CN114779209A - Laser radar point cloud voxelization method and device - Google Patents
Laser radar point cloud voxelization method and device Download PDFInfo
- Publication number
- CN114779209A CN114779209A CN202210311588.7A CN202210311588A CN114779209A CN 114779209 A CN114779209 A CN 114779209A CN 202210311588 A CN202210311588 A CN 202210311588A CN 114779209 A CN114779209 A CN 114779209A
- Authority
- CN
- China
- Prior art keywords
- voxel
- points
- point cloud
- reserved
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention discloses a laser radar point cloud voxelization method, which comprises the following steps: step 1: acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels; and 2, step: setting a down-sampling threshold and a dynamic increment; and step 3: for any voxel, if the number of points in the voxel is judged to exceed the down-sampling threshold, determining the last reserved point number of the voxel according to the dynamic increment, and performing down-sampling processing on the voxel according to the reserved point number; otherwise, all points in the current voxel are reserved, and the down-sampling processing is not carried out on the points; and 4, step 4: and (4) repeating the step (3) until all voxels are processed to obtain the voxelized point cloud. The laser radar point cloud voxelization method provided by the invention can be used for keeping original information as much as possible under the condition that the original point cloud is extremely unevenly distributed, so that a better voxelization effect is obtained.
Description
Technical Field
The invention belongs to the technical field of laser point cloud data processing, and particularly relates to a laser radar point cloud voxelization method and device.
Background
With the development of artificial intelligence, the convolutional neural network is increasingly applied to the field of two-dimensional image target detection, and the precision of a target detection task is greatly improved by virtue of the strong feature extraction capability of the convolutional neural network. The laser radar point cloud data is disordered data obtained by scanning the laser radar, so that the characteristic extraction is difficult to directly carry out by using a space convolution method, and due to disordered storage of the data in a memory, more random memory accesses are generated during use, and the data operation efficiency is further reduced. In order to solve the problem, a processing mode of performing voxelization on the point cloud is generally adopted, and the voxelized point cloud data is stored in order based on the position relation of the voxelized point cloud data in the three-dimensional middle, so that the reduction of random memory access is facilitated, the data operation efficiency is increased, and the point cloud data with larger magnitude can be processed.
At present, the existing laser radar point cloud voxelization method can be mainly divided into two types, namely voxelization in three-dimensional space and voxelization in two-dimensional space. After voxel segmentation is performed on a point cloud space, most of the three-dimensional space voxelization methods simply and roughly perform downsampling on points in voxels in a random selection mode. However, the voxelization process inevitably brings about a certain loss of information, and the points selected by the random selection strategy are not necessarily capable of well representing the original voxels. The voxelization of the two-dimensional space adopts a projection method to project the three-dimensional point cloud to a two-dimensional plane to obtain a depth map, and adopts a two-dimensional image convolution mode to process. Although simple and fast, such methods lose more three-dimensional information in the original point cloud.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a laser radar point cloud voxelization method and device. The technical problem to be solved by the invention is realized by the following technical scheme:
in a first aspect, the invention provides a laser radar point cloud voxelization method, which comprises the following steps:
step 1: acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels;
and 2, step: setting a down-sampling threshold and a dynamic increment;
and 3, step 3: for any voxel, if the number of points in the voxel is judged to exceed the down-sampling threshold, determining the last reserved point number of the voxel according to the dynamic increment, and performing down-sampling processing on the voxel according to the reserved point number;
otherwise, all points in the current voxel are reserved, and down-sampling processing is not carried out on the points;
and 4, step 4: and (5) repeating the step (3) until all voxels are processed, and obtaining the point cloud after the voxels are processed.
In one embodiment of the present invention, in step 3, the formula for determining the last remaining point number of the voxel according to the dynamic increment is as follows:
wherein c represents the number of points which need to be reserved after the current voxel is subjected to down-sampling, a represents a down-sampling threshold, b represents a dynamic increment, n represents the total number of points in the current voxel,
meaning rounding down.
