CN109492658A - A kind of point cloud classifications method and terminal - Google Patents
A kind of point cloud classifications method and terminal Download PDFInfo
- Publication number
- CN109492658A CN109492658A CN201811107718.5A CN201811107718A CN109492658A CN 109492658 A CN109492658 A CN 109492658A CN 201811107718 A CN201811107718 A CN 201811107718A CN 109492658 A CN109492658 A CN 109492658A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- cloud data
- disaggregated model
- data
- 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
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The present invention discloses a kind of point cloud classifications method and terminal, and wherein method includes: the disaggregated model learnt using the point cloud data by known type, carries out classification processing, acquisition classification results to point cloud data to be sorted；According to classification results determine point cloud data to be sorted include disaggregated model can not Accurate classification the first point cloud data in the case where, obtain the first point cloud data the first kind；Using the first point cloud data and the first kind, the second point cloud data and Second Type, disaggregated model is updated；Using updated disaggregated model, classification processing is carried out to point cloud data to be sorted, obtains classification results.In this way, terminal does not need pre- to first pass through mass data learning classification rule, it only needs to learn to obtain disaggregated model according to the point cloud data of known type, disaggregated model can be used to classify point cloud data, when detecting new type point, learn the feature of new type point and update disaggregated model, improves the universality and maintainability of point cloud classifications.
Description
Technical field
The present invention relates to technical field of data processing more particularly to a kind of point cloud classifications methods and terminal.
Background technique
With the development of data processing technique, the application of point cloud classifications method is also more universal.Existing point cloud classifications method
The method of generally rule-based algorithm, rule-based algorithm are mainly geometry, density, the Reflectivity by object to be identified
And spatial topotaxy between different objects to be identified.And be based partially on the point cloud classifications method of learning method, then it is that will put cloud
It is divided into a certain size voxel, by calculating some features of each voxel, then constitutive characteristic vector passes through learning method
It carries out by voxel classification.Existing scheme needs are the case where not considering preceding for algorithm design, pervasive premised on certain rule
Property is poor.In the point cloud for the new type that can not classify in face of existing algorithm, need to change even restructing algorithm to be adapted to new class
The class object of type, it is maintainable poor.
As it can be seen that there are universality and maintainable poor technical problems for existing point cloud classifications method.
Summary of the invention
The embodiment of the present invention provides a kind of point cloud classifications method and terminal, and to solve existing point cloud classifications method, there are pervasive
Property and maintainable poor technical problem.
In order to achieve the above object, concrete scheme provided by the invention is as follows:
In a first aspect, the embodiment of the invention provides a kind of point cloud classifications methods, comprising:
The disaggregated model learnt using the point cloud data by known type divides point cloud data to be sorted
Class processing, obtains classification results；
Determining that the point cloud data to be sorted includes that the disaggregated model can not accurately divide according to the classification results
In the case where first point cloud data of class, the first kind of first point cloud data is obtained；
Using first point cloud data and the first kind, the second point cloud data and Second Type, described point is updated
Class model, wherein second point cloud data is the point cloud data that the disaggregated model is capable of Accurate classification, the Second Type
For type belonging to second point cloud data；
Using the updated disaggregated model, classification processing is carried out to the point cloud data to be sorted, is classified
As a result.
Optionally, before described the step of carrying out classification processing to point cloud data to be sorted, the method also includes:
Receive the point cloud data of known type, wherein the point cloud data of the known type includes at least two data
Collect, the corresponding type of the point in each data subset；
The feature for learning the point at least two data subset, obtains the disaggregated model.
Optionally, the feature of the point in study at least two data subset, obtains the step of the disaggregated model
Suddenly, comprising:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, passed through
It crosses and trains, obtain the disaggregated model.
It is optionally, described to utilize first point cloud data and the first kind, the second point cloud data and Second Type,
The step of updating the disaggregated model, comprising:
By the point and the corresponding first kind and second point cloud data and the second class in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
Optionally, the point by first point cloud data and the corresponding first kind and the second point cloud
Data and Second Type are input to the deep neural network for being loaded with the disaggregated model, by training, update the classification
The step of model, comprising:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, every number
According to the corresponding type of point in subset；
By the point and second point cloud data and at least one data subset in first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, by training, update the disaggregated model.
Optionally, the point cloud data of the known type and/or the point cloud data to be sorted are that vehicle was being driven a vehicle
Cheng Zhong, the laser radar acquired image data on the vehicle.
Optionally, after the step of acquisition classification results, the method also includes:
Determining that the point cloud data to be sorted includes that the disaggregated model can not accurately divide according to the classification results
In the case where first point cloud data of class, prompt information is exported, wherein the prompt information is for prompting the point to be sorted
Cloud data include the disaggregated model can not Accurate classification the first point cloud data.
Second aspect, the embodiment of the invention provides a kind of terminals, comprising:
First categorization module, for using the disaggregated model learnt by the point cloud data of known type, treating point
The point cloud data of class carries out classification processing, obtains classification results；
Module is obtained, for determining that the point cloud data to be sorted includes the classification mould according to the classification results
Type can not obtain the first kind of first point cloud data in the case where the first point cloud data of Accurate classification；
Update module, for utilizing first point cloud data and the first kind, the second point cloud data and the second class
Type updates the disaggregated model, wherein second point cloud data is the point cloud number that the disaggregated model is capable of Accurate classification
According to the Second Type is type belonging to second point cloud data；
Second categorization module carries out the point cloud data to be sorted for utilizing the updated disaggregated model
Classification processing obtains classification results.
