A point cloud encoding method, system, device and storage medium

Abstract

The present disclosure relates to the technical field of computer vision, and provides a point cloud encoding method, a system, a device and a storage medium, wherein the point cloud encoding method comprises: obtaining a to-be-encoded point cloud sample, and dividing the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns; performing feature representation on points in each point cloud column, and determining a target tensor representation of the to-be-encoded point cloud sample according to the feature representation; determining a target feature map of the to-be-encoded point cloud sample in a target dimension according to the target tensor representation of the to-be-encoded point cloud sample; and determining a pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map. Through the division of the point cloud column, the present disclosure can reduce the interference of high and low points on other position points, and solves the problem of low point cloud encoding accuracy in the prior art.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of computer vision, and particularly relates to a point cloud encoding method, system, device and storage medium. BACKGROUND

[0002] Point cloud data is a point set obtained by using a measuring instrument to collect data on the surface of an object. Point cloud data is widely used in automatic driving, high-precision maps, virtual reality and other aspects. The key points of point cloud encoding are two aspects, one is the compression algorithm: point cloud data is usually very large, and compressing point cloud can effectively reduce the relevant calculation amount; the other is point cloud feature extraction: the core of point cloud encoding is to extract the features of point cloud, so as to perform effective representation and processing.

[0003] In the current automatic driving scene, the encoding result may be disturbed by high point cloud and low point cloud in the process of point cloud encoding, thereby affecting the effect of point cloud encoding, and the accuracy of point cloud encoding is low.

[0004] At present, there is no good solution to the problem of low accuracy of point cloud encoding in the existing related technology. SUMMARY

[0005] In view of the above problems, the present disclosure provides a point cloud encoding method, system, device and storage medium to solve the problem of low accuracy of point cloud encoding in the existing related technology.

[0006] The present disclosure provides a point cloud encoding method, comprising:

[0007] Obtaining a to-be-encoded point cloud sample, and dividing the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns;

[0008] Performing feature representation on points in each point cloud column, and determining a target tensor representation of the to-be-encoded point cloud sample according to the feature representation;

[0009] Determining a target feature map of the to-be-encoded point cloud sample in a target dimension according to the target tensor representation of the to-be-encoded point cloud sample;

[0010] Determining a pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map.

[0011] Preferably, before performing feature representation on points in each point cloud column, the method comprises:

[0012] A preset point cloud quantity standard is set, and the number of points in each point cloud column is adjusted according to the point cloud quantity standard.

[0013] Preferably, adjusting the number of points in each point cloud column according to the point cloud quantity standard comprises:

[0014] For the point cloud column whose number of points exceeds the point cloud number standard, the points in the point cloud column are sampled so that the number of points in the point cloud column is equal to the point cloud number standard;

[0015] For the point cloud column whose number of points is less than the point cloud number standard, the point cloud column is completed with "0" so that the number of points in the point cloud column is equal to the point cloud number standard.

[0016] Preferably, the feature representation of the points in the point cloud column includes the coordinates, reflection intensity, normal vector of the points in the point cloud column, the geometric center of the point cloud column, the relative position of the points to the geometric center of the point cloud column, and the difference between the distance of the points to the geometric center of the point cloud column and the average distance of the points to the remaining points in the point cloud column.

[0017] Preferably, determining the target tensor representation of the to-be-encoded point cloud sample according to the feature representation includes:

[0018] generating a first tensor representation of the to-be-encoded point cloud sample according to the feature representation of each point in the point cloud column;

[0019] dimensionally adjusting the first tensor representation of the to-be-encoded point cloud sample to determine a second tensor representation of the to-be-encoded point cloud sample in the target dimension;

[0020] on the basis of the second tensor representation of the to-be-encoded point cloud sample, assigning a weight to each point in the point cloud column according to a predefined weighting function to determine the target tensor representation of the to-be-encoded point cloud sample.

[0021] Preferably, determining the target feature map of the to-be-encoded point cloud sample in the target dimension according to the target tensor representation of the to-be-encoded point cloud sample includes:

[0022] max-pooling the points in each point cloud column to adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a first feature map;

[0023] average-pooling the points in each point cloud column to adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a second feature map;

[0024] stitching the first feature map and the second feature map to obtain a third feature map;

[0025] dimensionally adjusting the third feature map to obtain the target feature map of the encoded point cloud sample in the target dimension.