In an embodiment of the present invention, in step 3, the down-sampling the voxel according to the number of the reserved points includes:
for each point i in the current voxel, calculating the sum L of Euclidean distances between the point and other pointsi；
With LiThe magnitude of the value is according toAnd selecting a corresponding number of points from the reserved points to finish the downsampling processing of the current voxel.
In one embodiment of the invention, L isiSelecting a corresponding number of points according to the reserved points according to the value, wherein the method comprises the following steps:
mixing L withiArranging according to the size sequence;
if the reserved number c is an even number, respectively reserving c/2 larger LiCorresponding points and c/2 smaller LiA corresponding point;
if the number of reserved points c is odd, 1+ (c-1)/2 larger L are reserved respectivelyiCorresponding point sum (c-1)/2 smaller LiThe corresponding point.
In a second aspect, the present invention provides a laser radar point cloud voxelization apparatus, including:
the system comprises a voxelization module, a data processing module and a data processing module, wherein the voxelization module is used for acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels;
the parameter setting module is used for setting a down-sampling threshold value and a dynamic increment;
the data processing module is used for determining the last reserved point number of the voxel according to the dynamic increment and carrying out downsampling processing on the voxel according to the reserved point number if the number of points in the voxel is judged to exceed the downsampling threshold value for any voxel;
otherwise, all points in the current voxel are reserved and are not subjected to downsampling processing.
In a third aspect, the present invention provides an electronic device, which includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;
a memory for storing a computer program;
the processor is configured to implement the method steps described in the above embodiments when executing the program stored in the memory.
In a fourth aspect, the present invention provides a computer-readable storage medium, in which a computer program is stored, and the computer program, when executed by a processor, implements the method steps of the above embodiments.
The invention has the beneficial effects that:
1. according to the laser radar point cloud voxelization method provided by the invention, dynamic increment is introduced, more original points are reserved for voxels with a larger number of points, and all original points are reserved for voxels with a smaller number of points, so that original information is reserved as much as possible under the condition that the original point cloud is extremely non-uniform in distribution, and a better voxelization effect is obtained;
2. the invention selects the reserve points based on Euclidean distance when the voxel is down sampled, compared with the traditional random selection method, the points selected by the method of the invention have better representation effect on the original data.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a schematic flow chart of a laser radar point cloud voxelization method provided by an embodiment of the present invention;
fig. 2 is a schematic flow chart of another laser radar point cloud voxelization method provided by the embodiment of the invention;
fig. 3 is a schematic structural diagram of a laser radar point cloud voxelization apparatus according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
Referring to fig. 1, fig. 1 is a schematic flow chart of a laser radar point cloud voxelization method according to an embodiment of the present invention, which includes:
step 1: the method includes acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels.
First, raw point cloud data may be obtained by lidar scanning, which may be represented as (x, y, z) using three-dimensional cartesian coordinates.
Then, voxel division is performed on the obtained original point cloud data.
Specifically, voxel division may be performed according to the following method.
1) Firstly, determining a required distance resolution f according to the requirements of downstream tasks;
2) traversing the point cloud, and recording the maximum value and the minimum value X of the data in the X, y and z directionsmax，Xmin，Ymax，Ymin，Zmax，ZminSix parameters in total;
3) let the length, width and height of three-dimensional space be L, W, H respectively, and calculate L ═ Xmax-Xmin，W＝Ymax-Ymin，H＝Zmax-Zmin；
4) Definition of
Is the smallest integer of n, calculate
The three-dimensional space is divided into l × w × h voxels of size f × f and all points in the original point cloud are assigned to the respective voxels.
And 2, step: a down-sampling threshold and a dynamic delta are set.
In this embodiment, the down-sampling threshold a and the dynamic increment b can be determined according to the processing power and real-time requirement of the computer. The downsampling threshold value a represents a condition for downsampling the voxels, the voxels with the point number less than a are not downsampled, the points with the point number more than a are reserved for a point, and the last reserved point number is determined according to the dynamic increment b. The dynamic increment b refers to the number of points reserved after down-sampling according to the parameter when the number of points in the voxel is higher than the down-sampling threshold value a.
