CN115598614A - Three-dimensional point cloud target detection method and device and storage medium - Google Patents

Three-dimensional point cloud target detection method and device and storage medium Download PDF

Info

Publication number
CN115598614A
CN115598614A CN202211496628.6A CN202211496628A CN115598614A CN 115598614 A CN115598614 A CN 115598614A CN 202211496628 A CN202211496628 A CN 202211496628A CN 115598614 A CN115598614 A CN 115598614A
Authority
CN
China
Prior art keywords
cluster
vector
principal component
coordinate axis
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211496628.6A
Other languages
Chinese (zh)
Inventor
徐刚
裴昊
张慧
郭坤鹏
张燎
严涵
冯友怀
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Nanjing Hawkeye Electronic Technology Co Ltd
Original Assignee
Southeast University
Nanjing Hawkeye Electronic Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University, Nanjing Hawkeye Electronic Technology Co Ltd filed Critical Southeast University
Priority to CN202211496628.6A priority Critical patent/CN115598614A/en
Publication of CN115598614A publication Critical patent/CN115598614A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G01S13/9004—SAR image acquisition techniques
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G01S13/9021—SAR image post-processing techniques

Abstract

The invention discloses a three-dimensional point cloud target detection method, a device and a storage medium, which are used for a synthetic aperture radar, wherein the method comprises the following steps: obtaining a three-dimensional data point set based on the echo signal fed back by the target; removing abnormal points from the three-dimensional data point set to obtain a target data point set, and clustering the target data point set to obtain a plurality of clusters; determining a bounding box coordinate system which takes the cluster center of each cluster as a coordinate origin for each cluster, determining a plurality of target data points which have the maximum distance with each coordinate axis of the bounding box coordinate system, and further determining a bounding box corresponding to each cluster; the position and size of the target are determined based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters. The technical scheme provided by the invention can solve the technical problems that in the prior art, the point cloud imaging density is sparse, the positioning precision is low, the radar detection result is accompanied by a false alarm, and a machine learning classifier cannot be used on data of a single scene.

Description

Three-dimensional point cloud target detection method and device and storage medium
Technical Field
The invention relates to the technical field of radars, in particular to a method and a device for detecting a three-dimensional point cloud target and a storage medium.
Background
The traditional millimeter wave radar transmits frequency modulation continuous waves in each pulse period and mixes echoes at a receiving end. Echo data of a plurality of targets are aliased in finally acquired data. The aliasing data is processed by distance dimension Fast Fourier Transform (FFT) and velocity dimension FFT to obtain a distance-Doppler map (RDM). A Constant False Alarm Rate (CFAR) algorithm is used on a range-doppler plot (RDM) to obtain a 2D millimeter wave point cloud with a distance dimension and a velocity dimension. When the millimeter wave radar has a plurality of receiving antennas, each receiving antenna can obtain a range-doppler plot (RDM). The intensity distribution of the same target on the range-doppler plots (RDM) of different channels is basically consistent, but the phase depends on the angle of the target, so the complex echoes of the target point in each range-doppler plot (RDM) are extracted for beamforming, i.e. angle measurement. According to the arrangement of the millimeter wave radar antenna array, the traditional angle measurement algorithm can measure angles in the azimuth direction and the altitude direction, and finally 4D millimeter wave point cloud containing a distance dimension, an azimuth dimension, an altitude dimension and a time dimension can be obtained. In clustering data points detected by radar, a density clustering algorithm is generally used to cluster the entire imaged scene, thereby separating objects within the scene from the background.
In the prior art, in the process of detecting detection points of a target by a radar, and performing signal analysis and imaging scene clustering on the detection points, at least the following technical problems exist:
1. due to the limited array resolution and other factors, the traditional millimeter wave radar has sparse point cloud imaging density and low positioning precision.
2. Constant false alarm detection (CFAR) detection techniques are difficult to work in low signal-to-noise scenarios, and the detection result must be accompanied by a false alarm.
3. After the traditional millimeter wave radar obtains an echo signal, machine learning classification is needed when a target is detected based on the radar signal, and a machine learning classifier needs to prepare data of a certain scale and cannot be directly used on data of a single scene.
Disclosure of Invention
The invention provides a three-dimensional point cloud target detection method, a three-dimensional point cloud target detection device and a storage medium, and aims to effectively solve the technical problems that in the prior art, point cloud imaging density is sparse, positioning accuracy is low, radar detection results are accompanied by false alarms, and a machine learning classifier cannot be used on data of a single scene.
According to an aspect of the present invention, there is provided a method for detecting a three-dimensional point cloud target, the method comprising:
carrying out three-dimensional imaging processing on echo signals fed back by at least one target aiming at detection signals of the synthetic aperture radar to obtain a three-dimensional data point set under a world coordinate system;
excluding outliers from the three-dimensional data point set based on the box plot to obtain a target data point set, and performing clustering operation on the target data point set to obtain a plurality of clusters;
for each cluster, determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determining a plurality of target data points with maximum distances to coordinate axes of the bounding box coordinate system from the data points of the cluster, and determining a bounding box corresponding to the cluster according to the plurality of target data points;
determining a location and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
Further, the determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method includes:
calculating a covariance matrix of the cluster based on the clustered data points and the cluster center;
performing orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector;
and determining coordinate axis vectors corresponding to the coordinate axes of the bounding box coordinate system based on the first principal component vector and the second principal component vector.
Further, the computing a covariance matrix for the cluster based on the clustered data points and the cluster center comprises:
calculating a variance between each of the clustered data points and the cluster center, and constructing the covariance matrix based on the variances.
Further, the calculating a variance between each of the clustered data points and the cluster center, the constructing the covariance matrix based on the variances comprising:
calculating the variance based on:
Figure 319778DEST_PATH_IMAGE001
，
wherein M is I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Coordinates representing each data point in the set of data points for the cluster;
calculating the covariance matrix based on:
Figure 41746DEST_PATH_IMAGE002
，
wherein, cov I Representing said covariance matrix, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Representing the coordinates of each data point in the clustered set of data points, T representing a matrix transpose.
Further, the performing an orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector includes:
and carrying out orthogonal linear transformation on the covariance matrix, determining the principal component vector with the largest variance as the first principal component vector, and determining the principal component vector with the second largest variance as the second principal component vector.
Further, the determining the coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector;
determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector.
Determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector;
further, the determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining the third coordinate axis vector based on:
Figure 62792DEST_PATH_IMAGE003
，
wherein the content of the first and second substances,
Figure 688070DEST_PATH_IMAGE004
represents the third coordinate axis vector and the third coordinate axis vector,
Figure 769159DEST_PATH_IMAGE005
representing the first principal component vector, and,
Figure 775161DEST_PATH_IMAGE006
representing the second principal component vector;
the determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector comprises:
determining the second coordinate axis vector based on:
Figure 757155DEST_PATH_IMAGE007
，
wherein the content of the first and second substances,
Figure 266633DEST_PATH_IMAGE008
represents the second coordinate axis vector and the second coordinate axis vector,
Figure 285667DEST_PATH_IMAGE009
representing the first principal component vector, and,
Figure 982228DEST_PATH_IMAGE004
representing the third coordinate axis vector;
the determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector comprises:
determining the first coordinate axis vector based on:
Figure 79497DEST_PATH_IMAGE010
，
wherein the content of the first and second substances,
Figure 443482DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 335215DEST_PATH_IMAGE012
representing the second principal component vector, and,
Figure 20536DEST_PATH_IMAGE004
representing the third coordinate axis vector.
Further, the method further comprises:
after determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method, constructing a rotation matrix based on the first coordinate axis vector, the second coordinate axis vector and the third coordinate axis vector, constructing a translation vector based on the rotation matrix and the cluster center, and performing rigid body transformation on data points of the cluster based on the translation vector so that coordinates of the data points of the cluster are converted from coordinates of the world coordinate system to coordinates of the bounding box coordinate system.
Further, the constructing a rotation matrix based on the first, second, and third coordinate axis vectors comprises:
constructing the rotation matrix based on:
Figure 187075DEST_PATH_IMAGE013
,
Figure 874409DEST_PATH_IMAGE008
,
Figure 468201DEST_PATH_IMAGE014
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 873775DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 578425DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 152888DEST_PATH_IMAGE008
represents the second coordinate axis vector and is,
Figure 652003DEST_PATH_IMAGE004
represents the third coordinate axis vector, and T represents a matrix transpose.
Further, the constructing a translation vector based on the rotation matrix and the cluster center comprises:
constructing the translation vector based on:
Figure 544873DEST_PATH_IMAGE016
，
wherein t represents the translation vector,
Figure 53214DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 715140DEST_PATH_IMAGE017
coordinates representing the cluster center.
Further, the rigid-body transforming the clustered data points based on the translation vector comprises:
performing the rigid body transformation based on:
Figure 886620DEST_PATH_IMAGE018
，
wherein the content of the first and second substances,
Figure 266786DEST_PATH_IMAGE019
，
wherein the content of the first and second substances,
Figure 313240DEST_PATH_IMAGE020
coordinates of data points representing the cluster in the bounding box coordinate system;
Figure 95251DEST_PATH_IMAGE021
coordinates of data points representing the cluster in the world coordinate system,
Figure 936168DEST_PATH_IMAGE015
representing the rotation matrix, t representing the translation vector, C I Set of data points, P, representing said cluster i Representing each data point in the set of data points of the cluster in the world coordinate systemAnd (4) coordinates.
Further, the three-dimensional imaging processing of the echo signal fed back by at least one target for the detection signal of the synthetic aperture radar to obtain a three-dimensional data point set in a world coordinate system includes:
and performing direction dimensional imaging and distance dimensional imaging on the echo signals to obtain a two-dimensional data point set, and performing elevation direction imaging on the two-dimensional data point set to obtain a three-dimensional data point set.
Further, the method includes excluding outliers from the three-dimensional data point set based on the box plot to obtain a target data point set, and performing a clustering operation on the target data point set to obtain a plurality of clusters:
and respectively constructing box line graphs of the three-dimensional data point set based on each dimensional coordinate of the three-dimensional coordinates and the signal amplitude, and excluding abnormal points in each box line graph to obtain the target data point set.
Further, the determining the position and the size of the target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters comprises:
and for each cluster, determining coordinate values of the cluster center of the cluster in the world coordinate system as a central position point of a target associated with the cluster, and determining the length, width and height of a bounding box corresponding to the cluster as three-dimensional dimensions of the target associated with the cluster.
According to another aspect of the present invention, the present invention further provides a three-dimensional point cloud target detection apparatus for a synthetic aperture radar, the apparatus comprising:
the three-dimensional data point set generating module is used for carrying out three-dimensional imaging processing on echo signals fed back by at least one target aiming at detection signals of the synthetic aperture radar so as to obtain a three-dimensional data point set under a world coordinate system;
the clustering determination module is used for excluding abnormal points from the three-dimensional data point set based on the box line graph to obtain a target data point set, and carrying out clustering operation on the target data point set to obtain a plurality of clusters;
a bounding box generating module, configured to determine, for each of the clusters, a bounding box coordinate system using a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determine, among data points of the cluster, a plurality of target data points having a maximum distance from respective coordinate axes of the bounding box coordinate system, and determine, according to the plurality of target data points, a bounding box corresponding to the cluster;
a target determination module to determine a location and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
According to another aspect of the present invention, there is also provided a storage medium having stored therein a plurality of instructions adapted to be loaded by a processor to perform any of the three-dimensional point cloud target detection methods described above.
Through one or more of the above embodiments in the present invention, at least the following technical effects can be achieved:
according to the technical scheme, a traditional millimeter wave point cloud imaging method is not used, an imaging mode of a Synthetic Aperture Radar (SAR) is used for reference, a time domain BP algorithm is used for carrying out two-dimensional SAR imaging, and then a high-resolution and dense three-dimensional data point set is obtained by matching with an array three-dimensional imaging algorithm based on compressed sensing. And then, preprocessing the three-dimensional data point set by using box line graph analysis, screening abnormal points in the data point set, and clustering the target data point set by using a clustering method. And finally, calculating a bounding box of each cluster by using a Principal Component Analysis (PCA) method and coordinate Euclidean transformation to estimate geometrical information such as the position and the size of the interested target in the scene. The invention adopts the imaging mode of the Synthetic Aperture Radar (SAR), is not limited by factors such as array resolution and the like, can obtain a dense data point set, has no false alarm in a detection result, and improves the positioning precision. And clustering is carried out after abnormal points are screened out through the box diagram, so that a better clustering effect can be obtained. The cluster is processed into the OBB bounding box, the determined target shape is closer to a real object, the number of bounding bodies can be obviously reduced, and the target detection accuracy is improved.
Drawings
The technical solution and other advantages of the present invention will become apparent from the following detailed description of specific embodiments of the present invention, which is to be read in connection with the accompanying drawings.
Fig. 1 is a flowchart illustrating steps of a method for detecting a three-dimensional point cloud target according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a three-dimensional data point cloud according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a plurality of clusters according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a bounding box corresponding to a single cluster according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a bounding box corresponding to a plurality of clusters according to an embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating an elevation resolution principle according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of a three-dimensional point cloud target detection apparatus according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it should be noted that, unless explicitly stated or limited otherwise, the term "and/or" herein is only one kind of association relationship describing the associated object, and means that there may be three kinds of relationships, for example, a and/or B, and may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in this document generally indicates that the preceding and following related objects are in an "or" relationship, unless otherwise specified.
In a radar system, the principle of detecting a target by a radar is to obtain an accurate distance between the radar and the target by using a time difference between a transmission pulse and a reception pulse and a propagation speed (light speed) of an electromagnetic wave. The principle of measuring the angular position of the target is to use the directivity of the antenna, when the antenna beam is directed at the target, the echo signal is strongest, and the direction of the target can be determined according to the direction of the antenna beam when the received echo is strongest. The principle of measuring speed is the frequency Doppler effect generated by relative motion between the radar and a target. The target echo frequency received by the radar is different from the radar transmitting frequency, and the difference between the target echo frequency and the radar transmitting frequency is called Doppler frequency. One of the main information that can be extracted from the doppler frequency is the rate of change of the distance between the radar and the target, and hence the velocity of movement of the target. Therefore, the information of the movement speed, the movement direction, the distance and the like of the target can be obtained.
Synthetic Aperture Radar (SAR) is a high resolution imaging Radar, which uses a small antenna to move at a constant speed along the track of a long linear array and radiate coherent signals, and performs coherent processing on echoes received at different positions, thereby obtaining an imaging Radar with higher resolution and obtaining a high resolution Radar image similar to optical photography under the meteorological condition with extremely low visibility. The relative motion between the radar and the target is used to synthesize the real antenna aperture with small size into the radar with larger equivalent antenna aperture by data processing method, which is also called synthetic aperture radar. The synthetic aperture radar has the characteristics of high resolution, all-weather operation and effective identification of camouflage and penetration masks. The high azimuthal resolution obtained with synthetic aperture radars is comparable to that provided by a large aperture antenna. As with most other radars, synthetic aperture radars determine range by the time difference between the transmission of an electromagnetic pulse and the reception of a target echo, with resolution related to pulse width or pulse duration, with narrower pulse widths giving higher resolution.
Fig. 1 is a flowchart illustrating steps of a three-dimensional point cloud target detection method according to an embodiment of the present invention, and according to an aspect of the present invention, the present invention provides a three-dimensional point cloud target detection method for a synthetic aperture radar, as shown in fig. 1, the method includes:
step 101: carrying out three-dimensional imaging processing on echo signals fed back by at least one target aiming at detection signals of the synthetic aperture radar to obtain a three-dimensional data point set under a world coordinate system;
step 102: excluding outliers from the three-dimensional data point set based on the box plot to obtain a target data point set, and performing clustering operation on the target data point set to obtain a plurality of clusters;
step 103: for each cluster, determining a bounding box coordinate system which takes a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determining a plurality of target data points which have the maximum distance with each coordinate axis of the bounding box coordinate system from the clustered data points, and determining a bounding box corresponding to the cluster according to the plurality of target data points;
step 104: determining a position and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
The following describes steps 101 to 104 specifically.
In step 101, performing three-dimensional imaging processing on echo signals fed back by at least one target aiming at detection signals of the synthetic aperture radar to obtain a three-dimensional data point set in a world coordinate system;
exemplarily, the invention does not use the traditional millimeter wave point cloud imaging method, but uses the imaging mode of Synthetic Aperture Radar (SAR) for reference, and can obtain denser three-dimensional data point cloud through the SAR, and fig. 2 is a schematic diagram of the three-dimensional data point cloud provided by the embodiment of the invention.
The transmitting antenna of the synthetic aperture radar transmits radar detection signals to a target, the target reflects the detection signals and feeds back echo signals to the receiving antenna of the synthetic aperture radar. The antenna comprises a plurality of transmitting antennas and a plurality of receiving antennas, and one channel corresponds to any one transmitting antenna and any one receiving antenna. One transmitting antenna corresponds to a plurality of receiving antennas, and accordingly, a channel array is formed between one transmitting antenna and the plurality of receiving antennas. Therefore, the radar can receive echo signals of a plurality of channel arrays.
For echo signals of all channels, a time domain Back Propagation (BP) algorithm is used for two-dimensional Synthetic Aperture Radar (SAR) imaging. Then, a three-dimensional SAR point cloud with high resolution and density, namely a three-dimensional data point set, is obtained by matching with an array three-dimensional SAR imaging algorithm based on compressed sensing.
In step 102, excluding outliers from the three-dimensional data point set based on a box plot to obtain a target data point set, and performing a clustering operation on the target data point set to obtain a plurality of clusters;
illustratively, a box plot is used to analyze the distribution of characteristics such as data X, Y, Z coordinates and signal amplitude in a three-dimensional data point set. And limiting the upper and lower limits of X, Y and Z coordinates of the data points and the lower limit of the signal amplitude by using the result of the box plot to screen out abnormal points with characteristic distribution out of the limiting range, and reducing the influence of abnormal scattering points on the clustering result. After abnormal points are screened out through a box plot, a target data point set is obtained based on the remaining data points, then the target data point set is clustered, and all data points are subjected to data processing to obtain a plurality of clusters which are independent of each other, fig. 3 is a schematic diagram of the plurality of clusters provided by the embodiment of the invention, and as shown in fig. 3, the target data point set corresponding to a building of a certain cell is clustered into the plurality of clusters.
The box diagram is also called box whisker diagram, box diagram and box diagram, and is used for reflecting the central position and the dispersion range of one or more groups of continuous quantitative data distribution. The box diagram contains mathematical statistics, and can analyze the level difference of each layer of different types of data and reveal the dispersion degree, abnormal value, distribution difference and the like among the data.
The clustering method can adopt spectral clustering, and the spectral clustering is a clustering algorithm based on a graph theory. The spectral clustering takes the data points as one point in a feature space, the data points are connected in pairs to form edges, the weight of the edges is determined by the feature distance between the points, and the closer the two points are, the higher the weight is. A graph is finally constructed that describes the distribution of the entire data set. The purpose of spectral clustering is to cut the generated graph, so that the edge weight in the cut subgraph is as high as possible and the edge weight between subgraphs is as low as possible. The radar system continuously learns and optimizes the algorithm model, a plurality of cluster quantities are preset in advance according to the model learning result before clustering is carried out, spectral clustering is carried out on the target data point set respectively based on each cluster quantity, and finally the optimal clustering result is obtained.
In step 103, for each of the clusters, a bounding box coordinate system with a cluster center of the cluster as a coordinate origin is determined according to a principal component analysis method, a plurality of target data points having a maximum distance from each coordinate axis of the bounding box coordinate system are determined among the data points of the cluster, and a bounding box corresponding to the cluster is determined according to the plurality of target data points.
Illustratively, after processing by a clustering algorithm, a three-dimensional data point set is clustered into a plurality of clusters, each cluster is taken as a point set, for each cluster, a new coordinate system is established according to data points in the cluster, each cluster corresponds to one coordinate system, and then a suitable OBB bounding box is planned to be found to cover all points in the cluster. And converting the coordinates of the data points from the original world coordinate system into respective bounding box coordinate systems. The target data point is then determined for each coordinate axis in the respective bounding box coordinate system, which is the greatest distance from the coordinate axis. Specifically, the two target data points farthest from the x-axis are determined
Figure 803630DEST_PATH_IMAGE022
、
Figure 143520DEST_PATH_IMAGE023
Determining two target data points farthest from the y-axis
Figure 780038DEST_PATH_IMAGE024
、
Figure 57435DEST_PATH_IMAGE025
Determining two target data points farthest from the z-axis
Figure 881035DEST_PATH_IMAGE026
、
Figure 800449DEST_PATH_IMAGE027
. And then determining a bounding box corresponding to the cluster according to the 6 target data points. Fig. 4 is a schematic diagram of a bounding box corresponding to a single cluster according to an embodiment of the present invention, and as shown in fig. 4, for one cluster, a rectangular bounding box is finally determined according to a target data point. Wherein, the Bounding Box is directed Bounding Box (OBB), is the cuboid that is the most hugged closely object, and this cuboid can be according to the first moment arbitrary rotation of object. The OBB is closer to the object than the surrounding ball and the AABB, and the number of the surrounding bodies can be obviously reduced. Fig. 5 is a schematic diagram of bounding boxes corresponding to multiple clusters according to an embodiment of the present invention, where the multiple clusters in fig. 3 are subjected to data processing to obtain multiple bounding boxes corresponding to fig. 5.
In step 104, a position and a size of the at least one target are determined based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
For example, after all the bounding boxes are determined, the positions and the sizes of one or more targets in the radar detection scene can be determined through the corresponding clustering centers and the sizes of the multiple bounding boxes, and then the shapes of the three-dimensional targets can be determined.
Further, in step 103, the determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method includes:
calculating a covariance matrix of the cluster based on the clustered data points and the cluster center;
performing orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector;
and determining coordinate axis vectors corresponding to the coordinate axes of the bounding box coordinate system based on the first principal component vector and the second principal component vector.
Illustratively, a corresponding cluster Analysis is performed on an interested target data point, a cluster center is obtained as a center coordinate (x, y, z) of the target, and then data processing is performed on the cluster based on a Principal Component Analysis (PCA) and a euclidean transformation to obtain a bounding box. Firstly, determining a Bounding Box coordinate system based on a principal component analysis algorithm, further converting the coordinates of data points from the world coordinate system to the Bounding Box coordinate system, and then determining a Bounding Box of a point set OBB (organized Bounding Box) in a cluster as a three-dimensional shape of a target.
Specifically, three mutually perpendicular axis vectors in the euclidean space, that is, axis vectors of a bounding box coordinate system are calculated according to a principal component analysis method, the clustering center of the cluster is used as the origin of the bounding box coordinate system, and 3 mutually perpendicular axis vectors are found in the euclidean space and used as the X, Y, and Z axes of the bounding box.
The principal component analysis method is a statistical method, a group of variables possibly having correlation are converted into a group of linearly uncorrelated variables through orthogonal transformation, and the converted variables are called principal components. In the present invention, a covariance matrix of the cluster is first calculated based on the clustered data points and the cluster center; then, carrying out orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector; and finally, determining coordinate axis vectors corresponding to all coordinate axes of the bounding box coordinate system based on the first principal component vector and the second principal component vector.
Further, in step 103, the calculating a covariance matrix of the cluster based on the data points of the cluster and the cluster center comprises:
calculating a variance between each of the clustered data points and the cluster center, and constructing the covariance matrix based on the variances.
Further, in step 103, the calculating a variance between each of the clustered data points and the cluster center, the constructing the covariance matrix based on the variances includes:
calculating the variance based on:
Figure 792938DEST_PATH_IMAGE028
，
wherein, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Coordinates representing each data point in the set of data points for the cluster;
calculating the covariance matrix based on:
Figure 975658DEST_PATH_IMAGE029
，
wherein, cov I Represents the covariance matrix, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Coordinates of each data point in the set of data points representing the cluster, and T represents a matrix transpose.
Illustratively, two conditions to be satisfied for dimensionality reduction are implemented by principal component analysis methods: recent reconstructability and maximum separability. Achieving both conditions requires the covariance matrix of the sample points in the low-dimensional space to be maximized, i.e., the covariance matrix of the sample points in the low-dimensional space to be maximized. In the present invention, the variance between each of the clustered data points and the cluster center is first calculated, and then a covariance matrix is constructed based on the variances of all the data points within the cluster and the cluster center.
Further, in step 103, the performing an orthorhombic linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector includes:
and carrying out orthogonal linear transformation on the covariance matrix, determining the principal component vector with the largest variance as the first principal component vector, and determining the principal component vector with the second largest variance as the second principal component vector.
The covariance matrix is illustratively a symmetric matrix, so an orthogonalized linear transformation can be found using principal component analysis methods to transform the raw data into a new coordinate system such that the first largest variance of any projection of the raw data is on the first coordinate (called the first principal component), the second largest variance is on the second coordinate (the second principal component), and so on. The principal component analysis method determines two principal component vectors with the largest variance as a first principal component vector and a second principal component vector from the first three principal components.
Further, in step 103, the determining the coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector;
determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector.
Determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector;
further, in step 103, the determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining the third coordinate axis vector based on:
Figure 817712DEST_PATH_IMAGE003
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 9659DEST_PATH_IMAGE004
represents the third coordinate axis vector and is,
Figure 355190DEST_PATH_IMAGE005
representing the first principal component vector, and,
Figure 239969DEST_PATH_IMAGE006
representing the second principal component vector;
said determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector comprises:
determining the second coordinate axis vector based on:
Figure 805205DEST_PATH_IMAGE007
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 535264DEST_PATH_IMAGE008
represents the second coordinate axis vector and the second coordinate axis vector,
