CN108268740A - A kind of dummy acquisition method and apparatus of automatic Pilot training big data - Google Patents
A kind of dummy acquisition method and apparatus of automatic Pilot training big data Download PDFInfo
- Publication number
- CN108268740A CN108268740A CN201810122666.2A CN201810122666A CN108268740A CN 108268740 A CN108268740 A CN 108268740A CN 201810122666 A CN201810122666 A CN 201810122666A CN 108268740 A CN108268740 A CN 108268740A
- Authority
- CN
- China
- Prior art keywords
- gene
- target
- environment
- former generation
- environment target
- 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
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The embodiment of the invention discloses a kind of dummy acquisition method and apparatus of automatic Pilot training big data.This method includes：Determine the gene code of each candidate driving environmental goals of automatic virtual driving；Automatic Pilot assessment is carried out to each candidate driving environmental goals, and former generation's environment target is selected from each candidate driving environmental goals according to assessment result；For each group of former generation's environment target, gene crossover operation is carried out to the gene code of this group of former generation's environment target and/or genetic mutation operates, generates offspring's environment target of this group of former generation's environment target；If being unsatisfactory for end condition, using each offspring's environment target of generation as new candidate driving environmental goals, and return and perform automatic Pilot evaluation operation, and new former generation's environment target is selected from new candidate driving environmental goals according to assessment result.The embodiment of the present invention can reduce the difficulty of automatic Pilot big data acquisition, reduce acquisition cost and increase operability.
Description
Technical field
The present embodiments relate to automatic Pilot technical fields more particularly to a kind of automatic Pilot to train the virtual of big data
Acquisition method and device.
Background technology
With the promotion that quality of life is pursued in the development and people of science and technology, it is desirable to be obtained from the driving-activity of fatigue
To liberation, automatic Pilot technology is come into being.Automatic Pilot, it is also known as unmanned, it is that one kind realizes nobody by intelligence system
The technology of driving, and safe and reliable road driving ability is needed during automatic Pilot, it just must be to the information of road environment
There is comprehensive understanding.
Existing automatic Pilot technology, generally use multi-sensor technology obtains environmental information, such as is adopted using camera
Collect the image information of road environment and the three-dimensional information of laser radar acquisition road etc..In order to obtain under various complex environments
Mass data is general using the method that environmental data is acquired in real road, generally use vehicle loading multiple sensors row
Sail the gathered data under real roads environment.Ensure the diversity of data by acquiring enough road informations.
It is existing this usually influenced by environmental conditions for the road environment big data acquisition method of automatic Pilot, very
Difficulty is collected than more comprehensive data.For example, the environmental data of snowy day, rainy day and hail day, the ring of different illumination conditions
Border data etc..In the case of these, since uncontrollable natural environmental condition influences, it is difficult to obtain sufficiently comprehensive environment number
According to.It is acquired even for the data under common natural environment, in order to ensure the diversity of data, it usually needs when spending a large amount of
Between and financial cost acquisition data on the spot.In addition, the acquisition of real road data also has certain danger.Such as automatic Pilot
On the spot in data acquisition, it may occur however that some traffic accidents.Meanwhile data set should include the emergency of some bursts
Data.For example, the pedestrian accident of burst and the vehicle rear-end collision accident of burst etc..The data of these accidents, in actual environment
It can not be obtained in data acquisition.
Invention content
An embodiment of the present invention provides a kind of dummy acquisition method and apparatus of automatic Pilot training big data, can be simpler
It is single advantageously to realize the acquisition of more fully automatic Pilot big data, and reduce the difficulty of acquisition, reduce acquisition cost and increasing
Add operability.
In a first aspect, an embodiment of the present invention provides a kind of dummy acquisition method of automatic Pilot training big data, including：
Determine the gene code of each candidate driving environmental goals of automatic virtual driving；
Carry out automatic Pilot assessment to each candidate driving environmental goals, and according to assessment result from each candidate row
Former generation's environment target is selected in vehicle environmental goals；
For each group of former generation's environment target, gene friendship is carried out to the gene code of this group of former generation's environment target
Fork operation and/or genetic mutation operation generate offspring's environment target of this group of former generation's environment target；
If being unsatisfactory for end condition, using each offspring's environment target of generation as new candidate environment
Target, and return and perform automatic Pilot evaluation operation, and selected newly from new candidate driving environmental goals according to assessment result
Former generation's environment target.
