectively denoted as and

[0021] ; ;

[0022] After the corresponding sampling surface is rotated around the z-axis, the new coordinates are calculated by the following formula: x

[0023] ,

[0024] ,

[0025] ,

[0026] wherein is the detection range size of the i-th pixel at this distance, which is represented as:

[0027] ,

[0028] The rotation angle of the corresponding sampling surface around the z-axis is calculated from and , denoted as: x

[0029] .

[0030] Further, the preferred solution is provided: the S3 includes:

[0031] The normal vector of the corresponding sampling surface is represented as:​​​​​

[0032] ,

[0033] wherein, and are the rotation matrices around the z-axis and the x-axis respectively;

[0034] The normal vector of the corresponding sampling surface is also expressed as:

[0035] , (P) is an arbitrary point on the plane, wherein: x, y, z

[0036] ;

[0037] Thus, the problem of solving is converted into the problem of minimizing the plane fitting error under the constraint condition that e , that is:

[0038]

[0039] wherein, N is the number of points of the fitting plane.

[0040] Further, the preferred scheme is provided: the nearest neighbor algorithm is used to find the fitting plane of the adjacent points, the number of points of the fitting plane is selected as 8, that is k = 8, and the simulated annealing algorithm is used to obtain the optimal solution of N .

[0041] Further, the preferred scheme is provided: the three-dimensional coordinates are obtained through the conversion formula of the spherical coordinate system and the rectangular coordinate system.

[0042] The application further provides a computer device, including a memory and a processor, the memory stores a computer program, when the processor runs the computer program stored in the memory, the processor executes the three-dimensional point cloud reconstruction method of the array GM-APD laser radar based on echo waveform decomposition according to any one of the above schemes.

[0043] The application further provides a computer readable storage medium, the computer readable storage medium is used for storing a computer program, the computer program executes the steps of the three-dimensional point cloud reconstruction method of the array GM-APD laser radar based on echo waveform decomposition according to any one of the above schemes.

[0044] The application further provides a three-dimensional point cloud reconstruction system of the array GM-APD laser radar based on echo waveform decomposition, the system includes:

[0045] ​​Fitting module: used for collecting echo waveform of each pixel in each sampling surface of the target to be detected, and decomposing to obtain several response distance values of each pixel, and fitting several sub-waveforms;

[0046] First calculation module: used for calculating distance difference between several sub-waveforms on each sampling surface, and calculating rotation angle of corresponding sampling surface around x axis according to the distance difference;

[0047] Second calculation module: used for calculating minimum plane fitting error of each sampling surface, and calculating rotation angle of corresponding sampling surface around z axis according to the minimum plane fitting error;

[0048] Reconstruction module: used for calculating three-dimensional coordinates of multiple response points in each pixel through rotation and translation change, and reconstructing three-dimensional point cloud.

[0049] Compared with the prior art, the advantages of the present application are that:

[0050] The three-dimensional point cloud reconstruction method provided by the present application directly reconstructs three-dimensional point cloud using array GM-APD laser radar echo waveform, the embodiment proposes a waveform decomposition algorithm, extracts multiple response distances from single peak waveform, uses decomposed distance difference, calculates rotation angle by minimizing plane fitting error, can avoid the problem that multiple response distance values of each pixel share the same azimuth angle and polar angle, and the calculated three-dimensional coordinates are approximately the same, and sufficiently improves the density, uniformity and characterization ability of target structure characteristics of three-dimensional point cloud.

[0051] The present application is suitable for automatic driving and vehicle navigation, urban planning and building surveying scenes. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 The flow chart of the array GM-APD laser radar three-dimensional point cloud reconstruction method based on echo waveform decomposition according to the first embodiment of the present application is shown in the figure;

[0054] Figure 2 The flow chart of the waveform decomposition in S1 according to the first embodiment of the present application is shown in the figure;

[0055] Figure 3The original echo waveform in S1 and its EMD decomposition result as described in the second embodiment of the present application;

[0056] Figure 4 The YOZ side view as described in the second embodiment of the present application;

[0057] Figure 5 The experimental simulation model schematic diagram as described in the second embodiment of the present application, wherein (a) is ABRAMS, (b) is Chrysler, (c) is K2, (d) is M60, and (e) is T90;

[0058] Figure 6 The CD and HD curves of the reconstructed target point cloud at different angles and the target model matching as described in the second embodiment of the present application;

