CN114332795A - Laser radar target identification method and device and electronic equipment - Google Patents

Abstract

The invention provides a laser radar target identification method, a laser radar target identification device and electronic equipment, which are applied to the technical field of automobiles. According to the method, the amount of data to be processed of the identification model can be obviously reduced through the structuralization processing and the clustering processing of the point cloud to be processed, so that the network structure of the identification model can be simplified, the hardware performance requirement on electronic equipment is lowered, and the identification efficiency of the radar target is effectively improved.

Description

Laser radar target identification method and device and electronic equipment

Technical Field

The invention belongs to the technical field of automobiles, and particularly relates to a laser radar target identification method and device and electronic equipment.

Background

In the field of driving assistance and intelligent driving, vehicle-mounted laser radars are widely used as a main means for detecting obstacles around a vehicle. In each detection period, the vehicle-mounted laser radar feeds back point clouds consisting of a plurality of laser points, and the vehicle-mounted controller completes the identification of radar targets around the vehicle by analyzing the point clouds.

With the development of neural network technology, the point cloud is subjected to target recognition through a target recognition model obtained based on neural network training, which is a widely used technical means.

However, the inventor researches and discovers that the point cloud data volume fed back by the laser radar is huge, and radar target identification is directly carried out based on the point cloud, so that the network model structure of the target identification model is complex, the calculated amount is extremely large, the requirement on the hardware performance of the vehicle-mounted controller is high, and the radar target identification efficiency is low.

Disclosure of Invention

In view of this, an object of the present invention is to provide a method, an apparatus, and an electronic device for identifying a laser radar target, which reduce the amount of calculation in the radar target identification process, reduce the requirements on hardware devices, and improve the identification efficiency, and the specific scheme is as follows:

in a first aspect, the present invention provides a laser radar target identification method, including:

acquiring point cloud to be processed of the laser radar;

structuring the point cloud to be processed to obtain a structured point cloud;

performing invalidation processing on the laser points corresponding to the ground points in the structured point cloud to obtain a processed structured point cloud;

clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster;

inputting each point cloud cluster into a feature extraction network to obtain corresponding point cloud cluster feature vectors;

the characteristic extraction network is obtained by training a neural network by taking the performance parameters of each laser point in the point cloud cluster as input and taking the characteristic vector of the point cloud cluster as output;

inputting the point cloud cluster feature vectors into a target identification network to obtain a radar target;

and the target identification network is obtained by training a neural network by taking the characteristic vector of each point cloud cluster as input and taking a radar target in each point cloud cluster as output.

Optionally, the structuring the point cloud to be processed to obtain a structured point cloud includes:

taking the beam value and the horizontal angle value corresponding to each laser point in the point cloud to be processed as the sequencing index of the corresponding stress light spot;

and sequencing the laser points in the point cloud to be processed according to the sequencing index to obtain a structured point cloud.

Optionally, the clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster includes:

respectively taking each effective laser point in the processed structured point cloud as a target laser point;

the effective laser points are laser points except for laser points corresponding to the ground point in the processed structured point cloud;

acquiring a target clustering threshold corresponding to the target laser point;

taking the laser points in the processed structured point cloud which take the target laser point as a center and are in a preset screening range in an ordered index manner as candidate laser points;

respectively calculating Euclidean distances between each candidate laser point and the target laser point to obtain a screening distance corresponding to each candidate laser point;

and clustering the candidate laser points with the screening distance smaller than the target clustering threshold value with the target laser points to obtain corresponding point cloud clusters.

Optionally, the obtaining of the target clustering threshold corresponding to the target laser point includes:

calculating the Euclidean distance between the target laser point and a preset origin of the laser radar to obtain a reference distance corresponding to the target laser point;

and determining a target clustering threshold according to the size relation between the reference distance and the reference distance threshold.

