CN109101980A - A kind of automobile weightlessness Detection & Controling method based on machine learning - Google Patents
A kind of automobile weightlessness Detection & Controling method based on machine learning Download PDFInfo
- Publication number
- CN109101980A CN109101980A CN201810789221.XA CN201810789221A CN109101980A CN 109101980 A CN109101980 A CN 109101980A CN 201810789221 A CN201810789221 A CN 201810789221A CN 109101980 A CN109101980 A CN 109101980A
- Authority
- CN
- China
- Prior art keywords
- bridge
- automobile
- picture
- gradient
- image
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The automobile weightlessness Detection & Controling method based on machine learning that the present invention provides a kind of generates weightless sense for controlling car deceleration in vehicle bridge-crossing to avoid passenger, has the feature that, comprising the following steps: step 1, the identification model of bridge in foundation；Step 2, the identification model of lower bridge is established；Step 3, the Optimized model of ideal speed when establishing bridge under automobile using artificial neural network algorithm；Step 4, speed control when automobile is by bridge to be measured.It can be realized the automatic identification of bridge by this method and automatic and accurate when lower bridge slowed down, the unmanned process especially suitable for automobile.
Description
Technical field
The present invention relates to vehicular fields, and in particular to a kind of automobile weightlessness Detection & Controling method based on machine learning.
Background technique
With the continuous development of automobile industry, especially in unmanned technology, ride comfort and automobile it is smooth
Property is increasingly taken seriously.In daily life, bridge is typically done arched, because thus focusing on the stress of bridge
The both ends (bridge pier) of bridge can be transmitted to strength both ends (bridge pier) in this way when stress on bridge.Automobile passes through arch bridge process
In, automobile does centrifugal movement, and it is centripetal force that gravity, which subtracts holding power, so automobile is in state of weightlessness, brings discomfort to people
Feel.
In vehicle bridge-crossing, we are generally adopted by artificial slow down to alleviate the feeling of rider's discomfort.But manually
Slow down to exist and operates improper, adjustment overscale problems.In addition, with unpiloted development, when how to realize vehicle bridge-crossing
Adaptive slow down becomes one of urgent problem to be solved in vehicular field.
Summary of the invention
The present invention is to carry out to solve the above-mentioned problems, and it is an object of the present invention to provide a kind of automobile based on machine learning loses
Re-detection and control method.
The automobile weightlessness Detection & Controling method based on machine learning that the present invention provides a kind of, in vehicle bridge-crossing
It controls car deceleration and generates weightless sense to avoid passenger, have the feature that, comprising the following steps: step 1, bridge in foundation
Identification model: step 1-1, the positive sample and negative sample for collecting the picture about bridge of predetermined quantity obtain collecting data set, step
Rapid 1-2, sets the type of the characteristic point of bridge, and step 1-3 is obtained according to the characteristic point of setting to data set progress image zooming-out is collected
To fisrt feature point image, fisrt feature point image is handled to obtain first and is shown by step 1-4 by marking area detection method
Work property image, step 1-5 are detected to obtain object picture to the object of vehicle front in vehicle traveling process, step 1-6,
Image zooming-out is carried out to object picture according to the characteristic point of setting and obtains object features point image, step 1-7, by object feature point
Image is handled to obtain object Saliency maps picture, step 1-8, respectively to object Saliency maps picture by marking area detection method
And first Saliency maps picture carry out feature extraction and be compared identification to result is extracted, step 1-9, within the set time
Repeat step 1-5~step 1-8, and is accurately identified to obtain the identification mould of bridge using machine learning training pattern
Type；Step 2, establish the identification model of lower bridge: step 2-1 sets the type of the characteristic point of bridge, step 2-2, according to the spy of setting
Sign point obtains second feature point image to data set progress image zooming-out is collected, and step 2-3 passes through second feature point image aobvious
Region detection method is write to be handled to obtain the second Saliency maps picture, step 2-4, to the object of vehicle front in vehicle traveling process
It is detected to obtain object picture, step 2-5 carries out image zooming-out to object picture according to the characteristic point of setting and obtains object spy
Point image is levied, step 2-6 is handled object features point image by marking area detection method to obtain object Saliency maps
Picture, step 2-7 carry out feature extraction to object Saliency maps picture and the second Saliency maps picture respectively and carry out to result is extracted