[0059] Figure 7 The experimental scene distance image as described in the second embodiment of the present application, wherein (a) is a building, (b) is a building group, (c) is a vehicle, and (d) is a multi-vehicle scene;

[0060] Figure 8 The building target three-dimensional point cloud reconstruction result as described in the second embodiment of the present application, wherein (a) is Original Point Cloud, (b) is Proposed, (c) is Nearest Neighbor Interpolation, and (d) is Cubic interpolation;

[0061] Figure 9 The vehicle target three-dimensional point cloud reconstruction result as described in the second embodiment of the present application, wherein (a) is Original Point Cloud, (b) is Proposed, (c) is Nearest Neighbor Interpolation, and (d) is Cubic interpolation; DETAILED DESCRIPTION

[0062] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0063] It is to be understood that the terminology "including", "comprising", "consisting" and "consisting essentially of" used in the specification and the appended claims, indicate the presence of the stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0064] It is also to be understood that the terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0065] The technical solutions in the embodiments of the present application are clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0066] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may, however, be practiced without the specific details. In other instances, well-known methods, procedures and components have not been described in detail so as not to obscure aspects of the present application.

[0067] Embodiment one:

[0068] Referring to Figure 1 , Figure 2 The present embodiment is described.

[0069] The array GM-APD laser radar three-dimensional point cloud reconstruction method based on echo waveform decomposition comprises:

[0070] S1: collect the echo waveform of each pixel in each sampling surface of the target to be detected, and decompose to obtain a plurality of response distance values of each pixel, and fit a plurality of sub-waveforms; wherein the model of the echo waveform is composed of Gaussian function and generalized Gaussian function;

[0071] Further, the step adopts an empirical mode decomposition method to decompose the echo waveform of each pixel, including:

[0072] Obtain the peak position of all IMF signals, denoted as , wherein i is the IMF signal index, k is the index of the peak position in the i th IMF signal, the peak position of the echo waveform and the full width at half maximum of the transmit signal ;

[0073] According to and , the peak positions of the IMF signals in the range of are retained ;

[0074] Select the peak position with the highest frequency of occurrence as the initial position of the first signal component ;

[0075] Calculate the distance difference between the remaining peak positions and , and select the peak position with the highest frequency of occurrence and the distance difference less than as the initial position of the second signal component ;

[0076] Two sub-waveforms are obtained by fitting using the Levenberg-Marquardt algorithm.

[0077] S2: Calculate the distance difference values between a plurality of sub-waveforms on each sampling surface, and calculate the rotation angle of the corresponding sampling surface around the x-axis according to the distance difference values;

[0078] Further, in this step, two distance difference values of the corresponding sampling surface are calculated according to the two obtained sub-waveforms.

[0079] Let the center coordinates of the sampling surface where the th pixel is located be , and the sampling surface where the th pixel is located is between the two sampling surfaces corresponding to the two sub-waveforms.

[0080] The coordinates of the two sampling surfaces corresponding to the two sub-waveforms are denoted as and ;

[0081] The corresponding sampling surface rotates around x ​After the rotation of the axis, the new coordinates are calculated using the following equations:

[0082] ,

[0083] ,

[0084] ,

[0085] wherein, is the size of the detection range of the i-th pixel at this distance, and is expressed as:

[0086] ,

[0087] The rotation angle of the corresponding sampling surface around the z-axis is calculated from and , and is expressed as: x

[0088] .

[0089] S3: Calculate the minimum plane fitting error of each sampling surface, and calculate the rotation angle of the corresponding sampling surface around the z-axis according to the minimum plane fitting error;

[0090] Further, in this step, the normal vector of the corresponding sampling surface is expressed as:

[0091] ,

[0092] wherein, and are the rotation matrices around the z-axis and the x-axis, respectively;

[0093] The normal vector of the corresponding sampling surface is also expressed as:

[0094] , (P) is an arbitrary point on the plane, wherein: x, y, z

[0095] ,

[0096] Thus, the problem of solving is converted into the problem of minimizing the plane fitting error under the constraint condition that e , that is:

[0097] ,

[0098] wherein, N is the number of points of the fitting plane, and k ​​​​The nearest neighbor algorithm finds the nearest points to fit a plane, and the number of points used to fit the plane is selected to be 8, i.e. N = 8, and an annealing algorithm is used to obtain the optimal solution of .

