WO2021217518A1 - 雷达点云聚类方法和装置 - Google Patents

雷达点云聚类方法和装置 Download PDF

Info

Publication number
WO2021217518A1
WO2021217518A1 PCT/CN2020/087874 CN2020087874W WO2021217518A1 WO 2021217518 A1 WO2021217518 A1 WO 2021217518A1 CN 2020087874 W CN2020087874 W CN 2020087874W WO 2021217518 A1 WO2021217518 A1 WO 2021217518A1
Authority
WO
WIPO (PCT)
Prior art keywords
target point
points
point cloud
cluster
geometric center
Prior art date
Application number
PCT/CN2020/087874
Other languages
English (en)
French (fr)
Inventor
李晓波
劳大鹏
刘劲楠
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Priority to CN202080004810.8A priority Critical patent/CN112689775B/zh
Priority to PCT/CN2020/087874 priority patent/WO2021217518A1/zh
Publication of WO2021217518A1 publication Critical patent/WO2021217518A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO 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
    • G01S13/00Systems 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
    • G01S13/88Radar or analogous systems specially adapted for specific applications
    • G01S13/89Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO 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
    • G01S13/00Systems 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
    • G01S13/88Radar or analogous systems specially adapted for specific applications
    • G01S13/93Radar or analogous systems specially adapted for specific applications for anti-collision purposes
    • G01S13/931Radar or analogous systems specially adapted for specific applications for anti-collision purposes of land vehicles
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO 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
    • G01S7/00Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41Details 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