Optionally, the determining a target clustering threshold according to the size relationship between the reference distance and the reference distance threshold includes:

if the reference distance is smaller than a first reference distance threshold value, taking the first clustering threshold value as a target clustering threshold value;

if the reference distance is larger than a second reference distance threshold, taking a second clustering threshold as a target clustering threshold;

if the reference distance is greater than or equal to the first reference distance threshold and the reference distance is less than or equal to the second reference distance threshold, taking the product of the reference distance and a preset proportionality coefficient as a target clustering threshold;

wherein the first reference distance threshold is less than the second reference distance threshold;

the first clustering threshold is less than the second clustering threshold.

Optionally, the respectively using each effective laser point in the structured point cloud as a target laser point includes:

and according to the sequence of the sequencing indexes of the effective laser points in the processed structured point cloud, respectively taking the effective laser points as target laser points.

Optionally, the respectively inputting each point cloud cluster into the feature extraction network to obtain a corresponding point cloud cluster feature vector, including:

combining the three-dimensional coordinates and the reflection intensity corresponding to each laser point in each point cloud cluster into a two-dimensional vector matrix;

and inputting the two-dimensional matrix into a feature extraction network, and converting the two-dimensional vector matrix into point cloud cluster feature vectors with the same length after passing through the feature extraction network, wherein the point cloud cluster feature vectors are one-dimensional vectors.

Optionally, the inputting each point cloud cluster feature vector into a target identification network to obtain a radar target includes:

combining the point cloud cluster feature vectors into a point cloud cluster feature matrix;

and inputting the point cloud cluster characteristic matrix into a target identification network to obtain a radar target.

In a second aspect, the present invention provides a laser radar target recognition apparatus, including:

the acquisition unit is used for acquiring point clouds to be processed of the laser radar;

the first processing unit is used for carrying out structuring processing on the point cloud to be processed to obtain a structured point cloud;

the second processing unit is used for carrying out invalidation processing on the laser points corresponding to the ground points in the structured point cloud to obtain the processed structured point cloud;

the clustering unit is used for clustering the laser points in the processed structured point cloud to obtain at least one point cloud cluster;

the characteristic extraction unit is used for inputting each point cloud cluster into a characteristic extraction network to obtain corresponding point cloud cluster characteristic vectors;

the characteristic extraction network is obtained by training a neural network by taking the performance parameters of each laser point in the point cloud cluster as input and taking the characteristic vector of the point cloud cluster as output;

the identification unit is used for inputting the point cloud cluster feature vectors into a target identification network to obtain a radar target;

and the target identification network is obtained by training a neural network by taking the characteristic vector of each point cloud cluster as input and taking a radar target in each point cloud cluster as output.

In a third aspect, the present invention provides an electronic device comprising: a memory and a processor;

the memory stores a program adapted to be executed by the processor to implement the lidar target identification method according to any of the first aspects of the present invention.

Based on the technical scheme, the laser radar target identification method provided by the invention comprises the steps of obtaining point clouds to be processed of a laser radar, carrying out structuring processing on the point clouds to be processed to obtain the structured point clouds, carrying out invalidation processing on laser points corresponding to ground points in the structured point clouds to obtain the processed structured point clouds, then clustering the laser points in the processed structured point clouds to obtain at least one point cloud cluster, respectively inputting each point cloud cluster into a feature extraction network to obtain corresponding point cloud cluster feature vectors, and finally inputting each point cloud cluster feature vector into a target identification network to obtain a radar target. Compared with the prior art, the method can obviously reduce the data volume required to be processed by the identification model through the structuralization processing and the clustering processing of the point cloud to be processed, further can simplify the network structure of the identification model, reduces the hardware performance requirement on the electronic equipment through the adjustment of the data volume and the network structure, and correspondingly can effectively improve the identification efficiency of the radar target through the method under the condition of using the electronic equipment with the same performance.

Furthermore, when the target is identified, the characteristic extraction is firstly carried out on each point cloud cluster, then the radar target identification is carried out, the point cloud cluster characteristics obtained through the characteristic extraction model are finer than the three-dimensional coordinates and the reflection intensity in the point cloud to be processed, and compared with the prior art, the identification effect is better than that of a network of the same scale.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.