[0099] S4: The three-dimensional coordinates of multiple response points in each pixel are calculated by rotation and translation changes, and the three-dimensional coordinates are obtained by conversion formulas of spherical coordinate system and rectangular coordinate system, and a three-dimensional point cloud is reconstructed.

[0100] The three-dimensional point cloud reconstruction method described in the embodiment uses array GM-APD laser radar echo waveform to directly reconstruct a three-dimensional point cloud, and the embodiment proposes a waveform decomposition algorithm to extract multiple response distance values from a single peak waveform, and uses the decomposed distance difference to calculate the rotation angle by minimizing the plane fitting error. This method extracts multiple response distance values from the echo waveform of each pixel, and uses these values to reconstruct a three-dimensional point cloud, aiming to improve the density, uniformity and characterization ability of the three-dimensional point cloud to the target structure characteristics.

[0101] Embodiment two:

[0102] The embodiment is described with reference to Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 .

[0103] The embodiment is a further illustration of the array GM-APD laser radar three-dimensional point cloud reconstruction method based on echo waveform decomposition described in embodiment one, and the three-dimensional point cloud reconstruction method described in the embodiment includes:

[0104] S1: The echo waveform of each pixel is decomposed by using the empirical mode decomposition method, and the multiple response distance values of the target in each pixel are determined and the sub-waveform is fitted in combination with the array GM-APD laser radar echo signal characteristics, so as to solve the problem of sparse point cloud caused by the recovery of only one response distance value for each pixel;

[0105] S2: The rotation angle of the target around the x axis in the field of view of the pixel is calculated by calculating the distance difference between different sub-waveforms;

[0106] S3: The rotation angle of the target around the z axis in the field of view of the pixel is estimated by calculating the minimum plane fitting error, so as to solve the problem that the multiple response distance values of each pixel share the same azimuth angle and polar angle, and the calculated three-dimensional coordinates are approximately the same;

[0107] ​​​​​​S4: Calculate the three-dimensional coordinates of multiple response points in each pixel by rotation and translation variation, and reconstruct the three-dimensional point cloud.

[0108] Further, in the embodiment, the coordinate calculation process of the original three-dimensional point cloud is:

[0109] The time-of-flight histogram is constructed by accumulating the timing information of multiple laser pulses as the echo waveform of the array GM-APD lidar. It is generally believed that the position of the highest peak of the echo waveform is the distance between the target and the array GM-APD lidar. Combined with the divergence angle of the laser and the imaging position of each pixel, the three-dimensional coordinates of the target can be obtained, and the three-dimensional point cloud of the target can be generated.

[0110] In the array GM-APD lidar system, the optical axis of the laser is parallel to the y axis of the geodetic coordinate system. The laser divergence angle of the system is , and the array detector size is MxM. The azimuth angle and the polar angle of the pixel with imaging position in the spherical coordinate system can be represented as:

[0111] ,

[0112] ,

[0113] According to the position of the highest peak of the echo waveform, the distance of the target is , and the three-dimensional coordinates of the target in the pixel can be obtained by the conversion formula between the spherical coordinate system and the rectangular coordinate system:

[0114] ,

[0115] ,

[0116] ,

[0117] The array GM-APD lidar can obtain the three-dimensional point cloud of the detected target through the above calculation of each pixel.

[0118] Further, S1 in the embodiment includes:

[0119] To simplify the calculation, it is assumed that the targets in the field of view of a single pixel contain two sampling surfaces with different distances, and the distances of the two surfaces are obtained by decomposing the echo signal, so as to reconstruct the three-dimensional point cloud.

[0120] The echo waveform of each pixel is decomposed by Empirical Mode Decomposition (EMD). EMD is an adaptive signal processing method that can decompose a given signal into several Intrinsic Mode Functions (IMFs) and a residual term. The present embodiment uses all the IMF signals to determine the positions of the two waveforms after decomposition, thereby more accurately obtaining the target distance values.

[0121] For the decomposition result of the original echo waveform of a single pixel, first, the positions of the peaks of all the IMF signals are counted, denoted as . Wherein i is the index of the IMF signal, k is the index of the position of the peak in the i th IMF signal. In combination with the position of the peak of the original echo waveform and the full width at half maximum of the system transmission signal , the is screened. The positions of the peaks of the IMF signals within the range of are retained .