Definitions

  • This application relates to the field of radar imaging, and in particular to a method and device for clustering radar point clouds.
  • the current vehicle-mounted radar detects multiple measurement points when imaging a target to form a high-resolution target point cloud.
  • This vehicle-mounted radar is also called a point cloud imaging radar.
  • the millimeter wave radar measures the same target to form a point cloud.
  • the traditional noisy density-based clustering density-based spatial clustering of applications with noise, DBSCAN
  • clusters point clouds based on the point cloud density within a certain distance range When multiple targets are close in distance There will be a situation where the point clouds of multiple targets are grouped into one class.
  • the embodiments of the present application provide a radar point cloud clustering method and device, which are used to distinguish point clouds of different targets with similar distances.
  • a radar point cloud clustering method including: calculating the Euclidean distance between any two points according to the angle information and distance information of any two points in the M points of the point cloud data relative to the radar. Distance; where M is an integer greater than or equal to 2; make the difference between the radial velocity of any two points relative to the radar to obtain the radial velocity difference between any two points; for meeting the Euclidean distance less than the first distance threshold,
  • any two points whose radial velocity difference is less than the first velocity threshold are clustered to obtain N target point clouds; where N is an integer greater than 1.
  • the Euclidean distance between any two points in the M points of the point cloud data is calculated according to the angle information and distance information of any two points relative to the radar;
  • the difference between any two points relative to the radial velocity of the radar is obtained, and the radial velocity difference between any two points is obtained;
  • the Euclidean distance is less than the first distance threshold, and the radial velocity difference is less than the first velocity threshold
  • This application is based on the principle that the point clouds of different targets with similar distances are similar in Euclidean distance, but the radial velocities of the corresponding point clouds are also different due to the different moving speeds of different targets.
  • the Euclidean distance in the point cloud data is similar and the radial Points with a smaller speed difference are clustered into a point cloud of a target, so that the point clouds of different targets with similar distances can be distinguished.
  • the method further includes: calculating the respective moving speeds of the N target point clouds according to the angle information and the radial velocity of all points of each target point cloud in the N target point clouds relative to the radar; According to the angle information, distance information and radial velocity difference of the N geometric center points corresponding to the N target point clouds with respect to the radar, calculate the Euclidean distance and radius between any two geometric center points of the N geometric center points Directional velocity difference; among N target point clouds, the Euclidean distance between geometric center points is less than the second distance threshold, and the target point cloud whose radial velocity difference is less than the second velocity threshold is clustered to obtain the first cluster ; In the first cluster, the target point cloud whose moving speed meets the condition is clustered to obtain the second cluster.
  • calculating the respective moving speeds of the N target point clouds includes: For any target point cloud in the N target point clouds, calculate the K step and K step angles of the T points according to the K step angles, the angle information and the radial velocity of all T points of any target point cloud relative to the radar The moving speed corresponding to the advance angle; where K is a positive integer; according to the moving speed of the T points corresponding to the K step angles, the moving speed of any target point cloud is calculated.
  • the angle information of all T points of any target point cloud relative to the radar and the radial velocity calculate the corresponding K step angles of the T points
  • the moving speed includes: for the t-th point in the T points, the K step angles are subtracted from the angle information of the t-th point respectively to obtain K candidate angles of the t-th point; K candidate angles are expressed K moving directions of the t-th point, 1 ⁇ t ⁇ T, t is a positive integer; back-project the radial velocity of the t-th point to the K candidate angles of the t-th point to obtain the t-th point The movement speed corresponding to K step angles.
  • calculating the moving speed of any target point cloud according to the moving speeds of the T points corresponding to the K step angles includes: for the kth step of the K step angles Angle, average the movement speed of T points corresponding to the k-th step angle to obtain the mean value; 1 ⁇ k ⁇ K, and k is an integer; calculate the movement of T points corresponding to the k-th step angle The mean square error between the speed and the mean value; calculate the target step angle according to the smallest mean square error; average the moving speeds of the T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • the back projection of the radial velocity of the point corresponding to the target step angle to the corresponding candidate angle has the smallest deviation from the true velocity.
  • any two geometric centers among the N geometric center points are calculated.
  • the Euclidean distance and radial velocity difference between points include: calculating the Euclidean distance between any two geometric center points according to the angle information and distance information of any two geometric center points relative to the radar; for any two The geometric center point is compared with the radial velocity of the radar, and the radial velocity difference between any two geometric center points is obtained.
  • the Euclidean distance between the geometric center points is less than the second distance threshold, and the target point cloud whose radial velocity difference is less than the second speed threshold is clustered.
  • To obtain the first cluster including: adding the target point cloud corresponding to the second geometric center point out of the N target point clouds that does not belong to the first cluster into the first cluster, where the second geometric center point and the first geometric center point
  • the Euclidean distance between the center points is less than the second distance threshold and the radial velocity difference is less than the second speed threshold.
  • the first geometric center point is the geometric center point of any target point cloud belonging to the first cluster among the N target point clouds .
  • the second distance threshold is greater than the first distance threshold, and the second speed threshold is greater than the first speed threshold.
  • clustering the target point cloud whose moving speed satisfies the condition in the first cluster to obtain the second cluster includes: setting the difference between the moving speeds belonging to the first cluster to be smaller than the third speed threshold Any two target point clouds from join the second cluster. Since the target point clouds in the second cluster meet the close distance and small radial velocity difference, the second cluster is further clustered according to the conditions of moving speed, so that the distance is close and the radial velocity difference is small. And the target point cloud cluster with the smaller moving speed difference is a target cluster.
  • clustering the target point clouds whose moving speed meets the condition in the first cluster to obtain the second cluster includes: according to the geometric center points of any two target point clouds belonging to the first cluster Calculate the cosine similarity based on the distance information, angle information, and moving speed, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud; add any two target point clouds whose cosine similarity is less than the similarity threshold into the second Cluster.
  • clustering any two points where the Euclidean distance is less than the first distance threshold and the radial velocity difference is less than the first velocity threshold is clustered to obtain N target point clouds, including: For any target point cloud in two target point clouds, any two points that do not belong to any target point cloud, and the Euclidean distance to the first point is less than the first distance threshold and the radial velocity difference is less than the first
  • the second point of a speed threshold is added to any target point cloud; where the first point is a point belonging to any target point cloud among any two points.
  • any target point cloud of N target point clouds if you select a point, then you can find at least one point from any target point cloud, and the Euclidean distance from the selected point is less than the first distance threshold And the radial velocity difference is less than the first velocity threshold.
  • the first distance threshold and the first speed threshold satisfy at least one of the following conditions: the greater the distance information of the first point or the second point, the larger the corresponding first distance threshold; the first point Or the greater the distance information of the second point, the greater the corresponding first speed threshold.
  • the reason is that the farther the target is from the radar, the smaller the signal-to-noise ratio of the point, and the sparser the point cloud density.
  • the above-mentioned changes can be adapted to the above-mentioned changes by changing the above-mentioned threshold hierarchically, so that points belonging to the same target will not be missed during clustering.
  • a radar point cloud clustering method which includes: calculating the N target point clouds respectively according to the angle information and radial velocity of all points of each target point cloud in the N target point clouds relative to the radar Corresponding moving speed; according to the angle information, distance information and radial velocity difference of the N geometric center points corresponding to the N target point clouds with respect to the radar, calculate the distance between any two geometric center points of the N geometric center points Euclidean distance and radial velocity difference; among N target point clouds, the Euclidean distance between geometric center points is less than the second distance threshold, and the target point cloud whose radial velocity difference is less than the second velocity threshold is clustered , Obtain the first cluster; cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster.
  • the radar point cloud clustering method provided by the embodiments of the present application, based on the principle that the point cloud of the same target has a small difference in moving speed, by dividing the target point cloud with a similar distance, a small radial velocity difference, and a small moving speed difference
  • the clustering is a cluster of a target, which can prevent a target from being partially occluded or different materials at different locations, which may cause the point cloud of one target to be clustered into point clouds of multiple targets.
  • calculating the respective moving speeds of the N target point clouds includes: For any target point cloud in the N target point clouds, calculate the K step and K step angles of the T points according to the K step angles, the angle information and the radial velocity of all T points of any target point cloud relative to the radar The moving speed corresponding to the advance angle; where K is a positive integer; according to the moving speed of the T points corresponding to the K step angles, the moving speed of any target point cloud is calculated.
  • the angle information of all T points of any target point cloud relative to the radar and the radial velocity calculate the corresponding K step angles of the T points
  • the moving speed includes: for the t-th point in the T points, the K step angles are subtracted from the angle information of the t-th point respectively to obtain K candidate angles of the t-th point; K candidate angles are expressed K moving directions of the t-th point, 1 ⁇ t ⁇ T, t is a positive integer; back-project the radial velocity of the t-th point to the K candidate angles of the t-th point to obtain the t-th point The movement speed corresponding to K step angles.
  • calculating the moving speed of any target point cloud according to the moving speeds of the T points corresponding to the K step angles includes: for the kth step of the K step angles Angle, average the movement speed of T points corresponding to the k-th step angle to obtain the mean value; 1 ⁇ k ⁇ K, and k is an integer; calculate the movement of T points corresponding to the k-th step angle The mean square error between the speed and the mean value; calculate the target step angle according to the smallest mean square error; average the moving speeds of the T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • the Euclidean distance and radial velocity difference between any two geometric center points include: calculate the Euclidean distance between any two geometric center points according to the angle information and distance information of any two geometric center points relative to the radar; for any two geometric centers The radial velocity difference between the points relative to the radar is obtained, and the radial velocity difference between any two geometric center points is obtained.
  • the Euclidean distance between the geometric center points is less than the second distance threshold, and the target point cloud whose radial velocity difference is less than the second speed threshold is clustered.
  • To obtain the first cluster including: adding the target point cloud corresponding to the second geometric center point out of the N target point clouds that does not belong to the first cluster into the first cluster, where the second geometric center point and the first geometric center point
  • the Euclidean distance between the center points is less than the second distance threshold and the radial velocity difference is less than the second speed threshold.
  • the first geometric center point is the geometric center point of any target point cloud belonging to the first cluster among the N target point clouds .
  • the second distance threshold is greater than the first distance threshold, and the second speed threshold is greater than the first speed threshold.
  • clustering the target point cloud whose moving speed satisfies the condition in the first cluster to obtain the second cluster includes: setting the difference between the moving speeds belonging to the first cluster to be smaller than the third speed threshold Any two target point clouds from join the second cluster. Since the target point clouds in the second cluster meet the close distance and small radial velocity difference, the second cluster is further clustered according to the conditions of moving speed, so that the distance is close and the radial velocity difference is small. And the target point cloud cluster with the smaller moving speed difference is a target cluster.
  • clustering the target point clouds whose moving speed meets the condition in the first cluster to obtain the second cluster includes: according to the geometric center points of any two target point clouds belonging to the first cluster Calculate the cosine similarity based on the distance information, angle information, and moving speed, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud; add any two target point clouds whose cosine similarity is less than the similarity threshold into the second Cluster.
  • a radar point cloud clustering device including: a calculation module for calculating any two points according to the angle information and distance information of any two points in the M points of the point cloud data relative to the radar The Euclidean distance between; where M is an integer greater than or equal to 2; the calculation module is also used to make the difference between the radial velocity of any two points relative to the radar to obtain the radial velocity difference between any two points; The clustering module is used to cluster any two points that meet the Euclidean distance less than the first distance threshold and the radial velocity difference is less than the first velocity threshold to obtain N target point clouds; where N is greater than 1 Integer.
  • the calculation module is also used to calculate the respective corresponding N target point clouds according to the angle information and radial velocity of all points of each target point cloud in the N target point clouds relative to the radar. Movement speed; the calculation module is also used to calculate any two geometric centers of the N geometric center points corresponding to the N geometric center points of the N target point clouds relative to the radar angle information, distance information, and radial velocity difference The Euclidean distance between the points and the radial velocity difference; the clustering module is also used for N target point clouds, the Euclidean distance between geometric center points is less than the second distance threshold, and the radial velocity difference is less than The target point cloud with the second speed threshold is clustered to obtain the first cluster; the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster.
  • the calculation module is further used to calculate the respective corresponding N target point clouds according to the angle information and radial velocity of all points of each target point cloud in the N target point clouds relative to the radar.
  • Movement speed including: calculation module, specifically used for any target point cloud in N target point clouds, according to K step angles, angle information and path information of all T points of any target point cloud relative to the radar To speed, calculate the moving speed of T points corresponding to K step angles; where K is a positive integer; the calculation module is specifically used to calculate the moving speed of T points corresponding to K step angles. The moving speed of a target point cloud.
  • the calculation module is specifically configured to calculate the sum of the K step angles, the angle information and the radial velocity of all T points of any target point cloud relative to the radar, and
  • the moving speed corresponding to each step angle includes: a calculation module, which is specifically used to subtract the K step angles from the angle information of the t-th point for the t-th point in the T points to obtain the t-th point K candidate angles of the point; K candidate angles represent the K moving directions of the t-th point, 1 ⁇ t ⁇ T, and t is a positive integer; the calculation module is specifically used to change the radial velocity of the t-th point to the K candidate angles of t points are respectively back-projected to obtain the moving speed of the t-th point corresponding to the K step angles.
  • the calculation module is specifically configured to calculate the movement speed of any target point cloud according to the movement speeds of the T points corresponding to the K step angles, including: a calculation module, which is specifically used to calculate the movement speed of any target point cloud For the k-th step angle among the K step angles, average the moving speeds of the T points corresponding to the k-th step angle to obtain the average value; 1 ⁇ k ⁇ K, and k is an integer; calculation module, It is specifically used to calculate the mean square error between the moving speed corresponding to the k-th step angle and the mean value of the T points; the calculation module is specifically used to calculate the target step angle according to the smallest mean square error; the calculation module is specifically used to Average the moving speed of T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • the calculation module calculates any two of the N geometric center points relative to the radar angle information, distance information, and radial velocity difference of the N geometric center points corresponding to the N target point clouds.
  • the Euclidean distance and radial velocity difference between geometric center points including: calculation module, specifically used to calculate the distance between any two geometric center points relative to the radar angle information and distance information Euclidean distance; calculation module, specifically used to make the difference between the radial velocity of any two geometric center points relative to the radar, and obtain the radial velocity difference between any two geometric center points.
  • the clustering module is also used to determine the Euclidean distance between the geometric center points in the N target point cloud is less than the second distance threshold, and the radial velocity difference is less than the second speed threshold
  • Clustering the target point cloud of the N target point cloud to obtain the first cluster including: a clustering module, specifically used to add the target point cloud corresponding to the second geometric center point among the N target point clouds that does not belong to the first cluster A cluster, where the Euclidean distance between the second geometric center point and the first geometric center point is less than the second distance threshold and the radial velocity difference is less than the second speed threshold, and the first geometric center point is in the N target point cloud The geometric center point of any target point cloud belonging to the first cluster.
  • the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including: a clustering module, specifically used to group Any two target point clouds of the first cluster whose moving speed difference is less than the third speed threshold are added to the second cluster.
  • the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including: a calculation module, which is specifically used to cluster the target point cloud according to the Calculate the cosine similarity of the distance information, angle information and moving speed of the geometric center points of any two target point clouds in a cluster, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud; clustering module, specific It is used to add any two target point clouds whose cosine similarity is less than the similarity threshold into the second cluster.