Second aspect, the embodiment of the present invention additionally provide a kind of dummy acquisition device of automatic Pilot training big data, should
Device includes：
Gene code module, for determining the gene code of each candidate driving environmental goals of automatic virtual driving；
Evaluation module is driven, for carrying out automatic Pilot assessment, and according to assessment to each candidate driving environmental goals
As a result former generation's environment target is selected from each candidate driving environmental goals；
Evolution module, for being directed to each group of former generation's environment target, to the gene of this group of former generation's environment target
Coding carries out gene crossover operation and/or genetic mutation operation, generates offspring's environment of this group of former generation's environment target
Target；
Evaluation module is returned, if for being unsatisfactory for end condition, each offspring's environment target of generation is made
It for new candidate driving environmental goals, and returns and performs automatic Pilot evaluation operation, and according to assessment result from new candidate row
New former generation's environment target is selected in vehicle environmental goals.
Target in environment of the embodiment of the present invention by simulating automatic Pilot, and gene volume is carried out to the target
Code, automatic Pilot assessment and gene intersect and/or genetic mutation generates virtual automatic driving environment.Since the present invention is implemented
The technical solution that example provides can call the environment of the automatic Pilot of above-mentioned generation by virtual engine, generate it is different from
It is dynamic to drive virtual big data, the acquisition of more fully automatic Pilot big data can be realized more simple and easyly, and reduce acquisition
Difficulty, reduce acquisition cost and increase operability.
Description of the drawings
Fig. 1 is the flow chart of the dummy acquisition method of the automatic Pilot training big data in the embodiment of the present invention one；
Fig. 2 is the gene code schematic diagram of the automatic Pilot environment factor in the embodiment of the present invention one；
Fig. 3 is the distribution map of the target location in the embodiment of the present invention one；
Fig. 4 is the gene code schematic diagram of the automatic Pilot driving target factor in the embodiment of the present invention one；
Fig. 5 is the example schematic diagram of the automatic Pilot environment in the embodiment of the present invention one；
Fig. 6 is the flow chart of the dummy acquisition method of the automatic Pilot training big data in the embodiment of the present invention two；
Fig. 7 is the evolution exemplary plot of the gene code of the automatic Pilot environment factor in the embodiment of the present invention two；
Fig. 8 is the evolution phenotype example of the gene code of the automatic Pilot environment factor in this embodiment of the present invention two
Figure；
Fig. 9 is the evolution exemplary plot of the gene code of the automatic Pilot driving target factor in the embodiment of the present invention two；
Figure 10 is that the evolution phenotype of the gene code of the automatic Pilot driving target factor in this embodiment of the present invention two shows
Illustration；
Figure 11 is the flow chart of the dummy acquisition method of the automatic Pilot training big data in the embodiment of the present invention three；
Figure 12 is the structure diagram of the dummy acquisition device of the automatic Pilot training big data in the embodiment of the present invention four.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limitation of the invention.It also should be noted that in order to just
Part related to the present invention rather than entire infrastructure are illustrated only in description, attached drawing.
Embodiment one
Fig. 1 trains the flow chart of the dummy acquisition method of big data, this reality for the automatic Pilot in the embodiment of the present invention one
The situation that example is applicable to the dummy acquisition of automatic Pilot training big data is applied, this method can train big data by automatic Pilot
Dummy acquisition device perform, this method can specifically include：
Step 110, determine automatic virtual driving each candidate driving environmental goals gene code.
In the present embodiment, the virtual engine of high emulation, such as Unreal, to simulate the driving ring of automatic Pilot may be used
Border.Unreal is writing a Chinese character in simplified form for Unreal Engine (illusory engine), is one kind of game engine, employ instant ray tracing,
The technologies such as high dynamic range lighting and virtual displacement, and can 200,000,000 polygon operations of real-time operation each second, very
Real environment emulator is very high.The very high virtual emulation platform of other emulators can also be used to the row of simulation automatic Pilot
Vehicle environment.
Wherein, it is true can to include weather, road conditions, trees, building and traffic lights etc. for the candidate driving environmental goals
At least one of the environmental factor that environment can be potentially encountered environmental factor and including in environment vehicle and pedestrian extremely
A kind of few target factor.
Specifically, determine the gene code of each candidate driving environmental goals of automatic virtual driving, illustratively, Ke Yixuan
Weather and road conditions both environmental factors progress gene code are selected, as shown in table 1；It can be by both mesh of environment vehicle and pedestrian
Mark factor carries out gene code, as shown in table 2.
The gene code table of 1 weather of table and road conditions
The gene code table of 2 environment vehicle and pedestrian of table
Wherein, as shown in table 1, gene has been carried out to different weather conditions such as half cloudy day, cloudy day and the light rain in weather
Coding can also be added as needed on the gene code of other different weather conditions, such as greasy weather；It can also similarly add not
With the gene code of road conditions, you can be arranged as required to the gene code of different environmental factors, as shown in Fig. 2, Fig. 2 is this
The gene code schematic diagram of automatic Pilot environment factor in inventive embodiments one.
