CN110097077A - Point cloud data classification method, device, computer equipment and storage medium - Google Patents
Point cloud data classification method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN110097077A CN110097077A CN201910231062.6A CN201910231062A CN110097077A CN 110097077 A CN110097077 A CN 110097077A CN 201910231062 A CN201910231062 A CN 201910231062A CN 110097077 A CN110097077 A CN 110097077A
- Authority
- CN
- China
- Prior art keywords
- sorted
- point cloud
- cloud data
- unit
- eigenmatrix
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application involves a kind of point cloud data classification method, device, computer equipment and storage mediums.The described method includes: obtaining point cloud data to be sorted；Structuring pretreatment is carried out to the point cloud data to be sorted according to pre-set dimension, obtains eigenmatrix to be sorted；Disaggregated model trained in advance is called, is classified by the disaggregated model to the eigenmatrix to be sorted, obtains eigenmatrix classification results；Classified according to the eigenmatrix classification results to the point cloud data to be sorted.The classification accuracy of point cloud data can be effectively improved under automatic driving mode using this method.
Description
Technical field
This application involves artificial intelligence fields, more particularly to a kind of point cloud data classification method, device, computer equipment
And storage medium.
Background technique
The development of artificial intelligence technology promotes the development of automatic Pilot technology.During automatic Pilot, need in real time
Ambient condition information is monitored, this just needs to carry out real-time grading to the collected point cloud data of onboard sensor, before identifying dynamic
The object informations such as scape barrier, static foreground barrier, way line mark, background.In addition, raw according to sorted point cloud data
At high-precision map, it can preferably plan rational routes, avoiding barrier and observe traffic rules and regulations.Due to point cloud data amount
Huge, under automatic driving mode, computing resource is limited, and requirement of real-time is high, can not carry out Accurate classification to point cloud data.
Summary of the invention
Based on this, it is necessary to which in view of the above technical problems, providing one kind can be under automatic driving mode, and computing resource has
Limit and requirement of real-time it is high when, improve point cloud data classification method, device, the computer equipment of the classification accuracy of point cloud data
And storage medium.
A kind of point cloud data classification method, which comprises
Obtain point cloud data to be sorted；
Structuring pretreatment is carried out to the point cloud data to be sorted according to pre-set dimension, obtains eigenmatrix to be sorted；
Disaggregated model trained in advance is called, is classified by the disaggregated model to the eigenmatrix to be sorted,
Obtain eigenmatrix classification results；
Classified according to the eigenmatrix classification results to the point cloud data to be sorted.
In one of the embodiments, before the acquisition point cloud data to be sorted, the method also includes:
Obtain sample data；
Structuring pretreatment is carried out to the sample data according to pre-set dimension, obtains sample characteristics matrix；
Deep learning model is called, the deep learning model is trained according to the sample characteristics matrix；
When reaching default training condition, using the deep learning model as disaggregated model.
It is described in one of the embodiments, that the point cloud data progress structuring to be sorted is located in advance according to pre-set dimension
Reason includes:
The corresponding data area of the point cloud data to be sorted is determined according to the point cloud data to be sorted；
The data area is divided in X direction according to the length in pre-set dimension, obtains first structure unit；
The data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；
Target structural unit is generated according to the first structure unit and the second structuring unit；
The characteristic information of corresponding point cloud data to be sorted is extracted in each target structural unit；
The characteristic information is arranged in each target structural unit by row, eigenmatrix to be sorted is generated.
It is described in one of the embodiments, to include: to the point cloud data progress structuring pretreatment to be sorted
The corresponding data area of the point cloud data to be sorted is determined according to the point cloud data to be sorted；
The data area is divided in X direction according to the length in pre-set dimension, obtains first structure unit；
The data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；
The data area is subjected to impartial division along Z-direction according to the height in pre-set dimension, obtains third structuring
Unit；
Target structural list is generated according to the first structure unit, the second structuring unit and third structural unit
Member.
The eigenmatrix to be sorted includes matrix unit in one of the embodiments,；It is described to pass through the classification mould
Type carries out classification to the eigenmatrix to be sorted
Each matrix unit is calculated according to the characteristic information in the eigenmatrix to be sorted by the disaggregated model
Class probability；
Classified according to the class probability of the matrix unit to the matrix unit.
The eigenmatrix classification results include matrix unit type in one of the embodiments,；According to the feature
Matrix Classification result carries out classification to the point cloud data to be sorted
The index of the point cloud data to be sorted is searched in the eigenmatrix classification results；
The point to be sorted is searched in the eigenmatrix classification results according to the index of the point cloud data to be sorted
The corresponding matrix unit type of cloud data；
Classified according to the matrix unit type found to the point cloud data to be sorted.
A kind of point cloud data sorter, described device include:
Communication module, for obtaining point cloud data to be sorted；
Preprocessing module obtains feature square to be sorted for carrying out structuring pretreatment to the point cloud data to be sorted
Battle array；
Matrix Classification module, for calling disaggregated model trained in advance, by the disaggregated model to described to be sorted
Eigenmatrix is classified, and eigenmatrix classification results are obtained；
Data categorization module, for being divided according to the eigenmatrix classification results the point cloud data to be sorted
Class.
Described device in one of the embodiments, further include:
Model generation module, for obtaining sample data；It is pre- that structuring is carried out to the sample data according to pre-set dimension
Processing, obtains sample characteristics matrix；Deep learning model is called, according to the sample characteristics matrix to the deep learning model
It is trained；When reaching default training condition, using the deep learning model as disaggregated model.
The preprocessing module is also used to according to the point cloud data determination to be sorted in one of the embodiments,
The corresponding data area of point cloud data to be sorted；The data area is drawn in X direction according to the length in pre-set dimension
Point, obtain first structure unit；The data area is divided along Y-direction according to the width in pre-set dimension, is obtained
Second structuring unit；Target structural unit is generated according to the first structure unit and the second structuring unit；?
The characteristic information of corresponding point cloud data to be sorted is extracted in each target structural unit；By the characteristic information in each mesh
It is arranged in mark structuring unit by row, generates eigenmatrix to be sorted.
The preprocessing module is also used to according to the point cloud data determination to be sorted in one of the embodiments,
The corresponding data area of point cloud data to be sorted；The data area is drawn in X direction according to the length in pre-set dimension
Point, obtain first structure unit；The data area is divided along Y-direction according to the width in pre-set dimension, is obtained
Second structuring unit；The data area is subjected to impartial division along Z-direction according to the height in pre-set dimension, obtains third
Structuring unit；Object construction is generated according to the first structure unit, the second structuring unit and third structural unit
Change unit.
The Matrix Classification module is also used to through the disaggregated model according to described wait divide in one of the embodiments,
Characteristic information in category feature matrix calculates the class probability of each matrix unit；According to the class probability pair of the matrix unit
The matrix unit is classified.
The data categorization module is also used to search in the eigenmatrix classification results in one of the embodiments,
The index of the point cloud data to be sorted；According to the index of the point cloud data to be sorted in the eigenmatrix classification results
Search the corresponding matrix unit type of the point cloud data to be sorted；According to the matrix unit type found to described to be sorted
Point cloud data is classified.
A kind of computer equipment, including memory and processor, the memory are stored with and can run on a processor
Computer program, the processor realize the step in above-mentioned each embodiment of the method when executing the computer program.
A kind of computer readable storage medium, is stored thereon with computer program, and the computer program is held by processor
The step in above-mentioned each embodiment of the method is realized when row.
Above-mentioned point cloud data classification method, device, computer equipment and storage medium, by carrying out structure to point cloud data
Change pretreatment, so as to the disaggregated model that eigenmatrix to be sorted input is trained in advance.Pass through disaggregated model pair trained in advance
Eigenmatrix to be sorted is classified, and then is classified to point cloud data to be sorted.It can be calculated under automatic driving mode
When resource is limited and requirement of real-time is higher, the classification accuracy of point cloud data is effectively improved.
Detailed description of the invention
Fig. 1 is the schematic diagram for carrying out point cloud data classification in one embodiment during automatic Pilot；
Fig. 2 is the flow diagram of one embodiment midpoint cloud data classification method；
Fig. 3 is the flow diagram of another embodiment midpoint cloud data classification method；
Fig. 4 is the stream for carrying out structuring pre-treatment step in one embodiment to point cloud data to be sorted according to pre-set dimension
Journey schematic diagram；
Fig. 5 is the structural block diagram of one embodiment midpoint cloud device for classifying data；
Fig. 6 is the internal structure chart of computer equipment in one embodiment.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not
For limiting the application.
Point cloud data classification method provided by the present application can be applied to during automatic Pilot as shown in Figure 1 to point
In the schematic diagram that cloud data are classified.Wherein, collected point cloud data to be sorted is sent to vehicle-mounted by onboard sensor 102
Computer equipment 104.Vehicle computing machine equipment can be referred to as computer equipment.Computer equipment 104 is according to pre-set dimension
Structuring pretreatment is carried out to point cloud data to be sorted, obtains eigenmatrix to be sorted.Computer equipment 104 calls preparatory training
Disaggregated model, characteristic of division matrix is treated by disaggregated model and is classified, eigenmatrix classification results are obtained.Computer is set
Standby 104 classify to point cloud data to be sorted according to eigenmatrix classification results.Money can be calculated under automatic driving mode
When source is limited and requirement of real-time is high, the classification accuracy of point cloud data is improved.
In one embodiment, as shown in Fig. 2, providing a kind of point cloud data classification method, it is applied to Fig. 1 in this way
In computer equipment for be illustrated, comprising the following steps:
Step 202, point cloud data to be sorted is obtained.
Step 204, structuring pretreatment is carried out to point cloud data to be sorted according to pre-set dimension, obtains feature square to be sorted
Battle array.
For vehicle during automatic Pilot, collected point cloud data to be sorted is sent to computer by onboard sensor
Equipment.Onboard sensor can be laser radar.Point cloud data to be sorted is that onboard sensor is collected in the visible range
Point cloud data.The visual range of different onboard sensors can be different.
Computer equipment can carry out structure to point cloud data to be sorted according to pre-set dimension after obtaining point cloud classifications data
Change pretreatment, obtains eigenmatrix to be sorted.Structuring processing mode can there are many.Wherein it is possible to be rasterizing processing,
It is also possible to voxelization processing.It, can be with when the computing resource of computer equipment is less than preset threshold under automatic driving mode
Using rasterizing processing mode.When the computing resource of computer equipment is more than or equal to preset threshold, voxel can be used
Change processing mode.For example, rasterizing processing mode can be used under the scene of automatic Pilot real-time monitoring.For another example, automatic
It drives under the scene for generating high-precision map, voxelization processing mode can be used.
Specifically, computer equipment determines the number where point cloud data to be sorted according to the point cloud data to be sorted got
According to region.Data area can be the minimum data space comprising all point cloud datas to be sorted.For example, visual range is 200m
The collected point cloud data to be sorted of laser radar, data area be 400m*400m*10m (long * wide * high).Computer is set
It is standby to be divided the data area where point cloud data to be sorted according to pre-set dimension, obtain target structural unit.It is default
The size of size indicates the size of target structural unit.When computer equipment divides data area, by point to be sorted
Cloud data are divided into object construction unit.Object construction unit remains with the index of point cloud data to be sorted.Object construction list
It may include point cloud data to be sorted in member, be also possible to empty structural unit.Computer is carrying out structure to be sorted cloud
During change is handled, empty structural unit can be weeded out.
When structuring pretreatment mode is that rasterizing is handled, pre-set dimension can be long * wide.Wherein, in pre-set dimension
Length and width can be different.For example, under automatic Pilot real-time monitoring scene, data area 400m*400m*10m
When (long * wide * high), long and wide numerical value can be any value between 0.1m to 0.5m in object construction unit.Work as structure
When change processing mode is that voxelization is handled, pre-set dimension can be long * wide * high.Wherein, the length and width and Gao Ke in pre-set dimension
To be identical.For example, in high-precision map, when data area is 400m*400m*10m (long * wide * high), object construction list
Length and width and high numerical value can be 0.1m*0.1m*0.1m in member.
Computer equipment extracts the characteristic information of corresponding point cloud data to be sorted in each target structural unit.It mentions
Take characteristic information mode can there are many.Wherein it is possible to characteristic information is automatically extracted by autocoder, it can also be artificial
Statistical nature information.Characteristic information include: the number at point cloud data midpoint, the maximum height of point cloud data, point cloud data most
Low clearance, the average height of point cloud data, height variance of point cloud data etc..Computer equipment is by characteristic information in each target
It is arranged in structuring unit by row, generator matrix unit.Computer equipment arranges matrix unit according to preset rules
Column-generation eigenmatrix to be sorted.Putting in order for characteristic information in matrix unit is not construed as limiting.
Step 206, disaggregated model trained in advance is called, characteristic of division matrix is treated by disaggregated model and is classified,
Obtain eigenmatrix classification results.
Step 208, classified according to eigenmatrix classification results to point cloud data to be sorted.
Specifically, computer equipment is before obtaining point cloud data to be sorted, and training has disaggregated model in advance.Classification
Model is to be trained according to sample data to deep learning model.Parameter in disaggregated model is fixed, parameter
The size of structural unit, i.e. pre-set dimension when including to sample data progress structuring pretreatment.Disaggregated model can be two dimension
Convolutional neural networks model is also possible to Three dimensional convolution neural network model.When structuring processing mode is that rasterizing is handled,
Computer equipment calls two-dimensional convolution neural network model.When structuring processing mode is voxelization, computer equipment is called
Three dimensional convolution neural network model.Computer equipment calculates each matrix unit in eigenmatrix to be sorted by disaggregated model
Class probability.Computer equipment treats characteristic of division matrix according to class probability and classifies, and obtains eigenmatrix classification results.
It include a variety of matrix unit types in eigenmatrix classification results.
When computer equipment carries out structuring pretreatment to point cloud data to be sorted, the rope of point cloud data to be sorted is remained with
Draw.Computer equipment by searching the index of point cloud data to be sorted in the matrix unit of eigenmatrix classification results, according to
The corresponding matrix unit type of the index of point cloud data to be sorted classifies to point cloud data to be sorted, obtains to be sorted cloud
The classification results of data, to obtain the semantic information of each point in point cloud data.
In the present embodiment, computer equipment is by carrying out structuring pretreatment to point cloud data, so as to by spy to be sorted
The disaggregated model that sign Input matrix is trained in advance.Computer equipment treats characteristic of division matrix by disaggregated model trained in advance
Classify, and then classifies to point cloud data to be sorted.Can be under automatic driving mode, computing resource is limited and real-time
Property require it is high when, effectively improve the classification accuracy of point cloud data.
In another embodiment, as shown in figure 3, providing a kind of point cloud data classification method, it is applied in this way
It is illustrated for computer equipment in Fig. 1, comprising the following steps:
Step 302, sample data is obtained.
Step 304, structuring pretreatment is carried out to sample data according to pre-set dimension, obtains sample characteristics matrix.
Step 306, deep learning model is called, deep learning model is trained according to sample characteristics matrix.
Step 308, when reaching default training condition, using deep learning model as disaggregated model.
Step 310, point cloud data to be sorted is obtained.
Step 312, structuring pretreatment is carried out to point cloud data to be sorted according to pre-set dimension, obtains feature square to be sorted
Battle array.
Step 314, disaggregated model trained in advance is called, characteristic of division matrix is treated by disaggregated model and is classified,
Obtain eigenmatrix classification results.
Step 316, classified according to eigenmatrix classification results to point cloud data to be sorted.
Computer equipment needs preparatory train classification models before classifying to point cloud data.Computer equipment obtains
Notebook data is sampled, the semantic information of point cloud data is labeled in sample data, passes through the available point cloud data of semantic information
Type.Computer equipment carries out structuring pretreatment to sample data according to pre-set dimension.Specifically, computer equipment determines sample
The corresponding data area of notebook data, divides data area, obtains composition of sample unit.Computer equipment is in each sample
The characteristic information of corresponding sample data is extracted in this structuring unit.Characteristic information is pressed in each composition of sample unit
Row is arranged, and multiple matrix units are generated.Matrix unit is carried out arrangement according to preset rules and generates sample by computer equipment
Eigenmatrix.
Structuring processing mode can there are many.Wherein it is possible to be rasterizing processing, it is also possible to voxelization processing.
It is identical that computer equipment, which carries out the pre-set dimension of structuring processing to sample data and point cloud data to be sorted,.Computer
The deep learning model of equipment calls can be two-dimensional convolution neural network model, be also possible to Three dimensional convolution neural network.It is deep
Degree learning model can be selected according to structuring processing mode.For example, when computer equipment carries out grid to point cloud data
It formats when handling, computer equipment calls two-dimensional convolution neural network model.Computer equipment obtains after being handled according to rasterizing
Sample characteristics matrix two-dimensional convolution neural network model is trained.When computer equipment carries out voxelization to point cloud data
When processing, computer equipment calls Three dimensional convolution neural network model.The sample that computer equipment obtains after being handled according to voxelization
Eigen matrix is trained Three dimensional convolution neural network model.
During computer equipment is trained deep learning model, computer equipment is according in sample characteristics matrix
The characteristic information of sample data is trained deep learning model.The extracting mode of sample data characteristic information can have more
Kind.Wherein it is possible to be automatically extracted by autocoder, can also manually count.Characteristic information includes: point cloud data midpoint
Number, the maximum height of point cloud data, the minimum altitude of point cloud data, the average height of point cloud data, the height of point cloud data
Variance etc..
Computer equipment carries out the sample characteristics unit in sample characteristics matrix by the deep learning model after training
Classification.Computer equipment puts classification of the largest number of types as sample characteristics unit according to certain one kind in sample characteristics unit
Label.When sample characteristics unit is empty, which can be labeled as empty class or background classes.Work as deep learning
When the classification accuracy of model reaches preset condition, using deep learning model as disaggregated model.At this point, the ginseng in disaggregated model
Number is fixed, the size of structural unit, i.e. pre-set dimension when parameter includes to sample data progress structuring pretreatment.
Computer equipment can classify to point cloud data to be sorted by disaggregated model trained in advance.Computer is set
Standby to carry out structuring processing to point cloud data to be sorted, which is with structuring treatment process in training process
It is identical.Pre-set dimension is also identical.Computer equipment is handled structuring by disaggregated model trained in advance
Eigenmatrix to be sorted is classified, and then is classified to point cloud data to be sorted.
In the present embodiment, computer equipment is by carrying out structuring processing to sample data, so as to deep learning mould
Type is trained and predicts.Computer equipment calls deep learning model, and is trained to depth model, obtains meeting default
The disaggregated model of condition.It can guarantee that disaggregated model is suitable for predicting the classification of point cloud data, thus under automatic driving mode,
Further improve the accuracy of point cloud data classification.
In one embodiment, as shown in figure 4, carrying out structuring pretreatment to point cloud data to be sorted according to pre-set dimension
The step of include:
Step 402, the corresponding data area of point cloud data to be sorted is determined according to point cloud data to be sorted.
Step 404, data area is divided according to the length in pre-set dimension in X direction, obtains first structure
Unit.
Step 406, data area is divided according to the width in pre-set dimension along Y-direction, obtains the second structuring
Unit.
Step 408, target structural unit is generated according to first structure unit and the second structuring unit.
Step 410, the characteristic information of corresponding point cloud data to be sorted is extracted in each target structural unit.
Step 412, characteristic information is arranged in each target structural unit by row, generates feature square to be sorted
Battle array.
Computer equipment carries out structuring pretreatment to point cloud data, it is necessary first to determine that point cloud data to be sorted is corresponding
Data area.Computer equipment calculates point cloud data coordinate in X-direction, Y-direction and Z-direction three according to point cloud data to be sorted
The difference of maxima and minima on a direction.Computer equipment determines the length, width and height of data area according to three differences.Number
It can be the minimum data space comprising all point cloud datas to be sorted according to region.For example, visual range is the laser thunder of 200m
Up to collected point cloud data to be sorted, data area is 400m*400m*10m (long * wide * high).
When the computing resource of computer equipment be less than preset threshold when, computer equipment can to point cloud data to be sorted into
The processing of row rasterizing, obtains eigenmatrix to be sorted.Specifically, the pre-set dimension of rasterizing processing can be long * wide.Default ruler
Very little length and width can be different.Computer equipment is drawn data area according to the length in pre-set dimension in X direction
Point, obtain first structure unit.First structure unit can be multiple structuring units divided impartial in X direction.Meter
It calculates machine equipment to be divided data area along Y-direction according to the width in pre-set dimension, obtains the second structuring unit.Second
Structuring unit can be multiple structuring units divided along Y-direction equalization.The height of structuring unit can be identical.
The sequence for carrying out direction division to data area is not construed as limiting.For example, computer equipment can be first according to the length in pre-set dimension
Degree divides data area in X direction, carries out Y-direction to first structure unit further according to the width in pre-set dimension
It divides, obtains target structural unit.Computer equipment can also first according to the width in pre-set dimension by data area along the side Y
To being divided, the division of X-direction is carried out to first structure unit further according to the length in pre-set dimension, obtains object construction
Change unit.