Obviously, the less data in the voxel, the less calculation, and the higher the real-time performance. However, too little data does not retain the original point cloud data information well. Therefore, the size of the down-sampling threshold a and the dynamic increment b can be set according to actual requirements. For example, if the calculation amount is small and the real-time performance is high, the values of a and b are set to be small.
And 3, step 3: for any voxel, if the number of points in the voxel is judged to exceed the down-sampling threshold, determining the last reserved point number of the voxel according to the dynamic increment, and performing down-sampling processing on the voxel according to the reserved point number;
otherwise, all points in the current voxel are retained and are not down-sampled.
Specifically, referring to fig. 2, fig. 2 is a schematic flow chart of another laser radar point cloud voxelization method provided in the embodiment of the present invention.
For each voxel, firstly, judging whether the total number n of points in the voxel exceeds a sampling threshold a, if not, namely n is less than or equal to a, not performing down-sampling on the voxel, namely, retaining all original point cloud data in the current voxel for subsequent processing.
If the total number n of the points in the current voxel exceeds the down-sampling threshold value, namely n is larger than a, the down-sampling processing is needed. The method comprises the following specific steps:
firstly, determining the number of points to be reserved after down-sampling according to a dynamic increment b, wherein the specific principle is as follows: and while retaining a points, retaining one more point for each additional b points of the down-sampling result, and using the formula:
wherein c represents the number of points which need to be reserved after the current voxel is subjected to down-sampling, a represents a down-sampling threshold, b represents a dynamic increment, n represents the total number of points in the current voxel,
meaning rounding down.
Then, the voxel is down-sampled according to the reserved point number c.
This embodiment mainly carries out down-sampling processing based on european style distance, and it includes:
1) for each point i in the current voxel, calculating the sum L of Euclidean distances between the point and other pointsi；
2) With LiAnd selecting a corresponding number of points according to the number of the reserved points to finish the down-sampling processing of the current voxel on the basis of the value.
In this embodiment, all L's may be ordered from small to large or from large to smalliAnd (4) carrying out arrangement. Preferably, the present embodiment uses a large to small array.
Then, based on the number of reserved points c, from all LiIn (3), a corresponding number of points are selected.
For example, when the reserved number c is an even number, c/2 larger L are reserved respectivelyiCorresponding point and c/2 smaller LiA corresponding point; i.e. from LiC/2 points are taken at the head and the tail of the team respectively.
If the number of reserved points c is an odd number, 1+ (c-1)/2 larger L are reserved respectivelyiCorresponding point sum (c-1)/2 smaller LiCorresponding points, i.e. from LiThe head of the team takes 1+ (c-1)/2 points, and the tail of the team takes 1+ (c-1)/2 points.
In addition, two corresponding points may also be selected according to other selection strategies, which is not specifically limited in this embodiment.
And 4, step 4: and (4) repeating the step (3) until all voxels are processed to obtain the voxelized point cloud.
Specifically, for each voxel obtained by voxel division, the processing is performed by the method in step 3, so as to obtain data after voxel division. Therefore, for sparse point cloud voxels, original point cloud annual data can be preserved, for dense point clouds, some choices are made based on Euclidean distances, the data volume needing to be processed is reduced on the basis that the original point cloud data is reserved to the greatest extent, and the algorithm processing speed is improved.
According to the laser radar point cloud voxelization method provided by the embodiment of the invention, by introducing the dynamic increment, more original points are reserved for voxels with a larger number of points, and all original points are reserved for voxels with a smaller number of points, so that original information is reserved as much as possible under the condition that the original point cloud is extremely non-uniform in distribution, and a better voxelization effect is obtained. In addition, the invention selects the reserved points based on Euclidean distance when the voxels are down sampled, and compared with the traditional random selection method, the points selected by the method have better representation effect on the original data.