  • a calculation module which is specifically used to cluster the target point cloud according to the Calculate the cosine similarity of the distance information, angle information and moving speed of the geometric center points of any two target point clouds in a cluster, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud
  • clustering module specific It is used to add any two target point clouds whose cosine similarity is less than the similarity threshold into the second cluster.
  • the clustering module is used to cluster any two points where the Euclidean distance is less than the first distance threshold, and the radial velocity difference is less than the first velocity threshold, to obtain N target points Cloud, including: clustering module, specifically used for any target point cloud in N target point clouds, and any two points that do not belong to any target point cloud, and the Euclidean point between the first point The second point whose distance is less than the first distance threshold and the radial velocity difference is less than the first speed threshold is added to any target point cloud; where the first point is a point belonging to any target point cloud among any two points.
  • the first distance threshold and the first speed threshold satisfy at least one of the following conditions: the greater the distance information of the first point or the second point, the larger the corresponding first distance threshold; the first point Or the greater the distance information of the second point, the greater the corresponding first speed threshold.
  • a radar point cloud clustering device including: a calculation module for calculating N The moving speed of each target point cloud; the calculation module is also used to calculate the N geometric centers according to the angle information, distance information and radial velocity difference of the N geometric center points corresponding to the N target point clouds with respect to the radar.
  • clustering module is used for N target point clouds, the Euclidean distance between the geometric center points is less than the second distance threshold, and , The target point cloud whose radial velocity difference is less than the second speed threshold is clustered to obtain the first cluster; the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the first cluster Two clusters.
  • the calculation module is configured to calculate the respective movement of the N target point clouds according to the angle information and the radial velocity of all points of each target point cloud in the N target point clouds relative to the radar.
  • Speed including: calculation module, specifically used for any target point cloud in N target point clouds, according to K step angles, angle information and radial information of all T points of any target point cloud relative to the radar Speed, calculate the moving speed of T points corresponding to K step angles; where K is a positive integer; according to the moving speed of T points corresponding to K step angles, calculate the moving speed of any target point cloud .
  • the calculation module is specifically configured to calculate the sum of the K step angles, the angle information and the radial velocity of all T points of any target point cloud relative to the radar, and
  • the moving speed corresponding to each step angle includes: a calculation module, which is specifically used to subtract the K step angles from the angle information of the t-th point for the t-th point in the T points to obtain the t-th point K candidate angles of the point; K candidate angles represent the K moving directions of the t-th point, 1 ⁇ t ⁇ T, and t is a positive integer; the calculation module is specifically used to change the radial velocity of the t-th point to the K candidate angles of t points are respectively back-projected to obtain the moving speed of the t-th point corresponding to the K step angles.
  • the calculation module is specifically configured to calculate the movement speed of any target point cloud according to the movement speeds of the T points corresponding to the K step angles, including: a calculation module, which is specifically used to calculate the movement speed of any target point cloud For the k-th step angle among the K step angles, average the moving speeds of the T points corresponding to the k-th step angle to obtain the average value; 1 ⁇ k ⁇ K, and k is an integer; calculation module, It is specifically used to calculate the mean square error between the moving speed corresponding to the k-th step angle and the mean value of the T points; the calculation module is specifically used to calculate the target step angle according to the smallest mean square error; the calculation module is specifically used to Average the moving speed of T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • the calculation module calculates any two geometric centers among the N geometric center points of the N target point clouds based on the angle information, distance information, and radial velocity difference of the N geometric center points relative to the radar.
  • the Euclidean distance and radial velocity difference between the points including: a calculation module, specifically used to calculate the Euclidean distance between any two geometric center points based on the angle information and distance information of any two geometric center points relative to the radar Distance; calculation module, specifically used to make the difference between the radial velocity of any two geometric center points relative to the radar, and obtain the radial velocity difference between any two geometric center points.
  • the clustering module is used to determine the Euclidean distance between the geometric center points of the N target point clouds is less than the second distance threshold, and the radial velocity difference is less than the second speed threshold.
  • the target point cloud is clustered to obtain the first cluster, including: a clustering module, specifically used to add the target point cloud corresponding to the second geometric center point out of the N target point clouds that does not belong to the first cluster into the first Cluster, where the Euclidean distance between the second geometric center point and the first geometric center point is less than the second distance threshold and the radial velocity difference is less than the second speed threshold, and the first geometric center point belongs to the N target point cloud The geometric center point of any target point cloud of the first cluster.
  • the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including: a clustering module, specifically used to group Any two target point clouds of the first cluster whose moving speed difference is less than the third speed threshold are added to the second cluster.
  • the clustering module is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including: a calculation module, which is specifically used to cluster the target point cloud according to the Calculate the cosine similarity of the distance information, angle information and moving speed of the geometric center points of any two target point clouds in a cluster, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud; clustering module, specific It is used to add any two target point clouds whose cosine similarity is less than the similarity threshold into the second cluster.
  • a calculation module which is specifically used to cluster the target point cloud according to the Calculate the cosine similarity of the distance information, angle information and moving speed of the geometric center points of any two target point clouds in a cluster, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud
  • clustering module specific It is used to add any two target point clouds whose cosine similarity is less than the similarity threshold into the second cluster.
  • a radar point cloud clustering device which includes a processor, a memory, and a millimeter wave radar, the processor, the memory and the millimeter wave radar are coupled, the memory is used to store computer programs, and the processor is used to execute the data stored in the memory A computer program to cause the radar point cloud clustering device to execute the method according to the first aspect and any one of its implementations, or to execute the method according to the second aspect and any one of its implementations.
  • a computer-readable storage medium is provided, and a computer program is stored in the computer-readable storage medium.
  • a computer program is stored in the computer-readable storage medium.
  • a computer program product containing instructions is provided.
  • the instructions run on a computer or a processor, the method described in the first aspect and any one of the embodiments is executed, or, as in the second aspect and The method described in any one of its implementations is executed.
  • FIG. 1 is a schematic diagram of the working principle of a millimeter wave radar provided by an embodiment of the application;
  • FIG. 2 is a schematic diagram of clustering point clouds according to an embodiment of the application
  • FIG. 3 is a schematic flowchart 1 of a radar point cloud clustering method provided by an embodiment of this application;
  • FIG. 4 is a schematic diagram of distance information, angle information, and radial velocity of a point cloud provided by an embodiment of this application;
  • FIG. 5 is another schematic diagram of clustering point clouds according to an embodiment of the application.
  • FIG. 6 is a schematic diagram 2 of a flow chart of a radar point cloud clustering method provided by an embodiment of this application;
  • FIG. 7 is a third schematic flowchart of a radar point cloud clustering method provided by an embodiment of this application.
  • FIG. 8 is a schematic diagram of a step angle and a candidate angle provided by an embodiment of the application.
  • FIG. 9 is a fourth schematic flowchart of a radar point cloud clustering method provided by an embodiment of this application.
  • FIG. 10 is another schematic diagram of clustering point clouds according to an embodiment of the application.
  • FIG. 11 is a schematic structural diagram of a radar point cloud clustering device provided by an embodiment of the application.
  • FIG. 12 is a schematic structural diagram of another radar point cloud clustering device provided by an embodiment of this application.
  • Millimeter wave radar is a radar that works in the millimeter wave band (millimeter wave) detection.
  • millimeter wave refers to the frequency domain of 30 to 300 GHz (wavelength is 1 to 10 mm).
  • the working principle of millimeter-wave radar is to transmit millimeter-wave detection signals to the target, and then receive the reflected signal reflected from the target.
  • the distance information of the target relative to the radar can be obtained.
  • the position of the array antenna at the receiver the direction of arrival (DOA) of the received signal is estimated, and the angle information of the target relative to the radar can be obtained.
  • the Doppler frequency shift of the received signal the radial velocity of the target relative to the radar can be obtained, and so on.
  • the distance information, angle information, and radial velocity of the target corresponding to each reflected signal relative to the radar are taken as a data point (referred to as "point” in the embodiment of this application), and the set obtained by screening the points with higher energy is the point cloud .
  • point By clustering the points in the point cloud data and obtaining the centroids belonging to the same point cloud, it can be used to represent the target corresponding to the point cloud, and the centroid can be further used for target tracking and trajectory detection of the target.
  • the DBSCAN algorithm clusters point clouds only based on the point cloud density within a certain distance.
  • the point clouds of multiple targets will be clustered into one class. Partial occlusion or different materials at different locations will cause the point cloud of one target to be clustered into point clouds of multiple targets. If such a clustering result is used for track detection, it will cause multiple tracks for the same target; if it is used for target tracking, it will cause a sudden change in the centroid of the point cloud, which will seriously affect the tracking performance.
  • a large target is partially blocked or due to different materials at different locations, resulting in three point clouds. Due to the large distance between the three point clouds, the three point clouds are not The point cloud can be clustered into a target, and three centroids are obtained. As shown in B in Fig. 2, three smaller targets are recognized as a point cloud due to their close distance, and a centroid is obtained.
  • the millimeter wave radar point cloud clustering method and device provided by the embodiments of the present application not only consider the close distance between points, but also consider the small difference in radial velocity.
  • Such point clustering is a point cloud of a target, which can distinguish Multiple targets with similar distances and different radial velocities.
  • the moving speed of the point belonging to the same point cloud is obtained by the radial velocity of the point.
  • the radar point cloud clustering method provided by the embodiment of the present application includes S301-S303:
  • M is an integer greater than or equal to 2.
  • the distance information ⁇ and the angle information ⁇ are spherical coordinates
  • the moving speed v of the target relative to the radar is a vector
  • the angle between it relative to the line between the target and the radar is ⁇
  • the moving speed v is between the target and the radar.
  • the angle information ⁇ may include an azimuth angle ⁇
  • the angle information may also include a pitch angle ⁇ .
  • the rectangular coordinates of the point can be obtained according to the angle information and distance information of the point.
  • the Euclidean distance d ij between two points is shown in formula 1:
  • the Euclidean distance matrix can be obtained as shown in Table 1.
  • d ij d ji .
  • the radial velocity difference matrix can be obtained as shown in Table 2.
  • ⁇ v ij ⁇ v ji .
  • N is an integer greater than 1.
  • the point clouds of the same target are similar in distance, but if only the points that are close in distance (Euclidean distance is less than the first distance threshold) are clustered into a target point cloud, the point clouds of different targets that are close to each other will also be clustered.
  • the cluster is a target point cloud.
  • Clustering M points can distinguish point clouds of different targets with similar distances.
  • step S303 describes how to cluster a target point cloud. For the points that do not belong to the target point cloud among the M points, clustering can be performed again according to step S303.
  • the Euclidean distance between any two points in the M points of the point cloud data is calculated according to the angle information and distance information of any two points relative to the radar;
  • the difference between any two points relative to the radial velocity of the radar is obtained, and the radial velocity difference between any two points is obtained;
  • the Euclidean distance is less than the first distance threshold, and the radial velocity difference is less than the first velocity threshold
  • This application is based on the principle that the point clouds of different targets with similar distances are similar in Euclidean distance, but the radial velocities of the corresponding point clouds are also different due to the different moving speeds of different targets.
  • the Euclidean distance in the point cloud data is similar and the radial Points with a smaller speed difference are clustered into a point cloud of a target, so that the point clouds of different targets with similar distances can be distinguished.
  • any target point cloud among N target point clouds it is assumed that the point belonging to any target point cloud among any two points is the first point. If any two points are The second point does not belong to any target point cloud, and the Euclidean distance between the second point and the first point is less than the first distance threshold and the radial velocity difference is less than the first speed threshold, then the second point Join any target point cloud.
  • any target point cloud of N target point clouds if you select a point, then you can find at least one point from any target point cloud, and the Euclidean distance from the selected point is less than the first distance threshold And the radial velocity difference is less than the first velocity threshold.
  • Each of the M points belongs to at most one target point cloud in the N target point clouds, that is, there may be points that do not belong to any target point cloud, and such points are identified as noise. But the points in the N target point cloud all belong to M points.
  • cluster the first target point cloud select a point from M points sequentially or randomly as the first point of the first target point cloud. After adding the second point of the remaining M-1 points that meets the above conditions into the first target point cloud, it becomes the first point of the first target point cloud. As the second point is continuously added to the first target point Cloud, the number of second points that meet the above conditions is gradually reduced until it is zero. At this time, the clustering of the first target point cloud is completed. Assume that there are A points in the first target point cloud.
  • cluster the second target point cloud select a point from the remaining MA points in sequence or at random as the first point of the second target point, and then continuously collect the second point that meets the above conditions in the above manner The second target point cloud is added until the number of second points meeting the above conditions is zero, at which point the clustering of the second target point cloud is completed. And so on.
  • the number of second points is not less than the first number threshold, which can prevent a small number of discrete points from being clustered into a target point cloud.
  • the first number threshold is 2
  • a point for example, point a
  • the Euclidean distance between point b and point a is less than the first distance threshold and the radial velocity difference is less than the first speed threshold
  • the Euclidean distance between point c and point a is less than the first distance threshold
  • the radial velocity difference is less than the first velocity threshold
  • the number is 2 not less than the first number threshold, so point b and point c are added to the first target point cloud as the second points.
  • the Euclidean distance between point d and point a is less than the first distance threshold, the radial velocity difference between point d and point a is not less than the first speed threshold. Therefore, point d cannot be regarded as the second point.
  • Join the first target point cloud The Euclidean distance between point e, point f, point g and point a is not less than the first distance threshold, so point e, point f, and point g cannot be used as the second point, that is, the first target point cloud cannot be added.
  • point b in the first target point cloud is taken as the first point.
  • the Euclidean distance between point d, point e, point f and point b is less than the first distance threshold
  • point d, point e, point f and point The radial velocity difference between b is not less than the first velocity threshold, so point d, point e, and point f cannot be used as the second point, that is, the first target point cloud cannot be added.
  • the Euclidean distance between point g and point b is not less than the first distance threshold, so point g cannot be used as the second point, that is, it cannot be added to the first target point cloud.
  • point c in the first target point cloud is traversed as the first point.
  • point d, point e, point f, and point g cannot be used as the second point, that is, they cannot be added to the first target point cloud.
  • the first target point cloud obtained by the final clustering includes point a, point b, and point c.
  • a point for example, point d
  • point d is the first point first
  • the Euclidean distance between point e and point d is less than the first distance threshold and the radial velocity difference is less than the first speed threshold
  • point f and point d The Euclidean distance of is less than the first distance threshold and the radial velocity difference is less than the first speed threshold, and the number is 2 not less than the first number threshold, so point e and point f are added as second points to the second target point cloud.
  • the Euclidean distance between point g and point d is not less than the first distance threshold, so point g cannot be used as the second point, that is, the second target point cloud cannot be added.
  • point e and point f in the second target point cloud are taken as the first point in turn.
  • the Euclidean distance between point g and point e or point g and point f is not less than the first distance threshold, so the point cannot be g is the second point, that is, the second target point cloud cannot be added.
  • Points that do not belong to any target point cloud will be identified as noise.
  • centroid can be obtained to indicate the target corresponding to the point cloud, and the centroid can further be used for target tracking, trajectory detection, etc., for the target.
  • different point clouds may be displayed on the display screen, for example, points belonging to the same point cloud are included in the dotted frame.
  • the points belonging to different point clouds can be distinguished by using different colors or display modes.
  • the points of one point cloud are circular, and the points of the other point cloud are triangular.
  • the centroid of each point cloud can also be displayed.
  • the first quantity threshold, the first distance threshold, and the first speed threshold satisfy at least one of the following conditions: the greater the distance information of the first point or the second point, the larger the corresponding first distance threshold; the first point Or the greater the distance information of the second point, the greater the corresponding first speed threshold; the greater the distance information of the first point or the second point, the smaller the corresponding first quantity threshold.