Fig. 1 is a flowchart of a laser radar target identification method according to an embodiment of the present invention;

fig. 2 is a schematic diagram of a point cloud cluster feature vector extraction process according to an embodiment of the present invention;

fig. 3 is a schematic diagram of a radar target identification process according to an embodiment of the present invention;

fig. 4 is a block diagram of a lidar target recognition apparatus according to an embodiment of the present invention;

fig. 5 is a block diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The laser radar target identification method provided by the invention is applied to electronic equipment, the electronic equipment can be a controller corresponding to the laser radar, and can also be other controllers capable of acquiring laser radar feedback point clouds, for example, when the laser radar is applied to a vehicle, a vehicle controller or an automatic driving controller can be selected, and the laser radar target identification method can also be applied to a server on a network side under certain conditions.

Referring to fig. 1, fig. 1 is a flowchart of a laser radar target identification method provided in an embodiment of the present invention, where the flowchart of the laser radar target identification method provided in this embodiment may include:

s100, point cloud to be processed of the laser radar is obtained.

The point cloud to be processed mentioned in the embodiment of the invention refers to any frame of point cloud fed back in the working process of the laser radar, and is the original data fed back by the laser radar. Based on the related art, the point cloud to be processed includes a plurality of laser points, and of course, the number of the specific laser points is related to the number of objects in the detection range of the laser radar and the performance parameters of the laser radar.

And S110, carrying out structuralization processing on the point cloud to be processed to obtain a structuralized point cloud.

Suppose the point cloud to be processed is P₀，P₀Can be represented as [ N₀,M]Wherein N is₀M represents the three-dimensional coordinate and reflection of any laser point for the number of all laser points in the point cloud to be processedStrength. In addition, based on the principle related to radar, each laser spot may further correspond to unique identification information based on performance parameters of the laser radar, specifically, the performance parameters described in this embodiment include a beam value and a horizontal angle value of the laser radar, for example, a 32-line laser radar with a horizontal resolution of 0.2 °, the number of the corresponding laser spots should be 32 × 360/0.2 — 57600, and correspondingly, for any laser spot, the corresponding line value and horizontal angle value correspond to unique line value and horizontal angle value, so that the line value and horizontal angle value corresponding to each laser spot may be used as the number of the corresponding stress spot, and in this embodiment, the number is defined as a sorting index.

Based on the above, during the structuring process, the laser points in the point cloud to be processed are sorted according to the corresponding sorting indexes, and the structured point cloud can be obtained. For example, the corresponding laser points may be sorted in the order of the horizontal angle value from 0 ° to 360 ° for the first line beam; and then, aiming at the second wire harness, sequencing the corresponding laser points according to the sequence of the horizontal angle from 0 degree to 360 degrees, and so on until the sequencing of all the laser points is completed to obtain the structured point cloud.

And S120, carrying out invalidation treatment on the laser points corresponding to the ground points in the structured point cloud to obtain the processed structured point cloud.

Because the point cloud to be processed inevitably includes laser points corresponding to the ground points, in order to reduce the amount of calculation, the laser points corresponding to the ground points in the structured point cloud need to be marked as invalid laser points, and the processed structured point cloud is finally obtained. Optionally, in order to avoid not affecting the structure of the structured point cloud, marking the corresponding stress light spot as an invalid laser spot may be implemented by setting the laser spot corresponding to the ground point as a null value. Specifically, the foregoing process of setting the laser points corresponding to the ground points to be null may be represented as P0[ i ] - [ nan, nan, nan, nan ], where i is an ordering index corresponding to the ground points, and further, the processed structured point cloud may be labeled as P.

It should be noted that, the determination of the laser point corresponding to the ground point in the point cloud to be processed may be implemented by combining with the related technology, which is not limited in the present invention.

S130, clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster.

