CN117218629A - 3D target detection method, system, device and medium based on point cloud semantic information - Google Patents

Abstract

The invention discloses a 3D target detection method, system, device and storage medium based on point cloud semantic information. The method includes the following steps: obtaining point cloud data, inputting the point cloud data into a semantic segmentation model for processing, and obtaining a classification with classification Semantic segmentation results of the information; project the point cloud data after semantic segmentation to the bird's-eye view perspective; cluster the point clouds in the bird's-eye view perspective based on the semantic segmentation results to obtain the clustering results; perform central symmetry on each cluster Processing; generate target detection frames based on the symmetric clustering results and prior knowledge of object height. The present invention uses spatial prior knowledge and semantic segmentation algorithm to avoid problems such as large calculation amount and complex model of 3D target detection in autonomous driving scenes. It introduces point cloud semantic segmentation to provide more fine-grained scene perception for autonomous driving scenes, which can be widely used. Applied to the field of autonomous driving technology.

Description

3D target detection method, system, device and medium based on point cloud semantic information

Technical field

The present invention relates to the field of intelligent recognition technology, and in particular to a 3D target detection method, system, device and storage medium based on point cloud semantic information.

Background technique

A self-driving car is a vehicle that can automatically perceive the environment, position, make decisions and drive without manual control. The technology is divided into four key components: environment perception and modeling, localization and map construction, path planning and decision-making, and driving control. Among them, environmental perception has a crucial impact on subsequent steps and is also the most important step in determining the safety of autonomous vehicles. Target detection is a widely used technology in environmental perception. 2D object detection usually uses RGB images to predict the 2D bounding box, category and confidence of the object in the image. In recent years, the performance of 2D target detection methods for autonomous driving has been greatly improved, achieving an average accuracy higher than 90% in the KITTI target detection data set, but it lacks the direction and 3D position information of the target. Different from 2D target detection, 3D target detection usually uses RGB images, depth maps, lidar point cloud data to predict the 3D bounding box, category and other information of the target in 3D space, which makes up for the shortcomings of 2D target detection, but brings calculation Problems such as increased volume and complex models. Since 3D scene understanding is crucial for self-driving cars, more research is needed to remedy its shortcomings.

Contents of the invention

In order to solve one of the technical problems existing in the prior art to at least a certain extent, the purpose of the present invention is to provide a 3D target detection method, system, device and storage medium based on spatial prior knowledge and point cloud semantic information.

The technical solution adopted by the present invention is:

A 3D target detection method based on point cloud semantic information, including the following steps:

Obtain point cloud data, input the point cloud data into the semantic segmentation model for processing, and obtain semantic segmentation results with classification information;

Project the semantically segmented point cloud data to a bird's-eye view;

Cluster the point cloud in the bird's-eye view according to the semantic segmentation results to obtain the clustering results;

Perform center symmetry processing on each cluster;

Generate target detection frames based on the symmetric clustering results and prior knowledge of object height.

Further, the point cloud in the bird's-eye view perspective is clustered according to the semantic segmentation results to obtain clustering results, including:

Based on the semantic segmentation results, filter the point cloud in the bird's-eye view according to the target category that needs to be detected;

Use the density clustering algorithm to cluster the filtered point cloud data to obtain the clustering results.

Further, the density clustering algorithm is used to cluster the filtered point cloud data to obtain clustering results, including:

A1. Mark all objects in the point cloud data dataset as unprocessed;

A2. The objects are judged and processed based on the preset neighborhood radius Eps and the number of neighborhood objects MinPts in turn, and the objects are judged as core points, boundary points or noise points;

A3. Create a new cluster C based on the core point p, and add all objects within the neighborhood radius Eps of the core point p to the new cluster C;

A4. Obtain the unprocessed object q within the neighborhood radius Eps of the core point p. If the number of objects within the neighborhood radius Eps of the object q is greater than or equal to the number of neighborhood objects MinPts, add the neighborhood radius Eps of the object q Objects that are not classified into any cluster are added to new cluster C;

A5. Repeat the above steps A2-A4 until all objects are traversed and the clustering results are obtained.

Further, the A2 includes:

If an object contains points within its neighborhood radius Eps that exceed the number of neighborhood objects MinPts, the object is determined to be a core point;

If the number of points the object contains within its neighborhood radius Eps is less than the number of neighborhood objects MinPts, and the object falls within the neighborhood radius Eps of the core point, the object is determined to be a boundary point;

If an object is neither a core point nor a boundary point, the object is determined to be a noise point.