  • the reason is that the farther the target is from the radar, the smaller the signal-to-noise ratio of the point, and the sparser the point cloud density.
  • the above-mentioned changes can be adapted to the above-mentioned changes by changing the above-mentioned threshold hierarchically, so that points belonging to the same target will not be missed during clustering.
  • target point clouds After clustering the points in the point cloud data according to the above method to obtain multiple target point clouds, or after clustering the points in the point cloud data according to other methods in the prior art to obtain multiple target point clouds, return These target point clouds can be further clustered by combining the moving speed of each point cloud to obtain a cluster. Because a target is partially occluded or materials are different in different locations, the point cloud of a target is clustered into multiple target point clouds. However, the difference in the moving speed of these target point clouds is very small. Therefore, the target point clouds can be clustered into a cluster of target point clouds according to the difference in the moving speed of the target point clouds. It can solve the problem that the point cloud of a target is clustered into point clouds of multiple targets when a target is partially occluded or different materials at different locations.
  • an embodiment of the present application provides another radar point cloud clustering method, including S601-S602:
  • S601 Calculate the respective moving speeds of the N target point clouds according to the angle information and the radial velocity of all points of each target point cloud in the N target point clouds relative to the radar.
  • N is a positive integer.
  • the radial velocities at both ends may be different, but the overall moving speed of the target is consistent.
  • This application can estimate the moving speed of the target point cloud based on the minimum mean square error iteration.
  • the moving speed of the target point cloud can also be obtained by other methods, which is not limited in this application.
  • step S601 includes S6011-S6012:
  • T and K are positive integers.
  • the angles can be traversed in K step angles I.e. in the interval Within, select K step angles at equal intervals.
  • step S6011 includes S60111-S60112:
  • K candidate angles represent K moving directions of the t-th point, 1 ⁇ t ⁇ T, and t is a positive integer.
  • ⁇ t is the azimuth angle of the t-th point. 1 ⁇ k ⁇ K, and k is an integer.
  • S60112 Back-project the radial velocity of the t-th point to K candidate angles of the t-th point, respectively, to obtain the moving speed of the t-th point corresponding to the K step angles.
  • step S6012 includes S60121-S60124:
  • the target step angle is obtained according to formula 7
  • S60124 Average the moving speed of the T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • Step angle of target Substituting the azimuth angle ⁇ t of the t-th point into formula 4, the moving speed of the t-th point in the T points corresponding to the target step angle can be obtained, and the target step angle Substituting the azimuth angles of the T points into formula 4 can obtain the moving speeds of the T points corresponding to the target step angle. Taking the average of these moving speeds, that is, substituting into formula 5, the moving speed of any target point cloud can be obtained.
  • Target step angle The back projection of the radial velocity of the corresponding point to the corresponding candidate angle has the smallest deviation from the true velocity.
  • N geometric center points corresponding to the N target point clouds can be obtained. That is, each target point cloud corresponds to a geometric center point.
  • n the nth target point cloud
  • D n the nth target point cloud
  • n the nth target point cloud
  • n the number of points in the target point cloud
  • T and t are integers.
  • the Euclidean distance between any two geometric center points can be calculated according to the angle information and distance information of any two geometric center points relative to the radar. That is, according to formula 1 or formula 2, the Euclidean distance between the geometric center points of any two target point clouds in the N target point clouds can be obtained.
  • the radial velocity difference between any two geometric center points can be obtained by making a difference between the radial velocity of any two geometric center points relative to the radar. That is, according to formula 3, the radial velocity difference between the geometric center points of any two point clouds in the N target point clouds can be obtained.
  • the geometric center point of any target point cloud belonging to the first cluster among the N target point clouds is the first geometric center point
  • the target point cloud corresponding to the second geometric point does not belong to the first cluster
  • the Euclidean distance between the second geometric point and the first geometric center point is less than the second distance threshold and the radial velocity If the difference is less than the second speed threshold, the target point cloud corresponding to the second geometric point is added to the first cluster.
  • the second distance threshold is greater than the first distance threshold, and the second speed threshold is greater than the first speed threshold.
  • S604 Perform clustering on the target point cloud whose moving speed meets the condition in the first cluster to obtain a second cluster.
  • i and j are integers smaller than the number of target point clouds in the first cluster, and i ⁇ j.
  • the cosine similarity can be calculated according to the distance information, angle information, and moving speed of the geometric center points of any two target point clouds belonging to the first cluster.
  • the any two target point clouds whose cosine similarity is less than the similarity threshold are added to the second cluster.
  • is the distance information of the geometric center point
  • is the angle information of the geometric center point
  • v is the moving speed of the geometric center point
  • the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud.
  • the radar point cloud clustering method provided by the embodiments of the present application, based on the principle that the point cloud of the same target has a small difference in moving speed, by dividing the target point cloud with a similar distance, a small radial velocity difference, and a small moving speed difference
  • the clustering is a cluster of a target, which can prevent a target from being partially occluded or different materials at different locations, which may cause the point cloud of one target to be clustered into point clouds of multiple targets.
  • centroid can be obtained to indicate the target corresponding to the cluster, and the centroid can be further used for target tracking, trajectory detection, etc., for the target.
  • different clusters may be displayed on the display screen, for example, the point clouds belonging to the same cluster are included in the dotted frame.
  • the point clouds belonging to different clusters can be distinguished by using different colors or display modes.
  • the point cloud of one cluster adopts a circle
  • the point cloud of another cluster adopts a triangle.
  • centroid of each cluster can also be displayed.
  • a larger target is partially occluded or different materials at different positions lead to three point clouds.
  • the distance between the three point clouds is relatively large, the three points The moving speeds of the clouds are similar, so that the three point clouds can be clustered into a target cluster, so that a centroid can be obtained.
  • the three smaller targets are relatively close, their moving speeds are all different and cannot be recognized as a cluster, so that three centroids can be obtained.
  • the embodiment of the application provides a radar point cloud clustering device, which can be installed in smart cars, drones, cars, trucks, motorcycles, buses, boats, airplanes, helicopters, lawn mowers, recreational vehicles, Amusement park vehicles, construction equipment, trams, golf carts, trains and trolleys, etc., are not limited in this application.
  • the radar point cloud clustering device is used to implement the above radar point cloud clustering method.
  • the radar point cloud clustering device may be divided into functional modules according to the foregoing method embodiments.
  • each functional module may be divided corresponding to each function, or two or more functions may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of hardware or software functional modules. It should be noted that the division of modules in the embodiments of the present application is illustrative, and is only a logical function division, and there may be other division methods in actual implementation.
  • FIG. 11 shows a schematic structural diagram of a radar point cloud clustering device 110.
  • the radar point cloud clustering device 110 includes a calculation module 1101 and a clustering module 1102.
  • the calculation module 1101 may also be referred to as a calculation unit, and is used to implement the calculation function in the foregoing method embodiment. For example, steps S301-S302 in FIG. 3, steps S601-S602 in FIG. 6, steps S6011, S6012, S602 in FIG. 7, and steps S6011, S60121-S60124, and S602 in FIG. 9 are executed.
  • the clustering module 1102, which may also be referred to as a clustering unit, is used to implement the clustering function in the foregoing method embodiment. For example, step S303 in Fig. 3 and steps S603-S604 in Fig. 6, Fig. 7 and Fig. 9 are executed.
  • the calculation module 1101 is configured to calculate the Euclidean distance between any two points according to the angle information and distance information of any two points in the M points of the point cloud data relative to the radar; Among them, M is an integer greater than or equal to 2; the calculation module 1101 is also used to make the difference between the radial velocity of any two points relative to the radar to obtain the radial velocity difference between any two points; the clustering module 1102 uses To cluster any two points that satisfy the Euclidean distance less than the first distance threshold and the radial velocity difference is less than the first velocity threshold to obtain N target point clouds; where N is an integer greater than 1.
  • the calculation module 1101 is further configured to calculate the respective corresponding N target point clouds according to the angle information and radial velocity of all points of each target point cloud in the N target point clouds relative to the radar.
  • the calculation module 1101 is also used to calculate any two of the N geometric center points relative to the radar angle information, distance information, and radial velocity difference of the N geometric center points corresponding to the N target point clouds.
  • the calculation module 1101 is further configured to calculate the respective corresponding N target point clouds according to the angle information and radial velocity of all points of each target point cloud in the N target point clouds relative to the radar.
  • the moving speed of including: calculation module 1101, specifically used for any target point cloud in N target point clouds, according to K step angles, angle information of all T points of any target point cloud relative to the radar And radial velocity, calculate the movement speed of T points corresponding to K step angles; where K is a positive integer; the calculation module 1101 is specifically used for the movement speeds corresponding to K step angles of T points , Calculate the moving speed of any target point cloud.
  • the calculation module 1101 calculates the K step angles and the K step angles of the T points according to the K step angles, the angle information and the radial velocity of all T points of any target point cloud relative to the radar.
  • the moving speed corresponding to the angle includes: the calculation module 1101, which is specifically used for subtracting the K step angles from the angle information of the t-th point for the t-th point among the T points to obtain the t-th point K candidate angles; K candidate angles represent the K moving directions of the t-th point, 1 ⁇ t ⁇ T, and t is a positive integer; the calculation module 1101 is specifically used to change the radial velocity of the t-th point to the t-th point The K candidate angles of each point are respectively back-projected to obtain the moving speed of the t-th point corresponding to the K step angles.
  • the calculation module 1101 calculates the movement speed of any target point cloud according to the movement speeds of the T points corresponding to the K step angles, including: a calculation module 1101, which is specifically used to calculate the movement speed of any target point cloud For the k-th step angle in the step angle, average the moving speeds of the T points corresponding to the k-th step angle to obtain the average value; 1 ⁇ k ⁇ K, and k is an integer; calculation module 1101, specific It is used to calculate the mean square error between the moving speed corresponding to the k-th step angle of T points and the mean value; the calculation module 1101 is specifically used to calculate the target step angle according to the smallest mean square error; the calculation module 1101 is specifically used Take the average of the moving speeds of T points corresponding to the target step angle to obtain the moving speed of any target point cloud.
  • a calculation module 1101 which is specifically used to calculate the movement speed of any target point cloud For the k-th step angle in the step angle, average the moving speeds of the T points corresponding to the k-th step angle to obtain
  • the calculation module 1101 calculates any two of the N geometric center points relative to the radar angle information, distance information, and radial velocity difference corresponding to the N target point clouds.
  • the Euclidean distance and radial velocity difference between two geometric center points including: calculation module 1101, specifically used to calculate the angle information and distance information between any two geometric center points relative to the radar, and calculate the difference between any two geometric center points The Euclidean distance between any two points; the calculation module 1101 is specifically used to make the difference between the radial velocity of any two geometric center points relative to the radar, and obtain the radial velocity difference between any two geometric center points.
  • the clustering module 1102 is also used to determine the Euclidean distance between the geometric center points in the N target point cloud is less than the second distance threshold, and the radial velocity difference is less than the second velocity.
  • the threshold target point cloud is clustered to obtain the first cluster, including: a clustering module 1102, specifically used to cluster the target point cloud corresponding to the second geometric center point among the N target point clouds that do not belong to the first cluster, Join the first cluster, where the Euclidean distance between the second geometric center point and the first geometric center point is less than the second distance threshold and the radial velocity difference is less than the second speed threshold, and the first geometric center point is N target points The geometric center point of any target point cloud belonging to the first cluster in the cloud.
  • the clustering module 1102 is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including the clustering module 1102, which is specifically used for Add any two target point clouds that belong to the first cluster and whose moving speed difference is less than the third speed threshold into the second cluster.
  • the clustering module 1102 is also used to cluster the target point cloud whose moving speed meets the condition in the first cluster to obtain the second cluster, including: a calculation module 1101, which is specifically used to The distance information, angle information and moving speed of the geometric center points of any two target point clouds belonging to the first cluster are calculated for cosine similarity, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud; clustering module 1102, specifically configured to add any two target point clouds whose cosine similarity is less than the similarity threshold to the second cluster.
  • a calculation module 1101 which is specifically used to The distance information, angle information and moving speed of the geometric center points of any two target point clouds belonging to the first cluster are calculated for cosine similarity, where the moving speed of the geometric center point is equal to the moving speed of the corresponding target point cloud
  • clustering module 1102 specifically configured to add any two target point clouds whose cosine similarity is less than the similarity threshold to the second cluster.
  • the clustering module 1102 is configured to cluster any two points where the Euclidean distance is less than the first distance threshold, and the radial velocity difference is less than the first velocity threshold, to obtain N targets
  • the point cloud includes: a clustering module 1102, which is specifically used to classify any two points that do not belong to any target point cloud in any target point cloud among the N target point clouds, and those between the first point and the first point.
  • the second point whose Euclidean distance is less than the first distance threshold and the radial velocity difference is less than the first speed threshold is added to any target point cloud; where the first point is a point belonging to any target point cloud among any two points.
  • the first distance threshold and the first speed threshold satisfy at least one of the following conditions: the greater the distance information of the first point or the second point, the larger the corresponding first distance threshold; the first point Or the greater the distance information of the second point, the greater the corresponding first speed threshold.
  • the radar point cloud clustering device 110 is presented in the form of dividing various functional modules in an integrated manner.
  • the "module” here can refer to a specific ASIC, circuit, processor and memory that executes one or more software or firmware programs, integrated logic circuit, and/or other devices that can provide the above-mentioned functions.
  • the radar point cloud clustering device 110 provided in this embodiment can perform the above-mentioned method, the technical effects that can be obtained can refer to the above-mentioned method embodiment, which will not be repeated here.
  • an embodiment of the present application also provides a radar point cloud clustering device.
  • the radar point cloud clustering device 120 includes a processor 1201, a memory 1202, and a millimeter wave radar 1203.
  • the processor 1201, the memory 1202, and the The millimeter wave radar 1203 is coupled, and when the processor 1201 executes the computer program or instruction in the memory 1202, the method in FIG. 3, FIG. 6, FIG. 7 or FIG. 9 is executed.
  • the processor 1201 can perform the functions of the calculation module 1101 and the clustering module 1102 in FIG. 11.
  • the embodiment of the present application also provides a computer-readable storage medium in which a computer program is stored, and when it runs on a computer or a processor, as shown in Fig. 3, Fig. 6, Fig. 7 or Fig. 9 The method is executed.
  • the embodiment of the present application also provides a computer program product containing instructions.
  • the instructions run on a computer or a processor, the method in FIG. 3, FIG. 6, FIG. 7 or FIG. 9 is executed.
  • the embodiment of the present application provides a chip system, which includes a processor, which is used for the radar point cloud clustering apparatus to execute the method in FIG. 3, FIG. 6, FIG. 7 or FIG. 9.
  • the chip system also includes a memory for storing necessary program instructions and data.
  • the chip system may include a chip, an integrated circuit, or may include a chip and other discrete devices, which is not specifically limited in the embodiment of the present application.
  • the radar point cloud clustering device, chip, computer storage medium, computer program product, or chip system provided in this application are all used to execute the above-mentioned method. Therefore, the beneficial effects that can be achieved can refer to the above-mentioned The beneficial effects in the provided implementation manners are not repeated here.
  • the processor involved in the embodiment of the present application may be a chip.
  • it can be a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on chip (SoC), or a central processing unit.
  • the central processor unit (CPU) can also be a network processor (NP), a digital signal processing circuit (digital signal processor, DSP), or a microcontroller (microcontroller unit, MCU) It can also be a programmable logic device (PLD) or other integrated chips.
  • NP network processor
  • DSP digital signal processor
  • MCU microcontroller unit
  • PLD programmable logic device
  • the memory involved in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory.
  • the non-volatile memory can be read-only memory (ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), and electrically available Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory.
  • the volatile memory may be random access memory (RAM), which is used as an external cache.
  • RAM random access memory
  • static random access memory static random access memory
  • dynamic RAM dynamic RAM
  • DRAM dynamic random access memory
  • synchronous dynamic random access memory synchronous DRAM, SDRAM
  • double data rate synchronous dynamic random access memory double data rate SDRAM, DDR SDRAM
  • enhanced synchronous dynamic random access memory enhanced SDRAM, ESDRAM
  • synchronous connection dynamic random access memory serial DRAM, SLDRAM
  • direct rambus RAM direct rambus RAM
  • the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not correspond to the embodiments of the present application.
  • the implementation process constitutes any limitation.
  • the disclosed system, device, and method may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection between devices or units through some interfaces, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer or includes one or more data storage devices such as servers, data centers, etc. that can be integrated with the medium.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, and a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)).