Computer equipment divides the corresponding data area of point cloud data to be sorted, obtained target structural unit
In include point cloud data to be sorted.Computer equipment extracts the feature letter of point cloud data to be sorted in target structural unit
Breath.Characteristic information includes: the number at point cloud data midpoint, the maximum height of point cloud data, the minimum altitude of point cloud data, point cloud
Average height, height variance of point cloud data of data etc..Computer equipment is arranged characteristic information by row, generator matrix
Unit.Computer equipment part matrix unit carries out arrangement according to preset rules and generates eigenmatrix to be sorted.
In the present embodiment, when the computing resource of computer equipment is less than preset threshold, by to be sorted cloud number
The division in X-direction and Y-direction both direction is carried out according to corresponding data area, obtains target structural unit.Extract mesh
The characteristic information in structuring unit is marked, characteristic information is arranged, eigenmatrix to be sorted is generated.It can be in automatic Pilot
Under mode, when computing resource is limited and requirement of real-time is high, the classification effectiveness of point cloud data to be sorted is improved.
In one embodiment, carrying out structuring pretreatment to point cloud data to be sorted includes: according to be sorted cloud number
According to the corresponding data area of determination point cloud data to be sorted；Data area is carried out in X direction according to the length in pre-set dimension
It divides, obtains first structure unit；Data area is divided along Y-direction according to the width in pre-set dimension, obtains
Two structuring units.Data area is divided along Z-direction according to the height in pre-set dimension, obtains third structuring list
Member；Target structural unit is generated according to first structure unit, the second structuring unit and third structural unit.
When the computing resource of computer equipment is more than or equal to preset threshold, computer equipment can be to point to be sorted
Cloud data carry out voxelization processing, obtain eigenmatrix to be sorted.Specifically, computer equipment is according to point cloud data meter to be sorted
Calculate the difference of point cloud data coordinate maxima and minima on three X-direction, Y-direction and Z-direction directions.Computer equipment
The length, width and height of data area are determined according to three differences.It include all point cloud datas to be sorted in data area.Computer
Equipment carries out voxelization processing to point cloud data to be sorted according to pre-set dimension.Wherein, pre-set dimension can be long * wide * high.Meter
Machine equipment is calculated to carry out data area in X direction to divide and according to the width in pre-set dimension according to the length in pre-set dimension
After degree is divided data area along Y-direction, computer equipment can according to the height in pre-set dimension by data area along the side Z
To being divided, third structuring unit is obtained.The length and width and height of pre-set dimension can be identical.Data area is carried out
The sequence that direction divides is not construed as limiting.
For example, computer equipment can first divide data area according to the length in pre-set dimension in X direction, then
The division for carrying out Y-direction to data area according to the width in pre-set dimension, finally according to the height in pre-set dimension to data
Region carries out the division of Z-direction, obtains target structural unit, and then obtain eigenmatrix to be sorted.Computer equipment may be used also
Data area is divided in X direction according to the length in pre-set dimension with elder generation, further according to the height logarithm in pre-set dimension
The division of Z-direction is carried out according to region, is finally carried out the division of Y-direction to data area according to the width in pre-set dimension, is obtained
Target structural unit, and then obtain eigenmatrix to be sorted.
In the present embodiment, when the computing resource of computer equipment is more than or equal to preset threshold, computer equipment
To the division on three point cloud data to be sorted corresponding data area progress X-direction, Y-direction and Z-direction directions.It can be very
It handles well and there is the case where blocking other point cloud datas above target object, so as under automatic driving mode, into one
Step improves the classification accuracy of point cloud data.
It in one embodiment, include matrix unit in eigenmatrix to be sorted；Characteristic of division is treated by disaggregated model
It includes: to calculate each matrix unit according to the characteristic information in eigenmatrix to be sorted by disaggregated model that matrix, which carries out classification,
Class probability；Classified according to class probability to matrix unit.
Computer equipment carries out structuring pretreatment, obtained eigenmatrix to be sorted to point cloud data to be sorted.Wait divide
The characteristic information of corresponding point cloud data to be sorted is stored in the matrix unit of category feature matrix.Computer equipment is according to feature
Information calculates the class probability of each matrix unit.Matrix unit is divided to the corresponding square of maximum class probability by computer equipment
In array element type, realization classifies to matrix unit.
In the present embodiment, computer equipment calculates the class probability of each matrix unit, realization pair by disaggregated model
Matrix unit is classified.Since disaggregated model is deep learning model trained in advance, can be counted under automatic driving mode
When calculation resource is limited and requirement of real-time is high, the classification accuracy of point cloud data is effectively improved.
In one embodiment, eigenmatrix classification results include matrix unit type；According to eigenmatrix classification results
Carrying out classification to point cloud data to be sorted includes: the index that point cloud data to be sorted is searched in eigenmatrix classification results；Root
The corresponding matrix unit class of point cloud data to be sorted is searched in eigenmatrix classification results according to the index of point cloud data to be sorted
Type；Classified according to the matrix unit type found to point cloud data to be sorted.
When computer equipment carries out structuring pretreatment to point cloud data to be sorted, the rope of point cloud data to be sorted is remained with
Draw.Computer equipment treats characteristic of division classification matrix by disaggregated model and classifies, and obtains eigenmatrix classification results.Meter
The index that machine equipment searches point cloud data to be sorted in eigenmatrix classification results is calculated, corresponding matrix unit type is obtained.
Matrix unit type is the corresponding data type of point cloud data to be sorted.Computer equipment is corresponding according to point cloud data to be sorted
Data type obtain the semantic information of point cloud data.