Optionally, the terminal further include:
Receiving module, for receiving the point cloud data of known type, wherein the point cloud data of the known type includes extremely
Lack two data subsets, the corresponding type of the point in each data subset；
Study module obtains the disaggregated model for learning the feature of the point at least two data subset.
Optionally, the study module is used for:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, passed through
It crosses and trains, obtain the disaggregated model.
Optionally, the update module is used for:
By the point and the corresponding first kind and second point cloud data and the second class in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
Optionally, the update module is specifically used for:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, every number
According to the corresponding type of point in subset；
By the point and second point cloud data and at least one data subset in first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, by training, update the disaggregated model.
Optionally, the point cloud data of the known type and/or the point cloud data to be sorted are that vehicle was being driven a vehicle
Cheng Zhong, the laser radar acquired image data on the vehicle.
Optionally, the terminal further include:
Output module, for determining that the point cloud data to be sorted includes the classification mould according to the classification results
Type can not export prompt information in the case where the first point cloud data of Accurate classification, wherein the prompt information is for prompting institute
State point cloud data to be sorted include the disaggregated model can not Accurate classification the first point cloud data.
The third aspect, the embodiment of the invention provides a kind of terminal, the terminal includes: memory, processor and storage
On the memory and the computer program that can run on the processor；The processor executes the computer program
The point cloud classifications method of Shi Shixian as described in relation to the first aspect.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage
It is stored with computer program on medium, point cloud classifications as described in relation to the first aspect are realized when the computer sequence is executed by processor
The step of method.
5th aspect, the embodiment of the invention provides a kind of vehicles, including as any in second aspect or the third aspect
Terminal described in.
In the embodiment of the present invention, terminal does not need pre- to first pass through mass data learning classification rule, it is only necessary to according to known
The point cloud data of type learns to obtain disaggregated model, can be used to classify point cloud data by disaggregated model, and new detecting
When type point, learns the feature of new type point and update disaggregated model, improve the universality and maintainability of point cloud classifications.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, needed in being described below to the embodiment of the present invention
Attached drawing to be used is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention,
For those of ordinary skill in the art, without any creative labor, it can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is a kind of flow diagram of point cloud classifications method provided in an embodiment of the present invention；
Fig. 2 is the flow diagram of another point cloud classifications method provided in an embodiment of the present invention；
Fig. 3 is the structural schematic diagram of terminal provided in an embodiment of the present invention；
Fig. 4 is the structural schematic diagram of another terminal provided in an embodiment of the present invention；
Fig. 5 is the structural schematic diagram of another terminal provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
It is the flow diagram of point cloud classifications method provided in an embodiment of the present invention referring to Fig. 1.As shown in Figure 1, a kind of point
Cloud classification method is applied to terminal, mainly comprises the steps that
Step 101, the disaggregated model learnt using the point cloud data by known type, to point cloud number to be sorted
According to classification processing is carried out, classification results are obtained.
Point cloud classifications method provided in this embodiment is applied to terminal, for handling point cloud data.It is related
The point cloud data of known type and the point cloud data of UNKNOWN TYPE can be for vehicles when driving, the laser on the vehicle
Radar acquired image data.The point cloud data that the laser radar of vehicle where terminal directly acquires acquires in real time, carries out a little
Cloud classification processing and application.Certainly, point cloud data may be the point cloud data for the other equipment acquisition that terminal directly acquires, no
It limits.
Terminal can first receive the point cloud number of known type when needing to carry out classification processing to point cloud data to be sorted
According to being learnt by point cloud data to the known type, obtain disaggregated model.Disaggregated model stores the point cloud number of known type
The feature all put and the corresponding relationship for putting affiliated type, the disaggregated model in can be used for classifying to point cloud data
Processing.The disaggregated model that terminal is obtained using study carries out classification processing to point cloud data to be sorted, and obtains corresponding point
Class result.Classification results may include the affiliated type of whole points in the point cloud data to be sorted.
Step 102, according to the classification results determine the point cloud data to be sorted include the disaggregated model without
In the case where first point cloud data of method Accurate classification, the first kind of first point cloud data is obtained；
Terminal carries out classification processing to point cloud data to be sorted using the disaggregated model that study obtains, and obtains classification results
Afterwards, it may determine that the disaggregated model whether by the point cloud data Accurate classification to be sorted according to classification results.If terminal foundation
Classification results determine that the point cloud data to be sorted contains part point cloud data really and cannot be classified model Accurate classification,
It is the first point cloud data by the partial dot cloud data definition.Terminal obtains the first point cloud data and first point cloud data can
It can the corresponding first kind.