[0122] Then further screening is performed, the frequency of occurrence of each retained peak position is counted, and the position of the peak with the highest frequency of occurrence is selected as the initial position of the first signal component after decomposition . When the distance difference of the surface within the field of view of a single pixel is less than , the echo waveform thereof presents a single pulse distribution, based on this rule, the initial position of the second signal component is determined. The distance difference between the remaining peaks and is calculated, and the position of the peak with the highest frequency of occurrence and a difference less than is selected as the initial position of the second signal component .

[0123] Finally, the parameters of the two waveforms are fitted by using the Levenberg-Marquardt (LM) algorithm, thereby obtaining the two waveforms after decomposition, and determining the two distance values and . The model of the echo waveform adopts a model composed of a Gaussian function and a generalized Gaussian function, and the specific form is shown in the following formula.

[0124] ,

[0125] Wherein, β is an amplitude factor, is the time position corresponding to the peak value, is any positive number, and are the function widths on the left and right sides of the peak value respectively.

[0126] According to the final fitting result and , the two different distances of the target surface in the field of view of the pixel and can be calculated.

[0127] Further, the S2 of the embodiment includes:

[0128] In step S1, the distances of the two sampling surfaces have been determined according to the decomposition results of the echo signals of each pixel. However, when the three-dimensional coordinates are calculated by the original three-dimensional point cloud coordinate calculation formula, the azimuth and polar angle of the two different distances of the sampling surfaces are the same. This is not consistent with the law of laser detection, and the calculated three-dimensional coordinates are too close to the actual structure of the target, and there is a large difference. Therefore, the embodiment assumes that the two sampling surfaces form a target surface with rotation, and the rotation angle of the target surface around the x axis is , and the rotation angle of the target surface around the z axis is . The different azimuth and elevation angles of the two sampling surfaces in each pixel are simulated by rotating around the two axes.

[0129] The target center coordinates of the detection area of the first pixel are . Assuming that the distances obtained in step S1 are and , the sampling surfaces are evenly distributed on both sides of the target center, and the coordinates are respectively and . When the target surface rotates around the axis, the new coordinates can be calculated by the following formula: x

[0130] ,

[0131] ,

[0132] ,

[0133] wherein is the detection range size of the first pixel at this distance, which is represented as:

[0134] , The rotation angle of the target surface around the

[0135] axis can be calculated from and , which is represented as: x .

[0136] ​​​

[0137] Further, the S3 of the embodiment includes:

[0138] By calculating the best fitting plane of the adjacent points, the rotation angle of the target surface around the z axis of the pixel field of view can be determined . The plane defined by the dot product of the normal vector and the point on the plane is represented as:

[0139] ,

[0140] where the normal vector , ( x, y, z ) is any point on the plane. According to the rotation angles of the target surface around the three coordinate axes and the three-dimensional rotation matrix, the normal vector of the surface can also be represented as:

[0141] ,

[0142] where and are the rotation matrices around the z axis and the x axis, respectively. Combining the above formula, we get:

[0143] ,

[0144] Therefore, the problem of solving is converted to the problem of minimizing the plane fitting error under the constraint of e .

[0145] .

[0146] where N is the number of points of the fitting plane. The embodiment uses the k nearest neighbor algorithm to find the fitting plane of the adjacent points, and the number of adjacent points is selected to be 8, i.e. N = 8, and the simulated annealing algorithm is used to search for the optimal to minimize the plane fitting error.

[0147] Further, the S4 of the embodiment includes:

[0148] According to the rotation angles obtained above, the new coordinates of the target after rotation around the three coordinate axes can be calculated. Taking one of the points as an example, the coordinates after rotation are represented as:

[0149] ,

[0150] The final calculated three-dimensional coordinates after translation are:

[0151] .

[0152] The method described in this embodiment will now be described in detail with reference to the accompanying drawings:

[0153] The decomposition result of the original echo waveform of a single pixel is as follows: Figure 3 As shown. The waveform is decomposed into 8 IMFs and one residual term. First, the positions of the peaks of all IMF signals are counted, denoted as... Combined with the position of the peaks in the original echo waveform and the full width at half maximum (FWHM) of the system's transmitted signal ,right Perform filtering. Keep in Position of peaks in IMF signals within the range ,like Figure 3 As shown in the image.