Landscapes

  • Engineering & Computer Science (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

本申请公开了一种雷达点云聚类方法和装置,涉及雷达成像领域,用于区分距离相近的不同目标的点云。雷达点云聚类方法,包括:根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算任意两个点之间的欧氏距离;其中,M为大于等于2整数;对任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。

Description

雷达点云聚类方法和装置 技术领域
本申请涉及雷达成像领域,尤其涉及一种雷达点云聚类方法和装置。
背景技术
随着高级驾驶辅助系统(advanced driver assistance systems,ADAS)和无人驾驶技术的发展,对车载雷达的探测距离、角度分辨率等性能提出了更高的要求。目前的车载雷达(例如毫米波雷达)在对一个目标进行成像时会检测出多个测量点,形成高分辨率的目标点云,这种车载雷达也称之为点云成像雷达。
毫米波雷达对于同一目标进行测量会形成点云,为了对同一时刻检测的多个目标进行目标区分和状态估计,首先需要通过聚类方法对点云按照不同目标进行聚类。传统的带噪声的基于密度的聚类(density-based spatial clustering of applications with noise,DBSCAN)算法,是基于一定距离范围内的点云密度来对点云进行聚类的,当多个目标距离相近时会出现多个目标的点云被聚成一个类的情况。
发明内容
本申请实施例提供一种雷达点云聚类方法和装置,用于区分距离相近的不同目标的点云。
为达到上述目的,本申请的实施例采用如下技术方案:
第一方面,提供了一种雷达点云聚类方法,包括:根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算任意两个点之间的欧氏距离;其中,M为大于等于2整数;对任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。
本申请实施例提供的雷达点云聚类方法,根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算该任意两个点之间的欧氏距离;对该任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云。如果仅将距离相近的点聚类为一个目标的点云,会将距离相近的不同目标的点云聚类为一个目标的点云。本申请基于距离相近的不同目标的点云虽然欧氏距离相近,但是由于不同目标的运动速度不同导致对应的点云的径向速度也不同的原理,将点云数据中欧氏距离相近并且径向速度差较小的点聚类为一个目标的点云,从而能够区分距离相近的不同目标的点云。
在一种可能的实施方式中,还包括:根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度;根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小 于第二速度门限的目标点云进行聚类,得到第一集群;对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
在一种可能的实施方式中,根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度,包括:针对N个目标点云中的任一目标点云,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度;其中,K为正整数;根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度。
在一种可能的实施方式中,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度,包括:针对T个点中的第t个点,将K个步进角度分别与第t个点的角度信息相减,得到第t个点的K个候选角度;K个候选角度表示第t个点的K个移动方向,1≤t≤T,t为正整数;将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t个点的与K个步进角度对应的移动速度。
在一种可能的实施方式中,根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度,包括:针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;计算T个点的与第k个步进角度对应的移动速度与均值之间的均方差;根据最小的均方差计算目标步进角度;对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。此时目标步进角度对应的点的径向速度向对应的候选角度进行的逆投影,与真实速度的偏差最小。
在一种可能的实施方式中,根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离;对任意两个几何中心点相对于雷达的径向速度作差,得到任意两个几何中心点之间的径向速度差。
在一种可能的实施方式中,对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:将N个目标点云中不属于第一集群的,与第二几何中心点对应的目标点云,加入第一集群,其中,第二几何中心点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,第一几何中心点为N个目标点云中属于第一集群的任一目标点云的几何中心点。第二距离门限大于第一距离门限,第二速度门限大于第一速度门限。原因在于,与点云中的点之间的欧氏距离和径向速度差相比,点云之间的欧氏距离和径向速度差都会更大。
在一种可能的实施方式中,对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:将属于第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入第二集群。由于第二集群中的目标点云之间满足距离相近、径向速度差较小,所以根据移动速度满足条件对第二集群进一步进行聚类,则可以将距离相近、径向速度差较小,并且移动速度差别较小的目标点云聚类为一个目标的集群。
在一种可能的实施方式中,对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,几何中心点的移动速度等于对应的目标点云的移动速度;将余弦相似度小于相似度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,对欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云,包括:针对N个目标点云中任一目标点云,将任意两个点中不属于任一目标点云的,并且,与第一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限的第二点,加入任一目标点云;其中,第一点为任意两个点中属于任一目标点云的点。也就是说,对于N个目标点云中任一目标点云,选取一点,那么都可以从该任一目标点云中找到至少一点,与选取的一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限。
在一种可能的实施方式中,第一距离门限和第一速度门限满足以下条件中至少一个:第一点或第二点的距离信息越大,对应的第一距离门限越大;第一点或第二点的距离信息越大,对应的第一速度门限越大。原因在于,目标距离雷达越远,点的信噪比越小,点云密度越稀疏,通过分级改变上述门限可以适应上述变化,使得聚类时不会漏掉属于同一目标的点。
第二方面,提供了一种雷达点云聚类方法,包括:根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度;根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
本申请实施例提供的雷达点云聚类方法,根据同一目标的点云的移动速度差别较小的原理,通过将距离相近、径向速度差较小,并且移动速度差别较小的目标点云聚类为一个目标的集群,从而能够防止一个目标受到部分遮挡或不同位置材料不同而导致一个目标的点云被聚类为多个目标的点云。
在一种可能的实施方式中,根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度,包括:针对N个目标点云中的任一目标点云,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度;其中,K为正整数;根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度。
在一种可能的实施方式中,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度,包括:针对T个点中的第t个点,将K个步进角度分别与第t个点的角度信息相减,得到第t个点的K个候选角度;K个候选角度表示第t个点的K个移动方向,1≤t≤T,t为正整数;将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t 个点的与K个步进角度对应的移动速度。
在一种可能的实施方式中,根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度,包括:针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;计算T个点的与第k个步进角度对应的移动速度与均值之间的均方差;根据最小的均方差计算目标步进角度;对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。
在一种可能的实施方式中,根据N个目标点云的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离;对任意两个几何中心点相对于雷达的径向速度作差,得到任意两个几何中心点之间的径向速度差。
在一种可能的实施方式中,对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:将N个目标点云中不属于第一集群的,与第二几何中心点对应的目标点云,加入第一集群,其中,第二几何中心点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,第一几何中心点为N个目标点云中属于第一集群的任一目标点云的几何中心点。第二距离门限大于第一距离门限,第二速度门限大于第一速度门限。原因在于,与点云中的点之间的欧氏距离和径向速度差相比,点云之间的欧氏距离和径向速度差都会更大。
在一种可能的实施方式中,对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:将属于第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入第二集群。由于第二集群中的目标点云之间满足距离相近、径向速度差较小,所以根据移动速度满足条件对第二集群进一步进行聚类,则可以将距离相近、径向速度差较小,并且移动速度差别较小的目标点云聚类为一个目标的集群。
在一种可能的实施方式中,对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,几何中心点的移动速度等于对应的目标点云的移动速度;将余弦相似度小于相似度门限的任意两个目标点云加入第二集群。
第三方面,提供了一种雷达点云聚类装置,包括:计算模块,用于根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算任意两个点之间的欧氏距离;其中,M为大于等于2整数;计算模块,还用于对任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;聚类模块,用于对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。
在一种可能的实施方式中,计算模块,还用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度;计算模块,还用于根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信 息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;聚类模块,还用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
在一种可能的实施方式中,计算模块,还用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度,包括:计算模块,具体用于针对N个目标点云中的任一目标点云,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度;其中,K为正整数;计算模块,具体用于根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度。
在一种可能的实施方式中,计算模块,具体用于根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度,包括:计算模块,具体用于针对T个点中的第t个点,将K个步进角度分别与第t个点的角度信息相减,得到第t个点的K个候选角度;K个候选角度表示第t个点的K个移动方向,1≤t≤T,t为正整数;计算模块,具体用于将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t个点的与K个步进角度对应的移动速度。
在一种可能的实施方式中,计算模块,具体用于根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度,包括:计算模块,具体用于针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;计算模块,具体用于计算T个点的与第k个步进角度对应的移动速度与均值之间的均方差;计算模块,具体用于根据最小的均方差计算目标步进角度;计算模块,具体用于对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。
在一种可能的实施方式中,计算模块根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:计算模块,具体用于根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离;计算模块,具体用于对任意两个几何中心点相对于雷达的径向速度作差,得到任意两个几何中心点之间的径向速度差。
在一种可能的实施方式中,聚类模块,还用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:聚类模块,具体用于将N个目标点云中不属于第一集群的,与第二几何中心点对应的目标点云,加入第一集群,其中,第二几何中心点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,第一几何中心点为N个目标点云中属于第一集群的任一目标点云的几何中心点。
在一种可能的实施方式中,聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:聚类模块,具体用于将属于第一集群的、 移动速度之差小于第三速度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:计算模块,具体用于根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,几何中心点的移动速度等于对应的目标点云的移动速度;聚类模块,具体用于将余弦相似度小于相似度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,聚类模块,用于对欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云,包括:聚类模块,具体用于针对N个目标点云中任一目标点云,将任意两个点中不属于任一目标点云的,并且,与第一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限的第二点,加入任一目标点云;其中,第一点为任意两个点中属于任一目标点云的点。
在一种可能的实施方式中,第一距离门限和第一速度门限满足以下条件中至少一个:第一点或第二点的距离信息越大,对应的第一距离门限越大;第一点或第二点的距离信息越大,对应的第一速度门限越大。
第四方面,提供了一种雷达点云聚类装置,包括:计算模块,用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度;计算模块,还用于根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;聚类模块,用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
在一种可能的实施方式中,计算模块,用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度,包括:计算模块,具体用于针对N个目标点云中的任一目标点云,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度;其中,K为正整数;根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度。
在一种可能的实施方式中,计算模块,具体用于根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度,包括:计算模块,具体用于针对T个点中的第t个点,将K个步进角度分别与第t个点的角度信息相减,得到第t个点的K个候选角度;K个候选角度表示第t个点的K个移动方向,1≤t≤T,t为正整数;计算模块,具体用于将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t个点的与K个步进角度对应的移动速度。