Example two
On the basis of the first embodiment, the embodiment provides a laser radar point cloud voxelization device. Referring to fig. 3, fig. 3 is a schematic structural diagram of a laser radar point cloud voxelization apparatus according to an embodiment of the present invention, which includes:
the voxelization module 1 is used for acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels;
the parameter setting module 2 is used for setting a down-sampling threshold value and a dynamic increment;
the data processing module 3 is configured to, for any voxel, determine a last remaining point number of the voxel according to the dynamic increment if it is determined that the number of points in the voxel exceeds the downsampling threshold, and perform downsampling processing on the voxel according to the remaining point number;
otherwise, all points in the current voxel are retained and are not down-sampled.
The laser radar point cloud voxelization device provided by the embodiment can realize the laser radar point cloud voxelization method provided by the first embodiment, and the detailed process refers to the first embodiment.
Therefore, the device can keep the original information as much as possible under the condition that the original point cloud distribution is extremely uneven, thereby obtaining better voxelization effect.
EXAMPLE III
An embodiment of the present invention further provides an electronic device, please refer to fig. 4, where fig. 4 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention, and the electronic device includes: the system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;
a memory for storing a computer program;
the processor is configured to implement the method steps provided in the first embodiment when executing the program stored in the memory, and the detailed implementation process refers to the first embodiment.
The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.
The communication interface is used for communication between the electronic equipment and other equipment.
The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Alternatively, the memory may be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.
The electronic device may be: desktop computers, laptop computers, intelligent mobile terminals, servers, and the like. Without limitation, any electronic device capable of implementing the present invention is within the scope of the present invention.
Example four
Fig. 5 shows a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention, where fig. 5 is a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention. The computer-readable storage medium provided in this embodiment stores a computer program thereon, and the computer program, when executed by a processor, implements the method steps provided in the first embodiment, and the implementation principle and the technical effect are similar, which are not described herein again.
It should be noted that the apparatus, the electronic device, and the storage medium according to the embodiments of the present invention are respectively an apparatus, an electronic device, and a storage medium to which the laser radar point cloud voxelization method is applied, and all embodiments of the laser radar point cloud voxelization method are applicable to the apparatus, the electronic device, and the storage medium, and can achieve the same or similar beneficial effects.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, numerous simple deductions or substitutions may be made without departing from the spirit of the invention, which shall be deemed to belong to the scope of the invention.
Claims (7)
1. A laser radar point cloud voxelization method is characterized by comprising the following steps:
step 1: acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels;
step 2: setting a down-sampling threshold and a dynamic increment;
and step 3: for any voxel, if the number of points in the voxel is judged to exceed the down-sampling threshold, determining the last reserved point number of the voxel according to the dynamic increment, and carrying out down-sampling processing on the voxel according to the reserved point number;
otherwise, all points in the current voxel are reserved, and the down-sampling processing is not carried out on the points;
and 4, step 4: and (4) repeating the step (3) until all voxels are processed to obtain the voxelized point cloud.
2. The lidar point cloud voxelization method of claim 1, wherein in step 3, a formula for determining a final reserve point number of the voxel according to the dynamic increment is as follows:
wherein c represents the number of points which need to be reserved after the current voxel is subjected to down-sampling, a represents a down-sampling threshold value, b represents a dynamic increment, n represents the total number of points in the current voxel,
indicating a rounding down.
3. The lidar point cloud voxelization method according to claim 1, wherein in step 3, the down-sampling processing of the voxel according to the reserved point number comprises:
for each point i in the current voxel, calculating the sum L of Euclidean distances between the point and other pointsi；
With LiAnd selecting a corresponding number of points according to the number of the reserved points to finish the down-sampling processing of the current voxel on the basis of the value.
4. The lidar point cloud voxelization method of claim 3, wherein L isiSelecting a corresponding number of points according to the reserved points according to the value, wherein the method comprises the following steps:
will LiArranging according to the size sequence;
if the reserved number c is an even number, respectively reserving c/2 larger LiCorresponding point and c/2 smaller LiA corresponding point;
if the number of reserved points c is odd, 1+ (c-1)/2 larger L are reserved respectivelyiCorresponding point sum (c-1)/2 smaller LiThe corresponding point.