[0026] Preferably, determining the pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map includes:

[0027] According to the position of each point cloud column, the target feature map is unfolded in the target dimension to obtain the pseudo image feature of the to-be-encoded point cloud sample.

[0028] The present disclosure provides a point cloud encoding system, comprising:

[0029] A division module is configured to obtain a to-be-encoded point cloud sample and divide the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns.

[0030] An identification module is configured to perform feature representation on points in each point cloud column and determine a target tensor representation of the to-be-encoded point cloud sample according to the feature representation.

[0031] A processing module is configured to determine a target feature map of the to-be-encoded point cloud sample in a target dimension according to the target tensor representation of the to-be-encoded point cloud sample.

[0032] A determination module is configured to determine a pseudo image feature of the to-be-encoded point cloud sample according to the target feature map.

[0033] The present disclosure provides an electronic device, comprising:

[0034] A processor and a memory.

[0035] The processor invokes a computer program stored in the memory to execute the point cloud encoding method described above.

[0036] The present disclosure provides a computer readable storage medium,

[0037] The computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor can execute the point cloud encoding method described above.

[0038] The present disclosure has the following beneficial effects:

[0039] The present disclosure provides a point cloud encoding method. After obtaining a to-be-encoded point cloud sample, the present disclosure can conveniently separate points at different positions through point cloud column division, which helps to better process and encode points at different positions in subsequent processes, reduces the interference of high and low points on other position points, and solves the problem of low point cloud encoding accuracy in the prior art.

[0040] Other features and advantages of the present disclosure will be described in the following description, and some will become apparent from the description, or will be understood from the practice of the present disclosure. The purpose and other advantages of the present disclosure can be achieved and obtained by the structure indicated in the specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present disclosure, and for those skilled in the art, other drawings can also be obtained from these drawings without creative effort.

[0042] Figure 1 A flow chart of a point cloud encoding method in an embodiment of the present disclosure is shown.

[0043] Figure 2 A structural block diagram of a point cloud encoding system in an embodiment of the present disclosure is shown.

[0044] Figure 3 An electronic device diagram in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0045] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations can be implemented in any number of ways, and are not limited to the examples described herein. Rather, examples are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example implementations to those skilled in the art. The described features, structures, or characteristics can be combined in one or more implementations. In the following description, numerous specific details are provided to give a thorough understanding of implementations of the present disclosure. One skilled in the relevant art will recognize, however, that the implementations of the present disclosure can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures have not been described in detail to avoid obscuring the understanding of this description. The examples described herein are not intended to be exhaustive or to be limited to the examples presented in this description.

[0046] In addition, the drawings are merely schematic and are non-essential to the present disclosure. Like reference numerals designate like elements throughout the drawings, and thus a repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, and do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in software, or in one or more hardware units or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0047] The flowcharts shown in the drawings are merely illustrative, and do not necessarily include all the steps. For example, some steps can be further divided, and some steps can be combined or partially combined, so the actual execution order can be changed according to the actual situation.

[0048] The terms "first", "second", and the like, in the description and in the claims of the present application, as well as above-mentioned drawings, are used to distinguish similar objects and are not necessarily used to describe a specific sequential or chronological order. It should be understood that the data thus used can be interchanged, where appropriate, so that the embodiments of the present application described herein can be carried out in other than the order shown or described herein.

[0049] Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or apparatus that comprises a list of steps or sub-modules not necessarily limited to those explicitly listed, but can include other steps or sub-modules not expressly listed or inherent to such processes, methods, products or apparatus.

[0050] A point cloud encoding method, as shown in Figure 1 , Figure 1 is a flowchart of the point cloud encoding method proposed by the present disclosure, which comprises the following steps:

[0051] Obtain a point cloud sample to be encoded, and divide the point cloud sample to be encoded into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns.

[0052] Perform feature representation on the points in each point cloud column, and determine a target tensor representation of the point cloud sample to be encoded according to the feature representation.

[0053] Determine a target feature map of the point cloud sample to be encoded in a target dimension according to the target tensor representation of the point cloud sample to be encoded.

[0054] Determine a pseudo-image feature of the point cloud sample to be encoded according to the target feature map.