Figure 735301DEST_PATH_IMAGE009
representing the first principal component vector, and,
Figure 790981DEST_PATH_IMAGE004
representing the third coordinate axis vector;
the determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector comprises:
determining the first coordinate axis vector based on:
Figure 607628DEST_PATH_IMAGE010
，
wherein the content of the first and second substances,
Figure 705159DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 556440DEST_PATH_IMAGE012
representing the second principal component vector, and,
Figure 783022DEST_PATH_IMAGE004
representing the third coordinate axis vector.
Illustratively, the first three principal components output by the principal component analysis method are not necessarily perpendicular to each other. Therefore, after the first principal component vector and the second principal component vector are determined, three coordinate axis vectors perpendicular to each other are determined by the two principal component vectors
Figure 555806DEST_PATH_IMAGE011
,
Figure 394711DEST_PATH_IMAGE008
,
Figure 834920DEST_PATH_IMAGE004
A bounding box coordinate system of the cluster is determined based on the three coordinate axis vectors and the cluster center.
Further, the method further comprises:
after the bounding box coordinate system which takes the cluster center of the cluster as a coordinate origin is determined according to the principal component analysis method, constructing a rotation matrix based on the first coordinate axis vector, the second coordinate axis vector and the third coordinate axis vector, constructing a translation vector based on the rotation matrix and the cluster center, and performing rigid body transformation on the clustered data points based on the translation vector, so that the coordinates of the clustered data points are converted from the coordinates of the world coordinate system to the coordinates of the bounding box coordinate system.
Illustratively, the euclidean transformation between coordinates is performed by constructing a rotation matrix and a translation vector, and similarly to the rotation between vectors, the rotation and translation between two coordinates are collectively referred to as a transformation relationship between coordinate systems. An inertial coordinate (world coordinate system) is set, the coordinate value of a data point in the world coordinate system is obtained first, then rotation and translation are carried out to convert the coordinate value into a bounding box coordinate system, and the coordinate system is converted into rigid motion, so that the length and the included angle of the same vector under each coordinate system are ensured not to change.
Further, the constructing a rotation matrix based on the first, second, and third coordinate axis vectors comprises:
constructing the rotation matrix based on:
Figure 701245DEST_PATH_IMAGE013
,
Figure 226904DEST_PATH_IMAGE008
,
Figure 102456DEST_PATH_IMAGE014
，
wherein the content of the first and second substances,
Figure 898636DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 732600DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 276714DEST_PATH_IMAGE008
represents the second coordinate axis vector and is,
Figure 955957DEST_PATH_IMAGE004
represents the third coordinate axis vector, and T represents a matrix transpose.
Further, the constructing a translation vector based on the rotation matrix and the cluster center comprises:
constructing the translation vector based on:
Figure 606643DEST_PATH_IMAGE016
，
wherein t represents the translation vector,
Figure 345929DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 580602DEST_PATH_IMAGE017
coordinates representing the cluster center.
Further, the rigid-body transforming the clustered data points based on the translation vector comprises:
performing the rigid body transformation based on:
Figure 797956DEST_PATH_IMAGE018
，
wherein the content of the first and second substances,
Figure 67264DEST_PATH_IMAGE030
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 478916DEST_PATH_IMAGE020
coordinates of data points representing the cluster in the bounding box coordinate system;
Figure 466463DEST_PATH_IMAGE021
coordinates of data points representing the cluster in the world coordinate system,
Figure 487509DEST_PATH_IMAGE015
representing the rotation matrix, t representing the translation vector, C I Set of data points, P, representing said cluster i Coordinates of each data point in the set of data points representing the cluster in the world coordinate system.
Illustratively, a first coordinate axis vector is used
Figure 611323DEST_PATH_IMAGE011
Second coordinate axis vector
Figure 692411DEST_PATH_IMAGE008
And a third coordinate axis vector
Figure 636096DEST_PATH_IMAGE004
Rotation matrix capable of constructing world coordinate system to bounding box coordinate system
Figure 974016DEST_PATH_IMAGE013
,
Figure 952337DEST_PATH_IMAGE008
,
Figure 204327DEST_PATH_IMAGE014
And a translation vector
Figure 900887DEST_PATH_IMAGE031
. Further, rigid body transformation can be carried out on the clustered data points, and the data points in the clusters are collected
Figure 263735DEST_PATH_IMAGE032
And transforming to a bounding box coordinate system. The coordinate transformation relation of an Euclidean space can be completely described by using a rotation matrix and a translation vector.
Further, in step 101, the three-dimensional imaging processing of the echo signal fed back by at least one target with respect to the detection signal of the synthetic aperture radar to obtain a three-dimensional data point set in a world coordinate system includes:
and performing direction dimension imaging and distance dimension imaging on the echo signals to obtain a two-dimensional data point set, and performing elevation direction imaging on the two-dimensional data point set to obtain a three-dimensional data point set.
Illustratively, two times of imaging processing are carried out on echo signals, and three-dimensional imaging is three-dimensional SAR imaging and can be decomposed into two-dimensional imaging and elevation direction imaging. First focused in both the direction and distance dimensions to generate 2D high resolution images in the azimuth range, and then the phase difference between the complex images is measured using a Digital Elevation Model (DEM), determined from different perspectives in the elevation direction to recover the elevation information.
Fig. 6 is a schematic diagram illustrating an elevation resolution principle according to an embodiment of the present invention, where d is an array element pitch of an array, H is a target height,
Figure 332448DEST_PATH_IMAGE033
is the pitch angle of the radar,
Figure 755339DEST_PATH_IMAGE034
is the scene center action distance. After the echo signal of the three-dimensional SAR imaging is subjected to two-dimensional imaging processing, a certain pixel of the SAR two-dimensional image consists of N scatterer echoes at different heights. The pixel value g (n) can be regarded as the backscattering coefficient integral of the echo in the elevation direction h, which is calculated based on the following equation:
Figure 939195DEST_PATH_IMAGE035
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 840155DEST_PATH_IMAGE036
is the back-scattering coefficient of the light,
Figure 793068DEST_PATH_IMAGE037
is the spatial frequency corresponding to elevation h.
Spatial frequency of elevation h
Figure 121281DEST_PATH_IMAGE037
The specific form of (A) is represented by the following formula:
Figure 293899DEST_PATH_IMAGE038
，
wherein N is the total array element number of the array, d is the array element spacing of the array, and lambda is the wavelength of the radar detection signal,
Figure 998549DEST_PATH_IMAGE033
is the pitch angle of the radar.
SAR processes discrete samples, so that the pixel value of a certain channel in an image sequence is actually the target backscattering function
Figure 805968DEST_PATH_IMAGE036
Spectrum at spatial frequencies
Figure 570662DEST_PATH_IMAGE037
Taking into account discrete sampled values ofNoise(s)
Figure 463532DEST_PATH_IMAGE039
Discrete sampling is performed according to the following equation:
Figure 207759DEST_PATH_IMAGE040
，
wherein the content of the first and second substances,
Figure 135264DEST_PATH_IMAGE041
is composed of
Figure 70859DEST_PATH_IMAGE042
The observation vector of the dimension(s),
Figure 451025DEST_PATH_IMAGE043
is that
Figure 497478DEST_PATH_IMAGE044
The observation matrix of the dimension(s),
Figure 780954DEST_PATH_IMAGE043
is represented by the following formula:
Figure 153030DEST_PATH_IMAGE045
，
the above equation shows that the essence of SAR elevation focusing is a process of recovering original signals from discrete sampling frequency spectrum, SAR elevation imaging can be converted into a problem of array radar direction of arrival (DOA), and elevation resolution can be realized by using a spectrum estimation method based on compressed sensing, so as to generate three-dimensional point cloud as shown in fig. 2.
Further, in step 102, the method excludes outliers from the three-dimensional data point set based on the box plot to obtain a target data point set, and performs a clustering operation on the target data point set to obtain a plurality of clusters:
and respectively constructing box line graphs of the three-dimensional data point set based on each dimensional coordinate of the three-dimensional coordinates and the signal amplitude, and excluding abnormal points in each box line graph to obtain the target data point set.
Illustratively, for each dimension coordinate and signal amplitude of the three-dimensional data point set in the cartesian coordinate system, an upper quartile, a lower quartile and a quartile distance value of the box plot are calculated, respectively, and an upper limit value and a lower limit value are determined based on the upper quartile, the lower quartile and the quartile distance value. And (4) screening abnormal coordinate points based on the upper limit value and the lower limit value corresponding to each dimensional coordinate aiming at the three-dimensional coordinates of the three-dimensional data point set. And screening out abnormal points of the signal amplitude based on the lower limit value of the signal amplitude aiming at the signal amplitude of the three-dimensional data point set.
Further, in step 104, the determining the position and the size of the target based on the plurality of cluster centers and the plurality of bounding boxes corresponding to the plurality of clusters comprises:
and for each cluster, determining coordinate values of the cluster center of the cluster in the world coordinate system as a central position point of a target associated with the cluster, and determining the length, width and height of a bounding box corresponding to the cluster as three-dimensional dimensions of the target associated with the cluster.
Exemplarily, under the bounding box coordinate system corresponding to each cluster, the point coordinate with the farthest distribution at two ends of the three axes is determined
Figure 20491DEST_PATH_IMAGE022
、
Figure 136215DEST_PATH_IMAGE023
、
Figure 772733DEST_PATH_IMAGE024
、
Figure 551595DEST_PATH_IMAGE025
、
Figure 640774DEST_PATH_IMAGE026
、
Figure 294609DEST_PATH_IMAGE027
And then calculating the length, width and height of the bounding box:
Figure 785633DEST_PATH_IMAGE046
，
the cluster center is also the center of the bounding box, and a single bounding box is determined based on the bounding box size and the cluster center, with all bounding boxes corresponding to the three-dimensional size of the one or more targets.
Through one or more of the above embodiments of the present invention, at least the following technical effects can be achieved:
in the technical scheme disclosed by the invention, a three-dimensional point cloud target detection method is provided. According to the method, a traditional millimeter wave point cloud imaging method is not used, an imaging mode of a Synthetic Aperture Radar (SAR) is used for reference, a time domain BP algorithm is used for carrying out two-dimensional SAR imaging, and then an array three-dimensional imaging algorithm based on compressed sensing is matched to obtain a high-resolution and dense three-dimensional data point set. Then, the box plot analysis is used for preprocessing the three-dimensional data point set, abnormal points in the data point set are screened out, and a clustering method is used for clustering the target data point set. And finally, calculating a bounding box of each cluster by using a Principal Component Analysis (PCA) method and coordinate Euclidean transformation to estimate geometrical information such as the position and the size of the interested target in the scene. The invention adopts the imaging mode of the Synthetic Aperture Radar (SAR), is not limited by factors such as array resolution and the like, can obtain a dense data point set, has no false alarm in a detection result, and improves the positioning precision. And clustering is carried out after abnormal points are screened out through the box plot, so that a better clustering effect can be obtained. The cluster processing is performed to form the OBB bounding box, the determined target shape is closer to a real object, the number of the bounding bodies can be obviously reduced, and the target detection accuracy is improved.
Based on the same inventive concept as the method for detecting a three-dimensional point cloud target of the embodiment of the present invention, the embodiment of the present invention provides a three-dimensional point cloud target detection apparatus for a synthetic aperture radar, please refer to fig. 7, the apparatus includes:
a three-dimensional data point set generating module 201, configured to perform three-dimensional imaging processing on an echo signal fed back by at least one target with respect to a detection signal of the synthetic aperture radar to obtain a three-dimensional data point set in a world coordinate system;
a cluster determining module 202, configured to exclude outliers from the three-dimensional data point set based on a box plot to obtain a target data point set, and perform a clustering operation on the target data point set to obtain a plurality of clusters;
a bounding box generating module 203, configured to, for each of the clusters, determine a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determine, among data points of the cluster, a plurality of target data points having a maximum distance from each coordinate axis of the bounding box coordinate system, and determine, according to the plurality of target data points, a bounding box corresponding to the cluster;
a target determination module 204, configured to determine a position and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
Further, the bounding box generating module 203 is further configured to:
calculating a covariance matrix of the cluster based on the clustered data points and the cluster center;
performing orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector;
and determining coordinate axis vectors corresponding to the coordinate axes of the bounding box coordinate system based on the first principal component vector and the second principal component vector.
Further, the bounding box generating module 203 is further configured to:
calculating a variance between each of the clustered data points and the cluster center, and constructing the covariance matrix based on the variances.
Further, the bounding box generating module 203 is further configured to:
calculating the variance based on:
Figure 233932DEST_PATH_IMAGE047
，
wherein M is I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Coordinates representing each data point in the clustered set of data points;
calculating the covariance matrix based on:
Figure 577451DEST_PATH_IMAGE048
，
wherein, cov I Represents the covariance matrix, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Representing the coordinates of each data point in the clustered set of data points, T representing a matrix transpose.
Further, the bounding box generating module 203 is further configured to:
and carrying out orthogonal linear transformation on the covariance matrix, determining the principal component vector with the largest variance as the first principal component vector, and determining the principal component vector with the second largest variance as the second principal component vector.
Further, the determining the coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector;
determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector.
Determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector;
further, the bounding box generating module 203 is further configured to:
determining the third coordinate axis vector based on:
Figure 769398DEST_PATH_IMAGE003
，
wherein the content of the first and second substances,
Figure 114929DEST_PATH_IMAGE004
represents the third coordinate axis vector and the third coordinate axis vector,
Figure 999708DEST_PATH_IMAGE005
representing the first principal component vector, and,
Figure 63479DEST_PATH_IMAGE006
representing the second principal component vector;
said determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector comprises:
determining the second coordinate axis vector based on:
Figure 793538DEST_PATH_IMAGE007
，
wherein the content of the first and second substances,
Figure 495040DEST_PATH_IMAGE008
represents the second coordinate axis vector and the second coordinate axis vector,
Figure 550720DEST_PATH_IMAGE009
representing the first principal component vector, and,
Figure 836208DEST_PATH_IMAGE004
representing the third coordinate axis vector;
the determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector comprises:
determining the first coordinate axis vector based on:
Figure 369958DEST_PATH_IMAGE010
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 690081DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 651083DEST_PATH_IMAGE012
representing the second principal component vector, and,
Figure 179193DEST_PATH_IMAGE004
representing the third coordinate axis vector.
Further, the apparatus is further configured to:
after the bounding box coordinate system which takes the cluster center of the cluster as a coordinate origin is determined according to the principal component analysis method, constructing a rotation matrix based on the first coordinate axis vector, the second coordinate axis vector and the third coordinate axis vector, constructing a translation vector based on the rotation matrix and the cluster center, and performing rigid body transformation on the clustered data points based on the translation vector, so that the coordinates of the clustered data points are converted from the coordinates of the world coordinate system to the coordinates of the bounding box coordinate system.
Further, the apparatus is further configured to:
constructing the rotation matrix based on:
Figure 516633DEST_PATH_IMAGE013
,
Figure 425683DEST_PATH_IMAGE008
,
Figure 823166DEST_PATH_IMAGE014
，
wherein the content of the first and second substances,
Figure 83247DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 725843DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 20558DEST_PATH_IMAGE008
represents the second coordinate axis vector and is,
Figure 323363DEST_PATH_IMAGE004
representing the third coordinate axis vector, T represents a matrix transpose.
Further, the apparatus is further configured to:
constructing the translation vector based on:
Figure 336319DEST_PATH_IMAGE016
，
wherein t represents the translation vector,
Figure 15562DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 899204DEST_PATH_IMAGE017
coordinates representing the cluster center.
Further, the apparatus is further configured to:
performing the rigid body transformation based on:
Figure 874375DEST_PATH_IMAGE018
，
wherein the content of the first and second substances,
Figure 109048DEST_PATH_IMAGE049
，
wherein the content of the first and second substances,
Figure 326402DEST_PATH_IMAGE020
coordinates of data points representing the cluster in the bounding box coordinate system;
Figure 330131DEST_PATH_IMAGE021
coordinates of data points representing the cluster in the world coordinate system,
Figure 240318DEST_PATH_IMAGE015
representing the rotation matrix, t representing the translation vector, C I Set of data points, P, representing said cluster i Coordinates of each data point in the set of data points representing the cluster in the world coordinate system.
Further, the three-dimensional data point set generating module 201 is further configured to:
and performing direction dimension imaging and distance dimension imaging on the echo signals to obtain a two-dimensional data point set, and performing elevation direction imaging on the two-dimensional data point set to obtain a three-dimensional data point set.
Further, the cluster determining module 202 is further configured to:
and respectively constructing box line graphs of the three-dimensional data point set based on each dimensional coordinate of the three-dimensional coordinates and the signal amplitude, and excluding abnormal points in each box line graph to obtain the target data point set.
Further, the goal determination module 204 is further configured to:
and for each cluster, determining coordinate values of the cluster center of the cluster in the world coordinate system as a central position point of a target associated with the cluster, and determining the length, width and height of a bounding box corresponding to the cluster as three-dimensional dimensions of the target associated with the cluster.
Other aspects and implementation details of the three-dimensional point cloud target detection device are the same as or similar to those of the three-dimensional point cloud target detection method described above, and are not described herein again.
According to another aspect of the present invention, there is also provided a storage medium having a plurality of instructions stored therein, the instructions being adapted to be loaded by a processor to perform any of the three-dimensional point cloud object detection methods described above.
In summary, although the present invention has been described with reference to the preferred embodiments, the above-described preferred embodiments are not intended to limit the present invention, and those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, therefore, the scope of the present invention shall be determined by the appended claims.

Claims (16)

1. A three-dimensional point cloud target detection method for a synthetic aperture radar, the method comprising:
carrying out three-dimensional imaging processing on echo signals fed back by at least one target aiming at detection signals of the synthetic aperture radar to obtain a three-dimensional data point set under a world coordinate system;
excluding outliers from the three-dimensional data point set based on the box plot to obtain a target data point set, and performing clustering operation on the target data point set to obtain a plurality of clusters;
for each cluster, determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determining a plurality of target data points with maximum distances to coordinate axes of the bounding box coordinate system from the data points of the cluster, and determining a bounding box corresponding to the cluster according to the plurality of target data points;
determining a position and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
2. The method of claim 1, wherein determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method comprises:
calculating a covariance matrix for the cluster based on the clustered data points and the cluster center;
performing orthogonal linear transformation on the covariance matrix to obtain a first principal component vector and a second principal component vector;
and determining coordinate axis vectors corresponding to the coordinate axes of the bounding box coordinate system based on the first principal component vector and the second principal component vector.
3. The method of claim 2, wherein the computing the covariance matrix for the cluster based on the data points for the cluster and the cluster center comprises:
calculating a variance between each of the clustered data points and the cluster center, and constructing the covariance matrix based on the variances.
4. The method of claim 3, wherein the calculating a variance between each of the clustered data points and the cluster center, the constructing the covariance matrix based on the variances comprises:
calculating the variance based on:
Figure 189556DEST_PATH_IMAGE001
，
wherein, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Coordinates representing each data point in the clustered set of data points;
calculating the covariance matrix based on:
Figure 778801DEST_PATH_IMAGE002
，
wherein, cov I Represents the covariance matrix, M I Coordinates representing the center of the cluster, N represents the total number of data points of the cluster, C I Set of data points, P, representing said cluster i Representing the coordinates of each data point in the clustered set of data points, T representing a matrix transpose.
5. The method of claim 4, wherein the orthogonalizing a linear transformation of the covariance matrix to obtain a first principal component vector and a second principal component vector comprises:
and carrying out orthogonal linear transformation on the covariance matrix, determining the principal component vector with the largest variance as the first principal component vector, and determining the principal component vector with the second largest variance as the second principal component vector.
6. The method of claim 5, wherein the determining the coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector;
determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector;
determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector.
7. The method of claim 6, wherein the determining a third coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the second principal component vector comprises:
determining the third coordinate axis vector based on:
Figure 27379DEST_PATH_IMAGE003
，
wherein, the first and the second end of the pipe are connected with each other,
Figure 124386DEST_PATH_IMAGE004
represents the third coordinate axis vector and the third coordinate axis vector,
Figure 142021DEST_PATH_IMAGE005
representing the first principal component vector, and,
Figure 902166DEST_PATH_IMAGE006
representing the second principal component vector;
the determining a second coordinate axis vector of the bounding box coordinate system based on the first principal component vector and the third coordinate axis vector comprises:
determining the second coordinate axis vector based on:
Figure 841303DEST_PATH_IMAGE007
，
wherein the content of the first and second substances,
Figure 467633DEST_PATH_IMAGE008
represents the second coordinate axis vector and the second coordinate axis vector,
Figure 74195DEST_PATH_IMAGE009
representing the first principal component vector, and,
Figure 270821DEST_PATH_IMAGE004
representing the third coordinate axis vector;
the determining a first coordinate axis vector of the bounding box coordinate system based on the second principal component vector and the third coordinate axis vector comprises:
determining the first coordinate axis vector based on:
Figure 228412DEST_PATH_IMAGE010
，
wherein the content of the first and second substances,
Figure 168687DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 393869DEST_PATH_IMAGE012
representing the second principal component vector, and,
Figure 26976DEST_PATH_IMAGE004
representing the third coordinate axis vector.
8. The method of claim 7, wherein the method further comprises:
after determining a bounding box coordinate system with a cluster center of the cluster as a coordinate origin according to a principal component analysis method, constructing a rotation matrix based on the first coordinate axis vector, the second coordinate axis vector and the third coordinate axis vector, constructing a translation vector based on the rotation matrix and the cluster center, and performing rigid body transformation on data points of the cluster based on the translation vector so that coordinates of the data points of the cluster are converted from coordinates of the world coordinate system to coordinates of the bounding box coordinate system.
9. The method of claim 8, wherein constructing a rotation matrix based on the first, second, and third coordinate axis vectors comprises:
constructing the rotation matrix based on:
Figure 471864DEST_PATH_IMAGE013
,
Figure 950250DEST_PATH_IMAGE008
,
Figure 796983DEST_PATH_IMAGE014
，
wherein the content of the first and second substances,
Figure 836877DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 503481DEST_PATH_IMAGE011
represents the first coordinate axis vector and the second coordinate axis vector,
Figure 785558DEST_PATH_IMAGE008
represents the second coordinate axis vector and the second coordinate axis vector,
Figure 486798DEST_PATH_IMAGE004
represents the third coordinate axis vector, and T represents a matrix transpose.
10. The method of claim 9, wherein the constructing a translation vector based on the rotation matrix and the cluster center comprises:
constructing the translation vector based on:
Figure 461707DEST_PATH_IMAGE016
，
wherein t represents the translation vector,
Figure 848564DEST_PATH_IMAGE015
a representation of the rotation matrix is provided,
Figure 934331DEST_PATH_IMAGE017
coordinates representing the cluster center.
11. The method of claim 10, wherein the rigid-body transforming the clustered data points based on the translation vector comprises:
performing the rigid body transformation based on:
Figure 224498DEST_PATH_IMAGE018
，
wherein the content of the first and second substances,
Figure 370309DEST_PATH_IMAGE019
，
wherein the content of the first and second substances,
Figure 11506DEST_PATH_IMAGE020
coordinates of data points representing the cluster in the bounding box coordinate system;
Figure 136850DEST_PATH_IMAGE021
coordinates of data points representing the cluster in the world coordinate system,
Figure 812682DEST_PATH_IMAGE015
representing the rotation matrix, t representing the translation vector, C I Set of data points, P, representing said cluster i Coordinates of each data point in the set of data points representing the cluster in the world coordinate system.
12. The method of claim 1, wherein the three-dimensional imaging processing of the echo signals fed back by at least one target for the detection signals of the synthetic aperture radar to obtain a three-dimensional set of data points in a world coordinate system comprises:
and performing direction dimensional imaging and distance dimensional imaging on the echo signals to obtain a two-dimensional data point set, and performing elevation direction imaging on the two-dimensional data point set to obtain a three-dimensional data point set.
13. The method of claim 1, wherein the bin plot-based exclusion of outliers from the set of three-dimensional data points to obtain a set of target data points, and the clustering operation on the set of target data points to obtain a plurality of clusters:
and respectively constructing a box line graph of the three-dimensional data point set based on each dimensional coordinate of the three-dimensional coordinates and the signal amplitude, and excluding abnormal points in each box line graph to obtain the target data point set.
14. The method of claim 1, wherein the determining the location and the size of the target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters comprises:
and for each cluster, determining coordinate values of the cluster center of the cluster in the world coordinate system as a central position point of a target associated with the cluster, and determining the length, the width and the height of a bounding box corresponding to the cluster as three-dimensional dimensions of the target associated with the cluster.
15. A three-dimensional point cloud target detection apparatus for a synthetic aperture radar, the apparatus comprising:
the three-dimensional data point set generating module is used for carrying out three-dimensional imaging processing on echo signals fed back by at least one target aiming at the detection signals of the synthetic aperture radar so as to obtain a three-dimensional data point set under a world coordinate system;
the clustering determination module is used for excluding abnormal points from the three-dimensional data point set based on the box line graph to obtain a target data point set, and carrying out clustering operation on the target data point set to obtain a plurality of clusters;
a bounding box generating module, configured to determine, for each of the clusters, a bounding box coordinate system using a cluster center of the cluster as a coordinate origin according to a principal component analysis method, determine, among data points of the cluster, a plurality of target data points having a maximum distance from each coordinate axis of the bounding box coordinate system, and determine, according to the plurality of target data points, a bounding box corresponding to the cluster;
a target determination module to determine a location and a size of the at least one target based on a plurality of cluster centers and a plurality of bounding boxes corresponding to the plurality of clusters.
16. A storage medium having stored therein a plurality of instructions adapted to be loaded by a processor to perform the three-dimensional point cloud object detection method of any one of claims 1 to 14.
CN202211496628.6A 2022-11-28 2022-11-28 Three-dimensional point cloud target detection method and device and storage medium Pending CN115598614A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211496628.6A CN115598614A (en) 2022-11-28 2022-11-28 Three-dimensional point cloud target detection method and device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211496628.6A CN115598614A (en) 2022-11-28 2022-11-28 Three-dimensional point cloud target detection method and device and storage medium

