CN109543572A - A kind of traveling pavement condition evaluation method - Google Patents
A kind of traveling pavement condition evaluation method Download PDFInfo
- Publication number
- CN109543572A CN109543572A CN201811320148.8A CN201811320148A CN109543572A CN 109543572 A CN109543572 A CN 109543572A CN 201811320148 A CN201811320148 A CN 201811320148A CN 109543572 A CN109543572 A CN 109543572A
- Authority
- CN
- China
- Prior art keywords
- bat
- algorithm
- image
- initial
- traveling
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The application belongs to technical field of image processing, more particularly to a kind of traveling pavement condition evaluation method.If BP neural network initial weight and threshold value selection are inappropriate, it will so that BP neural network convergence is unsuccessful, prediction effect is poor.The application provides a kind of traveling pavement condition evaluation method, includes the following steps: to obtain traveling road conditions image；Road image pretreatment；Extract characteristics of image；Initial reverse Propagation Neural Network structure is established, and sets the network-related parameters and function, the characteristics of image based on extraction generates initial parameter using coding mode, while obtaining best initial weights and threshold value using bat algorithm；The best initial weights and threshold optimization reverse transmittance nerve network obtained by bat algorithm-back-propagation algorithm, and the optimal reverse Propagation Neural Network structure for meeting performance requirement is obtained by training；It establishes based on bat algorithm-reverse transmittance nerve network road condition evaluation model, traveling road conditions is evaluated.Unstructured road road conditions are predicted.
Description
Technical field
The application belongs to technical field of image processing, more particularly to a kind of traveling pavement condition evaluation method.
Background technique
Environment sensing has become the new trend of domestic and international intelligent vehicle development, passes through vehicle-mounted sensor-based system and information system
The intelligent information exchange for realizing vehicle, road and driver etc. makes vehicle have certain intelligent environment sensing capability, identifies path,
Present road situation is analyzed, in conjunction with vehicle location, barrier is detected, improves the safety of traveling.As intelligent carriage is being lived
In it is universal and in some special occasions such as anti-terrorism, inspection using more and more.
Currently, being largely to be directed to typical structured road for the information extraction of carriage way situation, in face of small
The unstructured road of type, such as city lane, staggered small streets etc. it is seldom, and the influence about road conditions to driving had no
More researchs.The evaluation of carriage way road conditions is largely to be directed to highway service performance simultaneously, as pavement strength index, antiskid refer to
The analysis of mark etc., there is no too many researchs for the influence about road conditions to driving.
BP neural network is one of current most widely used neural network model, but that there are convergence rates is slow for BP algorithm；
Local minimization problem.If BP neural network initial weight and threshold value selection are inappropriate, it will so that BP neural network
Restrain unsuccessful, and then prediction effect is very poor.
Summary of the invention
1. technical problems to be solved
Based on currently, being largely to be directed to typical structured road, face for the information extraction of carriage way situation
To small-sized unstructured road, such as city lane, staggered small streets etc. it is seldom, and the influence about road conditions to driving is simultaneously
Without excessive research.BP neural network is one of current most widely used neural network model, but there are convergence rates for BP algorithm
Slowly；Local minimization problem.If BP neural network initial weight and threshold value selection are inappropriate, it will so that BP nerve net
Network convergence is unsuccessful, and then the problem that prediction effect is very poor, and this application provides a kind of traveling pavement condition evaluation methods.
2. technical solution
To achieve the above object, this application provides a kind of traveling pavement condition evaluation method, the method includes as follows
Step:
Step 1): traveling road conditions image is obtained；
Step 2): road image pretreatment；
Step 3): characteristics of image is extracted；
Step 4): establishing initial reverse Propagation Neural Network structure, and set the network-related parameters and function, based on mentioning
The characteristics of image taken generates initial parameter using coding mode, while obtaining best initial weights and threshold value using bat algorithm；
Step 5): the best initial weights and threshold optimization Back propagation neural obtained by bat algorithm-back-propagation algorithm
Network, and the optimal reverse Propagation Neural Network structure for meeting performance requirement is obtained by training；
Step 6): it establishes based on bat algorithm-reverse transmittance nerve network evaluation model, traveling road conditions is commented
Valence.