[0055] In the above method, after obtaining the to-be-encoded point cloud sample, the size of the point cloud column is set, and the point cloud space of the to-be-encoded point cloud sample is divided into a point cloud column grid. For example, the horizontal and vertical sizes of the point cloud column grid can be the range of the point cloud space in the X-axis direction / the size of the point cloud column and the range of the point cloud space in the Y-axis direction / the size of the point cloud column, respectively. The point cloud column grid contains a plurality of point cloud columns, and each point in the to-be-encoded point cloud sample is also contained in each point cloud column, wherein a point cloud column without points can be regarded as a null point cloud column. Through the division of the point cloud column, the points at different positions can be conveniently separated and processed, which helps to better process and encode the points at different positions in the subsequent process, reduces the interference of high and low points on other position points, and then performs feature representation on the points in each point cloud column. The feature representation is mainly used to represent the position information of the points in the point cloud column. Then, the target tensor representation of the entire to-be-encoded point cloud sample is determined according to the feature representation, and the target tensor representation can represent the attribute information of the to-be-encoded point cloud sample. Next, the target feature map of the to-be-encoded point cloud sample in the target dimension is determined according to the target tensor representation, and the target dimension can be adjusted according to the application scenario. Finally, the to-be-encoded point cloud sample is encoded according to the target feature map to determine the pseudo-image feature of the to-be-encoded point cloud sample. Through the above method, the to-be-encoded point cloud sample can be divided into regions, the interference of high and low points on other point encoding is reduced, and the problem of low point cloud encoding accuracy in the prior art is solved.

[0056] In some embodiments, before the feature representation of the points in each point cloud column, a point cloud quantity standard is preset, and the number of points in each point cloud column is adjusted according to the point cloud quantity standard. In this embodiment, the number of points in each point cloud column is adjusted uniformly, which helps to improve the efficiency of encoding the to-be-encoded point cloud sample in combination with different point cloud columns.

[0057] Specifically, adjusting the number of points in each point cloud column according to the point cloud quantity standard includes: for the point cloud column whose number of points exceeds the point cloud quantity standard, sampling the points in the point cloud column so that the number of points in the point cloud column is equal to the point cloud quantity standard; and for the point cloud column whose number of points is less than the point cloud quantity standard, the point cloud column is completed with “0” so that the number of points in the point cloud column is equal to the point cloud quantity standard.

[0058] For example, assuming that the number of non-null point cloud columns contained in the to-be-encoded point cloud sample is P, and the point cloud quantity standard of the points in each point cloud column is N, if the number of points in a point cloud column is less than N, the point cloud column is completed with “0”, and if the number of points in a point cloud column exceeds N, N points are sampled from the points in the point cloud column.

[0059] The feature representation is mainly used to represent the position information of the point cloud column center point, and can be selected according to application scenarios and use requirements. In some embodiments, the feature representation of the point cloud column center point includes the coordinates of the point cloud column center point, the reflection intensity, the normal vector, the geometric center of the point cloud column, the relative position of the point to the geometric center of the point cloud column, and the difference between the point and the average distance of the point to the remaining points in the point cloud column.

[0060] In some embodiments, the target tensor representation of the to-be-encoded point cloud sample is determined according to the feature representation, including: generating a first tensor representation of the to-be-encoded point cloud sample according to the feature representation of each point in the point cloud column; performing dimension adjustment on the first tensor representation of the to-be-encoded point cloud sample to determine a second tensor representation of the to-be-encoded point cloud sample in the target dimension; and determining a target tensor representation of the to-be-encoded point cloud sample according to a pre-defined weighting function to assign weights to the points in each point cloud column based on the second tensor representation of the to-be-encoded point cloud sample.

[0061] For example, for a to-be-encoded point cloud sample containing P non-empty point clouds and N point cloud standards, the feature representation of each point in the point cloud column is denoted as D. The first tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, D). Then, convolution is used for the points in each point cloud column, so that the dimension of each point changes from D to C. The second tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, C). The weighting function is defined in the laser radar coordinate system, and the expression of the weighting function is as follows:

[0062]

[0063] wherein z is the coordinate of the point in the point cloud column on the Z axis, σ1 and σ2 are 1 in this embodiment, a is the lowest point coordinate of the main attention area in the actual application of autonomous driving, b is the highest point coordinate of the main attention area in the actual application of autonomous driving, μ1 is a, and μ2 is b. Based on the above weighting function, the weight f(z) is assigned to the points in the point cloud column. The tensor of the to-be-encoded point cloud sample can be represented as (P, N, C·f(z)), that is, the target tensor (P, N, K).