When clustering the laser points of the processed structured point cloud, it should be ensured that the laser points corresponding to each non-ground point in the processed structured point cloud are traversed, and meanwhile, it is ensured that the laser points corresponding to each non-ground point determine the point cloud cluster to which the laser points belong.

Optionally, based on the above principle, defining laser points other than the laser points corresponding to the ground point in the processed structured point cloud as effective laser points, respectively taking each effective laser point as a target laser point according to the sequence of the ranking index of each effective laser point in the processed structured point cloud, and then determining a target clustering threshold corresponding to the target laser point. It should be noted that, in the method provided in the embodiment of the present invention, based on the euclidean distance between the target laser point and the preset origin of the laser radar, as a selection basis of the target clustering threshold, it can be understood that the farther the target laser point is from the preset origin, the more sparse the point cloud density is, the larger the corresponding clustering threshold should be, and the closer the target laser point is to the preset origin, the denser the point cloud density is, and the smaller the corresponding clustering threshold should be.

Specifically, the euclidean distance between the target laser point and the preset origin of the laser radar is calculated, and this distance is defined as the reference distance corresponding to the target laser point, and then the target clustering threshold corresponding to the target laser point can be determined according to the following formula and the size relationship between the reference distance and the preset reference distance threshold.

Wherein, T_hRepresenting a target clustering threshold corresponding to the target laser point;

D_pa reference distance representing a target laser point;

D_minrepresents a first reference distance threshold;

D_maxrepresenting a second reference distance threshold, D_maxGreater than D_min；

R represents a preset proportionality coefficient;

T_minrepresenting a first clustering threshold;

T_maxindicating a second classification threshold, T_maxGreater than T_min。

In combination with the above formula, it can be seen that if the reference distance of the target laser point is smaller than the first reference distance threshold, the first clustering threshold is taken as the target clustering threshold; correspondingly, if the obtained reference distance is greater than the second reference distance threshold, the second clustering threshold is used as a target clustering threshold; and if the reference distance of the target laser point is greater than or equal to the first reference distance threshold and the reference distance is less than or equal to the second reference distance threshold, taking the product of the reference distance and a preset proportionality coefficient as a target clustering threshold. The preset proportionality coefficient can be flexibly set based on specific performance parameters of the laser radar and actual identification requirements, and specific values of the preset proportionality coefficient are not limited.

After the target clustering threshold corresponding to the target laser point is determined according to the above content, whether the target laser point can be clustered into one class or not can be judged according to the Euclidean distance between the target laser point and other laser points. Specifically, if the euclidean distance between any laser point and the target laser point is smaller than the target clustering threshold, the laser point and the target laser point may be clustered into one class, and conversely, if the euclidean distance between the laser point and the target laser point is greater than or equal to the target clustering threshold, the laser point and the target laser point may not be clustered into one class.

It can be understood from the foregoing that, the structured point cloud includes a large number of laser points, and if it is determined in a traversal manner whether other laser points except the target laser point and the target laser point can be grouped into one type, the calculation amount is very large, and a large amount of hardware resources of the electronic device is inevitably occupied.

In order to solve the above problem, in this embodiment, a preset screening range is set, laser points in the structured point cloud that are centered on the target laser point and are in the preset screening range in the sorting index are used as candidate laser points, and then only whether the candidate laser points and the target laser point can be grouped into one type is determined. Respectively calculating Euclidean distances between each candidate laser point and a target laser point to obtain a screening distance corresponding to each candidate laser point, and clustering the candidate laser points with the screening distance smaller than a target clustering threshold value with the target laser points to obtain corresponding point cloud clusters.

By setting the preset screening range, the calculated amount in the clustering process can be effectively reduced, and the efficiency of target identification is improved. In practical application, the specific determination of the preset screening range needs to be flexibly selected by combining the specific number of the laser points in the structured point cloud and the requirements on identification accuracy and efficiency, and the specific setting of the preset screening range is not limited by the invention.

After traversing all the laser points in the structured point cloud, at least one point cloud cluster can be obtained through clustering, and the coarse graining treatment of the point cloud to be treated is realized through clustering treatment.