In the present embodiment, computer equipment is classified by treating characteristic of division matrix, further according to find to
The index of classification point cloud data, obtains corresponding matrix unit type.Computer equipment can be according to eigenmatrix classification results
Realization classifies to point cloud data to be sorted, to further improve the classification of point cloud data under automatic driving mode
Accuracy.
It should be understood that although each step in the flow chart of Fig. 2 to 4 is successively shown according to the instruction of arrow,
It is these steps is not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, these steps
There is no stringent sequences to limit for rapid execution, these steps can execute in other order.Moreover, in Fig. 2 to 4 at least
A part of step may include that perhaps these sub-steps of multiple stages or stage are not necessarily in same a period of time to multiple sub-steps
Quarter executes completion, but can execute at different times, the execution in these sub-steps or stage be sequentially also not necessarily according to
Secondary progress, but in turn or can replace at least part of the sub-step or stage of other steps or other steps
Ground executes.
In one embodiment, as shown in figure 5, providing a kind of point cloud data sorter, comprising: communication module 502,
Preprocessing module 504, Matrix Classification module 506 and data categorization module 508, in which:
Communication module 502, for obtaining point cloud data to be sorted.
Preprocessing module 504 obtains feature square to be sorted for carrying out structuring pretreatment to point cloud data to be sorted
Battle array.
Matrix Classification module 506 treats characteristic of division square by disaggregated model for calling disaggregated model trained in advance
Battle array is classified, and eigenmatrix classification results are obtained.
Data categorization module 508, for being classified according to eigenmatrix classification results to point cloud data to be sorted.
In one embodiment, above-mentioned apparatus further include: model generation module.Above-mentioned model generation module is for obtaining sample
Notebook data；Structuring pretreatment is carried out to sample data according to pre-set dimension, obtains sample characteristics matrix；Call deep learning mould
Type is trained deep learning model according to sample characteristics matrix；When reaching default training condition, by deep learning model
As disaggregated model.
In one embodiment, preprocessing module 504 is also used to determine to be sorted cloud number according to point cloud data to be sorted
According to corresponding data area；Data area is divided in X direction according to the length in pre-set dimension, obtains first structure
Unit；Data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；According to
One structuring unit and the second structuring unit generate target structural unit；The extraction pair in each target structural unit
The characteristic information for the point cloud data to be sorted answered；Characteristic information is arranged in each target structural unit by row, it is raw
At eigenmatrix to be sorted.
In one embodiment, preprocessing module 504 is also used to determine to be sorted cloud number according to point cloud data to be sorted
According to corresponding data area；Data area is divided in X direction according to the length in pre-set dimension, obtains first structure
Unit；Data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；According to pre-
If the height in size divides data area along Z-direction, third structuring unit is obtained；According to first structure list
Member, the second structuring unit and third structural unit generate target structural unit.
In one embodiment, Matrix Classification module 506 is also used to through disaggregated model according in eigenmatrix to be sorted
Characteristic information calculate the class probability of each matrix unit；Matrix unit is divided according to the class probability of matrix unit
Class.
In one embodiment, data categorization module 508 is also used to search point to be sorted in eigenmatrix classification results
The index of cloud data；Point cloud data pair to be sorted is searched in eigenmatrix classification results according to the index of point cloud data to be sorted
The matrix unit type answered；Classified according to the matrix unit type found to point cloud data to be sorted.
Specific about point cloud data sorter limits the limit that may refer to above for point cloud data classification method
Fixed, details are not described herein.Modules in above-mentioned point cloud data sorter can fully or partially through software, hardware and its
Combination is to realize.Above-mentioned each module can be embedded in the form of hardware or independently of in the processor in computer equipment, can also be with
It is stored in the memory in computer equipment in a software form, in order to which processor calls the above modules of execution corresponding
Operation.
In one embodiment, a kind of computer equipment is provided, internal structure chart can be as shown in Figure 6.The calculating
Machine equipment includes processor, memory, communication interface and the database connected by system bus.Wherein, the computer equipment
Processor for provide calculate and control ability.The memory of the computer equipment includes non-volatile memory medium, memory
Reservoir.The non-volatile memory medium is stored with operating system, computer program and database.The built-in storage is non-volatile
The operation of operating system and computer program in storage medium provides environment.The database of the computer equipment is for storing a little
Cloud data.The communication interface of the computer equipment is communicated for being attached with onboard sensor.The computer program is processed
To realize a kind of point cloud data classification method when device executes.
It will be understood by those skilled in the art that structure shown in Fig. 6, only part relevant to application scheme is tied
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific computer equipment
It may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
In one embodiment, a kind of computer equipment, including memory and processor are provided, which is stored with
Computer program, the processor realize the step in above-mentioned each embodiment of the method when executing computer program.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program realizes the step in above-mentioned each embodiment of the method when being executed by processor.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of point cloud data classification method, which comprises
Obtain point cloud data to be sorted；
Structuring pretreatment is carried out to the point cloud data to be sorted according to pre-set dimension, obtains eigenmatrix to be sorted；
Disaggregated model trained in advance is called, is classified by the disaggregated model to the eigenmatrix to be sorted, is obtained
Eigenmatrix classification results；
Classified according to the eigenmatrix classification results to the point cloud data to be sorted.