[0064] In some embodiments, the target feature map of the to-be-encoded point cloud sample in the target dimension is determined according to the target tensor representation of the to-be-encoded point cloud sample, including: performing max-pooling on the points in each point cloud column to adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a first feature map; performing average-pooling on the points in each point cloud column to adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a second feature map; splicing the first feature map and the second feature map to obtain a third feature map; and performing dimension adjustment on the third feature map to obtain the target feature map of the encoded point cloud sample in the target dimension.

[0065] Exemplarily, using maximum pooling on the points in each point cloud column obtains the feature vector of each point cloud column, and also causes the N dimension in the target tensor of the to-be-encoded point cloud sample to disappear, obtaining a first feature map of the to-be-encoded point cloud sample in the (P, K) dimension. Using average pooling on the points in the point cloud column obtains the feature vector of each point cloud column, and also causes the N dimension in the target tensor of the to-be-encoded point cloud sample to disappear, obtaining a second feature map of the to-be-encoded point cloud sample in the (P, K) dimension. Then, the first feature map and the second feature map are spliced to obtain a third feature map of the to-be-encoded point cloud sample in the (P, 2K) dimension. Finally, the third feature map is dimensionally adjusted, and a fully connected layer is used to convert the dimension of the third feature map to (P, K) to obtain the target feature map of the to-be-encoded point cloud sample in the target dimension.

[0066] In some embodiments, determining the pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map comprises: unfolding the target feature map in the target dimension according to the position of each point cloud column to obtain the pseudo-image feature of the to-be-encoded point cloud sample.

[0067] Exemplarily, the target feature map in the (P, K) dimension is unfolded into a pseudo-image feature according to the position of the point cloud column, and specifically, P is unfolded into (H, W) to obtain a feature representation in the form of (K, H, W) similar to an image.

[0068] In summary, the present disclosure adopts a technical solution of dividing the point cloud sample space into point cloud columns, limiting the number of points in the point cloud column, and assigning weights to the points in the point cloud column according to the distribution of the point cloud on the Z axis, which solves the interference of high and low point clouds in the actual application of autonomous driving and effectively improves the effect of point cloud feature extraction.

[0069] The present disclosure also provides a point cloud encoding system, as shown in Figure 2 , Figure 2 is a structural block diagram of the point cloud encoding system proposed by the present disclosure, and the system comprises:

[0070] The division module is configured to obtain a to-be-encoded point cloud sample and divide the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns.

[0071] The identification module is configured to perform feature representation on the points in each point cloud column and determine a target tensor representation of the to-be-encoded point cloud sample according to the feature representation.

[0072] The processing module is configured to determine a target feature map of the to-be-encoded point cloud sample in a target dimension according to the target tensor representation of the to-be-encoded point cloud sample.

[0073] The determination module is configured to determine a pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map.

[0074] In the system, after obtaining the to-be-encoded point cloud sample, the size of the point cloud column is set, and the point cloud space of the to-be-encoded point cloud sample is divided into a point cloud column grid. For example, the horizontal and vertical sizes of the point cloud column grid can be the range of the point cloud space in the X-axis direction / the size of the point cloud column, and the range of the point cloud space in the Y-axis direction / the size of the point cloud column. The point cloud column grid contains a plurality of point cloud columns, and each point in the to-be-encoded point cloud sample is also contained in each point cloud column, wherein a point cloud column without points can be regarded as a null point cloud column. Through the division of the point cloud column, the points at different positions can be conveniently separated and processed, which helps to better process and encode the points at different positions in the subsequent process, reduces the interference of high and low points on other position points, and then the points in each point cloud column are represented by features. The feature representation is mainly used to represent the position information of the points in the point cloud column. Then, the target tensor representation of the entire to-be-encoded point cloud sample is determined according to the feature representation, and the target tensor representation can represent the attribute information of the to-be-encoded point cloud sample. Then, the target feature map of the to-be-encoded point cloud sample in the target dimension is determined according to the target tensor representation, and the target dimension can be adjusted according to the application scenario. Finally, the to-be-encoded point cloud sample is encoded according to the target feature map to determine the pseudo-image feature of the to-be-encoded point cloud sample. Through the above system, the to-be-encoded point cloud sample can be divided into regions, the interference of high and low points on other point encoding is reduced, and the problem of low point cloud encoding accuracy in the prior art is solved.

[0075] As shown in Figure 3 Corresponding to the above-mentioned point cloud encoding method, the present disclosure also provides an electronic device. Since the embodiments of the device are similar to the embodiments of the above-mentioned method, they are described more simply, and the relevant parts can be referred to the description of the above-mentioned method embodiment part. The device described below is only illustrative. The device can include a processor 1, a memory 2, a communication bus (i.e. the above-mentioned device bus), and a lookup engine. The processor 1 and the memory 2 complete mutual communication through the communication bus, and communicate with the outside through the communication interface. The processor 1 can call the logical instructions in the memory 2 to execute the point cloud encoding method.

[0076] In addition, the logic instructions in the memory 2 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present disclosure essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods of various embodiments of the present disclosure. The aforementioned storage medium includes: a storage chip, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0077] In another aspect, the embodiments of the present disclosure also provide a processor readable storage medium, and the processor readable storage medium stores a computer program 3. When the computer program 3 is executed by the processor 1, the point cloud encoding method provided by the above-mentioned embodiments is implemented.

[0078] The processor readable storage medium can be any available medium or data storage device that the processor 1 can access, including but not limited to a magnetic storage (such as a floppy disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical storage (such as a CD, a DVD, a BD, a HVD, etc.), and a semiconductor memory (such as a ROM, an EPROM, an EEPROM, a non-volatile memory (NAND FLASH), a solid state disk (SSD)), etc.

[0079] Those skilled in the art should understand that although the present disclosure is described in detail with reference to the foregoing embodiments, the technical solutions described in the foregoing embodiments can be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A point cloud encoding method, characterized by, The method comprises: obtaining a to-be-encoded point cloud sample, and dividing the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns; The features of the points in each of the point cloud columns are represented, and a target tensor representation of the to-be-encoded point cloud sample is determined according to the feature representations; the feature representation of the points in the point cloud column includes the coordinates, the reflection intensity, the normal vector, the geometric center of the point cloud column, the relative position of the point to the geometric center of the point cloud column, and the difference between the point and the average distance of the point to the remaining points in the point cloud column; the target tensor representation of the to-be-encoded point cloud sample is determined according to the feature representations, including: generating a first tensor representation of the to-be-encoded point cloud sample according to the feature representation of each point in the point cloud column; performing dimension adjustment on the first tensor representation of the to-be-encoded point cloud sample to determine a second tensor representation of the to-be-encoded point cloud sample in a target dimension; and determining the target tensor representation of the to-be-encoded point cloud sample according to a predefined weighting function on the basis of the second tensor representation of the to-be-encoded point cloud sample, and assigning a weight to each point in the point cloud column; for a to-be-encoded point cloud sample containing a non-empty point cloud number P and a point cloud number standard N, the feature representation of each point in the point cloud column is denoted as D, the first tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, D), convolution is used on each point in the point cloud column, so that the dimension of each point changes from D to C, and the second tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, C); the weighting function is defined in the laser radar coordinate system, and the expression of the weighting function is as follows: ; wherein, is the coordinate of the point in the point cloud column on the Z axis, and is 1, a is the lowest point coordinate of the main attention area in the actual application of autonomous driving, b is the highest point coordinate of the main attention area in the actual application of autonomous driving, is a, is b, and the weight of the point in the point cloud column is , the tensor of the to-be-encoded point cloud sample can be represented as , and the target tensor is ; determining a target feature map of the to-be-encoded point cloud sample in a target dimension according to a target tensor representation of the to-be-encoded point cloud sample, comprising: performing maximum pooling on points in each of the point cloud columns, and adjusting the target tensor representation of the to-be-encoded point cloud sample to obtain a first feature map; performing average pooling on the points in each of the point cloud columns, and adjusting the target tensor representation of the to-be-encoded point cloud sample to obtain a second feature map; splicing the first feature map and the second feature map to obtain a third feature map; and performing dimension adjustment on the third feature map to obtain the target feature map of the to-be-encoded point cloud sample in the target dimension; determining a pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map, comprising: unfolding the target feature map in the target dimension according to the positions of the point cloud columns to obtain the pseudo-image feature of the to-be-encoded point cloud sample.

2. The point cloud encoding method according to claim 1, wherein before performing feature representation on the points in each of the point cloud columns, the method comprises: predefining a point cloud quantity standard, and adjusting the number of points in each of the point cloud columns according to the point cloud quantity standard.

3. The point cloud encoding method according to claim 2, wherein adjusting the number of points in each of the point cloud columns according to the point cloud quantity standard comprises: for a point cloud column whose number of points exceeds the point cloud quantity standard, sampling the points in the point cloud column so that the number of points in the point cloud column is equal to the point cloud quantity standard; for a point cloud column whose number of points is less than the point cloud quantity standard, complementing the point cloud column with "0" so that the number of points in the point cloud column is equal to the point cloud quantity standard. The method comprises: a division module configured to obtain a to-be-encoded point cloud sample, and divide the to-be-encoded point cloud sample into a point cloud column grid; the point cloud column grid comprises a plurality of point cloud columns; 4. A point cloud encoding system, characterized by, a processing module configured to determine a target feature map of the to-be-encoded point cloud sample in a target dimension according to a target tensor representation of the to-be-encoded point cloud sample, comprising: the processing module is configured to perform maximum pooling on points in each of the point cloud columns, and adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a first feature map; perform average pooling on the points in each of the point cloud columns, and adjust the target tensor representation of the to-be-encoded point cloud sample to obtain a second feature map; splice the first feature map and the second feature map to obtain a third feature map; and perform dimension adjustment on the third feature map to obtain the target feature map of the to-be-encoded point cloud sample in the target dimension; a determination module configured to determine a pseudo-image feature of the to-be-encoded point cloud sample according to the target feature map, comprising: the determination module is configured to unfold the target feature map in the target dimension according to the positions of the point cloud columns to obtain the pseudo-image feature of the to-be-encoded point cloud sample. The identification module is used for feature representation of points in each point cloud column, and target tensor representation of the to-be-encoded point cloud sample is determined according to the feature representation. The feature representation of points in the point cloud column includes the coordinates, reflection intensity, normal vector, geometric center of the point cloud column, relative position of the point to the geometric center of the point cloud column, and difference between the point and the average distance of the point to the remaining points in the point cloud column. The target tensor representation of the to-be-encoded point cloud sample is determined according to the feature representation, including: generating the first tensor representation of the to-be-encoded point cloud sample according to the feature representation of each point in the point cloud column; performing dimension adjustment on the first tensor representation of the to-be-encoded point cloud sample to determine the second tensor representation of the to-be-encoded point cloud sample in the target dimension; and determining the target tensor representation of the to-be-encoded point cloud sample according to a predefined weighting function for assigning weights to points in each point cloud column based on the second tensor representation of the to-be-encoded point cloud sample. For a to-be-encoded point cloud sample containing P non-empty point clouds and having a point cloud number standard N, the feature representation of each point in the point cloud column is denoted as D, the first tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, D), convolution is used for points in each point cloud column, so that the dimension of each point changes from D to C, and the second tensor representation of the to-be-encoded point cloud sample can be represented as (P, N, C). The weighting function is defined in a laser radar coordinate system, and the expression of the weighting function is as follows: ; wherein, is the coordinate of the point in the point cloud column on the Z axis, and is 1, a is the lowest point coordinate of the main attention area in the actual application of autonomous driving, b is the highest point coordinate of the main attention area in the actual application of autonomous driving, is a, is b, and the weight of the point in the point cloud column is , the tensor of the to-be-encoded point cloud sample can be represented as , that is, the target tensor ; The method comprises: a processor and a memory; 5. An electronic device, comprising: ​ ​ The processor invokes a computer program stored in the memory to execute the point cloud encoding method in any one of claims 1 to 3.

6. A computer readable storage medium, comprising: The computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor can execute the point cloud encoding method in any one of claims 1 to 3.

Images