Optionally, the information of doubtful barrier has following expression in road conditions image in the step 1):
Wherein, the initial height/width value of the objects in front of acquisition is respectively h0And w0, the distance of car body and doubtful barrier is initial
Value is d0, dynamic barrier coefficient of determination is γ；
If acquiring, doubtful barrier object profile size in entire image is constant in area-of-interest in image or gets over
Carry out smaller, then d >=d0, i.e. movable body and doubtful obstacle distance remain unchanged or original remoter, and remove barriers object suspicion, γ
=0；Conversely, profile becomes larger, d < d0It is determined as dynamic barrier, at this time γ=1.
Optionally, characteristics of image includes surface evenness, road surface curvature, barrier the ratio of width to height, barrier in the step 2)
Hinder object effective area ratio and obstacle coefficient.
Optionally, the reverse transmittance nerve network is made of 3 node layers: input node n, hidden layer node p are a, defeated
Egress q.
Optionally, Dynamic gene is set in the bat algorithm-back-propagation algorithmWith
Wherein,For the spectrum of input variable.
Optionally, the bat algorithm-back-propagation algorithm are as follows:
By the parameter of reverse transmittance nerve network, parameter is encoded as a whole；N bat is generated at random, is formed
Initial bat population, the position x of random initializtion bat in search spacei, speed vi, loudness AiWith pulse frequency ri；It will
The position of every bat individual is connected with fitness f (x)；Judge whether algorithm reaches maximum number of iterations, if meeting item
Part, then algorithm terminates, and exports optimal solution, otherwise, carries out in next step；According to flying speed and location update formula fi=fmin+
(fmax-fmin)β、WithUpdate speed and the position of bat；Generate a random number
Rand1, if rand1 > ri, riFor the pulse frequency of i-th bat, formula x is utilizednew(i)=xold+εAtObtain xnew；To position
Quality assessed；A random number rand2 is generated, if rand2 < Ai, wherein AiFor the pulse loudness of i-th bat.
There is f (x simultaneouslyi) < f (x*) bat individual update is then carried out, pass through formula At+1(i)=α At(i) reduce Ai, pass through formula rt+1(i)=
r0(i) × [1-exp (- γ × t)] increase ri；If meeting termination condition, exports target value and terminator executes, otherwise
Steering connects the position of every bat individual with fitness f (x)；The individual decoding that bat algorithm optimization is obtained, knot
Pre-selection initial connection weight and threshold value of the fruit as reverse transmittance nerve network.According to resulting preset value, Dynamic gene is calculatedWithIf being unsatisfactory forOrIt then turns to and generates n bat at random, form initial bat population, searching
The position x of random initializtion bat in rope spacei, speed vi, loudness AiWith pulse frequency riRe-execute the program；If simultaneously
MeetAndThen preset value is the initial connection weight and threshold value for being used as reverse transmittance nerve network；
Wherein, β ∈ [0,1] is that an obedience is uniformly distributed random vector, xoldIt is randomly selected in current optimal solution set
One solution, AtFor the mean loudness of all bats under current iteration number, ε is that a D ties up random vector, ε ∈ [- 1,1], α
For the attenuation coefficient of pulse loudness, α is constant, and 0 < α < 1.
Optionally, the fitness function is as follows:
Wherein number of the l for institute's training sample, TimWithM training sample is respectively indicated determined by i-th bat
Weight and threshold value lower network reality output and desired output.
Optionally, the evaluation model of the bat algorithm-reverse transmittance nerve network are as follows:
Wherein, connection weight is { Wji, i=1,2 ..., n, j=1,2 ..., p；Middle layer to output layer connection weight
{Vtj, j=1,2 ..., p, t=1,2 ..., q；It is { θ that middle layer each unit, which exports threshold value,j, j=1,2 ..., p；aiFor input
Variable；γ is the increase coefficient of pulse frequency, and γ is constant, and γ > 0；The transmission function of network input layer to hidden layer is
Tan-sigmoid:The transmission function of hidden layer to output layer uses logarithm S type transmission function: Log-
Sigmoid type function
3. beneficial effect
Compared with prior art, a kind of beneficial effect travelling pavement condition evaluation method provided by the present application is:
Traveling pavement condition evaluation method provided by the present application makes up backpropagation using bat algorithm by extracting characteristics of image
The randomness defect being connected to the network on weight and threshold value selection, while the prominent barrier of Dynamic gene is added for traveling road conditions
It is main to influence, to achieve the purpose that mutual supplement with each other's advantages, obtain based on bat algorithm-back-propagation algorithm traveling evaluating pavement condition mould
Type, and then small-sized unstructured road road conditions are effectively predicted.
Detailed description of the invention
Fig. 1 is a kind of bat algorithm-back-propagation algorithm flow diagram of traveling pavement condition evaluation method of the application；
Fig. 2 is bat algorithm-counterpropagation network road condition evaluation model of the application.
Specific embodiment
Hereinafter, specific embodiment of the reference attached drawing to the application is described in detail, it is detailed according to these
Description, one of ordinary skill in the art can implement the application it can be clearly understood that the application.Without prejudice to the application principle
In the case where, the feature in each different embodiment can be combined to obtain new embodiment, or be substituted certain
Certain features in embodiment, obtain other preferred embodiments.
Back-propagation algorithm (English: Backpropagation algorithm, referred to as: BP algorithm) it is a kind of supervised learning
Algorithm is often used to train multi-layer perception (MLP).In 1974, Paul Werbos gave how to train general networking for the first time
Learning algorithm, and artificial neural network is special case therein.Unfortunately, in entire artificial neural network community at that time but without
People knows the learning algorithm that Paul is proposed.
Bat algorithm (Bat Algorithm, BA) algorithm is that bat is hunted using a kind of sonar to detect in simulation nature
Object avoids the random search algorithm of barrier from simulating bat carrying out most basic spy to barrier or prey using ultrasonic wave
Itself and optimization aim function are simultaneously associated by survey, stationkeeping ability.The bat individual that population quantity is by the bionic principle of BA algorithm
Be mapped as D dimension problem space in NP feasible solution, by optimization process and search be modeled to population bat individual moving process with
It quarters a prey using the fitness function value of Solve problems and measures the superiority and inferiority of bat present position, by the survival of the fittest of individual
Journey analogizes to the iterative process that the feasible solution made good use of in optimization and search process substitutes poor feasible solution.In bat searching algorithm
In, in order to simulate bat detection prey, avoid barrier, it need to assume that following three is approximate or Utopian rule:
1) all bats utilize the method perceived distance of echolocation, and they are distinguished using a kind of cleverly mode
Difference between prey and background barrier.
2) bat is in position xiWith speed viRandom flight, with fixed frequency fmin, variable wavelength X and volume A0To search
Rope prey.The pulse wavelength (or frequency) and adjust pulse that bat emit according to the proximity of itself and target come adjust automatically
Emissivity r belongs to [0,1]
Although 3) there are many variation patterns of volume but in bat algorithm, it is assumed that volume A is from a maximum value A0It is (whole
Number) change to fixed minimum value Amin。
Jiang Hongyan etc. establishes the road condition grade Evaluation model of road multi objective performance parameter by the method for Matter Analysis,
And this new life subject road engineering reality is applied to successfully.
Antti et al. is extracted the structure factor of highway pavement condition, the coefficient of roughness and transversal inhomogeneity coefficient three because
Element calculates condition of road surface using weighted sum.
Referring to Fig. 1~2, the application provides a kind of traveling pavement condition evaluation method, and described method includes following steps:
Step 1): traveling road conditions image is obtained；
Step 2): road image pretreatment；
Step 3): characteristics of image is extracted；
Step 4): establishing initial reverse Propagation Neural Network structure, and set the network-related parameters and function, based on mentioning
The characteristics of image taken generates initial parameter using coding mode, while obtaining best initial weights and threshold value using bat algorithm；
Step 5): the best initial weights and threshold optimization Back propagation neural obtained by bat algorithm-back-propagation algorithm
Network, and the optimal reverse Propagation Neural Network structure for meeting performance requirement is obtained by training；
Step 6): it establishes based on bat algorithm-reverse transmittance nerve network evaluation model, traveling road conditions is commented
Valence.
Optionally, the information of doubtful barrier has following expression in road conditions image in the step 1):
Wherein, the initial height/width value of the objects in front of acquisition is respectively h0And w0, the distance of car body and doubtful barrier is initial
Value is d0, dynamic barrier coefficient of determination is γ；
If acquiring, doubtful barrier object profile size in entire image is constant in area-of-interest in image or gets over
Carry out smaller, then d >=d0, i.e. movable body and doubtful obstacle distance remain unchanged or original remoter, and remove barriers object suspicion, γ
=0；Conversely, profile becomes larger, d < d0It is determined as dynamic barrier, at this time γ=1.
Optionally, characteristics of image includes surface evenness, road surface curvature, barrier the ratio of width to height, barrier in the step 2)
Hinder object effective area ratio and obstacle coefficient.
(1) surface evenness
The present invention gives the analyses of the surface evenness of suitable evaluating pavement condition demand.Road surface is excessively coarse or rugged
The defects of it is many when, it may appear that vehicle is difficult to travel, the relatively low situation of safety；And road surface it is moderately flat and smooth when then vehicle row
Sail very clear and coherent, road conditions are good；When excessively smooth further for road surface, such as ice-patch surface is also unsuitable for vehicle safe driving.
The case where considering smooth and excessive defect excessive there are road surface, is described herein surface evenness and is expressed as follows:
(2) road surface curvature
Correct influence of the evaluation path bend to vehicle driving is to travel the essential condition of evaluating pavement condition, takes two kinds herein
The scheme that method carries out simultaneously identifies road boundary.Acquired image is split, road is carried out using Hough transform
Pure road surface region is obtained with detection, lane boundary line is identified, obtains road area.If due to the various rings such as shade, water stain
Border factor is more sensitive and when being unable to get effective road boundary, the method for taking the both sides of the edge of identification travel is gone
Except noise, smooth edges line.Edge indicates connecting way contour line, the feasibility that ground protrusion and ground intersection are formed herein
Region.
Curvature is to describe the amount [8] of curved degree, and curve M corresponds to arc s, and M corresponds to s+ Δ s,M
→ M ' tangent line corner is | Δ α |.Segmental arcCurvature at point M of average curvature and curve C be respectively as follows:
The bending degree of the bigger curve of inscribe circle is with regard to smaller, and curvature is smaller, and radius of curvature is also bigger.Straight curvature of a curve
It is everywhere 0, therefore straight line is without radius of curvature.Then the road surface curvature mentioned by the present invention can be characterized by the curvature value in knee,
For road curvature section, corresponding bending section obtains r=1/k.
(3) barrier the ratio of width to height
When extracting barrier feature, edge processing is carried out to the doubtful barrier determined in area-of-interest first, is extracted
Shape feature, barrier contour feature are roughly divided into rectangle, ellipse, circle (similar round), polygon etc..Wherein complicated image
By taking circular contour detects as an example, discontinuous edge pixel point is connected into compositing area enclosed edge using Hough transform
Boundary's curve.
During car body moves ahead, high h and width w value of the doubtful barrier locked in area-of-interest in image can be sent out
Changing, but the configuration of an object will not be varied within the of short duration time before and after Image Acquisition, then have fixation at this time
The ratio of width to height:
Wherein, c is a constant.
(4) barrier effective area ratio
Whether object is determined as that dynamic barrier depends on the effective area ratio R of doubtful barrier in area-of-interest,
When effective area ratio R is greater than the threshold value T of setting, then it is assumed that detect doubtful barrier.Threshold value T set by this paper is one fixed
Value, depending on vehicle and the intrinsic parameter of video camera, sees later period specific experiment.
The expression formula of effective area ratio are as follows:
Molecule S in above-mentioned effective area ratio expression formulaoFor in the image in area-of-interest after binary conversion treatment
The valid pixel number of doubtful barrier is the number of pixels that pixel value is 1 in minimum external figure, denominator S in the present inventionrIt is fixed
Justice is the sum of all pixels that pixel value is 1 and 0 in region of interest area image.
(5) obstacle coefficient
Present invention research focuses on that barrier feature influences the degree of road surface running vehicle, therefore can pass through the shape to barrier
The high feature of shape, width is comprehensively considered, and show that the barrier on an influence coefficient of traveling road conditions, is defined herein as obstacle
Coefficient q.To intelligent vehicle is specifically tested in this article, the radius of vehicle is r, verified to know that r/4 is that intelligent vehicle cross protrusion
It is as follows to define q for critical point:
In formula, htAnd wtIndicate barrier actual height and width, q ∈ (0,1).The wide height of calculating barrier used herein
Method are as follows: with reference to the principle of reversing automobile image, the basic principle based on camera model and photography geometry extrapolates this ginseng
Number feature.
Optionally, the reverse transmittance nerve network is made of 3 node layers: input node n, hidden layer node p are a, defeated
Egress q.
If input mode vector Ak=(a1,a2,…,an), it is desirable to output vector Yk=(y1,y2,…,yq)；Middle layer elements
Input vector Sk=(s1,s2,…,sp), output vector Bk=(b1,b2,…,bp)；Output layer unit input vector Lk=(l1,
l2,…,lq), output vector Ck=(c1,c2,…,cq)；Connection weight is { Wji, i=1,2 ..., n, j=1,2 ..., p；It is intermediate
Layer to output layer connection weight { Vtj, j=1,2 ..., p, t=1,2 ..., q；It is { θ that middle layer each unit, which exports threshold value,j, j=
1,2,…,p；It is { γ that output layer each unit, which exports threshold value,t, t=1,2 ..., q.The above k=1,2 ..., m.
The index for influencing evaluating pavement condition result has L, K, c, R and q five, evaluation of each factor for traveling road conditions result
It influences different.Therefore the distribution of initial weight should be deviated, and the difference of prominent each index keeps later period training process more reasonable
Efficiently.The traveling road condition evaluation model for considering simple weighting method herein, substitutes into a certain number of experiment samples, obtains closest to the phase
One group of weight coefficient of prestige value is respectively as follows: wL=0.1156, wK=0.0893, wc=0.1744, wR=0.3915, wq=0.2292.
It is to a certain extent barrier effective area ratio on the maximum factor of traveling evaluating pavement condition result influence that the system, which is scolded clear,
Secondary is obstacle coefficient, influences the smallest to be surface evenness and road surface curvature, also demonstrates that traveling evaluating pavement condition is herein
It is based primarily upon dynamic barrier.
Optionally, Dynamic gene is set in the bat algorithm-back-propagation algorithmWith
Wherein,For the spectrum of input variable.Influence according to the analysis input of spectrum size to output, in order to protrude obstacle
The part initial weight of the effect of object feature, BP neural network also needs to meet Condition, the constraint item
Part is tested to obtain and be confirmed by weighted model, obtains more preferably search space with this.
Optionally, the bat algorithm-back-propagation algorithm are as follows:
By the parameter of reverse transmittance nerve network, parameter is encoded as a whole；N bat is generated at random, is formed
Initial bat population, the position x of random initializtion bat in search spacei, speed vi, loudness AiWith pulse frequency ri；It will
The position of every bat individual is connected with fitness f (x)；Judge whether algorithm reaches maximum number of iterations, if meeting item
Part, then algorithm terminates, and exports optimal solution, otherwise, carries out in next step；According to flying speed and location update formula fi=fmin+
(fmax-fmin)β、WithUpdate speed and the position of bat；Generate a random number
Rand1, if rand1 > ri, riFor the pulse frequency of i-th bat, formula x is utilizednew(i)=xold+εAtObtain xnew；To position
Quality assessed；A random number rand2 is generated, if rand2 < Ai, wherein AiFor the pulse loudness of i-th bat.
There is f (x simultaneouslyi) < f (x*) bat individual update is then carried out, pass through formula At+1(i)=α At(i) reduce Ai, pass through formula rt+1(i)=
r0(i) × [1-exp (- γ × t)] increase ri；If meeting termination condition, exports target value and terminator executes, otherwise
Steering connects the position of every bat individual with fitness f (x)；The individual decoding that bat algorithm optimization is obtained, knot
Pre-selection initial connection weight and threshold value of the fruit as reverse transmittance nerve network.According to resulting preset value, Dynamic gene is calculatedWithIf being unsatisfactory forOrIt then turns to and generates n bat at random, form initial bat population, searching
The position x of random initializtion bat in rope spacei, speed vi, loudness AiWith pulse frequency riRe-execute the program；If simultaneously
MeetAndThen preset value is the initial connection weight and threshold value for being used as reverse transmittance nerve network；
Wherein, β ∈ [0,1] is that an obedience is uniformly distributed random vector, xoldIt is randomly selected in current optimal solution set
One solution, AtFor the mean loudness of all bats under current iteration number, ε is that a D ties up random vector, ε ∈ [- 1,1], α
For the attenuation coefficient of pulse loudness, α is constant, and 0 < α < 1.
Optionally, the fitness function is as follows:
Wherein number of the l for institute's training sample, TimWithM training sample is respectively indicated determined by i-th bat
Weight and threshold value lower network reality output and desired output.
Optionally, the evaluation model of the bat algorithm-reverse transmittance nerve network are as follows:
Wherein, connection weight is { Wji, i=1,2 ..., n, j=1,2 ..., p；Middle layer to output layer connection weight
{Vtj, j=1,2 ..., p, t=1,2 ..., q；It is { θ that middle layer each unit, which exports threshold value,j, j=1,2 ..., p；aiFor input
Variable；γ is the increase coefficient of pulse frequency, and γ is constant, and γ > 0；The transmission function of network input layer to hidden layer is
Tan-sigmoid:The transmission function of hidden layer to output layer uses logarithm S type transmission function: Log-
Sigmoid type function
In conjunction with roadway characteristic invention definition and extracted, final herein by 5 sub-indicators such as L, K, c, R and q
The evaluation index E total to one.
Include 5 neurons in input layer in network structure, correspond to 5 points of indexs, hidden layer chooses 16 nerves
Member, associated nodes are 16, and output layer includes 1 neuron.The transmission function for enabling network input layer to hidden layer is tan-
The transmission function of sigmoid, hidden layer to output layer use logarithm S type transmission function: Log-sigmoid type function.Namely:
Traveling pavement condition evaluation method provided by the present application makes up backpropagation using bat algorithm by extracting characteristics of image
The randomness defect being connected to the network on weight and threshold value selection, while the prominent barrier of Dynamic gene is added for traveling road conditions
It is main to influence, to achieve the purpose that mutual supplement with each other's advantages, obtain based on bat algorithm-back-propagation algorithm traveling evaluating pavement condition mould
Type, and then small-sized unstructured road road conditions are effectively predicted.
Although the application is described above by referring to specific embodiment, one of ordinary skill in the art are answered
Work as understanding, in principle disclosed in the present application and range, many modifications can be made for configuration disclosed in the present application and details.
The protection scope of the application is determined by the attached claims, and claim is intended to technical characteristic in claim
Equivalent literal meaning or range whole modifications for being included.
Claims (8)
1. a kind of traveling pavement condition evaluation method, it is characterised in that: described method includes following steps:
Step 1): traveling road conditions image is obtained；
Step 2): road image pretreatment；
Step 3): characteristics of image is extracted；
Step 4): initial reverse Propagation Neural Network structure is established, and sets the network-related parameters and function, based on extraction
Characteristics of image generates initial parameter using coding mode, while obtaining best initial weights and threshold value using bat algorithm；
Step 5): the best initial weights and threshold optimization reverse transmittance nerve network obtained by bat algorithm-back-propagation algorithm,
And the optimal reverse Propagation Neural Network structure for meeting performance requirement is obtained by training；
Step 6): it establishes based on bat algorithm-reverse transmittance nerve network evaluation model, traveling road conditions is evaluated.
2. traveling pavement condition evaluation method as described in claim 1, it is characterised in that: doubtful in road conditions image in the step 1)
The information of barrier has following expression:
Wherein, the initial height/width value of the objects in front of acquisition is respectively h0And w0, car body and doubtful barrier are apart from initial value
d0, dynamic barrier coefficient of determination is γ；
If it is constant or increasingly to acquire in image doubtful barrier object profile size in entire image in area-of-interest
It is small, then d >=d0, i.e. movable body and doubtful obstacle distance remain unchanged or original remoter, and remove barriers object suspicion, γ=0；
Conversely, profile becomes larger, d < d0It is determined as dynamic barrier, at this time γ=1.
3. traveling pavement condition evaluation method as claimed in claim 2, it is characterised in that: characteristics of image includes road in the step 2)
Surface evenness, road surface curvature, barrier the ratio of width to height, barrier effective area ratio and obstacle coefficient.
4. traveling pavement condition evaluation method as described in claim 1, it is characterised in that: the reverse transmittance nerve network is by 3 layers
Node composition: input node n, hidden layer node p, output node q.
5. traveling pavement condition evaluation method as claimed in claim 4, it is characterised in that: the bat algorithm-back-propagation algorithm
Middle setting Dynamic geneWith
Wherein,For the spectrum of input variable.
6. traveling pavement condition evaluation method as claimed in claim 5, it is characterised in that: the bat algorithm-back-propagation algorithm
Are as follows:
By the parameter of reverse transmittance nerve network, parameter is encoded as a whole；N bat is generated at random, is formed initial
Bat population, the position x of random initializtion bat in search spacei, speed vi, loudness AiWith pulse frequency ri；By every
The position of bat individual is connected with fitness f (x)；Judge whether algorithm reaches maximum number of iterations, if meeting condition,
Algorithm terminates, and exports optimal solution, otherwise, carries out in next step；According to flying speed and location update formula fi=fmin+(fmax-
fmin)β、WithUpdate speed and the position of bat；A random number rand1 is generated,
If rand1 > ri, riFor the pulse frequency of i-th bat, formula x is utilizednew(i)=xold+εAtObtain xnew；To the quality of position
It is assessed；A random number rand2 is generated, if rand2 < Ai, wherein AiFor the pulse loudness of i-th bat.There is f simultaneously
(xi) < f (x*) bat individual update is then carried out, pass through formula At+1(i)=α At(i) reduce Ai, pass through formula rt+1(i)=r0(i)×
[1-exp (- γ × t)] increases ri；If meeting termination condition, export target value and terminator executes, otherwise turning to will
The position of every bat individual is connected with fitness f (x)；The individual decoding that bat algorithm optimization is obtained, as a result conduct
The pre-selection initial connection weight and threshold value of reverse transmittance nerve network.According to resulting preset value, Dynamic gene is calculatedWith
If being unsatisfactory forOrIt then turns to and generates n bat at random, initial bat population is formed, in search space
The position x of middle random initializtion bati, speed vi, loudness AiWith pulse frequency riRe-execute the program；If meeting simultaneouslyAndThen preset value is the initial connection weight and threshold value for being used as reverse transmittance nerve network；
Wherein, β ∈ [0,1] is that an obedience is uniformly distributed random vector, xoldFor randomly selected one in current optimal solution set
Solution, AtFor the mean loudness of all bats under current iteration number, ε is that a D ties up random vector, and ε ∈ [- 1,1], α is arteries and veins
Rush the attenuation coefficient of loudness, α is constant, and 0 < α < 1.
7. traveling pavement condition evaluation method as claimed in claim 6, it is characterised in that:
The fitness function is as follows:
Wherein number of the l for institute's training sample, TimWithRespectively indicate m training sample weight determined by i-th bat
With threshold value lower network reality output and desired output.
8. traveling pavement condition evaluation method as described in claim 1, it is characterised in that: the bat algorithm-Back propagation neural
The evaluation model of network are as follows:
Wherein, connection weight is { Wji, i=1,2 ..., n, j=1,2 ..., p；Middle layer to output layer connection weight { Vtj},j
=1,2 ..., p, t=1,2 ..., q；It is { θ that middle layer each unit, which exports threshold value,j, j=1,2 ..., p；aiFor input variable；γ
For the increase coefficient of pulse frequency, γ is constant, and γ > 0；The transmission function of network input layer to hidden layer is tan-
Sigmoid:The transmission function of hidden layer to output layer uses logarithm S type transmission function: Log-sigmoid
Type function
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811320148.8A CN109543572A (en) | 2018-11-07 | 2018-11-07 | A kind of traveling pavement condition evaluation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811320148.8A CN109543572A (en) | 2018-11-07 | 2018-11-07 | A kind of traveling pavement condition evaluation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109543572A true CN109543572A (en) | 2019-03-29 |
Family
ID=65844625
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811320148.8A Pending CN109543572A (en) | 2018-11-07 | 2018-11-07 | A kind of traveling pavement condition evaluation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543572A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111060119A (en) * | 2019-12-27 | 2020-04-24 | 中电海康集团有限公司 | City slow-travel navigation method and system based on street view visual element scoring |
| CN111860567A (en) * | 2020-05-09 | 2020-10-30 | 武汉理工大学 | Construction method of blade image recognition model |
| WO2020233063A1 (en) * | 2019-05-21 | 2020-11-26 | 深圳壹账通智能科技有限公司 | Intelligent path planning method and device, and computer readable storage medium |
| CN113221954A (en) * | 2021-04-15 | 2021-08-06 | 长春工业大学 | BP (Back propagation) classification algorithm based on improved bat algorithm |
| CN116879866A (en) * | 2023-09-06 | 2023-10-13 | 四川华腾公路试验检测有限责任公司 | Road surface flatness evaluation method based on laser radar three-dimensional data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN108765951A (en) * | 2018-06-11 | 2018-11-06 | 广东工业大学 | Method for identifying traffic status of express way based on bat algorithm support vector machines |
-
2018
- 2018-11-07 CN CN201811320148.8A patent/CN109543572A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN108765951A (en) * | 2018-06-11 | 2018-11-06 | 广东工业大学 | Method for identifying traffic status of express way based on bat algorithm support vector machines |
Non-Patent Citations (1)
| Title |
|---|
| 吕丹丹: "适用于小型智能车的行驶路况评价及自主避障控制研究", 《中国优秀硕士学位论文全文数据库工程科技II辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020233063A1 (en) * | 2019-05-21 | 2020-11-26 | 深圳壹账通智能科技有限公司 | Intelligent path planning method and device, and computer readable storage medium |
| CN111060119A (en) * | 2019-12-27 | 2020-04-24 | 中电海康集团有限公司 | City slow-travel navigation method and system based on street view visual element scoring |
| CN111060119B (en) * | 2019-12-27 | 2021-08-10 | 中电海康集团有限公司 | City slow-travel navigation method and system based on street view visual element scoring |
| CN111860567A (en) * | 2020-05-09 | 2020-10-30 | 武汉理工大学 | Construction method of blade image recognition model |
| CN111860567B (en) * | 2020-05-09 | 2024-04-16 | 武汉理工大学 | Construction method of blade image recognition model |
| CN113221954A (en) * | 2021-04-15 | 2021-08-06 | 长春工业大学 | BP (Back propagation) classification algorithm based on improved bat algorithm |
| CN116879866A (en) * | 2023-09-06 | 2023-10-13 | 四川华腾公路试验检测有限责任公司 | Road surface flatness evaluation method based on laser radar three-dimensional data |
| CN116879866B (en) * | 2023-09-06 | 2023-11-14 | 四川华腾公路试验检测有限责任公司 | Road surface flatness evaluation method based on laser radar three-dimensional data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543572A (en) | A kind of traveling pavement condition evaluation method | |
| CN106650913B (en) | A kind of vehicle density method of estimation based on depth convolutional neural networks | |
| CN103871029B (en) | A kind of image enhaucament and dividing method | |
| Alhussan et al. | Pothole and plain road classification using adaptive mutation dipper throated optimization and transfer learning for self driving cars | |
| CN105279966B (en) | A kind of congestion recognition methods of Campus transport | |
| Collado et al. | Model based vehicle detection for intelligent vehicles | |
| CN112561796B (en) | Laser point cloud super-resolution reconstruction method based on self-attention generation countermeasure network | |
| CN108909624A (en) | A kind of real-time detection of obstacles and localization method based on monocular vision | |
| CN107609522A (en) | A kind of information fusion vehicle detecting system based on laser radar and machine vision | |
| Akgüngör et al. | An artificial intelligent approach to traffic accident estimation: Model development and application | |
| CN104408481B (en) | Classification of Polarimetric SAR Image method based on depth wavelet neural network | |
| CN106599827A (en) | Small target rapid detection method based on deep convolution neural network | |
| CN110309732A (en) | Activity recognition method based on skeleton video | |
| CN108416783A (en) | Road scene dividing method based on full convolutional Neural network | |
| CN103440495A (en) | Method for automatically identifying hydrophobic grades of composite insulators | |
| CN113345237A (en) | Lane-changing identification and prediction method, system, equipment and storage medium for extracting vehicle track by using roadside laser radar data | |
| CN104504395A (en) | Method and system for achieving classification of pedestrians and vehicles based on neural network | |
| CN102306301A (en) | Motion identification system by simulating spiking neuron of primary visual cortex | |
| CN109800773B (en) | Cross-country road surface extraction method based on three-dimensional laser radar | |
| CN109919026A (en) | A kind of unmanned surface vehicle local paths planning method | |
| CN110525428A (en) | A kind of automatic parking method based on the study of fuzzy deeply | |
| CN114155270A (en) | Pedestrian trajectory prediction method, device, equipment and storage medium | |
| CN114299607A (en) | Human-vehicle collision risk degree analysis method based on automatic driving of vehicle | |
| CN105678275A (en) | Vehicle identification method | |
| CN112308005A (en) | Traffic video significance prediction method based on GAN |
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: 20190329 |
|
| RJ01 | Rejection of invention patent application after publication |