Further, performing central symmetry processing on each cluster includes:

For each cluster, obtain the leftmost point and the rightmost point from the origin of the point cloud in the cluster, and connect them into line segments;

Get the midpoint of the line segment;

All points of the cluster are processed centrally symmetrically based on the midpoint of the line segment.

Further, the target detection frame is generated based on the symmetric clustering results and prior knowledge of the object height, including:

Generate wrapping polygons for each cluster;

Draw the smallest enclosing rectangle based on the enclosing polygon;

Combined with the prior knowledge of the target height, the target detection frame is generated.

Another technical solution adopted by the present invention is:

A 3D target detection system based on point cloud semantic information, including:

The semantic segmentation module is used to obtain point cloud data, input the point cloud data into the semantic segmentation model for processing, and obtain semantic segmentation results with classification information;

The point cloud projection module is used to project the point cloud data after semantic segmentation to a bird's-eye view;

The point cloud clustering module is used to cluster point clouds in the bird's-eye view based on semantic segmentation results to obtain clustering results;

The symmetry processing module is used to perform central symmetry processing on each cluster;

The detection frame generation module is used to generate target detection frames based on the symmetric clustering results and prior knowledge of object height.

Further, performing central symmetry processing on each cluster includes:

For each cluster, find the leftmost and rightmost points from the origin of the point cloud and connect them into line segments;

Find the midpoint of the line segment;

All points of the cluster are processed centrally symmetrically based on the midpoint of the line segment.

Further, the target detection frame is generated based on the symmetric clustering results and prior knowledge of the object height, including:

Generate wrapping polygons for each cluster;

Draw the smallest enclosing rectangle based on the enclosing polygon;

Combined with the prior knowledge of the target height, the target detection frame is generated.

Another technical solution adopted by the present invention is:

A 3D target detection device based on point cloud semantic information, including:

at least one processor;

At least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the method as described above.

Another technical solution adopted by the present invention is:

A computer-readable storage medium in which a processor-executable program is stored, and the processor-executable program is used to perform the above method when executed by the processor.

The beneficial effects of the present invention are: the present invention applies the point cloud semantic segmentation algorithm to the target detection task, and reduces the problems of large calculation amount and complex model of 3D target detection in autonomous driving scenarios by fusing prior knowledge and semantic segmentation algorithm.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following is an introduction to the accompanying drawings of the embodiments of the present invention or the relevant technical solutions in the prior art. It should be understood that the drawings in the following introduction are only In order to facilitate and clearly describe some embodiments of the technical solutions of the present invention, those skilled in the art can also obtain other drawings based on these drawings without exerting creative efforts.

Figure 1 is a schematic flow chart of a 3D target detection method based on spatial prior knowledge and point cloud semantic information in an embodiment of the present invention;

Figure 2 is a schematic diagram of a bird's-eye view projection method in an embodiment of the present invention;

Figure 3 is a schematic diagram of each centrally symmetric cluster cluster in the embodiment of the present invention;

Figure 4 is a flow chart of a 3D target detection method based on spatial prior knowledge and point cloud semantic information in an embodiment of the present invention.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and cannot be understood as limiting the present invention. The step numbers in the following embodiments are only set for the convenience of explanation. The order between the steps is not limited in any way. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art. sexual adjustment.

In the description of the present invention, it should be understood that orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or position relationships shown in the drawings and are only In order to facilitate the description of the present invention and simplify the description, it is not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be construed as a limitation of the present invention.

In the description of the present invention, several means one or more, plural means two or more, greater than, less than, more than, etc. are understood to exclude the original number, and above, below, within, etc. are understood to include the original number. If there is a description of first and second, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the order of indicated technical features. relation.

Furthermore, in the description of the present invention, "plurality" means two or more unless otherwise stated. "And/or" describes the relationship between associated objects, indicating that there can be three relationships. For example, A and/or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the related objects are in an "or" relationship.

In the description of the present invention, unless otherwise explicitly limited, words such as setting, installation, and connection should be understood in a broad sense. Those skilled in the art can reasonably determine the specific meaning of the above words in the present invention in combination with the specific content of the technical solution.

Based on point cloud semantic segmentation, the road area can be output based on the point cloud of the bird's-eye view category, and the vehicle's feasible area can be obtained based on the position of the target frame, thereby providing scene information required for autonomous driving/assisted driving. While avoiding problems such as increased computational load and complex models caused by 3D target detection methods, it can also provide more fine-grained scene perception for autonomous driving scenarios. Based on this, this application provides a 3D target detection method, system, device and medium based on spatial prior knowledge and point cloud semantic information, so as to reduce the cost of 3D target detection in autonomous driving scenarios by integrating prior knowledge and semantic segmentation algorithms. Problems such as large amount of calculation and complex model.