Matching identification, step 2-8 repeat step 2-4~step 2-7 within the set time, and to using machine learning training
Model is accurately identified to obtain the identification model of lower bridge；Step 3, the Optimized model of ideal speed when establishing bridge under automobile:
Step 3-1 detects the object of vehicle front to obtain object picture and Gradient, and step 3-2 will be in object picture
Characteristic point carries out feature extraction and obtains image data, and step 3-3, when bridge under automobile, controlling car deceleration and recording makes passenger
There is no target vehicle speed when weightless sense, step 3-4 controls car deceleration to target vehicle speed, obtain at this time when bridge under automobile
Image data and the weight of Gradient simultaneously update target vehicle speed according to the weight, and step 3-5 repeats step 3-4 and passes through people
Artificial neural networks algorithm carries out model training to image data and Gradient and obtains the optimal of image data and Gradient
Weight；Step 4, speed control when automobile passes through bridge to be measured, step 4-1, to the object of vehicle front detected to obtain to
Object picture and Gradient to be measured are surveyed, step 4-2 obtains the characteristic point progress feature extraction in object under test picture to be measured
Testing image data are compared with the identification model of the identification model of upper bridge and lower bridge respectively by image data, step 4-3
Whether judge automobile in Xia Qiao, step 4-4, according to the judging result being judged as YES, and according to Gradient to be measured, to mapping
The ideal speed needed when obtaining bridge under automobile as data and optimal weights；Step 4-5, by speed in a manner of brake deceleration
Reduce to ideal speed, wherein in step 1-2, the characteristic point of coarse positioning bridge includes at least water, railing, the range of grade on road, road
The height of width, bridge, in step 2-1, the characteristic point of coarse positioning bridge is included at least under land change, railing variation and road surface
It is recessed.
In the automobile weightlessness Detection & Controling method provided by the invention based on machine learning, can also have such
Feature: where in step 4-3, judge whether automobile is sentenced in the situation of change that Xia Qiao can be combined with Gradient to be measured
Not.
In the automobile weightlessness Detection & Controling method provided by the invention based on machine learning, can also have such
Feature: where subject image and object under test image are examined by using object of the imaging sensor to vehicle front
It surveys and obtains, Gradient and Gradient to be measured are detected by using object of the Slope Transducer to vehicle front
And it obtains.
In the automobile weightlessness Detection & Controling method provided by the invention based on machine learning, can also have such
Feature: where step 1-8, in step 2-7, step 3-2 and step 4-2, carrying out the method that feature takes includes that direction gradient is straight
Square figure feature extraction and support vector machine classifier facture.
In the automobile weightlessness Detection & Controling method provided by the invention based on machine learning, can also have such
Feature: where in step 1-9 and step 2-8, the method for machine learning includes histograms of oriented gradients feature extraction and branch
Hold vector machine classifier facture.
The action and effect of invention
Automobile weightlessness Detection & Controling method based on machine learning involved according to the present invention, because this method is first
The identification model of bridge and the identification model of lower bridge is established in foundation, then, the optimization mould of ideal speed when establishing bridge under automobile
Type finally passes through speed control when bridge to be measured, so automobile can be made to arrive according to actually detected by this method to automobile
Object under test picture and Gradient to be measured are compared to identify automobile with the identification model of the identification model of upper bridge and lower bridge
Whether in Xia Qiao, and ideal speed needed for bridge under reality is obtained according to the Optimized model of the ideal speed under automobile when bridge,
Last automatic retarding avoids bridge under passenger from generating sense of discomfort to ideal speed, and such speed adjusts compared with artificial slow down more
Accurately, especially suitable for the unmanned process of automobile.
In addition, passing through during the identification model of bridge and in automobile under the identification model and foundation for establishing upper bridge to be measured
Machine learning training pattern has been used in speed control process when bridge to identify to Shang Qiao, lower bridge, obtained in this way
Recognition result is more accurate.It is calculated during the Optimized model of ideal speed in bridge under establishing automobile using artificial neural network
Method carries out model training to image data and Gradient to obtain optimal weights, and such calculated result is more accurate.
Further, since the identification model of upper bridge and the identification model of lower bridge have been pre-established, so automobile passes through bridge to be measured
When the model that only needs and pre-establish carry out aspect ratio to can differentiate whether automobile is passing a bridge and in lower bridge like state, from
And operand is greatly simplified, operation time is shortened, the burden of vehicle processor is reduced.
In addition, avoiding the possibility that overhead equal objects are mistaken for bridge since this method establishes the identification model of lower bridge
Property, to improve the accuracy rate identified to bridge.
Detailed description of the invention
Fig. 1 is the flow chart of the automobile weightlessness Detection & Controling method in the embodiment of the present invention based on machine learning；
Fig. 2 is to carry out model instruction to image data and Gradient by artificial neural network in the embodiment of the present invention
Experienced schematic diagram.
Specific embodiment
It is real below in order to be easy to understand the technical means, the creative features, the aims and the efficiencies achieved by the present invention
Example combination attached drawing is applied to be specifically addressed the automobile weightlessness Detection & Controling method the present invention is based on machine learning.
Fig. 1 is the flow chart of the automobile weightlessness Detection & Controling method in the embodiment of the present invention based on machine learning.
As shown in Figure 1, the automobile weightlessness Detection & Controling method based on machine learning in the present embodiment includes following step
It is rapid:
Step S1, the identification model of bridge in foundation:
Step S1-1 collects the positive sample and negative sample of the picture about bridge of predetermined quantity by the way of supervised learning
It obtains collecting data set.In the present embodiment, about the positive sample of the picture of bridge and each mobile phone of negative sample 1000.The picture of bridge
Positive sample is bridge, and negative sample is non-bridge.
Step S1-2 sets the type of the characteristic point of bridge.In the present embodiment, the characteristic point of the identification model jackshaft of upper bridge
Including at least the height of water, railing, the range of grade on road, the width on road, bridge.
Step S1-3 obtains fisrt feature point image to data set progress image zooming-out is collected according to the characteristic point of setting.
Wherein, there are the features such as water, railing is indicated with " 1 ", and there is no this category features to be indicated with " 0 "；The range of grade on road, road width
And the features such as height of bridge use one-hot coded representation.
Step S1-4 is handled fisrt feature point image by marking area detection method to obtain the first Saliency maps
Picture.Its detailed process are as follows: fisrt feature point image is divided by 200 super-pixel using SLIC algorithm, and according to its colour-difference
Different, point shape difference and spatial distribution acquire the matrix of similitude between a characterization super-pixel.Then, it according to this matrix, utilizes
Similar propagation algorithm (AP algorithm) to super-pixel cluster, and by measurement class between color contrast, class compact-sized degree with
Deviate the significance that centrad evaluates each class.Finally, the color difference and positional relationship of compared pixels and every class update pixel
Significance, the first Saliency maps that finally obtain pixel precision, full resolution.
Step S1-5 is detected to obtain object using object of the imaging sensor to vehicle front in vehicle traveling process
Picture.
Step S1-6 carries out image zooming-out to object picture according to the characteristic point of setting and obtains object features point image.Its
In, there are the features such as water, railing to be indicated with " 1 ", and there is no this category features to be indicated with " 0 "；The range of grade on road, road width with
And the features such as height of bridge use one-hot coded representation.
Step S1-7 is handled object features point image by marking area detection method to obtain object Saliency maps
Picture.Use SLIC algorithm by object feature point image segmentation for 200 super-pixel, and according to its color difference, divide shape difference and
Spatial distribution acquires the matrix of similitude between a characterization super-pixel.Then, according to this matrix, similar propagation algorithm is utilized
(AP algorithm) clusters super-pixel, and is commented by the compact-sized degree of color contrast, class between measurement class with centrad is deviateed
The significance of each class of valence.Finally, the color difference and positional relationship of compared pixels and every class update the significance of pixel, finally
Object Saliency maps that obtain pixel precision, full resolution.
Step S1-8 carries out feature extraction to object Saliency maps picture and the first Saliency maps picture respectively and ties to extracting
Identification is compared in fruit.Wherein, the process of feature extraction is carried out are as follows: histograms of oriented gradients is being extracted around characteristic point
(HOG) then feature uses support vector machines (SVM classifier) using extracted feature as matrix element composition characteristic square
Battle array, is finally normalized eigenmatrix, excludes the influence of the environmental factors such as illumination.
Step S1-9 repeats step S1-5~step S1-8 within the set time, and using machine learning training
Model is accurately identified to obtain the identification model of bridge.Wherein, machine learning training pattern includes Adaboost iterative algorithm
And harr classifier training method.
Step S2 establishes the identification model of lower bridge:
Step S2-1 sets the type of the characteristic point of bridge.In the present embodiment, the characteristic point of the identification model jackshaft of lower bridge
Including at least land change, railing variation and dip.
Step S2-2 obtains second feature point image to data set progress image zooming-out is collected according to the characteristic point of setting.
Wherein, there are the features such as dip is indicated with " 1 ", and there is no this category features to be indicated with " 0 "；Land change, railing variation etc.
Feature uses one-hot coded representation.
Step S2-3 is handled second feature point image by marking area detection method to obtain the second Saliency maps
Picture.Its detailed process are as follows: second feature point image is divided by 200 super-pixel using SLIC algorithm, and according to its colour-difference
Different, point shape difference and spatial distribution acquire the matrix of similitude between a characterization super-pixel.Then, it according to this matrix, utilizes
Similar propagation algorithm (AP algorithm) to super-pixel cluster, and by measurement class between color contrast, class compact-sized degree with
Deviate the significance that centrad evaluates each class.Finally, the color difference and positional relationship of compared pixels and every class update pixel
Significance, the second Saliency maps that finally obtain pixel precision, full resolution.
Step S2-4 is detected to obtain object using object of the imaging sensor to vehicle front in vehicle traveling process
Picture.
Step S2-5 carries out image zooming-out to object picture according to the characteristic point of setting and obtains object features point image.Its
In, there are the features such as dip is indicated with " 1 ", and there is no this category features to be indicated with " 0 "；The spies such as land change, railing variation
Sign uses one-hot coded representation.
Step S2-6 is handled object features point image by marking area detection method to obtain object Saliency maps
Picture uses SLIC algorithm by object feature point image segmentation for 200 super-pixel, and according to its color difference, divide shape difference and
Spatial distribution acquires the matrix of similitude between a characterization super-pixel.Then, according to this matrix, similar propagation algorithm is utilized
(AP algorithm) clusters super-pixel, and is commented by the compact-sized degree of color contrast, class between measurement class with centrad is deviateed
The significance of each class of valence.Finally, the color difference and positional relationship of compared pixels and every class update the significance of pixel, finally
Object Saliency maps that obtain pixel precision, full resolution.
Step S2-7 carries out feature extraction to object Saliency maps picture and the second Saliency maps picture respectively and ties to extracting
Identification is compared in fruit.Wherein, the process of feature extraction is carried out are as follows: histograms of oriented gradients is being extracted around characteristic point
(HOG) then feature uses support vector machines (SVM classifier) using extracted feature as matrix element composition characteristic square
Battle array, is finally normalized eigenmatrix, excludes the influence of the environmental factors such as illumination.
Step S2-8 is repeated step S2-4~step S2-7 within the set time, and is instructed to using machine learning
Practice model to be accurately identified to obtain the identification model of lower bridge.Wherein, machine learning training pattern includes that Adaboost iteration is calculated
Method and harr classifier training method.
Step S3, the Optimized model of ideal speed when establishing bridge under automobile:
Step S3-1 is detected to obtain object picture using object of the imaging sensor to vehicle front, meanwhile, it utilizes
Slope Transducer is detected to obtain Gradient to the object of vehicle front.
Step S3-2, according to the characteristic point set in the identification model of upper bridge and the identification model of lower bridge to object picture into
Row image zooming-out obtains feature point image.Wherein, there are water, railing, there are the features such as dip and indicated with " 1 ", be not present
This category feature is indicated with " 0 "；The features such as the range of grade on road, the width on road, the height of bridge, land change, railing variation use
One-hot coded representation.
Step S3-3 carries out feature extraction to the characteristic point of feature point image and obtains image data.Wherein, feature is carried out to mention
The process taken are as follows: histograms of oriented gradients (HOG) feature is being extracted around characteristic point, then using (SVM points of support vector machines
Class device) using extracted feature as matrix element composition characteristic matrix, finally eigenmatrix is normalized, is excluded
The influence of the environmental factors such as illumination.
Step S3-4 controls car deceleration and records target vehicle speed when passenger being made not have weightless sense when bridge under automobile.
Step S3-5 controls car deceleration to target vehicle speed, obtains image data and the gradient at this time when bridge under automobile
The weight of data simultaneously updates target vehicle speed according to the weight.
Step S3-6 repeats step 3-5 and by artificial neural network (BP) algorithm to image data and Gradient
It carries out model training and obtains the optimal weights of image data and Gradient.
Wherein, two inputs (image data X1 and Gradient X2) of BP neural network obtain by three-layer neural network
One output.
Fig. 2 is to carry out model instruction to image data and Gradient by artificial neural network in the embodiment of the present invention
Experienced schematic diagram.
As shown in Fig. 2, inputting X1, X2 during the training neural network.First hiding layer functions are set as f1(e)、f2
(e)、f3(e), the second hiding layer functions are set as f4(e)、f5(e), the hiding layer functions of third are set as f6(e)。
f1(e) the corresponding weight of node is w(X1)1、w(X2)1, then have y1=f1(w(X1)1X1+w(X2)1X2)；f2(e) node pair
The weight answered is w(X1)2、w(X2)2, then have y2=f2(w(X1)2X1+w(X2)2X2)；f3(e) the corresponding weight of node is w(X1)3、
w(X2)3, then have y3=f3(w(X1)3X1+w(X2)3X2).Wherein, y1、y2、y3Respectively indicate three outputs of the first hidden layer.
f4(e) the corresponding weight of node is w14、w24、w34, then have y4=f4(w14y1+w24y2+w34y3)；f5(e) node pair
The weight answered is w15、w25、w35, then have y5=f5(w15y1+w25y2+w35y3).Wherein, y4、y5Respectively indicate the second hidden layer
Two outputs.
f6(e) the corresponding weight of node is w46、w56, then have y6=f6(w46y4+w56y5).Obtain output y6。
Step S4, automobile pass through speed control when bridge to be measured:
Step S4-1 is detected to obtain object under test picture using object of the imaging sensor to vehicle front, meanwhile,
It is detected to obtain Gradient to be measured using object of the Slope Transducer to vehicle front.
Step S4-2, according to the characteristic point set in the identification model of upper bridge and the identification model of lower bridge to object picture into
Row image zooming-out obtains feature point image to be measured.Wherein, there are water, railing, there are the features such as dip and indicated with " 1 ", no
There are this category features to be indicated with " 0 "；The features such as the range of grade on road, the width on road, the height of bridge, land change, railing variation
Use one-hot coded representation.
Characteristic point in object under test picture to be measured is carried out feature extraction and obtains testing image data by step S4-3.Its
In, carry out the process of feature extraction are as follows: histograms of oriented gradients (HOG) feature is being extracted around characteristic point, then using branch
Vector machine (SVM classifier) is held using extracted feature as matrix element composition characteristic matrix, finally eigenmatrix is carried out
Normalized excludes the influence of the environmental factors such as illumination.
Testing image data are compared with the identification model of the identification model of upper bridge and lower bridge step S4-4 respectively
To judge automobile whether in Xia Qiao.Meanwhile judging the automobile whether in Xia Qiao in conjunction with the situation of change of Gradient to be measured.
Step S4-5, according to the judging result being judged as YES, and according to Gradient to be measured, testing image data and most
Excellent weight obtains the ideal speed needed when bridge under automobile
Speed is reduced to ideal speed in a manner of brake deceleration by step S4-6.
The action and effect of embodiment
Automobile weightlessness Detection & Controling method according to involved in the present embodiment based on machine learning, because this method is first
It first establishes the identification model of upper bridge and establishes the identification model of lower bridge, then, the optimization of ideal speed when establishing bridge under automobile
Model finally passes through speed control when bridge to be measured, so automobile can be made to arrive according to actually detected by this method to automobile
Object under test picture and Gradient to be measured be compared with the identification model of the identification model of upper bridge and lower bridge to identify vapour
Whether vehicle according to the Optimized model of ideal speed automobile under when bridge obtains reality under bridge needed for dream car in Xia Qiao
Speed, last automatic retarding to ideal speed avoid bridge under passenger from generating sense of discomfort, and such speed adjustment is compared with artificial slow down
It is more accurate.
In addition, passing through during the identification model of bridge and in automobile under the identification model and foundation for establishing upper bridge to be measured
Machine learning training pattern has been used in speed control process when bridge to identify to Shang Qiao, lower bridge, obtained in this way
Recognition result is more accurate.It is calculated during the Optimized model of ideal speed in bridge under establishing automobile using artificial neural network
Method carries out model training to image data and Gradient to obtain optimal weights, and such calculated result is more accurate.
Further, since the identification model of upper bridge and the identification model of lower bridge have been pre-established, so automobile passes through bridge to be measured
When the model that only needs and pre-establish carry out aspect ratio to can differentiate whether automobile is passing a bridge and in lower bridge like state, from
And operand is greatly simplified, operation time is shortened, the burden of vehicle processor is reduced.
In addition, avoiding the possibility that overhead equal objects are mistaken for bridge since this method establishes the identification model of lower bridge
Property, to improve the accuracy rate identified to bridge.
Further, judge whether automobile in Xia Qiao can be combined with the situation of change of Gradient to be measured to differentiate, when
On automobile when bridge, Gradient is positive value；When bridge under automobile, Gradient is negative value.The method of discrimination energy combined in this way
It is enough to improve the accuracy differentiated.
Above embodiment is preferred case of the invention, the protection scope being not intended to limit the invention.
Claims (5)
1. a kind of automobile weightlessness Detection & Controling method based on machine learning, in vehicle bridge-crossing control car deceleration with
Passenger is avoided to generate weightless sense, which comprises the following steps:
Step 1, in foundation bridge identification model:
Step 1-1, the positive sample and negative sample for collecting the picture about bridge of predetermined quantity obtain collecting data set,
Step 1-2 sets the type of the characteristic point of bridge,
Step 1-3 obtains fisrt feature point image to data set progress image zooming-out is collected according to the characteristic point of setting,
Step 1-4 is handled the fisrt feature point image by marking area detection method to obtain the first Saliency maps picture,
Step 1-5 is detected to obtain object picture to the object of vehicle front in vehicle traveling process,
Step 1-6 carries out image zooming-out to object picture according to the characteristic point of setting and obtains object features point image,
Step 1-7 is handled the object features point image by marking area detection method to obtain object Saliency maps picture,
Step 1-8 carries out feature extraction to the object Saliency maps picture and the first Saliency maps picture respectively and to mentioning
Take result that identification is compared,
Step 1-9 is repeated step 1-5~step 1-8 within the set time, and is carried out using machine learning training pattern
It accurately identifies to obtain the identification model of bridge；
Step 2, the identification model of lower bridge is established:
Step 2-1 sets the type of the characteristic point of bridge,
Step 2-2 obtains second feature point image to data set progress image zooming-out is collected according to the characteristic point of setting,
Step 2-3 is handled the second feature point image by marking area detection method to obtain the second Saliency maps picture,
Step 2-4 is detected to obtain object picture to the object of vehicle front in vehicle traveling process,
Step 2-5 carries out image zooming-out to object picture according to the characteristic point of setting and obtains object features point image,
Step 2-6 is handled object features point image by marking area detection method to obtain object Saliency maps picture,
Step 2-7 carries out feature extraction to the object Saliency maps picture and the second Saliency maps picture respectively and to mentioning
Take result that identification is compared,
Step 2-8, repeats step 2-4~step 2-7 within the set time, and to using machine learning training pattern into
Row accurately identifies to obtain the identification model of lower bridge；
Step 3, the Optimized model of ideal speed when establishing bridge under automobile:
Step 3-1 is detected to obtain object picture and Gradient to the object of vehicle front,
Characteristic point in object picture is carried out feature extraction and obtains image data by step 3-2,
Step 3-3 controls car deceleration and records the target vehicle speed when passenger being made not have weightless sense when bridge under automobile,
Step 3-4 controls car deceleration to the target vehicle speed, obtains image data and the gradient at this time when bridge under automobile
The weight of data simultaneously updates target vehicle speed according to the weight,
Step 3-5 repeats step 3-4 and carries out model instruction to image data and Gradient by artificial neural network algorithm
Get the optimal weights of image data and Gradient；
Step 4, speed control when automobile is by bridge to be measured,
Step 4-1 detects the object of vehicle front to obtain object under test picture and Gradient to be measured,
Characteristic point in the object under test picture is carried out feature extraction and obtains testing image data by step 4-2,
The testing image data are compared with the identification model of the identification model of upper bridge and lower bridge step 4-3 respectively
Come judge automobile whether in Xia Qiao,
Step 4-4, according to the judging result being judged as YES, and according to the Gradient to be measured, the testing image data with
And the optimal weights obtain the ideal speed needed when bridge under automobile；
Speed is reduced to the ideal speed in a manner of brake deceleration by step 4-5,
Wherein, in step 1-2, in foundation when the identification model of bridge bridge characteristic point include at least water, railing, road gradient model
It encloses, the height of the width on road, bridge,
In step 2-1, the characteristic point of bridge includes at least land change, railing variation and road surface when establishing the identification model of lower bridge
It is recessed.
2. the automobile weightlessness Detection & Controling method according to claim 1 based on machine learning, it is characterised in that:
Wherein, in step 4-3, judge whether automobile is sentenced in the situation of change that Xia Qiao can be combined with Gradient to be measured
Not.
3. the automobile weightlessness Detection & Controling method according to claim 1 based on machine learning, it is characterised in that:
Wherein, the subject image and the object under test image are by using imaging sensor to the object of vehicle front
It is detected and is obtained,
The Gradient and the Gradient to be measured are carried out by using object of the Slope Transducer to vehicle front
It detects and obtains.
4. the automobile weightlessness Detection & Controling method according to claim 1 based on machine learning, it is characterised in that:
Wherein, step 1-8, in step 2-7, step 3-2 and step 4-2, carrying out the method that feature takes includes that direction gradient is straight
Square figure feature extraction and support vector machine classifier facture.
5. the automobile weightlessness Detection & Controling method according to claim 1 based on machine learning, it is characterised in that:
Wherein, in step 1-9 and step 2-8, the method for the machine learning includes Adaboost iterative algorithm and harr
Classifier training method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810789221.XA CN109101980B (en) | 2018-07-18 | 2018-07-18 | Automobile weightlessness detection and control method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810789221.XA CN109101980B (en) | 2018-07-18 | 2018-07-18 | Automobile weightlessness detection and control method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109101980A true CN109101980A (en) | 2018-12-28 |
| CN109101980B CN109101980B (en) | 2021-06-18 |
Family
ID=64846670
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810789221.XA Active CN109101980B (en) | 2018-07-18 | 2018-07-18 | Automobile weightlessness detection and control method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109101980B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950490A (en) * | 2020-08-18 | 2020-11-17 | 东软睿驰汽车技术（沈阳）有限公司 | Parking rod recognition method and recognition model training method and device |
| CN114199262A (en) * | 2020-08-28 | 2022-03-18 | 阿里巴巴集团控股有限公司 | Method for training position recognition model, position recognition method and related equipment |
| CN114412988A (en) * | 2022-02-11 | 2022-04-29 | 吉林大学 | Uphill predictive gear shifting control method based on road information and machine learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1656524A (en) * | 2002-03-22 | 2005-08-17 | 易卜拉欣·纳拉 | Vehicle navigation, collision avoidance and control system |
| US20140330094A1 (en) * | 2000-06-16 | 2014-11-06 | Bodymedia, Inc. | System for monitoring and presenting health, wellness and fitness utilizng multiple user trend data having user selectable parameters |
| CN105023022A (en) * | 2015-07-09 | 2015-11-04 | 深圳天珑无线科技有限公司 | Tumble detection method and system |
| CN107438754A (en) * | 2015-02-10 | 2017-12-05 | 御眼视觉技术有限公司 | Sparse map for autonomous vehicle navigation |
-
2018
- 2018-07-18 CN CN201810789221.XA patent/CN109101980B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140330094A1 (en) * | 2000-06-16 | 2014-11-06 | Bodymedia, Inc. | System for monitoring and presenting health, wellness and fitness utilizng multiple user trend data having user selectable parameters |
| US20150339946A1 (en) * | 2000-06-16 | 2015-11-26 | Bodymedia, Inc. | System for monitoring and presenting health, wellness and fitness trend data having user selectable parameters |
| CN1656524A (en) * | 2002-03-22 | 2005-08-17 | 易卜拉欣·纳拉 | Vehicle navigation, collision avoidance and control system |
| CN107438754A (en) * | 2015-02-10 | 2017-12-05 | 御眼视觉技术有限公司 | Sparse map for autonomous vehicle navigation |
| CN105023022A (en) * | 2015-07-09 | 2015-11-04 | 深圳天珑无线科技有限公司 | Tumble detection method and system |
Non-Patent Citations (1)
| Title |
|---|
| 张蕊: "考虑道路几何特征的车速自适应控制方法研究", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950490A (en) * | 2020-08-18 | 2020-11-17 | 东软睿驰汽车技术（沈阳）有限公司 | Parking rod recognition method and recognition model training method and device |
| CN111950490B (en) * | 2020-08-18 | 2024-04-09 | 东软睿驰汽车技术（沈阳）有限公司 | Parking rod identification method and training method and device of identification model thereof |
| CN114199262A (en) * | 2020-08-28 | 2022-03-18 | 阿里巴巴集团控股有限公司 | Method for training position recognition model, position recognition method and related equipment |
| CN114412988A (en) * | 2022-02-11 | 2022-04-29 | 吉林大学 | Uphill predictive gear shifting control method based on road information and machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109101980B (en) | 2021-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109101980A (en) | A kind of automobile weightlessness Detection & Controling method based on machine learning | |
| CN108280856A (en) | The unknown object that network model is inputted based on mixed information captures position and orientation estimation method | |
| CN104794463B (en) | The system and method for indoor human body fall detection is realized based on Kinect | |
| CN104848851B (en) | Intelligent Mobile Robot and its method based on Fusion composition | |
| CN109460709A (en) | The method of RTG dysopia analyte detection based on the fusion of RGB and D information | |
| CN104136900B (en) | Barycenter position measurement device, barycenter position measurement method | |
| CN108537197A (en) | A kind of lane detection prior-warning device and method for early warning based on deep learning | |
| CN106408015A (en) | Road fork identification and depth estimation method based on convolutional neural network | |
| CN108629323A (en) | A kind of integrated providing method of scenic spot tourist chain type trip | |
| CN109711288A (en) | Remote sensing ship detecting method based on feature pyramid and distance restraint FCN | |
| CN106372666B (en) | A kind of target identification method and device | |
| CN106845487A (en) | A kind of licence plate recognition method end to end | |
| CN109870223A (en) | A kind of bridge dynamic weighing method of vision technique auxiliary | |
| CN107909010A (en) | A kind of road barricade object detecting method and device | |
| CN109460738B (en) | A kind of road surface types evaluation method of the depth convolutional neural networks based on free of losses function | |
| CN106128115A (en) | A kind of fusion method based on twin camera detection Traffic Information | |
| CN106446949A (en) | Vehicle model identification method and apparatus | |
| CN109934452A (en) | Road Comfort Evaluation method based on multi-source data | |
| CN108280397A (en) | Human body image hair detection method based on depth convolutional neural networks | |
| CN108573277A (en) | A kind of pantograph carbon slide surface disease automatic recognition system and method | |
| CN110490936A (en) | Scaling method, device, equipment and the readable storage medium storing program for executing of vehicle camera | |
| CN108734219A (en) | A kind of detection of end-to-end impact crater and recognition methods based on full convolutional neural networks structure | |
| CN110378246A (en) | Ground detection method, apparatus, computer readable storage medium and electronic equipment | |
| CN109871728A (en) | A kind of model recognizing method and device | |
| CN108021879A (en) | A kind of vehicular traffic kind identification method based on video image |
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 |