在一种可能的实施方式中,计算模块,具体用于根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度,包括:计算模块,具体用于针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均, 得到均值;1≤k≤K,且k为整数;计算模块,具体用于计算T个点的与第k个步进角度对应的移动速度与均值之间的均方差;计算模块,具体用于根据最小的均方差计算目标步进角度;计算模块,具体用于对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。
在一种可能的实施方式中,计算模块根据N个目标点云的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:计算模块,具体用于根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离;计算模块,具体用于对任意两个几何中心点相对于雷达的径向速度作差,得到任意两个几何中心点之间的径向速度差。
在一种可能的实施方式中,聚类模块,用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:聚类模块,具体用于将N个目标点云中不属于第一集群的,与第二几何中心点对应的目标点云,加入第一集群,其中,第二几何中心点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,第一几何中心点为N个目标点云中属于第一集群的任一目标点云的几何中心点。
在一种可能的实施方式中,聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:聚类模块,具体用于将属于第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,聚类模块,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:计算模块,具体用于根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,几何中心点的移动速度等于对应的目标点云的移动速度;聚类模块,具体用于将余弦相似度小于相似度门限的任意两个目标点云加入第二集群。
第五方面,提供了一种雷达点云聚类装置,包括处理器、存储器和毫米波雷达,处理器、存储器和毫米波雷达耦合,存储器用于存储计算机程序,处理器用于执行存储器中存储的计算机程序,以使得雷达点云聚类装置执行如第一方面及其任一项实施方式所述的方法,或者,执行如第二方面及其任一项实施方式所述的方法。
第六方面,提供了一种计算机可读存储介质,计算机可读存储介质中存储有计算机程序,当其在计算机上运行时,如第一方面及其任一项实施方式所述的方法被执行,或者,如第二方面及其任一项实施方式所述的方法被执行。
第七方面,提供了一种包含指令的计算机程序产品,当指令在计算机或处理器上运行时,如第一方面及任一项实施方式所述的方法被执行,或者,如第二方面及其任一项实施方式所述的方法被执行。
第三方面到第七方面的技术效果参照第一方面至第二方面的内容,在此不再重复。
附图说明
图1为本申请实施例提供的一种毫米波雷达的工作原理的示意图;
图2为本申请实施例提供的一种对点云进行聚类的示意图;
图3为本申请实施例提供的一种雷达点云聚类方法的流程示意图一;
图4为本申请实施例提供的一种点云的距离信息、角度信息和径向速度的示意图;
图5为本申请实施例提供的另一种对点云进行聚类的示意图;
图6为本申请实施例提供的一种雷达点云聚类方法的流程示意图二;
图7为本申请实施例提供的一种雷达点云聚类方法的流程示意图三;
图8为本申请实施例提供的一种步进角度和候选角度的示意图;
图9为本申请实施例提供的一种雷达点云聚类方法的流程示意图四;
图10为本申请实施例提供的又一种对点云进行聚类的示意图;
图11为本申请实施例提供的一种雷达点云聚类装置的结构示意图;
图12为本申请实施例提供的另一种雷达点云聚类装置的结构示意图。
具体实施方式
毫米波雷达,是工作在毫米波波段(millimeter wave)探测的雷达。通常毫米波是指30~300GHz频域(波长为1~10mm)。
如图1所示,毫米波雷达的工作原理是向目标发射毫米波的探测信号,然后接收从目标反射回来的反射信号。根据发射信号与反射信号之间的时间差即可以得到目标相对于雷达的距离信息。根据接收机处的阵列天线的位置对接收信号进行波达方向(Direction Of Arrival,DOA)估计,即可得到目标相对于雷达的角度信息。根据接收信号的多普勒频移可以得到目标相对于雷达的径向速度,等等。将每个反射信号对应的目标相对于雷达的距离信息、角度信息、径向速度作为一个数据点(本申请实施例简称为“点”),筛选能量较高的点得到的集合即为点云。通过对点云数据中的点进行聚类并获取属于同一点云的质心,可以用于表示该点云对应的目标,该质心进一步可以用于对该目标进行目标跟踪、轨迹检测等。
现有技术中,DBSCAN算法仅根据一定距离范围内的点云密度来对点云进行聚类,当多个目标交汇时会出现多个目标的点云聚成一个类的情况,当一个目标受到部分遮挡或不同位置材料不同,会导致一个目标的点云被聚类为多个目标的点云的问题。如果将这样的聚类结果用于航迹检测,会导致同一目标出现多条航迹;如果用于目标跟踪,会导致点云的质心点突变,严重影响跟踪性能。
示例性的,如图2中A所示,一个较大目标由于部分受到遮挡或不同位置材料不同导致出现三个点云,由于三个点云之间距离较大,使得这三个点云未能被聚类成一个目标的点云,从而得到了三个质心。如图2中B所示,三个较小目标由于距离较近,使得被识别成一个点云,从而得到一个质心。
针对上述问题,考虑到对于运动的刚体目标,两端的径向速度可能不同,但是其整体的移动速度具有一致性。因此本申请实施例提供的毫米波雷达点云聚类方法和装置,不仅考虑点与点的距离相近,还考虑径向速度差异小,这样的点聚类为一个目标的点云,从而能够区分距离相近的径向速度不同的多个目标。另外,还通过属于同一点云的点的径向速度得到该点的移动速度,通过将距离较远但是移动速度一致的点云作为一个目标的集群,从而能够防止体积大的目标被识别成多个小目标。
具体的,如图3所示,本申请实施例提供的雷达点云聚类方法,包括S301-S303:
S301、根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息转换为第一直角坐标,计算第一直角坐标计算点云数据的M个点中任意两个点之间的欧 氏距离。
其中,M为大于等于2整数。
前文所述的毫米波雷达在完成信号处理后输出点云数据,如图4所示,每个点记为x=[ρ,φ,v γ],即包括(目标相对于雷达的)距离信息ρ、角度信息φ、径向速度v γ。其中,距离信息ρ、角度信息φ均为球面坐标,目标相对于雷达的移动速度v为矢量,其相对于目标与雷达之间连线的夹角为γ,移动速度v向目标与雷达之间连线的投影为径向速度v γ,v γ=v·cosγ。对于二维坐标来说,角度信息φ可以包括方位角θ,对于三维坐标来说,角度信息还可以包括俯仰角α。
具体的,可以根据点的角度信息和距离信息得到点的直角坐标,对于二维坐标来说,点的直角坐标为x=ρ·cosθ,y=ρ·sinθ,则第i个点与第j个点之间的欧氏距离d ij见公式1:
Figure PCTCN2020087874-appb-000001
其中,1≤i≤M,1≤j≤M,i≠j。
对于三维坐标来说,点的直角坐标为x=ρ·cosθ·cosα,y=ρ·sinθ·cosα,z=ρ·sinα,则第i个点与第j个点之间的欧氏距离d ij见公式2:
Figure PCTCN2020087874-appb-000002
根据点云数据的M个点中任意两个点之间的欧氏距离可以得到欧氏距离矩阵如表1所示。其中d ij=d ji
表1
点编号 1 2 …… j …… M
1   d 12 …… d 1j …… d 1M
2 d 21   …… d 2j …… d 2M
…… …… …… …… …… …… ……
i d i1 d i2 …… d ij …… d iM
…… …… …… …… …… …… ……
M d M1 d M2 …… d Mj ……  
S302、对点云数据的M个点中任意两个点相对于雷达的径向速度作差,得到该任意两个点之间的径向速度差。
具体的,第i个点的径向速度
Figure PCTCN2020087874-appb-000003
第j个点的径向速度
Figure PCTCN2020087874-appb-000004
则第i个点与第j个点之间的径向速度差的1范数(即绝对值)Δv ij见公式3:
Figure PCTCN2020087874-appb-000005
根据M个点中任意两个点之间的径向速度差可以得到径向速度差矩阵如表2所示。其中Δv ij=Δv ji
表2
点编号 1 2 …… j …… M
1   Δv 12 …… Δv 1j …… Δv 1M
2 Δv 21   …… Δv 2j …… Δv 2M
…… …… …… …… …… …… ……
i Δv i1 Δv i2 …… Δv ij …… Δv iM
…… …… …… …… …… …… ……
M Δv M1 Δv M2 …… Δv Mj ……  
S303、对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云。
其中,N为大于1的整数。
同一目标的点云在距离上是相近的,但是如果仅对距离相近(欧氏距离小于第一距离门限)的点聚类为一个目标点云,则会将距离相近的不同目标的点云也聚类为一个目标点云。
根据距离相近的不同目标其移动速度是不同的,测量得到的不同目标的点云的径向速度也不同,但是同一目标的点云的径向速度差别很小,因此,可以结合以上两个条件对M个点进行聚类,就可以区分出距离相近的不同目标的点云。
需要说明的是,步骤S303描述的是针对一个目标点云如何聚类。对于M个点中不属于目标点云的点,可以再次按照步骤S303进行聚类。
本申请实施例提供的雷达点云聚类方法,根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算该任意两个点之间的欧氏距离;对该任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云。如果仅将距离相近的点聚类为一个目标的点云,会将距离相近的不同目标的点云聚类为一个目标的点云。本申请基于距离相近的不同目标的点云虽然欧氏距离相近,但是由于不同目标的运动速度不同导致对应的点云的径向速度也不同的原理,将点云数据中欧氏距离相近并且径向速度差较小的点聚类为一个目标的点云,从而能够区分距离相近的不同目标的点云。
在一种可能的实施方式中,针对N个目标点云中任一目标点云,假设任意两个点中属于该任一目标点云的点为第一点,如果该任意两个点中的第二点不属于该任一目标点云,并且,该第二点与第一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限,则将该第二点加入该任一目标点云。
也就是说,对于N个目标点云中任一目标点云,选取一点,那么都可以从该任一目标点云中找到至少一点,与选取的一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限。
M个点中每个点最多属于N个目标点云中一个目标点云,即可能有的点不属于任何目标点云,这类点被识别为噪声。但是N个目标点云中的点都属于M个点。
目标点云中包括的点的数量以及目标点云的数量是随着聚类过程逐渐增加的。初始时,对第一个目标点云进行聚类:按顺序或随机从M个点中选择一个点作为第一个目标点云的第一点。将剩余M-1个点中满足上述条件的第二点加入第一个目标点云后即变为第一个目标点云的第一点,随着不断将第二点加入第一个目标点云,满足上述条件的第二点数目逐渐减少直到为零,此时即完成了对第一个目标点云的聚类,假设第一个目标点云中的点有A个。
然后,对第二个目标点云进行聚类:按照顺序或随机从剩余M-A个点中选择一个点作为第二个目标点的第一点,然后按照上述方式不断将满足上述条件的第二点加入 第二个目标点云,直到满足上述条件的第二点数目为零,此时即完成了对第二个目标点云的聚类。以此类推。
需要说明的是,第二点的数量不小于第一数量门限,这样可以防止离散的少量点被聚类成一个目标点云。
下面,示例性的对第一个目标点云如何聚类进行描述:
示例性的,如图5所示,假设第一数量门限为2,在对第一个目标点云初始化时按顺序或随机选择一个点(例如点a)加入第一个目标点云,点a首先作为第一点,点b与点a之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限,点c与点a之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限,并且数量为2不小于第一数量门限,因此将点b和点c作为第二点加入第一个目标点云。
点d虽然与点a之间的欧氏距离小于第一距离门限,但是点d与点a之间的径向速度差不小于第一速度门限,因此不能将点d作为第二点,即不能加入第一个目标点云。点e、点f、点g与点a之间的欧氏距离不小于第一距离门限,因此不能将点e、点f、点g作为第二点,即不能加入第一个目标点云。
然后将第一个目标点云中的点b作为第一点,虽然点d、点e、点f与点b的欧氏距离小于第一距离门限,但是点d、点e、点f与点b之间的径向速度差不小于第一速度门限,因此不能将点d、点e、点f作为第二点,即不能加入第一个目标点云。点g与点b之间的欧氏距离不小于第一距离门限,因此不能将点g作为第二点,即不能加入第一个目标点云。
依此类推再遍历第一个目标点云中的点c作为第一点,同理,点d、点e、点f、点g不能作为第二点,即不能加入第一个目标点云。最终聚类得到的第一个目标点云包括点a、点b和点c。
下面示例性的对第二个目标点云如何聚类进行描述:
示例性的,如图5所示,在对第二个目标点云初始化时,按顺序或随机从M个点云中不属于第一个目标点云的点中选择一个点(例如点d)加入第二个目标点云,点d首先作为第一点,点e与点d之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限,点f与点d之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限,并且数量为2不小于第一数量门限,因此将点e和点f作为第二点加入第二个目标点云。点g与点d之间的欧氏距离不小于第一距离门限,因此不能将点g作为第二点,即不能加入第二个目标点云。
然后依次将第二个目标点云中的点e和点f作为第一点,点g与点e或者点g与点f之间的欧氏距离均不小于第一距离门限,因此不能将点g作为第二点,即不能加入第二个目标点云。
对于不属于任何目标点云的点(例如图5中的点g)将识别为噪声。
对于任一点云可以求取其质心,用于表示该点云对应的目标,该质心进一步可以用于对该目标进行目标跟踪、轨迹检测等。
示例性的,如图5所示,可以通过显示屏显示不同的点云,例如通过虚线框将属于同一点云的点包括其中。或者,可以将属于不同点云的点采用不同的颜色或显示方式进行区别,例如一个点云的点采用圆形,另一个点云的点采用三角形。可选的,还 可以显示各个点云的质心。
可选的,第一数量门限、第一距离门限和第一速度门限满足以下条件中至少一个:第一点或第二点的距离信息越大,对应的第一距离门限越大;第一点或第二点的距离信息越大,对应的第一速度门限越大;第一点或第二点的距离信息越大,对应的第一数量门限越小。原因在于,目标距离雷达越远,点的信噪比越小,点云密度越稀疏,通过分级改变上述门限可以适应上述变化,使得聚类时不会漏掉属于同一目标的点。
在按照上述方法对点云数据中的点进行聚类得到多个目标点云后,或者,按照现有技术中其他方法对点云数据中的点进行聚类得到多个目标点云后,还可以结合各个点云的移动速度对这些目标点云进一步聚类得到集群。由于一个目标受到部分遮挡或不同位置材料不同,一个目标的点云被聚类为多个目标点云。但是这些目标点云的移动速度差异很小,因此可以根据目标点云的移动速度的差异性将这些目标点云聚类为一个目标的点云的集群。可以解决一个目标受到部分遮挡或不同位置材料不同会导致一个目标的点云被聚类为多个目标的点云的问题。
可选的,如图6所示,本申请实施例提供了另一种雷达点云聚类方法,包括S601-S602:
S601、根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度。
N为正整数。
对于运动的刚体目标,两端的径向速度可能不同,但是目标整体的移动速度具有一致性。本申请可以基于最小均方误差迭代来对目标点云的移动速度进行估计。还可以通过其他方法得到目标点云的移动速度,本申请不作限定。
具体的,如图7所示,针对N个目标点云中的任一目标点云,步骤S601包括S6011-S6012:
S6011、根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度。
其中,T和K为正整数。
示例性的,如图8所示,以二维坐标为例,可以以K个步进角度的方式遍历角度
Figure PCTCN2020087874-appb-000006
即在区间
Figure PCTCN2020087874-appb-000007
内,选择K个等间隔的步进角度。
具体的,如图9所示,步骤S6011包括S60111-S60112:
S60111、针对T个点中的第t个点,将K个步进角度分别第t个点的角度信息相减,得到第t个点的K个候选角度。
K个候选角度表示所述第t个点的K个移动方向,1≤t≤T,t为正整数。
示例性的,如图8所示,取第k个步进角度
Figure PCTCN2020087874-appb-000008
Figure PCTCN2020087874-appb-000009
则可以遍历K个不同值的候选角度γ。其中,θ t为第t个点的方位角。1≤k≤K,且k为整数。
S60112、将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t个点的与K个步进角度对应的移动速度。
假设任一目标点云的T个点中第t个点的方位角为θ t,第t个点的径向速度为
Figure PCTCN2020087874-appb-000010
则第t个点的与第k个步进角度对应的移动速度
Figure PCTCN2020087874-appb-000011
见公式4:
Figure PCTCN2020087874-appb-000012
S6012、根据T个点的与所述K个步进角度对应的移动速度,计算该任一目标点云的移动速度。
具体的,如图9所示,步骤S6012包括S60121-S60124:
S60121、针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均,得到与第k个步进角度对应的T个点的移动速度的均值。
与第k个步进角度对应的T个点的移动速度的均值
Figure PCTCN2020087874-appb-000013
见公式5:
Figure PCTCN2020087874-appb-000014
S60122、计算T个点的与第k个步进角度对应的移动速度
Figure PCTCN2020087874-appb-000015