[0154] Then, further filtering is performed, counting the frequency of each retained peak position, and selecting the position of the peak with the highest frequency as the initial position of the first decomposed signal component. .like Figure 3 As shown, peaks at this location are present in IMF1, IMF2, and IMF3. This location has the highest frequency of occurrence, therefore it is identified as [position missing]. When the distance difference between the inner surfaces of a single pixel's field of view is less than τ, its echo waveform exhibits a single-pulse distribution. Based on this pattern, the initial position of the second signal component is determined. The residual peak is then calculated. and The position of the peak with the highest frequency and a distance difference less than τ between the two signals is selected as the initial position of the second signal component. .like Figure 3 As shown, both IMF1 and IMF4 have peaks at this location.

[0155] Assume that the two sampling surfaces form a rotating target surface, and its orbital path is... x The rotation angle of the axis is , around z The rotation angle of the axis is The different azimuth and elevation angles of the two sampling surfaces within each pixel are approximated by rotating the target surface around two axes. x The side view of the axis after rotation in the YOZ direction is as follows: Figure 4 As shown.

[0156] To test the ability of different algorithms to reconstruct 3D point clouds while preserving the structural characteristics of the target, this implementation method conducted simulation experiments using an array GM-APD lidar to detect different types of targets. For example... Figure 5 As shown, this embodiment conducted simulation experiments on five different target models.

[0157] Chamfer Distance (CD) and Hausdorff Distance (HD) are used as evaluation indexes to measure the ability of the reconstructed 3D point cloud to preserve the structural characteristics of the target. Taking the Chrysler model as an example, Figure 6 The CD and HD curves of the reconstructed 3D point cloud obtained by various algorithms at 13 different angles are shown.

[0158] To evaluate the ability of the proposed method to reconstruct 3D point clouds in real-world scenarios, experiments were conducted. The target of the experiment includes buildings mainly composed of planes and vehicles composed of curved surfaces. The distance of a single building is about 1.4 km, the distance of a group of buildings is about 1 km, the distance of a single vehicle target is about 300 m, and the distance of a multi-vehicle target is between 880 m and 980 m. Figure 7 The range image of the four experimental scenarios is shown in FIG. 8. The nearest neighbor interpolation and cubic spline interpolation methods are used for comparative analysis, Figure 8 The 3D point clouds of the buildings and building groups reconstructed by different algorithms are shown.

[0159] Figure 9 The 3D point clouds of the vehicles and multi-vehicle scenarios reconstructed by different algorithms are shown.

[0160] Those skilled in the art can understand that the above description is only a preferred embodiment of the present application, and the features described in the various embodiments and / or claims of the present disclosure can be combined or combined, even if such combination or combination is not explicitly described in the present disclosure. It is not intended to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, and those skilled in the art can modify the technical solutions described in the foregoing embodiments or make equivalent replacements to part of the technical features, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0161] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0162] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

[0163] Those skilled in the art will appreciate that embodiments of the disclosure can be supplied as a method, a system, or a computer program product. Thus, the disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the disclosure can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0164] The disclosure is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the disclosure. It should be understood that each flow and / or block in the flowchart and / or block diagram and a combination of flows and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the flow Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks

[0165] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable data processing apparatus to produce a computer-implemented process such that the instructions executed by the computer or other programmable data processing apparatus provide the steps for implementing the flow Figure 1 one or more flows and / or blocks Figure 1 Figure 1 one or more flows and / or blocks

[0166] Finally, it should be noted that the above embodiments are merely used to illustrate the technical solutions of the present disclosure, but cannot be used to limit the scope of protection of the present disclosure. Although the present disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present disclosure, they can make various changes, modifications or equivalent replacements to the specific embodiments. However, these changes, modifications or equivalent replacements are within the scope of protection of the disclosed claims.