As shown in Figures 1 and 4, this embodiment provides a 3D target detection method based on spatial prior knowledge and point cloud semantic information, including the following steps:

S1. Obtain point cloud data, input the point cloud data into the semantic segmentation model for processing, and obtain semantic segmentation results with classification information. Realize semantic segmentation of point cloud data.

S2. Project the point cloud data after semantic segmentation to a bird's-eye view perspective. As shown in Figure 2, Figure 2 is a schematic diagram of the projection process.

S3. Cluster the point cloud in the bird's-eye view according to the semantic segmentation results to obtain the clustering results.

In this embodiment, step S3 specifically includes steps S31-S32:

S31. Based on the semantic segmentation results, filter the point clouds in the bird's-eye view according to the target categories that need to be detected; for example, if you need to detect car targets, keep the point clouds in the bird's-eye view whose semantic segmentation results are cars.

S32. Use density clustering algorithm to cluster the filtered point cloud data to obtain clustering results. The following uses the DBSCAN clustering algorithm as an example to illustrate:

1. Define parameters:

1) Eps: Neighborhood radius when defining density.

1)MinPts: The threshold when defining core points, the minimum number of core points required to form a cluster.

2. In the DBSCAN algorithm, data points are divided into the following three categories:

1) Core points: points inside dense areas

If an object contains more than MmPts points within its radius Eps, the object is a core point.

2) Boundary points: points on the edge of dense areas

If an object contains a number of points less than MinPts within its radius Eps, but the object falls within the neighborhood of the core point, the object is a boundary point.

3) Noise points: points in sparse areas

If an object is neither a core point nor a boundary point, the object is a noise point.

3. The relationship between data points in the DBSCAN algorithm

Density direct: If P is the core point and Q is in the R neighborhood of P, then it is called density direct from P to Q. Any core point is directly connected to its own density, and the density is direct without symmetry. If P is directly connected to Q, then Q is not necessarily directly connected to P.

Density reachability: If there are core points P2, P3,..., Pn, and the density from P1 to P2 is direct, and the density from P2 to P3 is direct,..., the density from P(n-1) to Pn is direct, and the density from Pn to Q is direct, Then the density from P1 to Q is reachable. Density attainable also has no symmetry.

Density connected: If there is a core point S such that S to P and Q are both density reachable, then P and Q are density connected. Density connection has symmetry. If P and Q are density connected, then Q and P must also be density connected. Two points that are densely connected belong to the same cluster.

Non-density connected: If two points do not belong to the density-connected relationship, then the two points are not density-connected. Two points that are not densely connected belong to different clusters, or there are noise points in them.

4. DBSCAN algorithm steps

1) Determine all core objects according to the given neighborhood parameters Eps and MinPts.

2) For each core object: select an unprocessed core object and find samples with reachable density to generate clustering "clusters".

3) Repeat the above process.

As an optional implementation, the pseudo code of the DBSCAN algorithm is as follows:

S4. Perform center symmetry processing on each cluster.

Referring to Figure 3, for each cluster (i.e., a group of points adjacent in space obtained through the clustering algorithm), the maximum distance between the cluster cluster and the point cloud origin (i.e., the coordinates of the lidar) is calculated. The left point and the rightmost point (take the origin as the origin, draw a circle clockwise with a straight line, the first point of contact is the leftmost point, and the last point left is the rightmost point), connect the rightmost point and the leftmost point , find the midpoint of the line segment point, and all points of the cluster will be centrally symmetrical based on the line segment midpoint.

Specifically, for each cluster, find the leftmost and rightmost points within the cluster from the origin of the point cloud. Connect the leftmost point and the rightmost point into a line segment. Find the midpoint of the line segment, and then center-symmetry all points in the cluster based on the midpoint of the line segment. This step utilizes the symmetry information of large objects (such as vehicles) to make subsequent target detection frame positions more accurate.

S5. Generate a target detection frame based on the symmetric clustering results and prior knowledge of the object height.

In this embodiment, step S5 specifically includes steps S51-S53:

S51. Generate outsourcing polygons for each cluster;

S52. Draw the smallest enclosing rectangle based on the enclosing polygon;

S53. Generate a target detection frame based on the prior knowledge of the target height.