Whether terminal judges in be sorted cloud to include the first point cloud data that cannot be accurately identified according to classification results
Method can be with are as follows: judges of the feature of the feature of the corresponding point cloud data of each type and the point cloud data of disaggregated model record
With degree, the point cloud data using characteristic matching degree lower than preset threshold is as the first point cloud data, alternatively, terminal can also will classify
As a result it exports, judges classification accuracy according to classification results by user, and the part point cloud data that subscriber frame is selected is as nothing
The first point cloud data that method accurately identifies.
After terminal determines the first point cloud data that exists and can not accurately identify, the first kind of first point cloud data is obtained
Type.Terminal can export first point cloud data, and adding first point cloud data by user may corresponding type.
In addition, terminal, after obtaining classification results, the method can also include:
Determining that the point cloud data to be sorted includes that the disaggregated model can not accurately divide according to the classification results
In the case where first point cloud data of class, prompt information is exported, wherein the prompt information is for prompting the point to be sorted
Cloud data include the disaggregated model can not Accurate classification the first point cloud data.
Terminal, which determines, to be existed and prompt information can not be exported, to prompt the user portion after the first point cloud data of Accurate classification
The case where point point cloud data can not be by Accurate classification, can also be determined whether to continue by user using current class result or after
Continuous study new type point.
Step 103, using first point cloud data and the first kind, the second point cloud data and Second Type, more
The new disaggregated model；
User continues after learning new type point by the way that terminal is determining, and terminal can obtain first cloud according to above-mentioned steps
Data and the corresponding first kind of the first point cloud data update disaggregated model.Disaggregated model is capable of the point cloud of Accurate classification
Data definition is the second point cloud data, and type definition belonging to the second point cloud data is Second Type.
Terminal learns the feature at the first point cloud data midpoint and the corresponding relationship of the first kind, and combines the second point cloud data
The feature at midpoint and the corresponding relationship of Second Type, update the disaggregated model, so that disaggregated model being capable of Accurate classification
One point cloud data, the second point cloud data.
Step 104, using the updated disaggregated model, classification processing is carried out to the point cloud data to be sorted,
Obtain classification results.
After terminal updates the disaggregated model according to above-mentioned steps, classification processing is carried out to point cloud data to be sorted, with
Obtain classification results, the accurate classification results of point cloud data as to be sorted.
The point cloud classifications method that the embodiments of the present invention provide, terminal do not need integrated in advance/configuration mass data
Practise classifying rules, it is only necessary to learn to obtain disaggregated model according to the point cloud data of known type, can be used to disaggregated model will
Point cloud data classification, and when detecting new type point, learn the feature of new type point and update disaggregated model, improves a cloud
The universality and maintainability of classification.
In a specific embodiment, described in step 101, the step of classification processing is carried out to point cloud data to be sorted
Before rapid, the method can also include:
Receive the point cloud data of known type；
The feature for learning the point at least two data subset, obtains the disaggregated model.
In present embodiment, point cloud classifications method further comprises the acquisition process of disaggregated model.Terminal receives known type
Point cloud data.Received known type point cloud data include at least two data subsets, at least two data subset
The corresponding type of interior point.
Specifically, terminal receives the point cloud data of user's input, known to the type of the point in the point cloud data.Terminal is by institute
Received point cloud data carries out three-dimensional rendering, then applies input to be labeled, i.e., from point cloud data from user to terminal
It clicks or frame selects the point that Partial Feature is consistent or feature is close, form a data subset, and be partial data
Collection adds corresponding known type.It repeats the above process, until whole points in point cloud data are divided into corresponding data
In subset, and a corresponding known type, the corresponding type of point in each data subset are added for each data subset
Uniquely, to avoid classification error.
After terminal receives at least two data subsets and corresponding types, learn the spy of the point at least two data subset
Sign, in conjunction with the corresponding type of point in each data subset, can be obtained the disaggregated model.
Further, the feature of the point in study at least two data subset, obtains the disaggregated model
Step may include:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, passed through
It crosses and trains, obtain the disaggregated model.
Point at least two data subsets in point cloud data is input to pre-designed depth nerve net by terminal
Then network is trained, disaggregated model can be obtained.When specific implementation, terminal needs the neural network using a multilayer, right
The point cloud data of input is pre-processed, and the format of point cloud data is converted to the manageable data format of neural network, with
Just neural network can carry out relevant treatment to the received point cloud data of institute.Pretreated point cloud data is input to pre- by terminal
First designed deep neural network, is trained, and passes through back-propagation algorithm learning network structure to a large amount of tape label data
In parameter.Specifically, initialization model parameter of the terminal based on one group of setting, input data passes through several convolutional layers and pond
Layer, can obtain a desired output by propagated forward, if this desired output and the concrete class label of data be not identical,
Error is successively then propagated back into input layer, every layer of neuron can carry out more the parameter in network structure according to the error
Newly.For convolutional neural networks, parameter to be learned includes but is not limited to convolution nuclear parameter, the Connecting quantity of interlayer and each
The biasing of layer.Specific training process may refer to existing training process, not repeat them here.
Multiple convolution operation, it is constantly excellent then by the backward feedback of the type of each data subset and convolution interlayer
Change the parameter of neural network, it is trained to achieve the effect that, a model file is finally obtained, that is, forms the disaggregated model.
Correspondingly, terminal learns after obtaining disaggregated model, classification processing is carried out to point cloud data to be sorted.Terminal will be to
The point cloud data of classification is input to the deep neural network for being loaded with the disaggregated model, then laggard by multilayer convolution operation
Enter full articulamentum, with output category result, the point as in the to be sorted cloud may corresponding type.Terminal is according to classification knot
Fruit judge whether there is can not Accurate classification the first point cloud data, the first point cloud data if it does not exist can then wait for this point
The classification results of the point cloud data of class are stored into data storage.If it is determined that in the presence of can not Accurate classification the first point cloud data,
Then terminal carries out the feature learning of new type point again, to update disaggregated model.
Specifically, utilizing first point cloud data and the first kind, the second point cloud data described in step 103
And Second Type, the step of updating the disaggregated model, may include:
By the point and the corresponding first kind and second point cloud data and the second class in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
In present embodiment, terminal is by the point and the first kind and the second point cloud data and in the first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, and repeat the training process of sorting algorithm, in conjunction with
The second point cloud data and corresponding Second Type of disaggregated model energy Accurate classification, can be obtained new model file, using new
Model file replace before model file, as have updated disaggregated model.
In a specific embodiment, the point by first point cloud data and the corresponding first kind, with
And second point cloud data and Second Type are input to the deep neural network for being loaded with the disaggregated model, by instruction
The step of practicing, updating the disaggregated model, comprising:
Receive first point cloud data；
By the point and second point cloud data and at least two data subsets in first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, by training, update the disaggregated model.
Specifically, the first point cloud data that can not be accurately identified in view of terminal may include the point cloud of more than one type
Data, terminal can also determine at least one data subset that first point cloud data may include.
Point at least one data subset in first point cloud data is input to and is loaded with the classification mould by terminal
The deep neural network of type carries out multiple convolution operation, can be obtained the feature of the point in first point cloud data, and successively comes
Update disaggregated model.
After terminal has updated disaggregated model, the full articulamentum of original neural network is modified, so that deep neural network can be with
Learn new type point.Hereafter, if terminal recycles the deep neural network for being loaded with updated disaggregated model to new point cloud
Data carry out classification processing, however, it is determined that there is the new type point that can not be accurately identified, terminal repeats above-mentioned study and update
Process further increases the identifiable point of disaggregated model, to improve the classification feature of disaggregated model.
To sum up, as shown in Fig. 2, the implementation process of point cloud classifications method provided in an embodiment of the present invention can be with are as follows:
Step 201 learns to obtain disaggregated model by the point cloud data of known type, using disaggregated model to be sorted
Point cloud data carries out classification processing, obtains classification results.
Step 202, judge point cloud to be sorted whether include disaggregated model can not Accurate classification new type point；
If it is determined that new type point is not present, then follow the steps 203, by the classification results of point cloud data to be sorted export to
Data storage is stored；
If it is determined that thening follow the steps 204 there are new type point, selecting mark for new type point as sample pane, to new type
Point is trained, and updates disaggregated model；
Step 205 carries out classification processing to point cloud data to be sorted using updated disaggregated model, obtains classification knot
Fruit, and execute step 203.
To sum up, point cloud classifications method provided in an embodiment of the present invention, for before classification design not in view of the case where, can also
To complete by increasing mark sample, there is stronger universality to classification task.Related algorithm is relatively simple, to more
The classification of type objects is completed at the same time by a kind of algorithm, and algorithm structure is fairly simple.In addition, facing what existing algorithm can not classify
When the point cloud of new type, without changing algorithm.Only need newly to mark a certain amount of sample, re -training model, it is maintainable compared with
It is good.Also, invention carries out point-by-point classification, and classification accuracy is higher.
It is a kind of structural schematic diagram of terminal provided in an embodiment of the present invention referring to Fig. 3.As shown in figure 3, the terminal
300 may include:
First categorization module 301, the disaggregated model for being learnt using the point cloud data by known type, is treated
The point cloud data of classification carries out classification processing, obtains classification results；
Module 302 is obtained, for determining that the point cloud data to be sorted includes described point according to the classification results
Class model can not obtain the first kind of first point cloud data in the case where the first point cloud data of Accurate classification；
Update module 303, for utilizing first point cloud data and the first kind, the second point cloud data and second
Type updates the disaggregated model, wherein second point cloud data is the point cloud number that the disaggregated model is capable of Accurate classification
According to the Second Type is type belonging to second point cloud data；
Second categorization module 304, for utilize the updated disaggregated model, to the point cloud data to be sorted into
Row classification processing obtains classification results.
Optionally, the point cloud data of the known type and/or the point cloud data to be sorted are that vehicle was being driven a vehicle
Cheng Zhong, the laser radar acquired image data on the vehicle.
Optionally, as shown in figure 4, the terminal 300 can also include:
Receiving module 305, for receiving the point cloud data of known type, wherein the point cloud data packet of the known type
At least two data subsets are included, the corresponding type of the point in each data subset；
Study module 306 obtains the classification mould for learning the feature of the point at least two data subset
Type.
Optionally, the study module 306 can be used for:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, passed through
It crosses and trains, obtain the disaggregated model.
Optionally, the update module 303 can be used for:
By the point and the corresponding first kind and second point cloud data and the second class in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
Optionally, the update module 303 is specifically used for:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, every number
According to the corresponding type of point in subset；
By the point and second point cloud data and at least one data subset in first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, by training, update the disaggregated model.
Optionally, as shown in figure 4, the terminal 300 can also include:
Output module 307, for determining that the point cloud data to be sorted includes described point according to the classification results
Class model can not export prompt information in the case where the first point cloud data of Accurate classification, wherein the prompt information is for mentioning
Show the point cloud data to be sorted include the disaggregated model can not Accurate classification the first point cloud data.
The terminal that the embodiments of the present invention provide does not need pre- to first pass through mass data learning classification rule, it is only necessary to
Learn to obtain disaggregated model according to the point cloud data of known type, disaggregated model can be used to classify point cloud data, and
When detecting new type point, learn new type point feature simultaneously update disaggregated model, improve point cloud classifications universality and can
Maintainability.The specific implementation process of terminal provided in this embodiment may refer to the point cloud that above-mentioned embodiment shown in FIG. 1 provides
The specific implementation process of classification method, this is no longer going to repeat them.
Referring to Fig. 5, be another embodiment of the present invention provides terminal structural schematic diagram.As shown in figure 5, to realize this hair
A kind of terminal of bright each embodiment, which includes but is not limited to: radio frequency unit 501, network module 502, audio output
Unit 503, input unit 504, sensor 505, display unit 506, user input unit 507, interface unit 508, memory
509, the components such as processor 510 and power supply 511.It will be understood by those skilled in the art that terminal structure shown in Fig. 5 is simultaneously
The not restriction of structure paired terminal, terminal may include perhaps combining certain components or not than illustrating more or fewer components
Same component layout.In embodiments of the present invention, terminal includes but is not limited to mobile phone, tablet computer, laptop, palm electricity
Brain, car-mounted terminal, wearable device and pedometer etc..
Wherein, processor 510 can be used for:
The disaggregated model learnt using the point cloud data by known type divides point cloud data to be sorted
Class processing, obtains classification results；
Determining that the point cloud data to be sorted includes that the disaggregated model can not accurately divide according to the classification results
In the case where first point cloud data of class, the first kind of first point cloud data is obtained；
Using first point cloud data and the first kind, the second point cloud data and Second Type, described point is updated
Class model, wherein second point cloud data is the point cloud data that the disaggregated model is capable of Accurate classification, the Second Type
For type belonging to second point cloud data；
Using the updated disaggregated model, classification processing is carried out to the point cloud data to be sorted, is classified
As a result.
Optionally, the processor 510 can be also used for:
Receive the point cloud data of known type, wherein the point cloud data of the known type includes at least two data
Collect, the corresponding type of the point in each data subset；
The feature for learning the point at least two data subset, obtains the disaggregated model.
Optionally, the processor 510 can be also used for:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, passed through
It crosses and trains, obtain the disaggregated model.
Optionally, the processor 510 can be also used for:
By the point and the corresponding first kind and second point cloud data and the second class in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
Optionally, the processor 510 can be also used for:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, every number
According to the corresponding type of point in subset；
By the point and second point cloud data and at least one data subset in first point cloud data
Two types are input to the deep neural network for being loaded with the disaggregated model, by training, update the disaggregated model.
Optionally, the point cloud data of the known type and/or the point cloud data to be sorted are that vehicle was being driven a vehicle
Cheng Zhong, the laser radar acquired image data on the vehicle.
Optionally, the processor 510 can be also used for: determine the point cloud to be sorted according to the classification results
Data include that the disaggregated model can not export prompt information, wherein described in the case where the first point cloud data of Accurate classification
Prompt information be used for prompt the point cloud data to be sorted include the disaggregated model can not Accurate classification first cloud number
According to.
Terminal 500 can be realized each process that terminal is realized in previous embodiment, no longer superfluous here to avoid repeating
It states.
It should be understood that the embodiment of the present invention in, radio frequency unit 501 can be used for receiving and sending messages or communication process in, signal
Send and receive, specifically, by from base station downlink data receive after, to processor 510 handle；In addition, by uplink
Data are sent to base station.In general, radio frequency unit 501 includes but is not limited to antenna, at least one amplifier, transceiver, coupling
Device, low-noise amplifier, duplexer etc..In addition, radio frequency unit 501 can also by wireless communication system and network and other set
Standby communication.
Terminal provides wireless broadband internet by network module 502 for user and accesses, and such as user is helped to receive and dispatch electricity
Sub- mail, browsing webpage and access streaming video etc..
Audio output unit 503 can be received by radio frequency unit 501 or network module 502 or in memory 509
The audio data of storage is converted into audio signal and exports to be sound.Moreover, audio output unit 503 can also provide and end
The relevant audio output of specific function (for example, call signal receives sound, message sink sound etc.) that end 500 executes.Sound
Frequency output unit 503 includes loudspeaker, buzzer and receiver etc..
Input unit 504 is for receiving audio or video signal.Input unit 504 may include graphics processor
(Graphics Processing Unit, GPU) 5041 and microphone 5042, graphics processor 5041 is in video acquisition mode
Or the image data of the static images or video obtained in image capture mode by image capture terminal (such as camera) carries out
Reason.Treated, and picture frame may be displayed on display unit 506.Through graphics processor 5041, treated that picture frame can be deposited
Storage is sent in memory 509 (or other storage mediums) or via radio frequency unit 501 or network module 502.Mike
Wind 5042 can receive sound, and can be audio data by such acoustic processing.Treated audio data can be
The format output that mobile communication base station can be sent to via radio frequency unit 501 is converted in the case where telephone calling model.
Terminal 500 further includes at least one sensor 505, such as optical sensor, motion sensor and other sensors.
Specifically, optical sensor includes ambient light sensor and proximity sensor, wherein ambient light sensor can be according to ambient light
Light and shade adjusts the brightness of display panel 5061, and proximity sensor can close display panel when terminal 500 is moved in one's ear
5061 and/or backlight.As a kind of motion sensor, accelerometer sensor can detect in all directions (generally three axis) and add
The size of speed can detect that size and the direction of gravity when static, can be used to identify terminal posture (such as horizontal/vertical screen switching,
Dependent game, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap) etc.；Sensor 505 can be with
Including fingerprint sensor, pressure sensor, iris sensor, molecule sensor, gyroscope, barometer, hygrometer, thermometer,
Infrared sensor etc., details are not described herein.
Display unit 506 is for showing information input by user or being supplied to the information of user.Display unit 506 can wrap
Display panel 5061 is included, liquid crystal display (Liquid Crystal Display, LCD), Organic Light Emitting Diode can be used
Forms such as (Organic Light-Emitting Diode, OLED) configure display panel 5061.
User input unit 507 can be used for receiving the number or character information of input, and generates and set with the user of terminal
It sets and the related key signals of function control inputs.Specifically, user input unit 507 include touch panel 5071 and other
Input equipment 5072.Touch panel 5071, also referred to as touch screen, collect user on it or nearby touch operation (such as
User is using any suitable objects or attachment such as finger, stylus on touch panel 5071 or near touch panel 5071
Operation).Touch panel 5071 may include two parts of touch detection terminal and touch controller.Wherein, touch detection terminal is examined
The touch orientation of user is surveyed, and detects touch operation bring signal, transmits a signal to touch controller；Touch controller from
Touch information is received in touch detection terminal, and is converted into contact coordinate, then gives processor 510, receives processor 510
The order sent simultaneously is executed.Furthermore, it is possible to using multiple types such as resistance-type, condenser type, infrared ray and surface acoustic waves
Realize touch panel 5071.In addition to touch panel 5071, user input unit 507 can also include other input equipments 5072.
Specifically, other input equipments 5072 can include but is not limited to physical keyboard, function key (such as volume control button, switch
Key etc.), trace ball, mouse, operating stick, details are not described herein.
Further, touch panel 5071 can be covered on display panel 5061, when touch panel 5071 is detected at it
On or near touch operation after, send processor 510 to determine the type of touch event, be followed by subsequent processing device 510 according to touching
The type for touching event provides corresponding visual output on display panel 5061.Although in Fig. 5, touch panel 5071 and display
Panel 5061 is the function that outputs and inputs of realizing terminal as two independent components, but in certain embodiments, it can
The function that outputs and inputs of terminal is realized so that touch panel 5071 and display panel 5061 is integrated, is not limited herein specifically
It is fixed.
Interface unit 508 is the interface that exterior terminal is connect with terminal 500.For example, exterior terminal may include it is wired or
Wireless head-band earphone port, external power supply (or battery charger) port, wired or wireless data port, memory card port,
For connecting port, the port audio input/output (I/O), video i/o port, ear port of the terminal with identification module
Etc..Interface unit 508 can be used for receiving the input (for example, data information, electric power etc.) from exterior terminal and will
One or more elements that the input received is transferred in terminal 500 or can be used for terminal 500 and exterior terminal it
Between transmit data.
Memory 509 can be used for storing software program and various data.Memory 509 can mainly include storing program area
The storage data area and, wherein storing program area can (such as the sound of application program needed for storage program area, at least one function
Sound playing function, image player function etc.) etc.；Storage data area can store according to mobile phone use created data (such as
Audio data, phone directory etc.) etc..In addition, memory 509 may include high-speed random access memory, it can also include non-easy
The property lost memory, a for example, at least disk memory, flush memory device or other volatile solid-state parts.
Processor 510 is the control centre of terminal, using the various pieces of various interfaces and the entire terminal of connection, is led to
It crosses operation or executes the software program and/or module being stored in memory 509, and call and be stored in memory 509
Data execute the various functions and processing data of terminal, to carry out integral monitoring to terminal.Processor 510 may include one
Or multiple processing units；Preferably, processor 510 can integrate application processor and modem processor, wherein application processing
The main processing operation system of device, user interface and application program etc., modem processor mainly handles wireless communication.It can manage
Solution, above-mentioned modem processor can not also be integrated into processor 510.
Terminal 500 can also include the power supply 511 (such as battery) powered to all parts, it is preferred that power supply 511 can be with
It is logically contiguous by power-supply management system and processor 510, thus by power-supply management system realize management charging, electric discharge, with
And the functions such as power managed.
In addition, terminal 500 includes some unshowned functional modules, details are not described herein.
Preferably, the embodiment of the present invention also provides a kind of terminal, including processor 510, and memory 509 is stored in storage
It is real when which is executed by processor 510 on device 509 and the computer program that can be run on the processor 510
Each process of existing above-mentioned point cloud classifications embodiment of the method, and identical technical effect can be reached, to avoid repeating, here no longer
It repeats.
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program, the computer program realize each process of above-mentioned point cloud classifications embodiment of the method, and energy when being executed by processor
Reach identical technical effect, to avoid repeating, which is not described herein again.Wherein, the computer readable storage medium, such as only
Read memory (Read-Only Memory, abbreviation ROM), random access memory (Random Access Memory, abbreviation
RAM), magnetic or disk etc..
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the terminal that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or terminal institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or terminal.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal (can be mobile phone, computer, service
Device, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form belongs within protection of the invention.
Claims (16)
1. a kind of point cloud classifications method characterized by comprising
The disaggregated model learnt using the point cloud data by known type carries out at classification point cloud data to be sorted
Reason obtains classification results；
Determining that the point cloud data to be sorted includes that the disaggregated model can not Accurate classification according to the classification results
In the case where first point cloud data, the first kind of first point cloud data is obtained；
Using first point cloud data and the first kind, the second point cloud data and Second Type, the classification mould is updated
Type, wherein second point cloud data is the point cloud data that the disaggregated model is capable of Accurate classification, and the Second Type is institute
State type belonging to the second point cloud data；
Using the updated disaggregated model, classification processing is carried out to the point cloud data to be sorted, obtains classification results.
2. point cloud classifications method according to claim 1, which is characterized in that described to divide point cloud data to be sorted
Before the step of class processing, the method also includes:
Receive the point cloud data of known type, wherein the point cloud data of the known type includes at least two data subsets, often
The corresponding type of point in a data subset；
The feature for learning the point at least two data subset, obtains the disaggregated model.
3. according to the method described in claim 2, it is characterized in that, point in the study at least two data subset
Feature, the step of obtaining the disaggregated model, comprising:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, by instruction
Practice, obtains the disaggregated model.
4. according to the method in any one of claims 1 to 3, which is characterized in that described to utilize first point cloud data
With the first kind, the second point cloud data and Second Type, the step of updating the disaggregated model, comprising:
By in first point cloud data point and the corresponding first kind and second point cloud data and Second Type it is equal
It is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
5. according to the method described in claim 4, it is characterized in that, the point by first point cloud data and corresponding
The first kind and second point cloud data and Second Type are input to the depth nerve net for being loaded with the disaggregated model
Network, by training, the step of updating the disaggregated model, comprising:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, each data
The corresponding type of point in collection；
By the point and second point cloud data and the second class at least one data subset in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
6. the method according to claim 1, wherein the point cloud data of the known type and/or it is described to point
The point cloud data of class be vehicle when driving, the laser radar acquired image data on the vehicle.
7. the method according to claim 1, wherein after the step of acquisition classification results, the method
Further include:
Determining that the point cloud data to be sorted includes that the disaggregated model can not Accurate classification according to the classification results
In the case where first point cloud data, prompt information is exported, wherein the prompt information is for prompting the point cloud number to be sorted
According to comprising the disaggregated model can not Accurate classification the first point cloud data.
8. a kind of terminal characterized by comprising
First categorization module, the disaggregated model for being learnt using the point cloud data by known type, to be sorted
Point cloud data carries out classification processing, obtains classification results；
Obtain module, for according to the classification results determine the point cloud data to be sorted include the disaggregated model without
In the case where first point cloud data of method Accurate classification, the first kind of first point cloud data is obtained；
Update module, for utilizing first point cloud data and the first kind, the second point cloud data and Second Type, more
The new disaggregated model, wherein second point cloud data is the point cloud data that the disaggregated model is capable of Accurate classification, described
Second Type is type belonging to second point cloud data；
Second categorization module classifies to the point cloud data to be sorted for utilizing the updated disaggregated model
Processing obtains classification results.
9. terminal according to claim 8, which is characterized in that the terminal further include:
Receiving module, for receiving the point cloud data of known type, wherein the point cloud data of the known type includes at least two
A data subset, the corresponding type of point in each data subset；
Study module obtains the disaggregated model for learning the feature of the point at least two data subset.
10. terminal according to claim 9, which is characterized in that the study module is used for:
At least two data subset and corresponding type label are inputted into pre-designed deep neural network, by instruction
Practice, obtains the disaggregated model.
11. the terminal according to any one of claim 8 to 10, which is characterized in that the update module is used for:
By in first point cloud data point and the corresponding first kind and second point cloud data and Second Type it is equal
It is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
12. terminal according to claim 11, which is characterized in that the update module is used for:
Receive first point cloud data, wherein first point cloud data includes at least one data subset, each data
The corresponding type of point in collection；
By the point and second point cloud data and the second class at least one data subset in first point cloud data
Type is input to the deep neural network for being loaded with the disaggregated model, by training, updates the disaggregated model.
13. terminal according to claim 8, which is characterized in that the point cloud data of the known type and/or it is described to point
The point cloud data of class be vehicle when driving, the laser radar acquired image data on the vehicle.
14. terminal according to claim 8, which is characterized in that the terminal further include:
Output module, for according to the classification results determine the point cloud data to be sorted include the disaggregated model without
In the case where first point cloud data of method Accurate classification, export prompt information, wherein the prompt information for prompt it is described to
The point cloud data of classification include the disaggregated model can not Accurate classification the first point cloud data.
15. a kind of terminal, which is characterized in that including memory, processor and be stored on the memory and can be at the place
The computer program run on reason device, the processor are realized when executing the computer program as any in claim 1 to 7
Point cloud classifications method described in.
16. a kind of vehicle, which is characterized in that including the terminal as described in any one of claim 8 to 14, or as right is wanted
Terminal described in asking 15.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811107718.5A CN109492658A (en) | 2018-09-21 | 2018-09-21 | A kind of point cloud classifications method and terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811107718.5A CN109492658A (en) | 2018-09-21 | 2018-09-21 | A kind of point cloud classifications method and terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109492658A true CN109492658A (en) | 2019-03-19 |
Family
ID=65689315
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811107718.5A Pending CN109492658A (en) | 2018-09-21 | 2018-09-21 | A kind of point cloud classifications method and terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109492658A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097077A (en) * | 2019-03-26 | 2019-08-06 | 深圳市速腾聚创科技有限公司 | Point cloud data classification method, device, computer equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184250A (en) * | 2015-08-31 | 2015-12-23 | 哈尔滨工业大学 | Ground object classification method based on electric power corridor airborne LiDAR point cloud data |
| CN105701498A (en) * | 2015-12-31 | 2016-06-22 | 腾讯科技（深圳）有限公司 | User classification method and server |
| 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 |
| CN107506799A (en) * | 2017-09-01 | 2017-12-22 | 北京大学 | A kind of opener classification based on deep neural network is excavated and extended method and device |
| EP3349146A1 (en) * | 2017-01-17 | 2018-07-18 | Delphi Technologies LLC | Ground classifier system for automated vehicles |
-
2018
- 2018-09-21 CN CN201811107718.5A patent/CN109492658A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184250A (en) * | 2015-08-31 | 2015-12-23 | 哈尔滨工业大学 | Ground object classification method based on electric power corridor airborne LiDAR point cloud data |
| CN105701498A (en) * | 2015-12-31 | 2016-06-22 | 腾讯科技（深圳）有限公司 | User classification method and server |
| 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 |
| EP3349146A1 (en) * | 2017-01-17 | 2018-07-18 | Delphi Technologies LLC | Ground classifier system for automated vehicles |
| CN107506799A (en) * | 2017-09-01 | 2017-12-22 | 北京大学 | A kind of opener classification based on deep neural network is excavated and extended method and device |
Non-Patent Citations (1)
| Title |
|---|
| 梁宇乾 等: ""基于LIDAR点云回光强度信息的KNN算法与BP神经网络算法分类对比研究"", 《软件导刊》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097077A (en) * | 2019-03-26 | 2019-08-06 | 深圳市速腾聚创科技有限公司 | Point cloud data classification method, device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111260665B (en) | Image segmentation model training method and device | |
| CN108090855A (en) | Method and mobile terminal are recommended in a kind of study plan | |
| CN107734179A (en) | A kind of message prompt method, mobile terminal | |
| CN107809526A (en) | End application sorting technique, mobile terminal and computer-readable recording medium | |
| CN109871246A (en) | A kind of message display method and terminal device | |
| CN107742072A (en) | Face identification method and mobile terminal | |
| CN108427873A (en) | A kind of biological feather recognition method and mobile terminal | |
| CN108229574A (en) | A kind of method, apparatus and mobile terminal of picture screening | |
| CN109269521A (en) | A kind of air navigation aid and terminal device | |
| CN107273035A (en) | Application program recommends method and mobile terminal | |
| CN107729540A (en) | Method, equipment and the computer of photo classification can storage mediums | |
| CN110049185A (en) | Image processing method and terminal device | |
| CN108541015A (en) | A kind of signal strength reminding method and mobile terminal | |
| CN107704182A (en) | A kind of barcode scanning method and mobile terminal | |
| CN107704812A (en) | A kind of face identification method and mobile terminal | |
| CN107734170A (en) | A kind of notification message processing method, mobile terminal and wearable device | |
| CN109947650A (en) | Script step process methods, devices and systems | |
| CN107730460A (en) | A kind of image processing method and mobile terminal | |
| CN107317918A (en) | Parameter setting method and related product | |
| CN107908770A (en) | A kind of photo searching method and mobile terminal | |
| CN107920384A (en) | A kind of method, terminal and computer-readable recording medium for searching for communication network | |
| CN109726303A (en) | A kind of image recommendation method and terminal | |
| CN107608613A (en) | A kind of method and terminal for preventing maloperation | |
| CN104915625B (en) | A kind of method and device of recognition of face | |
| CN109993234B (en) | Unmanned driving training data classification method and device and electronic 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190319 |
|
| RJ01 | Rejection of invention patent application after publication |