与均值
Figure PCTCN2020087874-appb-000016
之间的均方差(即标准差)δ k
该均方差δ k见公式6:
Figure PCTCN2020087874-appb-000017
S60123、根据最小的均方差计算目标步进角度。
即根据公式7得到目标步进角度
Figure PCTCN2020087874-appb-000018
Figure PCTCN2020087874-appb-000019
S60124、对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。
将目标步进角度
Figure PCTCN2020087874-appb-000020
和第t个点的方位角θ t代入公式4可以得到T个点中第t个点的与目标步进角度对应的移动速度,将目标步进角度
Figure PCTCN2020087874-appb-000021
和T个点的方位角代入公式4即可以得到T个点的与目标步进角度对应的移动速度。对这些移动速度取平均,即代入代入公式5,即可得到任一目标点云的移动速度。此时目标步进角度
Figure PCTCN2020087874-appb-000022
对应的点的径向速度向对应的候选角度进行的逆投影,与真实速度的偏差最小。
S602、根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差。
对N个目标点云中每个目标点云的所有点取平均,即可以得到与N个目标点云分别对应的N个几何中心点。即每个目标点云对应一个几何中心点。
假设第n个目标点云记为D n:{x n1,x n2,...,x nT},1≤n≤N,且n为整数。该目标点云中的第t个点记为
Figure PCTCN2020087874-appb-000023
其中,1≤t≤T,T为该目标点云中的点的数目,T和t为整数。
则第n个目标点云的几何中心点
Figure PCTCN2020087874-appb-000024
见公式8:
Figure PCTCN2020087874-appb-000025
与步骤S301类似的,可以根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离。即根据公式1或公式2可以得到N 个目标点云中任意两个目标点云的几何中心点之间的欧氏距离。
与步骤S302类似的,对任意两个几何中心点相对于雷达的径向速度作差,可以得到任意两个几何中心点之间的径向速度差。即根据公式3可以得到N个目标点云中任意两个点云的几何中心点之间的径向速度差。
S603、对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群。
具体的,假设N个目标点云中属于第一集群的任一目标点云的几何中心点为第一几何中心点,如果在与N个目标点云分别对应的N个几何中心点中存在一个第二几何点,与该第二几何点对应的目标点云不属于第一集群,并且,该第二几何点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,则将与该第二几何点对应的目标点云加入第一集群。
第二距离门限大于第一距离门限,第二速度门限大于第一速度门限。原因在于,与点云中的点之间的欧氏距离和径向速度差相比,点云之间的欧氏距离和径向速度差都会更大。
S604、对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
在一种可能的实施方式中,如果第一集群中存在任意两个目标点云,这任意两个目标点云的移动速度差小于第三速度门限,则将这任意两个目标点云加入第二集群。其中,属于第一集群的目标点云中任意两个目标点云的移动速度差Δv见公式9:
Figure PCTCN2020087874-appb-000026
其中,i、j为小于第一集群中的目标点云的数目的整数,且i≠j。
在另一种可能的实施方式中,可以根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度。将余弦相似度小于相似度门限的该任意两个目标点云加入第二集群。
假设属于第一集群的任意两个目标点云的几何中心点分别记为x i=[ρ ii,v i]、x j=[ρ jj,v j],则这两个几何中心点的余弦相似度ζ见公式10:
Figure PCTCN2020087874-appb-000027
其中,ρ为几何中心点的距离信息,φ为几何中心点的角度信息,v为几何中心点的移动速度,几何中心点的移动速度等于对应的目标点云的移动速度。
另外,这两个几何中心点也可以分别记为x i=[ρ ii,v ix,v iy]、x j=[ρ jj,v jx,v jy],其中,
Figure PCTCN2020087874-appb-000028
Figure PCTCN2020087874-appb-000029
为该几何中心点对应的目标点云的目标步进角度。
本申请实施例提供的雷达点云聚类方法,根据同一目标的点云的移动速度差别较小的原理,通过将距离相近、径向速度差较小,并且移动速度差别较小的目标点云聚类为一个目标的集群,从而能够防止一个目标受到部分遮挡或不同位置材料不同而导致一个目标的点云被聚类为多个目标的点云。
对于任一第二集群可以求取其质心,用于表示该集群对应的目标,该质心进一步可以用于对该目标进行目标跟踪、轨迹检测等。
示例性的,如图10所示,可以通过显示屏显示不同的集群,例如通过虚线框将属 于同一集群的点云包括其中。或者,可以将属于不同集群的点云采用不同的颜色或显示方式进行区别,例如一个集群的点云采用圆形,另一个集群的点云采用三角形。
可选的,还可以显示各个集群的质心。
如图10中A所示,与图2相比,一个较大目标由于部分受到遮挡或不同位置材料不同导致出现三个点云,虽然三个点云之间距离较大,但是这三个点云的移动速度相近,使得这三个点云可以被聚类成一个目标的集群,从而可以得到一个质心。如图10中B所示,三个较小目标虽然距离较近,但是移动速度均不相同,不能被识别成一个集群,从而可以得到三个质心。
本申请实施例提供了一种雷达点云聚类装置,可以安装在智能汽车、无人机、轿车、卡车、摩托车、公共汽车、船、飞机、直升飞机、割草机、娱乐车、游乐场车辆、施工设备、电车、高尔夫球车、火车和手推车等,本申请不作限定。
该雷达点云聚类装置用于执行上述雷达点云聚类方法。可以根据上述方法实施例对雷达点云聚类装置进行功能模块的划分,例如,可以对应各个功能划分各个功能模块,也可以将两个或两个以上的功能集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。需要说明的是,本申请实施例中对模块的划分是示意性的,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式。
图11示出了一种雷达点云聚类装置110的结构示意图。该雷达点云聚类装置110包括计算模块1101和聚类模块1102。计算模块1101也可以称为计算单元,用以实现上述方法实施例中的计算功能。例如执行图3中的步骤S301-S302,图6中的步骤S601-S602,图7中的步骤S6011、S6012、S602,图9中的步骤S6011、S60121-S60124、S602。聚类模块1102,也可以称为聚类单元,用以实现上述方法实施例中的聚类功能。例如执行图3中的步骤S303,图6、图7、图9中的步骤S603-S604。
在一种可能的实施方式中,计算模块1101,用于根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算任意两个点之间的欧氏距离;其中,M为大于等于2整数;计算模块1101,还用于对任意两个点相对于雷达的径向速度作差,得到任意两个点之间的径向速度差;聚类模块1102,用于对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。
在一种可能的实施方式中,计算模块1101,还用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动速度;计算模块1101,还用于根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;聚类模块1102,还用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;聚类模块1102,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
在一种可能的实施方式中,计算模块1101,还用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算N个目标点云分别对应的移动 速度,包括:计算模块1101,具体用于针对N个目标点云中的任一目标点云,根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度;其中,K为正整数;计算模块1101,具体用于根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度。
在一种可能的实施方式中,计算模块1101根据K个步进角度、任一目标点云的所有T个点相对于雷达的角度信息和径向速度,计算T个点的与K个步进角度对应的移动速度,包括:计算模块1101,具体用于针对T个点中的第t个点,将K个步进角度分别与第t个点的角度信息相减,得到第t个点的K个候选角度;K个候选角度表示第t个点的K个移动方向,1≤t≤T,t为正整数;计算模块1101,具体用于将第t个点的径向速度向第t个点的K个候选角度分别进行逆投影,得到第t个点的与K个步进角度对应的移动速度。
在一种可能的实施方式中,计算模块1101根据T个点的与K个步进角度对应的移动速度,计算任一目标点云的移动速度,包括:计算模块1101,具体用于针对K个步进角度中的第k个步进角度,对T个点的与第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;计算模块1101,具体用于计算T个点的与第k个步进角度对应的移动速度与均值之间的均方差;计算模块1101,具体用于根据最小的均方差计算目标步进角度;计算模块1101,具体用于对T个点的与目标步进角度对应的移动速度取平均,得到任一目标点云的移动速度。
在一种可能的实施方式中,计算模块1101根据N个目标点云分别对应的N个几何中心点相对于雷达的角度信息、距离信息和径向速度差,计算N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:计算模块1101,具体用于根据任意两个几何中心点相对于雷达的角度信息和距离信息,计算任意两个几何中心点之间的欧氏距离;计算模块1101,具体用于对任意两个几何中心点相对于雷达的径向速度作差,得到任意两个几何中心点之间的径向速度差。
在一种可能的实施方式中,聚类模块1102,还用于对N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:聚类模块1102,具体用于将N个目标点云中不属于第一集群的,与第二几何中心点对应的目标点云,加入第一集群,其中,第二几何中心点与第一几何中心点之间的欧氏距离小于第二距离门限并且径向速度差小于第二速度门限,第一几何中心点为N个目标点云中属于第一集群的任一目标点云的几何中心点。
在一种可能的实施方式中,聚类模块1102,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:聚类模块1102,具体用于将属于第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,聚类模块1102,还用于对第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:计算模块1101,具体用于根据属于第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,几何中心点的移动速度等于对应的目标点云的移动速度;聚类模块1102,具体用于将余弦相似度小于相似度门限的任意两个目标点云加入第二集群。
在一种可能的实施方式中,聚类模块1102,用于对欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的任意两个点进行聚类,得到N个目标点云,包括:聚类模块1102,具体用于针对N个目标点云中任一目标点云,将任意两个点中不属于任一目标点云的,并且,与第一点之间的欧氏距离小于第一距离门限并且径向速度差小于第一速度门限的第二点,加入任一目标点云;其中,第一点为任意两个点中属于任一目标点云的点。
在一种可能的实施方式中,第一距离门限和第一速度门限满足以下条件中至少一个:第一点或第二点的距离信息越大,对应的第一距离门限越大;第一点或第二点的距离信息越大,对应的第一速度门限越大。
在本实施例中,该雷达点云聚类装置110以采用集成的方式划分各个功能模块的形式来呈现。这里的“模块”可以指特定ASIC,电路,执行一个或多个软件或固件程序的处理器和存储器,集成逻辑电路,和/或其他可以提供上述功能的器件。
由于本实施例提供的雷达点云聚类装置110可执行上述方法,因此其所能获得的技术效果可参考上述方法实施例,在此不再赘述。
如图12所示,本申请实施例还提供了一种雷达点云聚类装置,该雷达点云聚类装置120包括处理器1201、存储器1202和毫米波雷达1203,处理器1201、存储器1202和毫米波雷达1203耦合,当处理器1201执行存储器1202中的计算机程序或指令时,图3、图6、图7或图9中的方法被执行。处理器1201可以执行图11中计算模块1101和聚类模块1102的功能。
本申请实施例还提供了一种计算机可读存储介质,该计算机可读存储介质中存储有计算机程序,当其在计算机或处理器上运行时,图3、图6、图7或图9中的方法被执行。
本申请实施例还提供了一种包含指令的计算机程序产品,当指令在计算机或处理器上运行时,图3、图6、图7或图9中的方法被执行。
本申请实施例提供了一种芯片系统,该芯片系统包括处理器,用于雷达点云聚类装置执行图3、图6、图7或图9中的方法。
在一种可能的设计中,该芯片系统还包括存储器,该存储器,用于保存必要的程序指令和数据。该芯片系统,可以包括芯片,集成电路,也可以包含芯片和其他分立器件,本申请实施例对此不作具体限定。
其中,本申请提供的雷达点云聚类装置、芯片、计算机存储介质、计算机程序产品或芯片系统均用于执行上文所述的方法,因此,其所能达到的有益效果可参考上文所提供的实施方式中的有益效果,此处不再赘述。
本申请实施例涉及的处理器可以是一个芯片。例如,可以是现场可编程门阵列(field programmable gate array,FPGA),可以是专用集成芯片(application specific integrated circuit,ASIC),还可以是系统芯片(system on chip,SoC),还可以是中央处理器(central processor unit,CPU),还可以是网络处理器(network processor,NP),还可以是数字信号处理电路(digital signal processor,DSP),还可以是微控制器(micro controller unit,MCU),还可以是可编程控制器(programmable logic device,PLD)或其他集成芯片。
本申请实施例涉及的存储器可以是易失性存储器或非易失性存储器,或可包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(read-only memory,ROM)、可编程只读存储器(programmable ROM,PROM)、可擦除可编程只读存储器(erasable PROM,EPROM)、电可擦除可编程只读存储器(electrically EPROM,EEPROM)或闪存。易失性存储器可以是随机存取存储器(random access memory,RAM),其用作外部高速缓存。通过示例性但不是限制性说明,许多形式的RAM可用,例如静态随机存取存储器(static RAM,SRAM)、动态随机存取存储器(dynamic RAM,DRAM)、同步动态随机存取存储器(synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM,DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM,SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM,DR RAM)。应注意,本文描述的系统和方法的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
应理解,在本申请的各种实施例中,上述各过程的序号的大小并不意味着执行顺序的先后,各过程的执行顺序应以其功能和内在逻辑确定,而不应对本申请实施例的实施过程构成任何限定。
本领域普通技术人员可以意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统、设备和方法,可以通过其它的方式实现。例如,以上所描述的设备实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,设备或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。
在上述实施例中,可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件程序实现时,可以全部或部分地以计算机程序产品的形式来实现。该计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行计算机程序指令时,全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用 计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中,或者从一个计算机可读存储介质向另一个计算机可读存储介质传输,例如,所述计算机指令可以从一个网站站点、计算机、服务器或者数据中心通过有线(例如同轴电缆、光纤、数字用户线(Digital Subscriber Line,DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可以用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如,软盘、硬盘、磁带),光介质(例如,DVD)、或者半导体介质(例如固态硬盘(Solid State Disk,SSD))等。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以所述权利要求的保护范围为准。