Publications (1)

Publication Number Publication Date
CN115598614A true CN115598614A (en) 2023-01-13

Family

ID=84852001

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211496628.6A Pending CN115598614A (en) 2022-11-28 2022-11-28 Three-dimensional point cloud target detection method and device and storage medium

Country Status (1)

Country Link
CN (1) CN115598614A (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180356213A1 (en) * 2016-09-14 2018-12-13 Hangzhou Scantech Co., Ltd Three-dimensional sensor system and three-dimensional data acquisition method
CN111652936A (en) * 2020-08-06 2020-09-11 天津迦自机器人科技有限公司 Three-dimensional sensing and stacking planning method and system for open container loading
CN111781611A (en) * 2020-06-16 2020-10-16 当家移动绿色互联网技术集团有限公司 Method and device for establishing model, storage medium and electronic equipment
CN113609914A (en) * 2021-07-09 2021-11-05 北京经纬恒润科技股份有限公司 Obstacle identification method and device and vehicle control system
CN114089377A (en) * 2021-10-21 2022-02-25 江苏大学 Point cloud processing and object identification system and method based on laser radar
CN115393818A (en) * 2022-09-06 2022-11-25 一汽解放汽车有限公司 Driving scene recognition method and device, computer equipment and storage medium

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180356213A1 (en) * 2016-09-14 2018-12-13 Hangzhou Scantech Co., Ltd Three-dimensional sensor system and three-dimensional data acquisition method
CN111781611A (en) * 2020-06-16 2020-10-16 当家移动绿色互联网技术集团有限公司 Method and device for establishing model, storage medium and electronic equipment
CN111652936A (en) * 2020-08-06 2020-09-11 天津迦自机器人科技有限公司 Three-dimensional sensing and stacking planning method and system for open container loading
CN113609914A (en) * 2021-07-09 2021-11-05 北京经纬恒润科技股份有限公司 Obstacle identification method and device and vehicle control system
CN114089377A (en) * 2021-10-21 2022-02-25 江苏大学 Point cloud processing and object identification system and method based on laser radar
CN115393818A (en) * 2022-09-06 2022-11-25 一汽解放汽车有限公司 Driving scene recognition method and device, computer equipment and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
唐启义 等, 中国农业出版社 *
朱飞祥: "基于三维激光雷达的沿海泊位码头识别算法", 《中国航海》 *

Similar Documents

Publication Publication Date Title
US6337654B1 (en) A-scan ISAR classification system and method therefor
Jacobs et al. Automatic target recognition using sequences of high resolution radar range-profiles
US8718323B2 (en) Batch detection association for enhanced target descrimination in dense detection environments
US6150972A (en) Process for combining multiple passes of interferometric SAR data
Smith et al. Robust through-the-wall radar image classification using a target-model alignment procedure
CN115061113B (en) Target detection model training method and device for radar and storage medium
Pandey et al. Classification of automotive targets using inverse synthetic aperture radar images
Martorella Introduction to inverse synthetic aperture radar
JPH0980146A (en) Radar apparatus
CN115291207A (en) Multi-target detection method for small rotor unmanned aerial vehicle based on MIMO radar
CN115546526B (en) Three-dimensional point cloud clustering method, device and storage medium
EP3444632A1 (en) Signal processing device and radar apparatus
CN113419238A (en) Mountain landslide monitoring method based on millimeter wave radar, electronic device and storage medium
CN111487621A (en) Sea surface flow field inversion method based on radar image and electronic equipment
Minvielle et al. Indoor 3-D radar imaging for low-RCS analysis
CN115754969A (en) Radar identification method and system based on wide-narrow band combination sum difference monopulse
CN115598614A (en) Three-dimensional point cloud target detection method and device and storage medium
Gao et al. Static background removal in vehicular radar: Filtering in azimuth-elevation-doppler domain
Cho et al. Spatiotemporal radar target identification using radar cross-section modeling and hidden Markov models
Omori et al. K-space decomposition based super-resolution three-dimensional imaging method for millimeter wave radar
Yuan et al. 3DRUDAT: 3D robust unambiguous Doppler beam sharpening using adaptive threshold for forward-looking region
CN113671494B (en) Radar scattering sectional area measurement method based on super-resolution imaging
Laubie et al. An empirical look at cross-target correlation in bistatic SAR images
Roldan et al. See Further Than CFAR: a Data-Driven Radar Detector Trained by Lidar
Wessling Radar target modelling based on RCS measurements

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20230113

RJ01 Rejection of invention patent application after publication