Claims

1. A method for array GM-APD lidar three-dimensional point cloud reconstruction based on echo waveform decomposition, characterized in that, The method comprises: S1: collecting echo waveforms of each pixel in each sampling surface of a target to be detected, decomposing the echo waveforms, obtaining a plurality of response distance values of each pixel, and fitting a plurality of sub-waveforms; in S1, the echo waveforms of each pixel are decomposed by using an empirical mode decomposition method, comprising: The peak position of all IMF signals is obtained and denoted as wherein i is the IMF signal index, k is the peak position index in the i th IMF signal, the peak position of the echo waveform and the full width at half maximum of the transmitted signal ; According to and , the is screened, and the peak position of the IMF signal in the range is reserved ; selecting the position of the peak having the highest occurrence frequency as the initial position of the first signal component ; the distance difference between the remaining peak positions and is selected as the initial position of the second signal component , and the peak position with the highest frequency of occurrence is selected as the initial position of the second signal component ; two sub-waveforms are fitted by using a Levenberg-Marquardt algorithm; S2: Calculate the distance difference between several sub-waveforms on each sampling surface, and obtain the corresponding sampling surface circumference based on the distance difference. x The rotation angle of the shaft, S2 includes: two distance difference values of the corresponding sampling surface are calculated according to the two obtained sub-waveforms; Record No. The center coordinates of the sampling surface where each pixel is located are The first The sampling surface containing each pixel is located between the two sampling surfaces corresponding to the two sub-waveforms; The coordinates of the two sampling surfaces corresponding to the two sub-waveforms are respectively denoted as and ; The corresponding sampling surface rotates around x the axis after which the new coordinates are calculated with the following equations: ， ， ， wherein, and are two different sampled surface-to-lidar distance values extracted from the echo waveform of a single pixel by a waveform decomposition algorithm, is the detection range size of the pixel at this distance, expressed as: , By and The rotation angle of the corresponding sampling surface around the axis is calculated x represented as:​ ； S3: Calculate the minimum plane fitting error for each sampling surface, and calculate the corresponding sampling surface's orbital path based on the minimum plane fitting error. z The rotation angle of the shaft, S3 includes: the normal vector of the corresponding sampling surface is represented as: ， wherein and Rzand Rxare the rotation matrices around the z-axis and x-axis, respectively; the normal vector of the corresponding sampling surface is also represented as: , ( x, y, z Let be any point on the plane, where: ； Thus the problem of solving is converted to the problem of minimizing the plane fitting error under the known e constraints, i.e. wherein, N is the number of points of the fitted plane; S4: calculating three-dimensional coordinates of a plurality of response points in each pixel by rotation and translation change, and reconstructing a three-dimensional point cloud.

2. The echo waveform decomposition based array GM-APD lidar three-dimensional point cloud reconstruction method according to claim 1, characterized in that, The model of the echo waveform is composed of a Gaussian function and a generalized Gaussian function.

3. The echo waveform decomposition based array GM-APD lidar three-dimensional point cloud reconstruction method according to claim 1, characterized in that, Adopting k The neighbor algorithm finds the adjacent points to fit the plane, the number of points fitting the plane is selected as 8, i.e. N =8, and the optimal solution of is obtained by adopting the simulated annealing algorithm.

4. The echo waveform decomposition based array GM-APD lidar three-dimensional point cloud reconstruction method according to claim 1, characterized in that, The three-dimensional coordinates are obtained by a conversion formula of a spherical coordinate system and a rectangular coordinate system.

5. Computer device, characterized in that The system comprises a memory and a processor, the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the three-dimensional point cloud reconstruction method according to any one of claims 1-4.

6. A computer readable storage medium, characterized in that, The computer readable storage medium is used to store a computer program, and the computer program executes the steps of the three-dimensional point cloud reconstruction method according to any one of claims 1-4.

7. A system for 3D point cloud reconstruction of an array GM-APD lidar based on echo waveform decomposition, employing the method for 3D point cloud reconstruction of an array GM-APD lidar based on echo waveform decomposition according to any one of claims 1 to 4, characterized in that, The system comprises: a fitting module for collecting echo waveforms of each pixel in each sampling surface of a target to be detected, decomposing the echo waveforms, obtaining a plurality of response distance values of each pixel, and fitting a plurality of sub-waveforms; The first calculation module is used for calculating distance difference values between a plurality of sub waveforms on each sampling surface respectively, and calculating rotation angle of the corresponding sampling surface around the axis according to the distance difference values. x The first calculation module is used for calculating distance difference values between a plurality of sub waveforms on each sampling surface respectively, and calculating rotation angle of the corresponding sampling surface around the axis according to the distance difference values. a second calculation module for calculating a minimum plane fitting error for each sampling surface, and calculating a rotation angle of the corresponding sampling surface around the axis according to the minimum plane fitting error; and z a third calculation module for calculating a rotation angle of the corresponding sampling surface around the axis according to the minimum plane fitting error. a reconstruction module for calculating three-dimensional coordinates of a plurality of response points in each pixel by rotation and translation change, and reconstructing a three-dimensional point cloud.

Images