As shown in table 2, in the gene code carried out to environment vehicle and pedestrian, there is provided the type of target, quantity and
Position.Wherein type may have different types between identical target, for example, vehicle can include trolley and lorry
It can include child and adult etc. Deng, pedestrian；Quantity is the number of different target；In the gene code of position, 10 can be set
Corresponding position of the target in virtual environment image, is divided into 1024 positions by position gene, can be that x and y directions distinguish 32
The position grouping of × 32 units, as shown in figure 3, Fig. 3 is the distribution map of the target location in the embodiment of the present invention one, according to figure
3 position distribution can generate different target locations by variation.It, can in addition to vehicle and pedestrian both target factors
To be added as needed on other target factors, such as pet, as shown in figure 4, Fig. 4 is driving automatically in the embodiment of the present invention one
Sail the gene code schematic diagram of driving target factor.
In the present embodiment, by setting the position of target, the foundation of special environment can be realized, such as burst
Pedestrian accident or the rear-end collision of burst etc., if can specifically show as pedestrian with vehicle or the position of vehicle and vehicle in short-term
It is interior that nearer distance is become from larger distance.Similarly, other special situations can also be configured as needed, than existing
Some technologies, significantly reduce cost.
Step 120 carries out each candidate driving environmental goals automatic Pilot assessment, and according to assessment result from each institute
It states in candidate driving environmental goals and selects former generation's environment target.
Wherein, the detailed process of the automatic Pilot assessment can be：By several environment targets (including different rings
Border factor and target factor) corresponding automatic virtual driving environment established by emulation platform, vehicle is assessed according to automatic Pilot
Performance in the automatic virtual driving environment carries out automatic Pilot assessment.Illustratively, if automatic Pilot assessment vehicle exists
It travels good in a kind of driving environment, does not collide or other accidents, then the corresponding automatic Pilot assessment of the driving environment
Value can be higher, for example, 100 points；If automatic Pilot assessment vehicle constantly stops or touches in another driving environment
It hits, then the corresponding automatic Pilot assessed value of the driving environment can be relatively low, for example, 0 point.The automatic Pilot assessed value can be with
Evaluation condition and assessment section are arranged as required to, gives a mark, and analysis statistical formula can be set, obtains to different manifestations
Final automatic Pilot assessed value.
Specifically, can randomly choose several candidate's driving environmental goals carries out automatic Pilot assessment, and according to assessment
As a result the candidate driving environmental goals of preset quantity is selected as former generation's environment target, such as can be according to assessment result
Automatic Pilot assessed value selects be arranged in front 5 candidate driving environmental goals as former generation's environment target.It is described default
Quantity can be configured as needed.
Step 130, for each group of former generation's environment target, to the gene code of this group of former generation's environment target into
Row gene crossover operation and/or genetic mutation operation generate offspring's environment target of this group of former generation's environment target.
Wherein, the gene crossover operation can be the gene code for exchanging two different locations, the genetic mutation behaviour
Make be that the gene code of some position is modified.
Specifically, the former generation's environment target determined in step 120 can be grouped two-by-two, and for each group of base
Because coding can carry out gene crossover operation and/or genetic mutation operation, the corresponding son of this group of former generation's environment target is generated
Generation's environment target.
If step 140 is unsatisfactory for end condition, using each offspring's environment target of generation as new candidate
Environment target, and return and perform automatic Pilot evaluation operation, and according to assessment result from new candidate driving environmental goals
The new former generation's environment target of middle selection.
Wherein, the end condition can be to generate the number of offspring's environment target more than frequency threshold value or right
Difference between the automatic Pilot assessed value of offspring's environment target of generation and assessment desired value is less than difference threshold, described
Frequency threshold value and difference threshold can be configured according to the requirement of different training big datas.
Specifically, if end condition is unsatisfactory for, using each offspring's environment target of generation as new candidate
Environment target, and return and perform step 120, that is, it returns and performs automatic Pilot evaluation operation, and according to assessment result from new
Candidate driving environmental goals in select new former generation's environment target, to the gene code of new former generation's environment target
Gene crossover operation and/or genetic mutation operation are carried out, corresponding new offspring's environment target is generated, is terminated until meeting
Condition.
In the present embodiment, can be that offspring's environment target sets certain constraint item if meeting end condition
Part so that offspring's environment target and true automatic Pilot environment are with uniformity.The constraints can be to drive automatically
Sail that assessment vehicle is different from the position of the target factor and/or the target factor in environment vehicle and pedestrian position difference
Deng.The constraints can be configured according to the requirement of different training big datas, such as constraints can also be two
A pedestrian cannot have identical position, but can have adjacent position etc..
It in the present embodiment, can be in emulation platform if after meeting end condition and the corresponding constraints of setting
Environmental factor and target factor in final offspring's environment target are called in (such as Unreal), generates different automatic drive
Sail training data.Fig. 5 is the example schematic diagram of the automatic Pilot environment in the embodiment of the present invention one, as shown in figure 5, for fine day
Under backroad environment, wherein have pedestrian, street lamp sum number wood.
Target in environment of the present embodiment by simulating automatic Pilot, and to the target carry out gene code,
Automatic Pilot is assessed and gene intersects and/or genetic mutation generates automatic virtual driving environment, and sets corresponding constraint
Condition and different automatic Pilot training datas is generated by emulation platform.Technical solution provided in an embodiment of the present invention can be with
Realize the acquisition of more fully automatic Pilot big data more simple and easyly, and reduce the difficulty of acquisition, reduce acquisition cost with
And increase operability.
Optionally, it determines the gene code of each candidate driving environmental goals of automatic virtual driving, can include：For every
One candidate driving environmental goals, determines the gene expression of each environmental goals factor included in candidate driving environmental goals；It will
The gene expression of each environmental goals factor is combined according to each putting in order for environmental goals factor, obtains the time
Select the gene code of environment target.
Embodiment two
Fig. 6 is the flow chart of the dummy acquisition method of the automatic Pilot training big data in the embodiment of the present invention two.This reality
Example is applied on the basis of above-described embodiment, has advanced optimized the dummy acquisition method of above-mentioned automatic Pilot training big data.Phase
It answers, the method for the present embodiment can specifically include：
Step 210, environmental goals of driving a vehicle for each candidate, determine each environment included in candidate driving environmental goals
The gene expression of target factor.
Specifically, several candidate's driving environmental goals can be randomly choosed, for each candidate driving environmental goals, really
The gene expression of each environmental goals factor included in fixed candidate driving environmental goals, specific gene code such as table 1 and Fig. 2
It is shown.Such as candidate driving environmental goals can select weather and road conditions both environmental factors.
It is step 220, the gene expression of each environmental goals factor is suitable according to the arrangement of each environmental goals factor
Sequence is combined, and obtains the gene code of candidate driving environmental goals.
It specifically, can be by each environmental goals factor in each the candidate's driving environmental goals determined in step 210
Gene expression be combined according to each putting in order for environmental goals factor, obtain the candidate drive a vehicle environmental goals base
Because of coding.Wherein, described put in order can be configured as needed.Illustratively, if a candidate driving environmental goals
Environmental factor for weather and road conditions, weather was half cloudy day, and road conditions are highway, reference table 1, then candidate's environment mesh
Target gene code is 00000.
Step 230 carries out each candidate driving environmental goals automatic Pilot assessment, and according to assessment result from each institute
It states in candidate driving environmental goals and selects former generation's environment target.
Step 240, for each group of former generation's environment target, to the gene code of this group of former generation's environment target into
Row gene crossover operation obtains offspring's environment target of this group of former generation's environment target.
Wherein, the process of the gene code progress gene crossover operation of described pair of this group of former generation's environment target can wrap
It includes：Adjacent at least one position coding gene is selected as crossover location；Exchange in this group of former generation's gene environment target
Gene expression of the one former generation's gene environment target at the crossover location intersects with second former generation's gene environment target described
Gene expression at position.
Specifically, adjacent at least one position coding gene can be selected, as crossover location, illustratively, Fig. 7 is
The evolution exemplary plot of the gene code of automatic Pilot environment factor in the embodiment of the present invention two, as shown in fig. 7, former generation can
To select the 2nd and the 3rd position coding gene as crossover location.And exchange in this group of former generation's gene environment target first
Gene expression of former generation's gene environment target at the crossover location intersects position with second former generation's gene environment target described
The gene expression at place is put, illustratively, as shown in fig. 7, by first former generation's gene environment target (left side) in former generation in the friendship
The gene 00 at the vent place of putting is swapped with gene 11 of the second former generation's gene environment target (right side) at the crossover location,
And genetic mutation operation is carried out to the gene 1 of the last one position of second former generation's gene environment target, that is, it modifies, obtains
Corresponding offspring's gene expression is 01100 and 10010.
It should be noted that for each group of former generation's environment target, it can be by this group of former generation's environment mesh
Target gene code carries out gene crossover operation and obtains offspring's environment target of this group of former generation's environment target.In addition,
Genetic mutation can also be carried out to any former generation's environment target or any offspring's environment target to operate to obtain new son
Generation's environment target.
Step 250, the gene expression changed at an at least position at least one offspring's environment target obtain newly
Offspring's environment target gene code.
Specifically, the gene code for the offspring's environment target obtained in step 240 carries out genetic mutation operation,
The gene expression at an at least position at least one offspring's environment target is changed, new offspring's driving can be obtained
The gene code of environmental goals.Illustratively, referring to Fig. 7, can to the former generation of offspring, the i.e. second generation, first offspring's gene
The gene of first position of environmental goals (left side) is modified, and it is 10010 and 11100 to obtain new offspring's gene expression.
It, can be to this group of former generation's environment target it should be noted that for each group of former generation's environment target
Gene code carries out gene crossover operation or genetic mutation operates to obtain offspring's environment of this group of former generation's environment target
Target can also carry out gene crossover operation to the gene code of this group of former generation's environment target and genetic mutation operates to obtain
Offspring's environment target of this group of former generation's environment target.
Fig. 8 is the evolution phenotype example of the gene code of the automatic Pilot environment factor in the embodiment of the present invention two
Figure is corresponded with the gene code in Fig. 7, it can be seen that the environmental factor of the environment target of this group of former generation is cloudy for half
It highway and the country road of fine day are driven a vehicle by the offspring that gene crossover operation and genetic mutation operate twice
The environmental factor of environmental goals is the urban road in slight snow and the highway of fine day, the environment with the environment target of former generation
Factor is entirely different.
Fig. 9 is the evolution exemplary plot of the gene code of the automatic Pilot driving target factor in the embodiment of the present invention two, such as
Shown in Fig. 9, target factor is vehicle and pedestrian.Figure 10 is the base of the automatic Pilot driving target factor in the embodiment of the present invention two
Because of the evolution phenotype exemplary plot of coding, corresponded with the gene code in Fig. 9, the target of the environment target of this group of former generation
The gene code of factor operates to obtain different offspring's environment targets by gene crossover operation and genetic mutation twice
Target factor.
If step 260 is unsatisfactory for end condition, using each offspring's environment target of generation as new candidate
Environment target, and return and perform automatic Pilot evaluation operation, and according to assessment result from new candidate driving environmental goals
The new former generation's environment target of middle selection.
Specifically, if end condition is unsatisfactory for, using each offspring's environment target of generation as new candidate
Environment target, and return and perform step 230, that is, it returns and performs automatic Pilot evaluation operation, and according to assessment result from new
Candidate driving environmental goals in select new former generation's environment target, to the gene code of new former generation's environment target
Gene crossover operation and/or genetic mutation operation are carried out, corresponding new offspring's environment target is generated, is terminated until meeting
Condition.
In the present embodiment, can be that offspring's environment target sets certain constraint item if meeting end condition
Part so that offspring's environment target and true automatic Pilot environment are with uniformity.
It in the present embodiment, can be in emulation platform if after meeting end condition and the corresponding constraints of setting
Environmental factor and target factor in final offspring's environment target are called in (such as Unreal), generates different automatic drive
Sail training data.
Target in environment of the present embodiment by simulating automatic Pilot, and to the target carry out gene code,
Automatic Pilot is assessed and gene intersects and/or genetic mutation generates automatic virtual driving environment, and sets corresponding constraint
Condition and different automatic Pilot training datas is generated by emulation platform.Technical solution provided in this embodiment can be simpler
It is single advantageously to realize the acquisition of more fully automatic Pilot big data, and reduce the difficulty of acquisition, reduce acquisition cost and increasing
Add operability.
Embodiment three
The present embodiment can provide a kind of example based on above-described embodiment, to the virtual of automatic Pilot training big data
The evolutionary process of acquisition method middle rolling car environment is further detailed.
In the present embodiment, the thought of evolution algorithm may be used, evolution algorithm is a kind of simulation the Nature biological evolution
Algorithm, process includes outstanding individual is selected to generate offspring from former generation by the method intersected and made a variation as former generation,
Illustratively, evolution algorithm may comprise steps of：1) one group of initial solution is given；2) performance of this current group solution of evaluation；3)
Basis of a certain number of solutions as the solution after iteration is selected from current this group of solution；4) it operates on it again, obtains iteration
Solution afterwards；5) stop if these solutions are met the requirements, otherwise re-operated the solution that these iteration obtain as current solution.
Figure 11 is the flow chart of the dummy acquisition method of the automatic Pilot training big data in the embodiment of the present invention three, is such as schemed
Shown in 11, the technical solution of the present embodiment can be a kind of evolutionary process, corresponding with the process of evolution algorithm, can specifically wrap
It includes：
Step 310 starts.
Step 320, road environment initialization, randomly generate several former generation, quantity can be configured as needed.
Step 330, automatic Pilot assessment, assess the former generation generated in step 320, obtain the property of former generation's individual
Energy.
Step 340, road environment individual choice, the former generation for the preset quantity for selecting performance higher, preset quantity can root
It is configured according to needs.
Step 350, individual intersection carry out gene crossover operation to former generation's individual of selected preset quantity.
Step 360, individual variation carry out genetic mutation behaviour to former generation's individual after the gene crossover operation in step 350
Make, generate new offspring's individual.
Step 370 terminates, and judges whether to meet end condition.End condition could be provided as defined evolution time, rule
The performance of fixed evolutionary generation or offspring's individual meets certain requirement, can be configured as needed.If meet eventually
Only condition, then evolutionary process completion, enters step 380.If conditions are not met, step 330 can be then returned to, circular flow,
Until meeting end condition.
Step 380 terminates.
The present embodiment is by selecting outstanding former generation's gene, by gene swapping and/or genetic mutation, by outstanding former generation
Gene generates outstanding offspring's gene, and by continuous iteration further by the new outstanding offspring of outstanding offspring's gene generation
Gene.Technical solution provided in this embodiment uses the thought of evolution algorithm, and it is virtual can to generate multiple and different automatic Pilots
Environment, and constantly evolve and learn so that the environment of generation is constantly evolved until the requirement for meeting training big data,
So as on the basis of the virtual environment of this automatic Pilot, more fully automatic Pilot big data is realized more simple and easyly
Acquisition, and reduce the difficulty of acquisition, reduce acquisition cost and increase operability.
Example IV
Figure 12 is the structure diagram of the dummy acquisition device of the automatic Pilot training big data in the embodiment of the present invention four,
Described device can include：
Gene code module 410, for determining the gene code of each candidate driving environmental goals of automatic virtual driving；
Evaluation module 420 is driven, for carrying out automatic Pilot assessment, and foundation is commented to each candidate driving environmental goals
Estimate result and select former generation's environment target from each candidate driving environmental goals；
Evolution module 430, for being directed to each group of former generation's environment target, to the base of this group of former generation's environment target
Because coding carries out gene crossover operation and/or genetic mutation operation, offspring's driving ring of this group of former generation's environment target is generated
Border target；
Evaluation module 440 is returned to, if for being unsatisfactory for end condition, by each offspring's environment target of generation
It as new candidate driving environmental goals, and returns and performs automatic Pilot evaluation operation, and according to assessment result from new candidate
New former generation's environment target is selected in environment target.
Further, the evolution module 430 can include gene cross unit, and the gene cross unit can be used
In：
Adjacent at least one position coding gene is selected as crossover location；
Exchange base of the first former generation's gene environment target in this group of former generation's gene environment target at the crossover location
Because of expression and gene expression of the second former generation's gene environment target at the crossover location.
Further, the evolution module 430 can also include gene cross and variation unit, the gene cross and variation list
Member can be used for：
Gene crossover operation is carried out to the gene code of this group of former generation's environment target, obtains this group of former generation's environment
Offspring's environment target of target；
It changes the gene expression at an at least position at least one offspring's environment target and obtains new offspring's row
The gene code of vehicle environmental goals.
Further, the end condition can be generate offspring's environment target number be more than frequency threshold value or
Difference of the person between the automatic Pilot assessed value of offspring's environment target of generation and assessment desired value is less than difference threshold.
Further, the gene code module 410 specifically can be used for：
For each candidate driving environmental goals, each environmental goals factor included in candidate driving environmental goals is determined
Gene expression；
The gene expression of each environmental goals factor is subjected to group according to the putting in order for each environmental goals factor
It closes, obtains the gene code of candidate driving environmental goals.
Further, weather, road conditions, trees, building and traffic lights can be included in the candidate driving environmental goals
At least one of environmental factor and including at least one of environment vehicle and pedestrian target factor.
Further, automatic Pilot assessment vehicle is different from the position of the target factor, environment in the target factor
The position of vehicle and pedestrian is different.
The dummy acquisition device of automatic Pilot training big data that the embodiment of the present invention is provided can perform of the invention arbitrary
The dummy acquisition method of automatic Pilot training big data that embodiment is provided, has the advantages that：It is at low cost, it does not need to
Road data is acquired on the spot using vehicular platform；Operability is high, not by limitations such as natural conditions, safeties, can adopt at any time
Collect the road data of the different natural conditions and different experiments situation etc. under simulated environment；Data are comprehensive, existing to drive automatically
Road data acquisition method is sailed, leads to that the number of enough vehicle acquisition various road conditions from all parts of the world and environment can not possibly be put into
According to, and the present invention can reach any virtual environment for acquiring automatic Pilot road using the method for virtual environment data acquisition
The big data of road traveling.
The dummy acquisition device of automatic Pilot training big data that the embodiment of the present invention is provided can perform of the invention arbitrary
The dummy acquisition method of automatic Pilot training big data that embodiment is provided, has the corresponding function module of execution method and has
Beneficial effect.
Note that it above are only presently preferred embodiments of the present invention and institute's application technology principle.It will be appreciated by those skilled in the art that
The present invention is not limited to specific embodiment described here, can carry out for a person skilled in the art various apparent variations,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above example to the present invention
It is described in further detail, but the present invention is not limited only to above example, without departing from the inventive concept, also
It can include other more equivalent embodiments, and the scope of the present invention is determined by scope of the appended claims.
Claims (14)
- A kind of 1. dummy acquisition method of automatic Pilot training big data, which is characterized in that including：Determine the gene code of each candidate driving environmental goals of automatic virtual driving；Carry out automatic Pilot assessment to each candidate driving environmental goals, and according to assessment result from each candidate driving ring Former generation's environment target is selected in the target of border；For each group of former generation's environment target, gene is carried out to the gene code of this group of former generation's environment target and intersects behaviour Make and/or genetic mutation operates, generate offspring's environment target of this group of former generation's environment target；If being unsatisfactory for end condition, using each offspring's environment target of generation as new candidate environment mesh It marks, and returns and perform automatic Pilot evaluation operation, and selected newly from new candidate driving environmental goals according to assessment result Former generation's environment target.
- 2. according to the method described in claim 1, it is characterized in that, the gene code to this group of former generation's environment target carries out Gene crossover operation operates, including：Adjacent at least one position coding gene is selected as crossover location；Exchange gene table of the first former generation's gene environment target in this group of former generation's gene environment target at the crossover location Up to the gene expression with second former generation's gene environment target at the crossover location.
- 3. according to the method described in claim 1, it is characterized in that, the gene code to this group of former generation's environment target carries out Gene crossover operation and genetic mutation operation, including：Gene crossover operation is carried out to the gene code of this group of former generation's environment target, obtains this group of former generation's environment target Offspring's environment target；It changes the gene expression at an at least position at least one offspring's environment target and obtains new offspring's driving ring The gene code of border target.
- 4. according to the method described in claim 1, it is characterized in that, the end condition is to generate offspring's environment target Number is more than frequency threshold value or between the automatic Pilot assessed value of offspring's environment target and assessment desired value of generation Difference be less than difference threshold.
- 5. according to the method described in claim 1, it is characterized in that, determine each candidate driving environmental goals of automatic virtual driving Gene code, including：For each candidate driving environmental goals, the base of each environmental goals factor included in candidate driving environmental goals is determined Because of expression；The gene expression of each environmental goals factor according to each putting in order for environmental goals factor is combined, is obtained The gene code for environmental goals of driving a vehicle to the candidate.
- 6. according to the method described in claim 5, it is characterized in that, the candidate driving environmental goals include weather, road conditions, At least one of trees, building and traffic lights environmental factor and including at least one of environment vehicle and pedestrian mesh Mark factor.
- 7. the according to the method described in claim 6, it is characterized in that, position of automatic Pilot assessment vehicle and the target factor Difference, the position of environment vehicle and pedestrian is different in the target factor.
- 8. a kind of dummy acquisition device of automatic Pilot training big data, which is characterized in that including：Gene code module, for determining the gene code of each candidate driving environmental goals of automatic virtual driving；Evaluation module is driven, for carrying out automatic Pilot assessment, and according to assessment result to each candidate driving environmental goals Former generation's environment target is selected from each candidate driving environmental goals；Evolution module, for being directed to each group of former generation's environment target, to the gene code of this group of former generation's environment target Gene crossover operation and/or genetic mutation operation are carried out, generates offspring's environment target of this group of former generation's environment target；Evaluation module is returned to, if for being unsatisfactory for end condition, using each offspring's environment target of generation as new Candidate driving environmental goals, and return and perform automatic Pilot evaluation operation, and according to assessment result from new candidate driving ring New former generation's environment target is selected in the target of border.
- 9. device according to claim 8, which is characterized in that the evolution module includes gene cross unit, the base Because cross unit is used for：Adjacent at least one position coding gene is selected as crossover location；Exchange gene table of the first former generation's gene environment target in this group of former generation's gene environment target at the crossover location Up to the gene expression with second former generation's gene environment target at the crossover location.
- 10. device according to claim 8, which is characterized in that the evolution module further includes gene cross and variation unit, The gene cross and variation unit is used for：Gene crossover operation is carried out to the gene code of this group of former generation's environment target, obtains this group of former generation's environment target Offspring's environment target；It changes the gene expression at an at least position at least one offspring's environment target and obtains new offspring's driving ring The gene code of border target.
- 11. device according to claim 8, which is characterized in that the end condition is generates offspring's environment target Number be more than frequency threshold value or to the automatic Pilot assessed value and assessment desired value of offspring's environment target of generation it Between difference be less than difference threshold.
- 12. according to the device described in claim 8, which is characterized in that the gene code module is specifically used for：For each candidate driving environmental goals, the base of each environmental goals factor included in candidate driving environmental goals is determined Because of expression；The gene expression of each environmental goals factor according to each putting in order for environmental goals factor is combined, is obtained The gene code for environmental goals of driving a vehicle to the candidate.
- 13. device according to claim 12, which is characterized in that the candidate driving environmental goals includes weather, road At least one of condition, trees, building and traffic lights environmental factor and including at least one of environment vehicle and pedestrian Target factor.
- 14. device according to claim 13, which is characterized in that automatic Pilot assesses the position of vehicle and the target factor Put difference, the position of environment vehicle and pedestrian is different in the target factor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810122666.2A CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810122666.2A CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108268740A true CN108268740A (en) | 2018-07-10 |
| CN108268740B CN108268740B (en) | 2021-11-16 |
Family
ID=62773863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810122666.2A Active CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108268740B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108959813A (en) * | 2018-07-26 | 2018-12-07 | 北京理工大学 | A kind of emulation modelling method of intelligent vehicle road guide environmental model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080297377A1 (en) * | 2007-05-28 | 2008-12-04 | High Tech Computer Corp. | Keypad structure and electronic device using the same |
| CN103218489A (en) * | 2013-04-11 | 2013-07-24 | 浙江大学 | Method for simulating vehicle personalized driving characteristic based on video sample |
| CN106154834A (en) * | 2016-07-20 | 2016-11-23 | 百度在线网络技术（北京）有限公司 | For the method and apparatus controlling automatic driving vehicle |
-
2018
- 2018-02-07 CN CN201810122666.2A patent/CN108268740B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080297377A1 (en) * | 2007-05-28 | 2008-12-04 | High Tech Computer Corp. | Keypad structure and electronic device using the same |
| CN103218489A (en) * | 2013-04-11 | 2013-07-24 | 浙江大学 | Method for simulating vehicle personalized driving characteristic based on video sample |
| CN106154834A (en) * | 2016-07-20 | 2016-11-23 | 百度在线网络技术（北京）有限公司 | For the method and apparatus controlling automatic driving vehicle |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108959813A (en) * | 2018-07-26 | 2018-12-07 | 北京理工大学 | A kind of emulation modelling method of intelligent vehicle road guide environmental model |
| CN108959813B (en) * | 2018-07-26 | 2021-01-15 | 北京理工大学 | Simulation modeling method for intelligent vehicle road navigation environment model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108268740B (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111061167B (en) | Intelligent network connection demonstration area-oriented mixed reality automatic driving test method and virtual test platform | |
| US7487074B2 (en) | Road traffic simulation apparatus | |
| CN108528458A (en) | System and method for vehicle dimension prediction | |
| Prendinger et al. | Tokyo virtual living lab: Designing smart cities based on the 3d internet | |
| CN111797001A (en) | Method for constructing automatic driving simulation test model based on SCANeR | |
| US8894413B2 (en) | Architecture, method, and program for generating realistic vehicular mobility patterns | |
| CN101251958A (en) | Method for implementing automobile driving analog machine facing to disciplinarian | |
| CN110716529A (en) | Automatic generation method and device for automatic driving test case | |
| CN101567017B (en) | Urban-evacuation simulation method based on multi-resolution images | |
| CN103646557B (en) | Based on the three-dimensional microform intelligent-traffic management system of Internet of Things | |
| CN115048972B (en) | Traffic scene deconstruction classification method and virtual-real combined automatic driving test method | |
| CN109143852A (en) | Intelligent driving vehicle environmental adaptively imports method under urban environment | |
| CN107590766A (en) | A kind of method of discrimination of the land used combination form related to road traffic accident risk | |
| CN112053433A (en) | Construction method of civil aviation airport dynamic scene and airport special vehicle training method | |
| CN103198182B (en) | The pedestrian guide sign design method of view-based access control model perception simulation technology | |
| Krueger et al. | SILAB-A task-oriented driving simulation | |
| CN108268740A (en) | A kind of dummy acquisition method and apparatus of automatic Pilot training big data | |
| CN117313931B (en) | Subway platform passenger evacuation time prediction method based on fire scene | |
| CN110647053A (en) | Automatic driving simulation method and system | |
| CN114398253A (en) | Method and system for generating test scene of automatic driving real vehicle | |
| CN108898284B (en) | Internet of vehicles management control strategy evaluation method and system | |
| CN102693329A (en) | Urban dispersion simulation method based on multiresolution chart | |
| CN111841012B (en) | Automatic driving simulation system and test resource library construction method thereof | |
| CN116127786B (en) | System and method for measuring and calculating security state of slow traffic group based on city slow traffic simulation | |
| Lebiedź et al. | Real terrain visualisation on the basis of GIS data |
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 |