5. A laser radar point cloud voxelization device is characterized by comprising:
the system comprises a voxelization module (1) for acquiring original point cloud data and dividing the original point cloud data into a plurality of voxels;
the parameter setting module (2) is used for setting a down-sampling threshold value and a dynamic increment;
the data processing module (3) is used for determining the last reserved point number of the voxel according to the dynamic increment and performing down-sampling processing on the voxel according to the reserved point number if the number of points in the voxel is judged to exceed the down-sampling threshold value for any voxel;
otherwise, all points in the current voxel are reserved and are not subjected to downsampling processing.
6. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor and the communication interface are used for realizing mutual communication by the memory through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any one of claims 1 to 4 when executing a program stored in the memory.
7. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, which computer program, when being executed by a processor, carries out the method steps of any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210311588.7A CN114779209A (en) | 2022-03-28 | 2022-03-28 | Laser radar point cloud voxelization method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210311588.7A CN114779209A (en) | 2022-03-28 | 2022-03-28 | Laser radar point cloud voxelization method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114779209A true CN114779209A (en) | 2022-07-22 |
Family
ID=82424690
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210311588.7A Pending CN114779209A (en) | 2022-03-28 | 2022-03-28 | Laser radar point cloud voxelization method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114779209A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116030104A (en) * | 2023-03-29 | 2023-04-28 | 山东港口渤海湾港集团有限公司 | Point cloud nearest neighbor data structure construction method and system thereof |
-
2022
- 2022-03-28 CN CN202210311588.7A patent/CN114779209A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116030104A (en) * | 2023-03-29 | 2023-04-28 | 山东港口渤海湾港集团有限公司 | Point cloud nearest neighbor data structure construction method and system thereof |
| CN116030104B (en) * | 2023-03-29 | 2023-08-11 | 山东港口渤海湾港集团有限公司 | Point cloud nearest neighbor data structure construction method and system thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10929746B2 (en) | Low-power hardware acceleration method and system for convolution neural network computation | |
| CN109545072B (en) | Map construction pose calculation method, map construction pose calculation device, map construction pose storage medium and map construction pose calculation system | |
| CN112562069B (en) | Method, device, equipment and storage medium for constructing three-dimensional model | |
| CN113077476B (en) | Height measurement method, terminal device and computer storage medium | |
| CN112270332A (en) | Three-dimensional target detection method and system based on sub-stream sparse convolution | |
| CN114779209A (en) | Laser radar point cloud voxelization method and device | |
| CN112613450A (en) | 3D target detection method for enhancing performance on difficult sample | |
| CN112465050A (en) | Image template selection method, device, equipment and storage medium | |
| CN112734827A (en) | Target detection method and device, electronic equipment and storage medium | |
| US20230196719A1 (en) | Method for cargo counting, computer equipment, and storage medium | |
| CN116543417A (en) | Human body posture estimation method, device, equipment and storage medium | |
| CN116227599A (en) | Inference model optimization method and device, electronic equipment and storage medium | |
| WO2022017129A1 (en) | Target object detection method and apparatus, electronic device, and storage medium | |
| CN113888635B (en) | Visual positioning method and related device | |
| CN112816959B (en) | Clustering method, device, equipment and storage medium for vehicles | |
| CN111009004B (en) | Hardware optimization method for accelerating image matching | |
| CN114384541A (en) | Point cloud target detection method, terminal device and computer readable storage medium | |
| CN113361601A (en) | Method for fusing perspective and aerial view characteristics based on unmanned vehicle laser radar data | |
| CN112905239B (en) | Point cloud preprocessing acceleration method based on FPGA, accelerator and electronic equipment | |
| CN113033256A (en) | Training method and device for fingertip detection model | |
| CN117217274B (en) | Vector processor, neural network accelerator, chip and electronic equipment | |
| CN115984583B (en) | Data processing method, apparatus, computer device, storage medium, and program product | |
| CN114913213B (en) | Method and device for learning aerial view characteristics | |
| CN115223110B (en) | Target detection method, device, computer equipment and storage medium | |
| CN114781634B (en) | Automatic mapping method and device of neural network array based on memristor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |