A laser radar point cloud integration target detection method, medium and system

Abstract

The present invention seeks to protect a laser radar point cloud integral target detection method, medium and system. The method improves the detection capability of the laser radar by performing time integration operations on the laser radar point clouds at multiple consecutive moments, and the steps include: firstly, collecting the laser radar point cloud data, performing ground segmentation on the original point cloud, and then constructing a point cloud time integration model for the segmented non-ground point cloud, and performing different time integration operations according to the point cloud distance; then clustering the point cloud after the integration processing, and tracking and filtering the clustered targets to obtain the target motion state. Among them, the time integration operation maps the point clouds at multiple consecutive moments to the current moment according to the target speed obtained by clustering, and then the union of the mapped multiple point cloud sets is calculated. The present invention can enable the low-beam radar to achieve the detection effect of the high-beam radar, and effectively improve the detection capability of the laser radar for long-distance targets and small targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous unmanned system environment perception, and specifically relates to a laser radar point cloud integration target detection method. Background Art

[0002] Environmental perception is one of the key technologies of autonomous unmanned systems (including unmanned vehicles, intelligent robots, etc.) and is the basis of decision-making, planning and control. As a major environmental perception sensor, LiDAR and its target detection method play an important role in autonomous unmanned systems.

[0003] Chinese patent application: Target detection clustering method, system, computer equipment and readable storage medium (application number: CN202010003576.9) discloses a target detection clustering method adaptive to the external environment, characterized in that it includes: obtaining a laser radar point cloud and preprocessing the laser radar point cloud; adjusting the neighborhood radius and density threshold according to the real-time external environment; clustering the preprocessed laser radar point cloud according to the neighborhood radius and density threshold to generate a clustering result; and detecting obstacles according to the clustering result. This method improves the problem that the traditional method does not specifically adjust the clustering radius and density threshold according to the external environment, and cannot solve the problem of missed detection of sparse point clouds in the case of long distances and small targets. Chinese patent application: Laser radar point cloud target detection method, system and device (application number: CN202011060317.6) discloses a laser radar point cloud target detection method, which constructs a front view grid for point cloud information, extracts features from the front view grid respectively, and extracts point features from the nearest point in each front view grid and combines them, and then detects obstacles through a three-dimensional target detector. This method only uses the lidar data at the current moment, and cannot solve the problem of missed detection of sparse point clouds in the case of long-distance and small targets for low-resolution lidar.

[0004] Aiming at the characteristics of laser radar point clouds that are dense near and sparse far away in the environment perception of autonomous unmanned systems, the present invention utilizes the contextual connection between multi-frame laser radar point clouds and adopts multi-time point cloud integration to increase target feature information and point cloud density, so as to improve the detection capability of laser radar for long-distance targets and small targets. Summary of the invention

[0005] The present invention aims to solve the above problems of the prior art and improve the detection capability of laser radar for long-distance targets and small targets. The technical solution of the present invention is as follows:

[0006] A laser radar point cloud integration target detection method, comprising the following steps:

[0007] (1) At the current time k, the laser radar collects point cloud data of the surrounding environment, performs ground segmentation on the original point cloud, classifies the original point cloud according to its z coordinate, and screens the ground point cloud set. and non-ground point cloud sets

[0008] (2) Non-ground point cloud According to the distance threshold D thr , filter out long-distance sparse point clouds and close-range dense point clusters satisfy

[0009]

[0010] (3) Long-distance sparse point cloud Join the remote point cloud queue Construct a point cloud time integration model, perform time integration operations according to the time of the point cloud and the corresponding target motion state, and obtain a new point cloud set

[0011] (4) New point cloud collection Clustering is performed, and tracking and filtering is performed on the targets obtained by clustering to obtain the target motion state.

[0012] Furthermore, the ground point cloud set is screened according to the z coordinate of the original point cloud. and non-ground point cloud sets The method is: for each point cloud P(x,y,z),

[0013]

[0014] Where Z thr It is the ground segmentation threshold.