Claims (40)

  1. 一种雷达点云聚类方法,其特征在于,包括:
    根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算所述任意两个点之间的欧氏距离;其中,M为大于等于2整数;
    对所述任意两个点相对于所述雷达的径向速度作差,得到所述任意两个点之间的径向速度差;
    对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。
  2. 根据权利要求1所述的方法,其特征在于,还包括:
    根据所述N个目标点云中每个目标点云的所有点相对于所述雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度;
    根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;
    对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;
    对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
  3. 根据权利要求2所述的方法,其特征在于,所述根据所述N个目标点云中每个目标点云的所有点相对于所述雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度,包括:
    针对所述N个目标点云中的任一目标点云,根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度;其中,K为正整数;
    根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度。
  4. 根据权利要求3所述的方法,其特征在于,所述根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度,包括:
    针对所述T个点中的第t个点,将所述K个步进角度分别与所述第t个点的角度信息相减,得到所述第t个点的K个候选角度;所述K个候选角度表示所述第t个点的K个移动方向,1≤t≤T,t为正整数;
    将所述第t个点的径向速度向所述第t个点的所述K个候选角度分别进行逆投影,得到所述第t个点的与所述K个步进角度对应的移动速度。
  5. 根据权利要求3或4所述的方法,其特征在于,所述根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度,包括:
    针对所述K个步进角度中的第k个步进角度,对所述T个点的与所述第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;
    计算所述T个点的与所述第k个步进角度对应的移动速度与所述均值之间的均方差;
    根据最小的所述均方差计算目标步进角度;
    对所述T个点的与所述目标步进角度对应的移动速度取平均,得到所述任一目标点云的移动速度。
  6. 根据权利要求2-5任一项所述的方法,其特征在于,所述根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:
    根据所述任意两个几何中心点相对于所述雷达的角度信息和距离信息,计算所述任意两个几何中心点之间的欧氏距离;
    对所述任意两个几何中心点相对于所述雷达的径向速度作差,得到所述任意两个几何中心点之间的径向速度差。
  7. 根据权利要求2-6任一项所述的方法,其特征在于,所述对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:
    将所述N个目标点云中不属于所述第一集群的,与第二几何中心点对应的目标点云,加入所述第一集群,其中,所述第二几何中心点与第一几何中心点之间的欧氏距离小于所述第二距离门限并且径向速度差小于所述第二速度门限,所述第一几何中心点为所述N个目标点云中属于所述第一集群的任一目标点云的几何中心点。
  8. 根据权利要求2-7任一项所述的方法,其特征在于,所述对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    将属于所述第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入所述第二集群。
  9. 根据权利要求2-7任一项所述的方法,其特征在于,所述对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    根据属于所述第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,所述几何中心点的移动速度等于对应的目标点云的移动速度;
    将余弦相似度小于相似度门限的所述任意两个目标点云加入所述第二集群。
  10. 根据权利要求1-9任一项所述的方法,其特征在于,所述对欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云,包括:
    针对所述N个目标点云中任一目标点云,将所述任意两个点中不属于所述任一目标点云的,并且,与第一点之间的欧氏距离小于所述第一距离门限并且径向速度差小于所述第一速度门限的第二点,加入所述任一目标点云;其中,所述第一点为所述任意两个点中属于所述任一目标点云的点。
  11. 根据权利要求10所述的方法,其特征在于,所述第一距离门限和所述第一速度门限满足以下条件中至少一个:所述第一点或所述第二点的距离信息越大,对应的第一距离门限越大;所述第一点或所述第二点的距离信息越大,对应的第一速度门限越大。
  12. 一种雷达点云聚类方法,其特征在于,包括:
    根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度;
    根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;
    对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;
    对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
  13. 根据权利要求12所述的方法,其特征在于,所述根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度,包括:
    针对所述N个目标点云中的任一目标点云,根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度;其中,K为正整数;
    根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度。
  14. 根据权利要求13所述的方法,其特征在于,所述根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度,包括:
    针对所述T个点中的第t个点,将所述K个步进角度分别与所述第t个点的角度信息相减,得到所述第t个点的K个候选角度;所述K个候选角度表示所述第t个点的K个移动方向,1≤t≤T,t为正整数;
    将所述第t个点的径向速度向所述第t个点的所述K个候选角度分别进行逆投影,得到所述第t个点的与所述K个步进角度对应的移动速度。
  15. 根据权利要求13或14所述的方法,其特征在于,所述根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度,包括:
    针对所述K个步进角度中的第k个步进角度,对所述T个点的与所述第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;
    计算所述T个点的与所述第k个步进角度对应的移动速度与所述均值之间的均方差;
    根据最小的所述均方差计算目标步进角度;
    对所述T个点的与所述目标步进角度对应的移动速度取平均,得到所述任一目标点云的移动速度。
  16. 根据权利要求12-15任一项所述的方法,其特征在于,所述根据所述N个目标点云的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:
    根据所述任意两个几何中心点相对于所述雷达的角度信息和距离信息,计算所述任意两个几何中心点之间的欧氏距离;
    对所述任意两个几何中心点相对于所述雷达的径向速度作差,得到所述任意两个 几何中心点之间的径向速度差。
  17. 根据权利要求12-16任一项所述的方法,其特征在于,所述对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:
    将所述N个目标点云中不属于所述第一集群的,与第二几何中心点对应的目标点云,加入所述第一集群,其中,所述第二几何中心点与第一几何中心点之间的欧氏距离小于所述第二距离门限并且径向速度差小于所述第二速度门限,所述第一几何中心点为所述N个目标点云中属于所述第一集群的任一目标点云的几何中心点。
  18. 根据权利要求12-17任一项所述的方法,其特征在于,所述对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    将属于所述第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入所述第二集群。
  19. 根据权利要求12-17任一项所述的方法,其特征在于,所述对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    根据属于所述第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,所述几何中心点的移动速度等于对应的目标点云的移动速度;
    将余弦相似度小于相似度门限的所述任意两个目标点云加入所述第二集群。
  20. 一种雷达点云聚类装置,其特征在于,包括:
    计算模块,用于根据点云数据的M个点中任意两个点相对于雷达的角度信息和距离信息,计算所述任意两个点之间的欧氏距离;其中,M为大于等于2整数;
    所述计算模块,还用于对所述任意两个点相对于所述雷达的径向速度作差,得到所述任意两个点之间的径向速度差;
    聚类模块,用于对满足欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云;其中,N为大于1的整数。
  21. 根据权利要求20所述的雷达点云聚类装置,其特征在于,
    所述计算模块,还用于根据所述N个目标点云中每个目标点云的所有点相对于所述雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度;
    所述计算模块,还用于根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;
    所述聚类模块,还用于对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;
    所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
  22. 根据权利要求21所述的雷达点云聚类装置,其特征在于,所述计算模块,还用于根据所述N个目标点云中每个目标点云的所有点相对于所述雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度,包括:
    所述计算模块,具体用于针对所述N个目标点云中的任一目标点云,根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度;其中,K为正整数;
    所述计算模块,具体用于根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度。
  23. 根据权利要求22所述的雷达点云聚类装置,其特征在于,所述计算模块,具体用于根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度,包括:
    所述计算模块,具体用于针对所述T个点中的第t个点,将所述K个步进角度分别与所述第t个点的角度信息相减,得到所述第t个点的K个候选角度;所述K个候选角度表示所述第t个点的K个移动方向,1≤t≤T,t为正整数;
    所述计算模块,具体用于将所述第t个点的径向速度向所述第t个点的所述K个候选角度分别进行逆投影,得到所述第t个点的与所述K个步进角度对应的移动速度。
  24. 根据权利要求22或23所述的雷达点云聚类装置,其特征在于,所述计算模块,具体用于根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度,包括:
    所述计算模块,具体用于针对所述K个步进角度中的第k个步进角度,对所述T个点的与所述第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;
    所述计算模块,具体用于计算所述T个点的与所述第k个步进角度对应的移动速度与所述均值之间的均方差;
    所述计算模块,具体用于根据最小的所述均方差计算目标步进角度;
    所述计算模块,具体用于对所述T个点的与所述目标步进角度对应的移动速度取平均,得到所述任一目标点云的移动速度。
  25. 根据权利要求21-24任一项所述的雷达点云聚类装置,其特征在于,所述计算模块,还用于根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:
    所述计算模块,具体用于根据所述任意两个几何中心点相对于所述雷达的角度信息和距离信息,计算所述任意两个几何中心点之间的欧氏距离;
    所述计算模块,具体用于对所述任意两个几何中心点相对于所述雷达的径向速度作差,得到所述任意两个几何中心点之间的径向速度差。
  26. 根据权利要求21-25任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,还用于对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:
    所述聚类模块,具体用于将所述N个目标点云中不属于所述第一集群的,与第二几何中心点对应的目标点云,加入所述第一集群,其中,所述第二几何中心点与第一几何中心点之间的欧氏距离小于所述第二距离门限并且径向速度差小于所述第二速度门限,所述第一几何中心点为所述N个目标点云中属于所述第一集群的任一目标点云 的几何中心点。
  27. 根据权利要求21-26任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    所述聚类模块,具体用于将属于所述第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入所述第二集群。
  28. 根据权利要求21-26任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    所述计算模块,具体用于根据属于所述第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,所述几何中心点的移动速度等于对应的目标点云的移动速度;
    所述聚类模块,具体用于将余弦相似度小于相似度门限的所述任意两个目标点云加入所述第二集群。
  29. 根据权利要求20-28任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,用于对欧氏距离小于第一距离门限,并且,径向速度差小于第一速度门限的所述任意两个点进行聚类,得到N个目标点云,包括:
    所述聚类模块,具体用于针对所述N个目标点云中任一目标点云,将所述任意两个点中不属于所述任一目标点云的,并且,与第一点之间的欧氏距离小于所述第一距离门限并且径向速度差小于所述第一速度门限的第二点,加入所述任一目标点云;其中,所述第一点为所述任意两个点中属于所述任一目标点云的点。
  30. 根据权利要求29所述的雷达点云聚类装置,其特征在于,所述第一距离门限和所述第一速度门限满足以下条件中至少一个:所述第一点或所述第二点的距离信息越大,对应的第一距离门限越大;所述第一点或所述第二点的距离信息越大,对应的第一速度门限越大。
  31. 一种雷达点云聚类装置,其特征在于,包括:
    计算模块,用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度;
    所述计算模块,还用于根据所述N个目标点云分别对应的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差;
    聚类模块,用于对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群;
    所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群。
  32. 根据权利要求31所述的雷达点云聚类装置,其特征在于,所述计算模块,用于根据N个目标点云中每个目标点云的所有点相对于雷达的角度信息和径向速度,计算所述N个目标点云分别对应的移动速度,包括:
    所述计算模块,具体用于针对所述N个目标点云中的任一目标点云,根据K个步 进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度;其中,K为正整数;
    根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度。
  33. 根据权利要求32所述的雷达点云聚类装置,其特征在于,所述计算模块,具体用于根据K个步进角度、所述任一目标点云的所有T个点相对于所述雷达的角度信息和径向速度,计算所述T个点的与所述K个步进角度对应的移动速度,包括:
    所述计算模块,具体用于针对所述T个点中的第t个点,将所述K个步进角度分别与所述第t个点的角度信息相减,得到所述第t个点的K个候选角度;所述K个候选角度表示所述第t个点的K个移动方向,1≤t≤T,t为正整数;
    所述计算模块,具体用于将所述第t个点的径向速度向所述第t个点的所述K个候选角度分别进行逆投影,得到所述第t个点的与所述K个步进角度对应的移动速度。
  34. 根据权利要求32或33所述的雷达点云聚类装置,其特征在于,所述计算模块,具体用于根据所述T个点的与所述K个步进角度对应的移动速度,计算所述任一目标点云的移动速度,包括:
    所述计算模块,具体用于针对所述K个步进角度中的第k个步进角度,对所述T个点的与所述第k个步进角度对应的移动速度取平均,得到均值;1≤k≤K,且k为整数;
    所述计算模块,具体用于计算所述T个点的与所述第k个步进角度对应的移动速度与所述均值之间的均方差;
    所述计算模块,具体用于根据最小的所述均方差计算目标步进角度;
    所述计算模块,具体用于对所述T个点的与所述目标步进角度对应的移动速度取平均,得到所述任一目标点云的移动速度。
  35. 根据权利要求31-34任一项所述的雷达点云聚类装置,其特征在于,所述计算模块,还用于根据所述N个目标点云的N个几何中心点相对于所述雷达的角度信息、距离信息和径向速度差,计算所述N个几何中心点中任意两个几何中心点之间的欧氏距离和径向速度差,包括:
    所述计算模块,具体用于根据所述任意两个几何中心点相对于所述雷达的角度信息和距离信息,计算所述任意两个几何中心点之间的欧氏距离;
    所述计算模块,具体用于对所述任意两个几何中心点相对于所述雷达的径向速度作差,得到所述任意两个几何中心点之间的径向速度差。
  36. 根据权利要求31-35任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,用于对所述N个目标点云中,几何中心点之间的欧氏距离小于第二距离门限,并且,径向速度差小于第二速度门限的目标点云进行聚类,得到第一集群,包括:
    所述聚类模块,具体用于将所述N个目标点云中不属于所述第一集群的,与第二几何中心点对应的目标点云,加入所述第一集群,其中,所述第二几何中心点与第一几何中心点之间的欧氏距离小于所述第二距离门限并且径向速度差小于所述第二速度门限,所述第一几何中心点为所述N个目标点云中属于所述第一集群的任一目标点云的几何中心点。
  37. 根据权利要求31-36任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    所述聚类模块,具体用于将属于所述第一集群的、移动速度之差小于第三速度门限的任意两个目标点云加入所述第二集群。
  38. 根据权利要求31-36任一项所述的雷达点云聚类装置,其特征在于,所述聚类模块,还用于对所述第一集群中,移动速度满足条件的目标点云进行聚类,得到第二集群,包括:
    所述计算模块,具体用于根据属于所述第一集群的任意两个目标点云的几何中心点的距离信息、角度信息和移动速度计算余弦相似度,其中,所述几何中心点的移动速度等于对应的目标点云的移动速度;
    所述聚类模块,具体用于将余弦相似度小于相似度门限的所述任意两个目标点云加入所述第二集群。
  39. 一种雷达点云聚类装置,其特征在于,包括:处理器、存储器和毫米波雷达,所述处理器、所述存储器和所述毫米波雷达耦合,所述存储器用于存储计算机程序,所述处理器用于执行所述存储器中存储的所述计算机程序,以使得所述雷达点云聚类装置执行如权利要求1-11任一项所述的方法,或者,执行如权利要求12-19任一项所述的方法。
  40. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有计算机程序,当其在计算机上运行时,如权利要求1-11任一项所述的方法被执行,或者,如权利要求12-19任一项所述的方法被执行。
PCT/CN2020/087874 2020-04-29 2020-04-29 雷达点云聚类方法和装置 WO2021217518A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202080004810.8A CN112689775B (zh) 2020-04-29 2020-04-29 雷达点云聚类方法和装置
PCT/CN2020/087874 WO2021217518A1 (zh) 2020-04-29 2020-04-29 雷达点云聚类方法和装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/087874 WO2021217518A1 (zh) 2020-04-29 2020-04-29 雷达点云聚类方法和装置

Publications (1)

Publication Number Publication Date
WO2021217518A1 true WO2021217518A1 (zh) 2021-11-04

Family

ID=75457661

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/087874 WO2021217518A1 (zh) 2020-04-29 2020-04-29 雷达点云聚类方法和装置

Country Status (2)

Country Link
CN (1) CN112689775B (zh)
WO (1) WO2021217518A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115171378A (zh) * 2022-06-28 2022-10-11 武汉理工大学 一种基于路侧雷达长距离多车辆高精度检测追踪方法
CN115439484A (zh) * 2022-11-10 2022-12-06 苏州挚途科技有限公司 基于4d点云的检测方法、装置、存储介质及处理器
CN115600118A (zh) * 2022-11-29 2023-01-13 山东亚历山大智能科技有限公司(Cn) 基于二维激光点云的托盘腿识别方法及系统
TWI819613B (zh) * 2022-05-19 2023-10-21 緯創資通股份有限公司 物件的雙感測方法及用於物件感測的運算裝置

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114460582B (zh) * 2021-12-14 2023-04-14 江苏航天大为科技股份有限公司 一种基于点云速度的毫米波雷达大车识别方法

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170307735A1 (en) * 2016-04-22 2017-10-26 Mohsen Rohani Object detection using radar and machine learning
CN109581312A (zh) * 2018-11-22 2019-04-05 西安电子科技大学昆山创新研究院 一种高分辨毫米波雷达多目标聚类方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109543704B (zh) * 2017-09-22 2021-02-23 比亚迪股份有限公司 汽车及车载雷达的多目标聚类方法、装置
CN108710828B (zh) * 2018-04-18 2021-01-01 北京汽车集团有限公司 识别目标物的方法、装置和存储介质以及车辆
CN109849930B (zh) * 2018-12-29 2020-08-04 百度在线网络技术(北京)有限公司 自动驾驶汽车的相邻车辆的速度计算方法和装置

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170307735A1 (en) * 2016-04-22 2017-10-26 Mohsen Rohani Object detection using radar and machine learning
CN109581312A (zh) * 2018-11-22 2019-04-05 西安电子科技大学昆山创新研究院 一种高分辨毫米波雷达多目标聚类方法

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI819613B (zh) * 2022-05-19 2023-10-21 緯創資通股份有限公司 物件的雙感測方法及用於物件感測的運算裝置
CN115171378A (zh) * 2022-06-28 2022-10-11 武汉理工大学 一种基于路侧雷达长距离多车辆高精度检测追踪方法
CN115171378B (zh) * 2022-06-28 2023-10-27 武汉理工大学 一种基于路侧雷达长距离多车辆高精度检测追踪方法
CN115439484A (zh) * 2022-11-10 2022-12-06 苏州挚途科技有限公司 基于4d点云的检测方法、装置、存储介质及处理器
CN115600118A (zh) * 2022-11-29 2023-01-13 山东亚历山大智能科技有限公司(Cn) 基于二维激光点云的托盘腿识别方法及系统
CN115600118B (zh) * 2022-11-29 2023-08-08 山东亚历山大智能科技有限公司 基于二维激光点云的托盘腿识别方法及系统

Also Published As

Publication number Publication date
CN112689775B (zh) 2022-06-14
CN112689775A (zh) 2021-04-20

Similar Documents

Publication Publication Date Title
WO2021217518A1 (zh) 雷达点云聚类方法和装置
WO2021104497A1 (zh) 基于激光雷达的定位方法及系统、存储介质和处理器
US20190361106A1 (en) Method of estimating a velocity magnitude of a moving target in a horizontal plane and radar detection system
CN107229033B (zh) 基于高度维分段搜索的多目标到达时间差定位方法
CN105487061B (zh) 目标数据关联的多特征信息融合方法
CN111045024A (zh) 一种基于光检测和测距的车辆追踪方法和系统
US20220215762A1 (en) System and method for fusing asynchronous sensor tracks in a track fusion application
CN110333480B (zh) 一种基于聚类的单无人机多目标aoa定位方法
CN110794392B (zh) 车辆定位方法、装置、车辆及存储介质
CN105787081B (zh) 一种基于辐射源空间位置的辐射平台关联方法
CN111045023A (zh) 一种基于光检测和测距的车辆追踪方法和系统
CN111045025A (zh) 一种基于光检测和测距的车辆追踪方法和系统
WO2023284705A1 (zh) 一种激光雷达点云聚类方法、装置、激光雷达及车辆
WO2021129483A1 (zh) 一种确定点云的边界框的方法及装置
CN107801168B (zh) 一种室外自适应的被动式目标的定位方法
CN111765883B (zh) 机器人蒙特卡罗定位方法、设备及存储介质
WO2020198973A1 (zh) 微波雷达检测栅栏附近静止物体的方法和毫米波雷达
CN109799477B (zh) 一种面向毫米波车联网的序贯车辆指纹定位方法及装置
CN109541541B (zh) 一种室内三角定位精度修正方法及装置
Hu et al. Accurate three-dimensional tracking method in bistatic forward scatter radar
JP2017116445A (ja) 物体検出装置
WO2021078283A1 (zh) 一种数据关联方法及装置
WO2021218347A1 (zh) 一种聚类的方法和装置
CN108334893A (zh) 一种多亮点聚类分析的水下细长体特征识别方法
CN113534130A (zh) 基于视线角度的多站雷达多目标数据关联方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20933720

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20933720

Country of ref document: EP

Kind code of ref document: A1