It can be seen from the above that the method provided by the embodiment of the present invention is applied to obtain point cloud data to be identified; the point cloud data is input into the semantic segmentation model for processing, and a prediction result with classification information is obtained; the point cloud data after semantic segmentation is Project to a bird's-eye view perspective; filter the bird's-eye view point cloud based on the target category detected on demand. Use the density clustering algorithm to cluster the point cloud on the bird's-eye view to obtain multiple clusters; for each cluster, perform central symmetry processing on the internal points based on the cluster center point; calculate the outsourcing of each cluster Rectangle, and then based on the high degree of prior knowledge of the category, a target detection frame is formed to obtain the prediction result. This method uses spatial prior knowledge and semantic segmentation algorithms to avoid the problems of large calculation amount and complex model of 3D target detection in autonomous driving scenarios. At the same time, the road area can be output based on the point cloud of the bird's-eye view category, and the vehicle's feasible area can be obtained based on the position of the target frame, providing more fine-grained scene perception for autonomous driving scenarios.

This embodiment also provides a 3D target detection system based on point cloud semantic information, including:

The semantic segmentation module is used to obtain point cloud data, input the point cloud data into the semantic segmentation model for processing, and obtain semantic segmentation results with classification information;

The point cloud projection module is used to project the point cloud data after semantic segmentation to a bird's-eye view;

The point cloud clustering module is used to cluster point clouds in the bird's-eye view based on semantic segmentation results to obtain clustering results;

The symmetry processing module is used to perform central symmetry processing on each cluster;

The detection frame generation module is used to generate target detection frames based on the symmetric clustering results and prior knowledge of object height.

As a further optional implementation, performing central symmetry processing on each cluster includes:

For each cluster, find the leftmost and rightmost points from the origin of the point cloud and connect them into line segments;

Find the midpoint of the line segment;

All points of the cluster are processed centrally symmetrically based on the midpoint of the line segment.

As a further optional implementation, generating a target detection frame based on the symmetric clustering results and prior knowledge of the object height includes:

Generate wrapping polygons for each cluster;

Draw the smallest enclosing rectangle based on the enclosing polygon;

Combined with the prior knowledge of the target height, the target detection frame is generated.

The 3D target detection system based on point cloud semantic information in this embodiment can execute the 3D target detection method based on point cloud semantic information provided by the method embodiment of the present invention, and can execute any combination of implementation steps of the method embodiment. , possessing the corresponding functions and beneficial effects of this method.

This embodiment also provides a computer-readable storage medium in which a processor-executable program is stored, and the processor-executable program is used to perform the method shown in Figure 4 when executed by the processor.

The vehicle destination prediction system based on user behavior of this embodiment can execute the vehicle destination prediction method based on user behavior provided by the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding requirements of the method. functions and beneficial effects.

The embodiment of the present application also discloses a computer program product or computer program. The computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the method shown in FIG. 4 .

This embodiment also provides a storage medium that stores instructions or programs that can execute a 3D target detection method based on point cloud semantic information provided by the method embodiment of the present invention. When the instructions or programs are run, Any combination of implementation steps of the method embodiments has the corresponding functions and beneficial effects of the method.

In some alternative embodiments, the functions/operations noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending on the functionality/operations involved. Furthermore, the embodiments presented and described in the flow diagrams of the present invention are provided by way of example for the purpose of providing a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of a larger operation are performed independently.

Furthermore, although the present invention has been described in the context of functional modules, it should be understood that, unless stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or or software modules, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion regarding the actual implementation of each module is not necessary to understand the invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be within the ordinary skill of an engineer, taking into account the properties, functions and internal relationships of the modules. Therefore, a person skilled in the art using ordinary skills can implement the invention set forth in the claims without undue experimentation. It will also be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the invention, which is to be determined by the full scope of the appended claims and their equivalents.

If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

The logic and/or steps represented in the flowcharts or otherwise described herein, for example, may be considered a sequenced list of executable instructions for implementing the logical functions, and may be embodied in any computer-readable medium, For use by, or in combination with, instruction execution systems, devices or devices (such as computer-based systems, systems including processors or other systems that can fetch instructions from and execute instructions from the instruction execution system, device or device) or equipment. For the purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wires (electronic device), portable computer disk cartridges (magnetic device), random access memory (RAM), Read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, and subsequently edited, interpreted, or otherwise suitable as necessary. process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if it is implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following technologies known in the art: a logic gate circuit with a logic gate circuit for implementing a logic function on a data signal. Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

In the above description of this specification, reference to the description of the terms "one embodiment/example", "another embodiment/example" or "certain embodiments/examples" etc. is meant to be described in connection with the embodiment or example Specific features, structures, materials, or characteristics are included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and purposes of the invention. The scope of the invention is defined by the claims and their equivalents.