In practical application, the clustering process can be implemented by using the following program segments:

in the above program segment, point represents a laser point, point's around is a laser point in a preset screening range around the point, Dist (around, point) is an euler distance between a target laser point and other laser points, and threshold (point) is a target clustering threshold corresponding to the target laser point.

By adopting the clustering method, the processed structured point cloud P is clustered to generate a point cloud cluster set (P0, P1, P2 …, pN), wherein N is the total number of the clustered point cloud clusters, each point cloud cluster can be represented as a two-dimensional point cloud matrix [ Npi, M ], wherein Npi is the number of laser points in the pi-th point cloud cluster and is the original information of the laser radar sensor.

And S140, inputting the point cloud clusters into a feature extraction network respectively to obtain corresponding point cloud cluster feature vectors.

The laser radar target identification method provided by this embodiment provides a pre-trained feature extraction network, where the feature extraction network is obtained by training a neural network with the performance parameters of each laser point in a point cloud cluster as input and the feature vectors of the point cloud cluster as output, where the performance parameters of the laser points mainly include three-dimensional coordinates and reflection intensity of the laser points.

After each point cloud cluster is obtained, combining the three-dimensional coordinates and the reflection intensity corresponding to each laser point in each point cloud cluster into a two-dimensional vector matrix, further inputting the two-dimensional matrix into the characteristic extraction network, and converting the two-dimensional vector matrix into point cloud cluster feature vectors with the same length after the characteristic extraction network, wherein the point cloud cluster feature vectors are one-dimensional vectors.

Specifically, referring to fig. 2, fig. 2 is a schematic diagram of a process of extracting a point cloud cluster feature vector by using the feature extraction network provided by the embodiment of the present invention.

The input of the feature extraction network is the three-dimensional coordinates and the reflection intensity of each laser point in the clustered point cloud clusters (p0, p1, p2 … pN), the output is the feature vector (Fp0, Fp1, Fp2, … FpN) of each point cloud cluster, N is the number of the clustered point cloud clusters, and the feature extraction process can be represented by the following formula:

Fpi＝f(pi),i＝0,1,2…N

optionally, referring to fig. 2, the feature extraction network provided in the embodiment of the present invention implements the process f (pi), the feature extraction network may extract by using a pointent-like method, and a network structure of the feature extraction network may be as shown in fig. 2.

After the characteristic extraction network is adopted, the characteristics of each point cloud cluster are changed into one-dimensional vectors Fpi with the same length from a two-dimensional matrix [ Npi, M ], so that the characteristic vectors of all the point cloud clusters can form a two-dimensional matrix [ N, F ], wherein N is the number of the clustered point cloud clusters, and F is the length of the characteristic vector output by the characteristic extraction network.

And S150, inputting the point cloud cluster feature vectors into a target identification network to obtain the radar target.

Optionally, an embodiment of the present invention provides a pre-trained target recognition network, where the target recognition network is obtained by training a neural network with feature vectors of cloud clusters of each point as input and radar targets included in the cloud clusters of each point as output.

Based on the above, after the point cloud cluster feature vectors of each point cloud cluster are output in S140, the feature vectors of each point cloud cluster are combined into a point cloud cluster feature matrix [ N, F ], where N is the number of the point cloud clusters after clustering, and F is the length of the point cloud cluster feature vector output by the feature extraction network. And (3) taking the point cloud cluster feature matrix as input, combining specific requirements, and utilizing a trained target identification network to classify, envelop and regress each cluster of point cloud, divide and the like, so as to complete radar target extraction in the whole point cloud to be processed. This process can be seen in fig. 3.

Optionally, if other information, such as heading angle information, needs to be included in the recognition result, a network layer for recognizing the heading angle may also be added during training of the model.

In summary, the laser radar target identification method provided by the invention can significantly reduce the data amount to be processed by the identification model through the structuralization processing and the clustering processing of the point cloud to be processed, so as to simplify the network structure of the identification model, reduce the hardware performance requirement on the electronic equipment through the adjustment of the data amount and the network structure, and accordingly, under the condition that the electronic equipment with the same performance is used, the identification efficiency of the radar target can be effectively improved through the method.

Taking a mechanical 32-line laser radar as an example, the number of the original point clouds is 1800 × 32, and after clustering, about 300 point cloud clusters can be obtained, and the number of the point clouds is reduced by nearly 200 times. Meanwhile, the number of the point cloud clusters is related to the scene, and has no obvious relation with the line beam and the model of the laser radar, namely if the laser radar with higher line beam is adopted, the reduced data volume is more obvious, and if the laser radar is compared with an 80-line laser radar, the number of the point cloud clusters is reduced by nearly 500 times.

Furthermore, the characteristic extraction is carried out on each point cloud cluster, then the radar target identification is carried out, compared with the prior art, the identification effect is better than that of a network with the same quantity level, the point cloud cluster characteristics obtained through the characteristic extraction model are finer than the three-dimensional coordinates and the reflection intensity in the point cloud to be processed, and the subsequent first target identification model is easier to realize better extraction and identification effects through a simpler network.

In the following, the laser radar target identification apparatus provided in the embodiment of the present invention is introduced, and the laser radar target identification apparatus described below may be regarded as a functional module architecture that needs to be set in the central device to implement the laser radar target identification method provided in the embodiment of the present invention; the following description may be cross-referenced with the above.

Fig. 4 is a block diagram of a lidar target recognition apparatus according to an embodiment of the present invention, where referring to fig. 4, the apparatus may include:

the acquisition unit is used for acquiring point clouds to be processed of the laser radar;

the first processing unit is used for carrying out structuralization processing on the point cloud to be processed to obtain a structuralized point cloud;

the second processing unit is used for carrying out invalidation processing on the laser points corresponding to the ground points in the structured point cloud to obtain the processed structured point cloud;

the clustering unit is used for clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster;

the characteristic extraction unit is used for inputting the cloud clusters of each point into a characteristic extraction network respectively to obtain corresponding point cloud cluster characteristic vectors;

the characteristic extraction network is obtained by training a neural network by taking the performance parameters of each laser point in a point cloud cluster as input and taking the characteristic vector of the point cloud cluster as output;

the identification unit is used for inputting the cloud cluster feature vectors of all points into a target identification network to obtain a radar target;

the target recognition network is obtained by training a neural network by taking the characteristic vectors of the cloud clusters of each point as input and taking radar targets included in the cloud clusters of each point as output.

Optionally, the first processing unit 20 is configured to perform structuring processing on the point cloud to be processed to obtain a structured point cloud, and includes:

taking the beam value and the horizontal angle value corresponding to each laser point in the point cloud to be processed as the sequencing index of the corresponding stress light spot;

and sequencing the laser points in the point cloud to be processed according to the sequencing index to obtain the structured point cloud.

Optionally, the clustering unit 40 is configured to cluster laser points in the processed structured point cloud to obtain at least one point cloud cluster, and includes:

respectively taking each effective laser point in the processed structured point cloud as a target laser point;

the effective laser points are laser points except for laser points corresponding to the ground point in the processed structured point cloud;

acquiring a target clustering threshold corresponding to a target laser point;

taking the target laser point as the center and sequencing and indexing the laser points in the preset screening range in the processed structured point cloud as candidate laser points;

respectively calculating the Euclidean distance between each candidate laser point and the target laser point to obtain the screening distance corresponding to each candidate laser point;

and clustering the candidate laser points with the screening distance smaller than the target clustering threshold value with the target laser points to obtain corresponding point cloud clusters.

Optionally, the clustering unit 40 is configured to obtain a target clustering threshold corresponding to the target laser point, and includes:

calculating the Euclidean distance between a target laser point and a preset origin of the laser radar to obtain a reference distance corresponding to the target laser point;

and determining a target clustering threshold according to the size relation between the reference distance and the reference distance threshold.

Optionally, the clustering unit 40 is configured to determine the target clustering threshold according to a size relationship between the reference distance and the reference distance threshold, and includes:

if the reference distance is smaller than the first reference distance threshold, taking the first clustering threshold as a target clustering threshold;

if the reference distance is greater than the second reference distance threshold, taking the second clustering threshold as a target clustering threshold;

if the reference distance is greater than or equal to a first reference distance threshold and the reference distance is less than or equal to a second reference distance threshold, taking the product of the reference distance and a preset proportionality coefficient as a target clustering threshold;

wherein the first reference distance threshold is less than the second reference distance threshold;

the first clustering threshold is less than the second clustering threshold.

Optionally, the clustering unit 40 is configured to take each effective laser point in the structured point cloud as a target laser point, and includes:

and respectively taking each effective laser point as a target laser point according to the sequence of the sequencing indexes of each effective laser point in the processed structured point cloud.

Optionally, the feature extraction unit 50 is configured to input each point cloud cluster into a feature extraction network, and obtain a corresponding point cloud cluster feature vector, where the feature extraction network includes:

combining the three-dimensional coordinates and the reflection intensity corresponding to each laser point in each point cloud cluster into a two-dimensional vector matrix;

and inputting the two-dimensional matrix into a feature extraction network, and converting the two-dimensional vector matrix into point cloud cluster feature vectors with the same length after the feature extraction network, wherein the point cloud cluster feature vectors are one-dimensional vectors.

Optionally, the identifying unit 60 is configured to input the point cloud cluster feature vectors into a target identification network to obtain a radar target, and includes:

combining the point cloud cluster feature vectors into a point cloud cluster feature matrix;

and inputting the point cloud cluster characteristic matrix into a target identification network to obtain a radar target.

Fig. 5 is a block diagram of an electronic device according to an embodiment of the present invention, which is shown in fig. 5 and may include: at least one processor 100, at least one communication interface 200, at least one memory 300, and at least one communication bus 400;

in the embodiment of the present invention, the number of the processor 100, the communication interface 200, the memory 300, and the communication bus 400 is at least one, and the processor 100, the communication interface 200, and the memory 300 complete the communication with each other through the communication bus 400; it is clear that the communication connections shown by the processor 100, the communication interface 200, the memory 300 and the communication bus 400 shown in fig. 5 are merely optional;

optionally, the communication interface 200 may be an interface of a communication module, such as an interface of a GSM module;

the processor 100 may be a central processing unit CPU or an application Specific Integrated circuit asic or one or more Integrated circuits configured to implement embodiments of the present invention.

The memory 300, which stores application programs, may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

The processor 100 is specifically configured to execute an application program in the memory to implement any embodiment of the laser radar target identification method described above.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.

Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A laser radar target identification method is characterized by comprising the following steps:

acquiring point cloud to be processed of the laser radar;

structuring the point cloud to be processed to obtain a structured point cloud;

performing invalidation processing on the laser points corresponding to the ground points in the structured point cloud to obtain a processed structured point cloud;

clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster;

inputting each point cloud cluster into a feature extraction network to obtain corresponding point cloud cluster feature vectors;

the characteristic extraction network is obtained by training a neural network by taking the performance parameters of each laser point in the point cloud cluster as input and taking the characteristic vector of the point cloud cluster as output;

inputting the point cloud cluster feature vectors into a target identification network to obtain a radar target;

and the target identification network is obtained by training a neural network by taking the characteristic vector of each point cloud cluster as input and taking a radar target in each point cloud cluster as output.

2. The lidar target identification method according to claim 1, wherein the structuring of the point cloud to be processed to obtain a structured point cloud comprises:

taking the beam value and the horizontal angle value corresponding to each laser point in the point cloud to be processed as the sequencing index of the corresponding stress light spot;

and sequencing the laser points in the point cloud to be processed according to the sequencing index to obtain a structured point cloud.

3. The lidar target identification method according to claim 2, wherein the clustering laser points in the processed structured point cloud to obtain at least one point cloud cluster comprises:

respectively taking each effective laser point in the processed structured point cloud as a target laser point;

the effective laser points are laser points except for laser points corresponding to the ground point in the processed structured point cloud;

acquiring a target clustering threshold corresponding to the target laser point;

taking the laser points in the processed structured point cloud which take the target laser point as a center and are in a preset screening range in an ordered index manner as candidate laser points;

respectively calculating Euclidean distances between each candidate laser point and the target laser point to obtain a screening distance corresponding to each candidate laser point;

and clustering the candidate laser points with the screening distance smaller than the target clustering threshold value with the target laser points to obtain corresponding point cloud clusters.

4. The lidar target identification method according to claim 3, wherein the obtaining of the target clustering threshold corresponding to the target laser point comprises:

calculating the Euclidean distance between the target laser point and a preset origin of the laser radar to obtain a reference distance corresponding to the target laser point;

and determining a target clustering threshold according to the size relation between the reference distance and the reference distance threshold.

5. The lidar target identification method according to claim 4, wherein the determining a target clustering threshold according to the magnitude relationship between the reference distance and a reference distance threshold comprises:

if the reference distance is smaller than a first reference distance threshold value, taking the first clustering threshold value as a target clustering threshold value;

if the reference distance is larger than a second reference distance threshold, taking a second clustering threshold as a target clustering threshold;

if the reference distance is greater than or equal to the first reference distance threshold and the reference distance is less than or equal to the second reference distance threshold, taking the product of the reference distance and a preset proportionality coefficient as a target clustering threshold;

wherein the first reference distance threshold is less than the second reference distance threshold;

the first clustering threshold is less than the second clustering threshold.

6. The lidar target identification method according to claim 3, wherein the taking each effective laser point in the structured point cloud as a target laser point comprises:

and according to the sequence of the sequencing indexes of the effective laser points in the processed structured point cloud, respectively taking the effective laser points as target laser points.

7. The lidar target identification method according to claim 6, wherein the respectively inputting each point cloud cluster into the feature extraction network to obtain a corresponding point cloud cluster feature vector comprises:

combining the three-dimensional coordinates and the reflection intensity corresponding to each laser point in each point cloud cluster into a two-dimensional vector matrix;

and inputting the two-dimensional matrix into a feature extraction network, and converting the two-dimensional vector matrix into point cloud cluster feature vectors with the same length after passing through the feature extraction network, wherein the point cloud cluster feature vectors are one-dimensional vectors.

8. The lidar target identification method according to claim 7, wherein the inputting each point cloud cluster feature vector into a target identification network to obtain a radar target comprises:

combining the point cloud cluster feature vectors into a point cloud cluster feature matrix;

and inputting the point cloud cluster characteristic matrix into a target identification network to obtain a radar target.

9. A lidar target recognition apparatus comprising:

the acquisition unit is used for acquiring point clouds to be processed of the laser radar;

the first processing unit is used for carrying out structuring processing on the point cloud to be processed to obtain a structured point cloud;

the second processing unit is used for carrying out invalidation processing on the laser points corresponding to the ground points in the structured point cloud to obtain the processed structured point cloud;

the clustering unit is used for clustering the laser points in the processed structured point cloud to obtain at least one point cloud cluster;

the characteristic extraction unit is used for inputting each point cloud cluster into a characteristic extraction network to obtain corresponding point cloud cluster characteristic vectors;

the characteristic extraction network is obtained by training a neural network by taking the performance parameters of each laser point in the point cloud cluster as input and taking the characteristic vector of the point cloud cluster as output;

the identification unit is used for inputting the point cloud cluster feature vectors into a target identification network to obtain a radar target;

and the target identification network is obtained by training a neural network by taking the characteristic vector of each point cloud cluster as input and taking a radar target in each point cloud cluster as output.

10. An electronic device, comprising: a memory and a processor;

the memory stores a program adapted to be executed by the processor to implement the lidar target identification method of any of claims 1-8.

Images