[0015] Furthermore, the non-ground point cloud Filter out long-distance sparse point clouds and close-range dense point clusters The method is: for each point cloud

[0016]

[0017] Furthermore, the point cloud time integration operation includes the following steps:

[0018] (1) For the current moment k, estimate the target motion state at the previous moment Make a one-step prediction to get the target motion state at time k Will Add to target status queue

[0019] (2) For each moment ki, the corresponding long-distance point cloud queue According to the target status queue The target set {T1,…,T m} to associate Divided into:

[0020]

[0021] in, is the target T j The corresponding point cloud, j = 1, ..., m, and It is an unassociated point cloud;

[0022] (3) Yes Any point in x l ,y l , z l is the coordinate of the point, according to the target T j The movement speed V j =[v x ,v y ,v z ] T Transform and map to the current moment k:

[0023]

[0024] Corresponding After transformation, it is recorded as Long-range point cloud become

[0025]

[0026] Among them, the point cloud is not associated The corresponding virtual target movement speed is V 0 =[0,0,0] T ;

[0027] (4) The near-distance point cloud and the mapped target point cloud and the far-distance point cloud are accumulated to obtain a new point cloud after time integration:

[0028]

[0029] Furthermore, the state equation and measurement equation of the target motion model are:

[0030] X(k,k)=ΦX(k-1,k-1)+Γw k (1)

[0031] Z(k)=HX(k,k)+v k (2)

[0032] in They represent the x position, x velocity, y position, y velocity, length, width, height, and azimuth of the target at time k respectively. The measurement at time k is represented by Z(k)={Z i (k)},Z i (k) = [x i ,y i ,L i ,W i ,H i ,θ i ], respectively represent the x, y position, length, width, height, and azimuth of target i at time k, Φ is the state transfer matrix, Γ is the noise matrix, H is the measurement matrix, ω k and k are process noise and measurement noise respectively.

[0033] Further, the target motion state estimation at the previous moment Make a one-step prediction to get the target motion state at time k The method is to follow formula (1):

[0034]

[0035] Furthermore, the method for associating the long-range point cloud with the target comprises the following steps:

[0036] (1) For each moment ki, the long-distance point cloud set Point cloud in And the target set {T1,…,T m}, repeat steps 3.2 and 3.3 to associate the point cloud with the target;

[0037] (2) For target T i , represented by a cube model as T i =[x i ,y i ,z i ,L i ,W i ,H i ,θ i ] T , where x i ,y i ,z i Its center coordinates, L i ,W i ,H i are length, width, height, θ i is the azimuth;

[0038] (3) For each point cloud satisfy:

[0039]

[0040] Where δ is the correlation threshold, then the point cloud P(x, y, z) and the target T i association, If the point cloud cannot be associated with any target, it is divided into point cloud sets.

[0041] The advantages and beneficial effects of the present invention are as follows:

[0042] The present invention proposes a laser radar point cloud integration target detection method. The method targets the characteristics of laser radar point clouds that are dense near and sparse far in the environment perception of autonomous unmanned systems, utilizes the contextual connection between multi-frame laser radar point clouds, increases target feature information and point cloud density by performing time integration operations on laser radar point clouds at multiple consecutive moments, and then performs clustering to extract targets. The method of the present invention can increase the density of long-distance point clouds, so that low-beam radar can achieve the detection effect of high-beam radar, effectively improving the detection range of laser radar and its ability to detect small targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a general architecture diagram of a laser radar point cloud integration target detection method according to a preferred embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the point cloud and target association principle of the present invention. DETAILED DESCRIPTION

[0045] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0046] The technical solution of the present invention to solve the above technical problems is:

[0047] like Figure 1 The figure shows the overall architecture of a laser radar point cloud integral target detection method of the present invention, and the method processing flow is as follows:

[0048] (1) At the current time k, the laser radar collects point cloud data of the surrounding environment, performs ground segmentation on the original point cloud, classifies the original point cloud according to its z coordinate, and screens the ground point cloud set. and non-ground point cloud sets

[0049] (2) Non-ground point cloud According to the distance threshold D thr , filter out long-distance sparse point clouds and close-range dense point clusters satisfy

[0050]

[0051] (3) Long-distance sparse point cloud Join the remote point cloud queue Construct a point cloud time integration model, perform time integration operations according to the time of the point cloud and the corresponding target motion state, and obtain a new point cloud set The steps of this method are:

[0052] (3.1) For the current moment k, estimate the target motion state at the previous moment Make a one-step prediction to get the target motion state at time k Will Add to target status queue

[0053] (3.2) For each moment ki, the corresponding long-distance point cloud queue According to the target status queue The target set {T1,…,T m} to associate Divided into:

[0054]

[0055] in, is the target T j The corresponding point cloud, j = 1, ..., m, and It is an unassociated point cloud;

[0056] (3.3) Yes Any point in x l ,y l , z l is the coordinate of the point, according to the target T j The movement speed V j =[v x ,v y ,v z ] T Transform and map to the current moment k:

[0057]

[0058] Corresponding After transformation, it is recorded as Long-range point cloud become

[0059]

[0060] Among them, the point cloud is not associated The corresponding virtual target movement speed is V 0 =[0,0,0] T .

[0061] (3.4) The near-distance point cloud and the mapped target point cloud and the far-distance point cloud are accumulated to obtain the new point cloud after time integration:

[0062]

[0063] (4) New point cloud collection Clustering is performed, and tracking and filtering is performed on the targets obtained by clustering to obtain the target motion state.

[0064] like Figure 2 Shown is a schematic diagram of the point cloud and target association principle of the present invention.

[0065] Assuming that the target does not roll over, let the target T i As shown in the solid rectangle in the figure, its length, width and azimuth are L, W and θ respectively. i , the association threshold is δ, and the association gate is shown by the dotted line in the figure. The point cloud falling into the dotted box is consistent with the target T i The dashed box is not related.

[0066] The boundary lines of the associated gate and the azimuth direction are l1 and l2, passing through the target center (x i ,y i ) The straight line perpendicular to the target heading intersects l1 and l2 at P x1 and P x2 . The normal vector N of l1 and l2 can be calculated x for

[0067] (-sinθ i ,cosθ i )

[0068] For any point cloud P(x,y) in the point cloud set, the vector for With vector N x Angle for

[0069]

[0070] vector With vector N x Angle for

[0071]

[0072] When the point cloud P(x,y) is within the associated gate, it satisfies

[0073]

[0074] Right now:

[0075]

[0076] For another set of boundaries of the association gate, the corresponding conditions of the point cloud P(x,y) within the association gate are:

[0077]

[0078] For the z direction, the condition for the point cloud P(x,y) to be within the associated gate is

[0079]

[0080] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0081] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0082] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0083] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A laser radar point cloud integral target detection method, characterized in that it includes the following steps: 1.1 At the current time k, the laser radar collects the surrounding environment point cloud data, performs ground segmentation on the original point cloud, classifies the original point cloud according to the z coordinate, and selects the ground point cloud set and non-ground point cloud sets 1.2 Non-ground point cloud According to the distance threshold D thr , filter out long-distance sparse point clouds and close-range dense point cloud satisfy 1.3 Distant sparse point clouds Join the remote point cloud queue Construct a point cloud time integration model, perform time integration operations according to the time of the point cloud and the corresponding target motion state, and obtain a new point cloud set 1.4 New Point Cloud Collection Clustering is performed, and tracking and filtering is performed on the targets obtained by clustering to obtain the target motion state; The point cloud time integration operation specifically includes the following steps: 2.1 For the current moment k, estimate the target motion state based on the previous moment Make a one-step prediction to get the target motion state at time k Will Add to target status queue 2.2 For each moment ki, the corresponding long-distance point cloud queue According to the target status queue The target set {T1,…,T m } to associate Divided into: in, is the target T j The corresponding point cloud, j = 1, ..., m, and It is an unassociated point cloud; 2.3 pairs Any point in x l ,y l , z l is the coordinate of the point, according to the target T j The movement speed V j =[v x ,v y ,v z ] T Transform and map to the current moment k: Corresponding After transformation, it is recorded as Long-range point cloud become Among them, the point cloud is not associated The corresponding virtual target movement speed is V 0 =[0,0,0] T ; 2.4 The near-distance point cloud and the mapped target point cloud are added together to obtain a new point cloud after time integration: The method for associating a long-distance point cloud with a target specifically comprises the following steps: 3.1 For each moment ki, the long-distance point cloud set Point cloud in And the target set {T1,…,T m }, repeat steps 3.2 and 3.3 to associate the point cloud with the target; 3.2 Target T i , represented by a cube model as T i =[x i ,y i ,z i ,L i ,W i ,H i ,θ i ] T , where x i ,y i ,z i Its center coordinates, L i ,W i ,H i are length, width, height, θ i is the azimuth; 3.3 For each point cloud satisfy: Where δ is the correlation threshold, then the point cloud P(x, y, z) and the target T i association, If the point cloud cannot be associated with any target, it is divided into point cloud sets.

2. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the laser radar point cloud integral target detection method as claimed in claim 1.

3. A laser radar point cloud integration target detection system based on the method of claim 1, characterized in that: include: Acquisition module: used to collect surrounding point cloud data through LiDAR, pre-process the point cloud data, filter out valid point clouds, divide the area of ​​interest, and then perform ground segmentation on the point cloud data in the area of ​​interest to filter out ground point cloud sets and non-ground point cloud sets; Detection module: used to filter out long-distance sparse point clouds and short-distance dense point clouds from non-ground point clouds, build a point cloud time integration model for long-distance sparse point clouds, perform time integration operations according to the time of the point cloud and the corresponding target motion state, and obtain a new point cloud set; perform clustering on the integrated point cloud to obtain the target position and shape information; Tracking module: used to track and filter the target position and shape data obtained by clustering to obtain the target motion state.

Images