The above is a detailed description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art can also make various equivalent modifications or substitutions without violating the spirit of the present invention. Equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims (10)

1. The 3D target detection method based on the point cloud semantic information is characterized by comprising the following steps of:

acquiring point cloud data, inputting the point cloud data into a semantic segmentation model for processing, and acquiring a semantic segmentation result with classification information;

projecting the semantically segmented point cloud data to a bird's eye view perspective;

clustering point clouds in the aerial view angle according to the semantic segmentation result to obtain a clustering result;

performing central symmetry processing on each cluster;

and generating a target detection frame based on the symmetrical clustering result and the priori knowledge of the object height.

2. The 3D object detection method based on the semantic information of the point cloud according to claim 1, wherein the clustering the point cloud in the bird's eye view perspective according to the semantic segmentation result to obtain a clustering result includes:

according to semantic segmentation results, filtering point clouds in the aerial view angle according to target categories to be detected;

and clustering the filtered point cloud data by using a density clustering algorithm to obtain a clustering result.

3. The 3D object detection method based on the point cloud semantic information according to claim 2, wherein the clustering the filtered point cloud data using the density clustering algorithm to obtain a clustering result includes:

a1, marking all objects in a data set of point cloud data as unprocessed states;

a2, judging the objects according to a preset neighborhood radius Eps and a neighborhood object number MinPts in sequence, and judging the objects as core points, boundary points or noise points;

a3, a new cluster C is established according to the core point p, and all objects in the neighborhood radius Eps of the core point p are added into the new cluster C;

a4, acquiring an object q which is not processed in the neighborhood radius Eps of the core point p, and adding an object which is not classified into any cluster in the neighborhood radius Eps of the object q into a new cluster C if the number of the objects in the neighborhood radius Eps of the object q is greater than or equal to the neighborhood object number MinPts;

a5, repeating the steps A2-A4 until all the objects are traversed, and obtaining a clustering result.

4. A 3D object detection method based on point cloud semantic information according to claim 3, wherein the A2 comprises:

if the object contains points exceeding the number MinPts of the neighborhood objects in the neighborhood radius Eps, judging that the object is a core point;

if the number of the points contained in the neighborhood radius Eps of the object is smaller than the number MinPts of the neighborhood objects and the object falls in the neighborhood radius Eps of the core points, judging the object as a boundary point;

if an object is neither a core point nor a boundary point, the object is determined to be a noise point.

5. The 3D object detection method based on the point cloud semantic information according to claim 1, wherein the performing central symmetry processing on each cluster includes:

for each cluster, obtaining the leftmost point and the rightmost point which are far from the point cloud origin of the cluster, and connecting the leftmost point and the rightmost point into a line segment;

acquiring the midpoint of the line segment;

and carrying out center symmetry processing on all points of the cluster based on the midpoints of the line segments.

6. The 3D object detection method based on point cloud semantic information according to claim 1, wherein the generating the object detection frame based on the symmetric clustering result and the priori knowledge of the object height comprises:

generating an outsourcing polygon of each cluster;

drawing a minimum circumscribed rectangle based on the outsourcing polygon;

and generating a target detection frame by combining the prior knowledge of the target height.

7. A 3D object detection system based on point cloud semantic information, comprising:

the semantic segmentation module is used for acquiring point cloud data, inputting the point cloud data into the semantic segmentation model for processing, and acquiring semantic segmentation results with classification information;

the point cloud projection module is used for projecting the point cloud data subjected to semantic segmentation to a bird's eye view perspective;

the point cloud clustering module is used for clustering point clouds in the aerial view according to the semantic segmentation result to obtain a clustering result;

the symmetrical processing module is used for carrying out central symmetrical processing on each cluster;

and the detection frame generation module is used for generating a target detection frame based on the symmetrical clustering result and the priori knowledge of the object height.

8. The 3D object detection system based on point cloud semantic information according to claim 7, wherein the performing a central symmetry process on each cluster includes:

for each cluster, finding the leftmost point and the rightmost point which are far away from the origin of the point cloud, and connecting the leftmost point and the rightmost point into a line segment;

finding the midpoint of the line segment;

and carrying out center symmetry processing on all points of the cluster based on the midpoints of the line segments.

9. 3D target detection device based on point cloud semantic information, characterized by comprising:

at least one processor;

at least one memory for storing at least one program;

the at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1-6.

10. A computer readable storage medium, in which a processor executable program is stored, characterized in that the processor executable program is for performing the method according to any of claims 1-6 when being executed by a processor.