2. the method according to claim 1, wherein it is described obtain point cloud data to be sorted before, the side
Method further include:
Obtain sample data；
Structuring pretreatment is carried out to the sample data according to pre-set dimension, obtains sample characteristics matrix；
Deep learning model is called, the deep learning model is trained according to the sample characteristics matrix；
When reaching default training condition, using the deep learning model as disaggregated model.
3. the method according to claim 1, wherein it is described according to pre-set dimension to the point cloud data to be sorted
Carrying out structuring pretreatment includes:
The corresponding data area of the point cloud data to be sorted is determined according to the point cloud data to be sorted；
The data area is divided in X direction according to the length in pre-set dimension, obtains first structure unit；
The data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；
Target structural unit is generated according to the first structure unit and the second structuring unit；
The characteristic information of corresponding point cloud data to be sorted is extracted in each target structural unit；
The characteristic information is arranged in each target structural unit by row, eigenmatrix to be sorted is generated.
4. the method according to claim 1, wherein described pre- to the point cloud data progress structuring to be sorted
Processing includes:
The corresponding data area of the point cloud data to be sorted is determined according to the point cloud data to be sorted；
The data area is divided in X direction according to the length in pre-set dimension, obtains first structure unit；
The data area is divided along Y-direction according to the width in pre-set dimension, obtains the second structuring unit；
The data area is divided along Z-direction according to the height in pre-set dimension, obtains third structuring unit；
Target structural unit is generated according to the first structure unit, the second structuring unit and third structural unit.
5. the method according to claim 1, wherein the eigenmatrix to be sorted includes matrix unit；It is described
Carrying out classification to the eigenmatrix to be sorted by the disaggregated model includes:
The classification of each matrix unit is calculated according to the characteristic information in the eigenmatrix to be sorted by the disaggregated model
Probability；
Classified according to the class probability of the matrix unit to the matrix unit.
6. the method according to claim 1, wherein the eigenmatrix classification results include matrix unit class
Type；Carrying out classification to the point cloud data to be sorted according to the eigenmatrix classification results includes:
The index of the point cloud data to be sorted is searched in the eigenmatrix classification results；
The to be sorted cloud number is searched in the eigenmatrix classification results according to the index of the point cloud data to be sorted
According to corresponding matrix unit type；
Classified according to the matrix unit type found to the point cloud data to be sorted.
7. a kind of point cloud data sorter, which is characterized in that described device includes:
Communication module, for obtaining point cloud data to be sorted；
Preprocessing module obtains eigenmatrix to be sorted for carrying out structuring pretreatment to the point cloud data to be sorted；
Matrix Classification module, for calling disaggregated model trained in advance, by the disaggregated model to the feature to be sorted
Matrix is classified, and eigenmatrix classification results are obtained；
Data categorization module, for being classified according to the eigenmatrix classification results to the point cloud data to be sorted.
8. device according to claim 7, which is characterized in that the preprocessing module is also used to according to the point to be sorted
Cloud data determine the corresponding data area of the point cloud data to be sorted；According to the length in pre-set dimension by the data area
It is divided in X direction, obtains first structure unit；According to the width in pre-set dimension by the data area along Y-direction
It is divided, obtains the second structuring unit；Target is generated according to the first structure unit and the second structuring unit
Structuring unit；The characteristic information of corresponding point cloud data to be sorted is extracted in each target structural unit；By the spy
Reference breath is arranged in each target structural unit by row, and eigenmatrix to be sorted is generated.
9. a kind of computer equipment, including memory and processor, the memory are stored with the meter that can be run on a processor
Calculation machine program, which is characterized in that the processor realizes any one of claims 1 to 6 institute when executing the computer program
The step of stating method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of any one of claims 1 to 6 the method is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910231062.6A CN110097077B (en) | 2019-03-26 | 2019-03-26 | Point cloud data classification method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910231062.6A CN110097077B (en) | 2019-03-26 | 2019-03-26 | Point cloud data classification method and device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110097077A true CN110097077A (en) | 2019-08-06 |
| CN110097077B CN110097077B (en) | 2021-12-10 |
Family
ID=67444009
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910231062.6A Active CN110097077B (en) | 2019-03-26 | 2019-03-26 | Point cloud data classification method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110097077B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111338336A (en) * | 2020-02-11 | 2020-06-26 | 腾讯科技（深圳）有限公司 | Automatic driving method and device |
| CN112444784A (en) * | 2019-08-29 | 2021-03-05 | 北京市商汤科技开发有限公司 | Three-dimensional target detection and neural network training method, device and equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150006117A1 (en) * | 2013-07-01 | 2015-01-01 | Here Global B.V. | Learning Synthetic Models for Roof Style Classification Using Point Clouds |
| CN106295586A (en) * | 2016-08-16 | 2017-01-04 | 长春理工大学 | Humanoid target identification method based on single line cloud data machine learning and device |
| CN106408011A (en) * | 2016-09-09 | 2017-02-15 | 厦门大学 | Laser scanning three-dimensional point cloud tree automatic classifying method based on deep learning |
| CN107092859A (en) * | 2017-03-14 | 2017-08-25 | 佛山科学技术学院 | A kind of depth characteristic extracting method of threedimensional model |
| CN108133227A (en) * | 2017-11-29 | 2018-06-08 | 北京数字绿土科技有限公司 | Laser radar point cloud data sorting technique and device |
| CN109118564A (en) * | 2018-08-01 | 2019-01-01 | 湖南拓视觉信息技术有限公司 | A kind of three-dimensional point cloud labeling method and device based on fusion voxel |
| CN109426828A (en) * | 2017-08-28 | 2019-03-05 | 中国科学院深圳先进技术研究院 | Point cloud classifications method, apparatus, equipment and storage medium |
| CN109492658A (en) * | 2018-09-21 | 2019-03-19 | 北京车和家信息技术有限公司 | A kind of point cloud classifications method and terminal |
-
2019
- 2019-03-26 CN CN201910231062.6A patent/CN110097077B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150006117A1 (en) * | 2013-07-01 | 2015-01-01 | Here Global B.V. | Learning Synthetic Models for Roof Style Classification Using Point Clouds |
| CN106295586A (en) * | 2016-08-16 | 2017-01-04 | 长春理工大学 | Humanoid target identification method based on single line cloud data machine learning and device |
| CN106408011A (en) * | 2016-09-09 | 2017-02-15 | 厦门大学 | Laser scanning three-dimensional point cloud tree automatic classifying method based on deep learning |
| CN107092859A (en) * | 2017-03-14 | 2017-08-25 | 佛山科学技术学院 | A kind of depth characteristic extracting method of threedimensional model |
| CN109426828A (en) * | 2017-08-28 | 2019-03-05 | 中国科学院深圳先进技术研究院 | Point cloud classifications method, apparatus, equipment and storage medium |
| CN108133227A (en) * | 2017-11-29 | 2018-06-08 | 北京数字绿土科技有限公司 | Laser radar point cloud data sorting technique and device |
| CN109118564A (en) * | 2018-08-01 | 2019-01-01 | 湖南拓视觉信息技术有限公司 | A kind of three-dimensional point cloud labeling method and device based on fusion voxel |
| CN109492658A (en) * | 2018-09-21 | 2019-03-19 | 北京车和家信息技术有限公司 | A kind of point cloud classifications method and terminal |
Non-Patent Citations (4)
| Title |
|---|
| HENG ZHANG等: ""Object Classification Based on 3D Point Clouds Covariance Descriptor"", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTATIONAL SCIENCE AND ENGINEERING (CSE) AND IEEE INTERNATIONAL CONFERENCE ON EMBEDDED AND UBIQUITOUS COMPUTING (EUC)》 * |
| 李峰等: "《机载LiDAR系统原理与点云处理方法》", 31 August 2017, 北京：煤炭工业出版社 * |
| 胡春梅等: "《地面激光雷达与近景摄影测量技术集成》", 31 May 2017 * |
| 赵刚等: ""基于Gradient Boosting的车载LiDAR点云分类"", 《地理信息世界》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112444784A (en) * | 2019-08-29 | 2021-03-05 | 北京市商汤科技开发有限公司 | Three-dimensional target detection and neural network training method, device and equipment |
| CN112444784B (en) * | 2019-08-29 | 2023-11-28 | 北京市商汤科技开发有限公司 | Three-dimensional target detection and neural network training method, device and equipment |
| CN111338336A (en) * | 2020-02-11 | 2020-06-26 | 腾讯科技（深圳）有限公司 | Automatic driving method and device |
| CN111338336B (en) * | 2020-02-11 | 2021-07-13 | 腾讯科技（深圳）有限公司 | Automatic driving method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110097077B (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264336A (en) | A kind of anti-system of intelligent case based on big data | |
| CN108230339A (en) | A kind of gastric cancer pathological section based on pseudo label iteration mark marks complementing method | |
| Junos et al. | An optimized YOLO‐based object detection model for crop harvesting system | |
| CN110348384B (en) | Small target vehicle attribute identification method based on feature fusion | |
| CN110135231A (en) | Animal face recognition methods, device, computer equipment and storage medium | |
| CN108664844A (en) | The image object semantics of convolution deep neural network identify and tracking | |
| CN107690660A (en) | Image-recognizing method and device | |
| CN112560722B (en) | Airplane target identification method and device, computer equipment and storage medium | |
| CN109886554A (en) | Unlawful practice method of discrimination, device, computer equipment and storage medium | |
| WO2021185121A1 (en) | Model generation method and apparatus, object detection method and apparatus, device, and storage medium | |
| CN106204642A (en) | A kind of cell tracker method based on deep neural network | |
| CN106845416A (en) | Obstacle recognition method and device, computer equipment and computer-readable recording medium | |
| CN110008853A (en) | Pedestrian detection network and model training method, detection method, medium, equipment | |
| Pinto et al. | Iris flower species identification using machine learning approach | |
| CN110097077A (en) | Point cloud data classification method, device, computer equipment and storage medium | |
| CN111210158B (en) | Target address determining method, device, computer equipment and storage medium | |
| US20150242676A1 (en) | Method for the Supervised Classification of Cells Included in Microscopy Images | |
| CN110163206A (en) | Licence plate recognition method, system, storage medium and device | |
| Lorentz et al. | Explaining defect detection with saliency maps | |
| CN112699858B (en) | Unmanned platform smoke fog sensing method and system, computer equipment and storage medium | |
| CN112766176A (en) | Training method of lightweight convolutional neural network and face attribute recognition method | |
| CN110287970B (en) | Weak supervision object positioning method based on CAM and covering | |
| Li et al. | Incremental learning of infrared vehicle detection method based on SSD | |
| US11755688B2 (en) | Apparatus and method for generating training data for a machine learning system | |
| CN115565115A (en) | Outfitting intelligent identification method and computer equipment |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |