CN109101980B - Automobile weightlessness detection and control method based on machine learning - Google Patents

Automobile weightlessness detection and control method based on machine learning Download PDF

Info

Publication number
CN109101980B
CN109101980B CN201810789221.XA CN201810789221A CN109101980B CN 109101980 B CN109101980 B CN 109101980B CN 201810789221 A CN201810789221 A CN 201810789221A CN 109101980 B CN109101980 B CN 109101980B
Authority
CN
China
Prior art keywords
automobile
bridge
image
data
machine learning
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.)
Active
Application number
CN201810789221.XA
Other languages
Chinese (zh)
Other versions
CN109101980A (en
Inventor
朱烨
丁顺
顾垚江
赵善政
汪百前
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Shanghai for Science and Technology
Original Assignee
University of Shanghai for Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Shanghai for Science and Technology filed Critical University of Shanghai for Science and Technology
Priority to CN201810789221.XA priority Critical patent/CN109101980B/en
Publication of CN109101980A publication Critical patent/CN109101980A/en
Application granted granted Critical
Publication of CN109101980B publication Critical patent/CN109101980B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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

Abstract

The invention provides an automobile weightlessness detection and control method based on machine learning, which is used for controlling the automobile to decelerate when an automobile passes a bridge so as to avoid the weightlessness of passengers, and has the characteristics that the method comprises the following steps: step 1, establishing an identification model of an upper bridge; step 2, establishing an identification model of the lower bridge; step 3, establishing an optimal model of the ideal vehicle speed when the vehicle gets off the bridge by using an artificial neural network algorithm; and 4, controlling the speed of the automobile when the automobile passes through the bridge to be measured. The method can realize automatic identification of the bridge and automatic and accurate deceleration when getting off the bridge, and is particularly suitable for the unmanned driving process of the automobile.

Description

Automobile weightlessness detection and control method based on machine learning
Technical Field
The invention relates to the field of vehicles, in particular to an automobile weightlessness detection and control method based on machine learning.
Background
With the continuous development of the automobile industry, particularly in the unmanned technology, riding comfort and smoothness of automobiles are more and more emphasized. In daily life, bridges are generally made into arch shapes, and stress on the bridges is concentrated to two ends (piers) of the bridges, so that when the bridges are stressed, force can be transmitted to the two ends (piers). When the automobile passes through the arch bridge, the automobile does centrifugal motion, and the gravity minus the supporting force is centripetal force, so that the automobile is in a weightless state, and uncomfortable feeling is brought to people.
When the automobile passes a bridge, manual deceleration is generally adopted to relieve discomfort of a passenger. However, the manual deceleration has the problems of improper operation and excessive adjustment. In addition, with the development of unmanned driving, how to realize adaptive deceleration when an automobile passes a bridge becomes one of the problems to be solved urgently in the field of vehicles.
Disclosure of Invention
The present invention is made to solve the above problems, and an object of the present invention is to provide a method for detecting and controlling a weight loss of an automobile based on machine learning.
The invention provides an automobile weightlessness detection and control method based on machine learning, which is used for controlling the automobile to decelerate when an automobile passes a bridge so as to avoid the weightlessness of passengers, and has the characteristics that the method comprises the following steps: step 1, establishing an identification model of an upper bridge: step 1-1, collecting positive samples and negative samples of a preset number of pictures related to a bridge to obtain a collected data set, step 1-2, setting the type of characteristic points of the bridge, step 1-3, extracting images of the collected data set according to the set characteristic points to obtain a first characteristic point image, step 1-4, processing the first characteristic point image through a salient region detection method to obtain a first salient image, step 1-5, detecting an object in front of an automobile in the driving process of the automobile to obtain an object image, step 1-6, extracting images of the object image according to the set characteristic points to obtain an object characteristic point image, step 1-7, processing the object characteristic point image through a salient region detection method to obtain an object salient image, step 1-8, respectively extracting characteristics of the object salient image and the first salient image and comparing and identifying the extraction results, 1-9, repeating the steps 1-5-1-8 within a set time, and accurately identifying by adopting a machine learning training model to obtain an identification model of the upper bridge; step 2, establishing an identification model of the lower bridge: step 2-1, setting the type of the characteristic points of the bridge, step 2-2, extracting images from a collected data set according to the set characteristic points to obtain a second characteristic point image, step 2-3, processing the second characteristic point image by a salient region detection method to obtain a second salient image, step 2-4, detecting an object in front of the automobile in the driving process of the automobile to obtain an object image, step 2-5, extracting images from the object image according to the set characteristic points to obtain an object characteristic point image, step 2-6, processing the object characteristic point image by a salient region detection method to obtain an object salient image, step 2-7, respectively extracting the characteristics of the object salient image and the second salient image and comparing and identifying the extraction results, step 2-8, repeating the steps 2-4 to 2-7 within a set time, and accurately identifying the training model by using machine learning to obtain an identification model of the lower bridge; step 3, establishing an optimization model of the ideal speed when the automobile is off-axle: step 3-1, detecting an object in front of an automobile to obtain an object picture and gradient data, step 3-2, performing feature extraction on feature points in the object picture to obtain image data, step 3-3, controlling the automobile to decelerate and recording a target speed when passengers do not feel weightless when the automobile is off-bridge, step 3-4, controlling the automobile to decelerate to the target speed when the automobile is off-bridge, obtaining the weight of the image data and the gradient data at the moment, updating the target speed according to the weight, step 3-5, repeating the step 3-4, and performing model training on the image data and the gradient data through an artificial neural network algorithm to obtain the optimal weight of the image data and the gradient data; step 4, controlling the speed of the automobile when passing through the bridge to be tested, step 4-1, detecting an object in front of the automobile to obtain an image of the object to be tested and gradient data to be tested, step 4-2, extracting features of feature points in the image of the object to be tested to obtain image data to be tested, step 4-3, comparing the image data to be tested with an identification model of an upper bridge and an identification model of a lower bridge respectively to judge whether the automobile is at the lower bridge, and step 4-4, obtaining the ideal automobile speed required when the automobile is at the lower bridge according to the judgment result that the automobile is yes, the gradient data to be tested, the image data to be tested and the optimal weight; and 4-5, reducing the speed of the vehicle to the ideal speed in a braking and speed reducing mode, wherein in the step 1-2, the characteristic points of the coarse positioning bridge at least comprise water, a railing, the gradient range of the road, the width of the road and the height of the bridge, and in the step 2-1, the characteristic points of the coarse positioning bridge at least comprise land change, railing change and road surface sinking.
The automobile weightlessness detection and control method based on machine learning provided by the invention can also have the following characteristics: and 4-3, judging whether the automobile is on the lower axle or not according to the change condition of the gradient data to be detected.
The automobile weightlessness detection and control method based on machine learning provided by the invention can also have the following characteristics: wherein, object image and the object image that awaits measuring all detect the object in the car the place ahead and obtain through using image sensor, and slope data and the slope data that awaits measuring all detect the object in the car the place ahead and obtain through using slope sensor.
The automobile weightlessness detection and control method based on machine learning provided by the invention can also have the following characteristics: in steps 1-8, 2-7, 3-2 and 4-2, the method for performing feature extraction includes histogram feature extraction and SVM classifier processing.
The automobile weightlessness detection and control method based on machine learning provided by the invention can also have the following characteristics: in steps 1-9 and 2-8, the method of machine learning includes histogram feature extraction and SVM classifier processing.
Action and Effect of the invention
According to the automobile weightlessness detection and control method based on machine learning, the method comprises the steps of firstly establishing an upper axle identification model and an lower axle identification model, then establishing an ideal automobile speed optimization model when the automobile is at the lower axle, and finally controlling the speed when the automobile passes through the bridge to be detected, so that the method can enable the automobile to identify whether the automobile is at the lower axle or not by comparing the actually detected image of the object to be detected and the gradient data to be detected with the upper axle identification model and the lower axle identification model, obtain the ideal automobile speed required by the actual lower axle according to the ideal automobile speed optimization model when the automobile is at the lower axle, and finally automatically decelerate to the ideal automobile speed to avoid discomfort caused by the passenger at the lower axle.
In addition, machine learning training models are used for recognizing the upper axle and the lower axle in the process of establishing the identification model of the upper axle and the identification model of the lower axle and in the process of controlling the speed of the automobile when the automobile passes through the axle to be detected, so that the obtained recognition result is more accurate. In the process of establishing an optimal model of ideal vehicle speed when the vehicle is off-axle, the artificial neural network algorithm is used for carrying out model training on the image data and the gradient data to obtain the optimal weight, and the calculation result is more accurate.
In addition, because the identification model of the upper bridge and the identification model of the lower bridge are established in advance, whether the automobile is in the bridge passing state or not can be judged only by comparing the characteristics of the automobile and the pre-established models when the automobile passes through the bridge to be tested, so that the operation amount is greatly simplified, the operation time is shortened, and the burden of an automobile processor is reduced.
In addition, because the method establishes the identification model of the lower bridge, the possibility of mistakenly judging objects such as an overhead bridge and the like as the bridge is avoided, and the accuracy rate of identifying the bridge is improved.
Drawings
FIG. 1 is a flow chart of a method for detecting and controlling automobile weightlessness based on machine learning according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of model training of image data and slope data by an artificial neural network in an embodiment of the invention.
Detailed Description
In order to make the technical means, the creation features, the achievement purposes and the effects of the invention easy to understand, the following embodiments specifically describe the machine learning-based automobile weight loss detection and control method of the invention with reference to the accompanying drawings.
Fig. 1 is a flowchart of an automobile weight loss detection and control method based on machine learning in an embodiment of the present invention.
As shown in fig. 1, the method for detecting and controlling automobile weightlessness based on machine learning in this embodiment includes the following steps:
step S1, establishing an identification model of the upper bridge:
step S1-1, collecting positive and negative samples of a predetermined number of pictures about the bridge in a supervised learning manner to obtain a collected data set. In this embodiment, the phone number is 1000 for positive and negative samples of the picture of the bridge. The positive sample of the picture of the bridge is the bridge and the negative sample is the non-bridge.
Step S1-2, the type of the characteristic point of the bridge is set. In this embodiment, the characteristic points of the bridge in the identification model of the upper bridge at least include water, a railing, a slope range of a road, a width of the road, and a height of the bridge.
And step S1-3, carrying out image extraction on the collected data set according to the set characteristic points to obtain a first characteristic point image. Wherein the presence of water, railing, etc. features is represented by "1" and the absence of such features is represented by "0"; the characteristics of the slope range of the road, the width of the road, the height of the bridge and the like are expressed by one-hot coding.
And step S1-4, processing the first feature point image by a salient region detection method to obtain a first salient image. The specific process comprises the following steps: and (3) segmenting the first feature point image into 200 superpixels by adopting an SLIC algorithm, and obtaining a matrix representing the similarity between the superpixels according to the color difference, the fractal difference and the spatial distribution of the superpixels. From this matrix, the superpixels are then clustered using a similarity propagation algorithm (AP algorithm), and the saliency of each class is evaluated by measuring the inter-class color contrast, the compactness of the class, and the off-center. And finally, comparing the color difference and the position relation between the pixel and each type to update the significance of the pixel, and finally obtaining a first significance map with pixel precision and full resolution.
And step S1-5, detecting an object in front of the automobile in the driving process of the automobile by using the image sensor to obtain an object picture.
And step S1-6, extracting the image of the object picture according to the set characteristic points to obtain an object characteristic point image. Wherein the presence of water, railing, etc. features is represented by "1" and the absence of such features is represented by "0"; the characteristics of the slope range of the road, the width of the road, the height of the bridge and the like are expressed by one-hot coding.
And step S1-7, processing the object feature point image by a salient region detection method to obtain an object salient image. And (3) dividing the object feature point image into 200 super pixels by adopting an SLIC algorithm, and obtaining a matrix representing the similarity between the super pixels according to the color difference, the fractal difference and the spatial distribution of the super pixels. From this matrix, the superpixels are then clustered using a similarity propagation algorithm (AP algorithm), and the saliency of each class is evaluated by measuring the inter-class color contrast, the compactness of the class, and the off-center. And finally, comparing the color difference and the position relation between the pixel and each type to update the significance of the pixel, and finally obtaining the object significance map with pixel precision and full resolution.
And step S1-8, respectively carrying out feature extraction on the object saliency image and the first saliency image and comparing and identifying the extraction results. Wherein, the process of carrying out the feature extraction is as follows: extracting Histogram of Oriented Gradient (HOG) features around the feature points, using a support vector machine (SVM classifier) to form a feature matrix by using the extracted features as matrix elements, and finally performing normalization processing on the feature matrix to eliminate the influence of environmental factors such as illumination and the like.
And S1-9, repeating the steps S1-5-S1-8 within set time, and accurately identifying by adopting a machine learning training model to obtain an identification model of the upper bridge. The machine learning training model comprises an Adaboost iterative algorithm and a harr classifier training method.
Step S2, establishing a lower bridge identification model:
step S2-1, the type of the characteristic point of the bridge is set. In this embodiment, the feature points of the bridge in the identification model of the lower bridge at least include land changes, railing changes, and road surface depressions.
And step S2-2, carrying out image extraction on the collected data set according to the set characteristic points to obtain a second characteristic point image. Wherein the presence of features such as road surface depressions is represented by "1", and the absence of such features is represented by "0"; characteristics such as land change, railing change and the like are represented by one-hot codes.
And step S2-3, processing the second feature point image by a salient region detection method to obtain a second salient image. The specific process comprises the following steps: and (3) dividing the second feature point image into 200 super pixels by adopting an SLIC algorithm, and obtaining a matrix representing the similarity between the super pixels according to the color difference, the fractal difference and the spatial distribution of the super pixels. From this matrix, the superpixels are then clustered using a similarity propagation algorithm (AP algorithm), and the saliency of each class is evaluated by measuring the inter-class color contrast, the compactness of the class, and the off-center. And finally, comparing the color difference and the position relation between the pixel and each type to update the significance of the pixel, and finally obtaining a second significance map with pixel precision and full resolution.
And step S2-4, detecting an object in front of the automobile in the driving process of the automobile by using the image sensor to obtain an object picture.
And step S2-5, extracting the image of the object picture according to the set characteristic points to obtain an object characteristic point image. Wherein the presence of features such as road surface depressions is represented by "1", and the absence of such features is represented by "0"; characteristics such as land change, railing change and the like are represented by one-hot codes.
And step S2-6, processing the object characteristic point image by a salient region detection method to obtain an object salient image, dividing the object characteristic point image into 200 superpixels by adopting an SLIC algorithm, and obtaining a matrix representing the similarity between the superpixels according to the color difference, the fractal difference and the spatial distribution of the superpixels. From this matrix, the superpixels are then clustered using a similarity propagation algorithm (AP algorithm), and the saliency of each class is evaluated by measuring the inter-class color contrast, the compactness of the class, and the off-center. And finally, comparing the color difference and the position relation between the pixel and each type to update the significance of the pixel, and finally obtaining the object significance map with pixel precision and full resolution.
And step S2-7, respectively carrying out feature extraction on the object saliency image and the second saliency image and comparing and identifying the extraction results. Wherein, the process of carrying out the feature extraction is as follows: extracting Histogram of Oriented Gradient (HOG) features around the feature points, using a support vector machine (SVM classifier) to form a feature matrix by using the extracted features as matrix elements, and finally performing normalization processing on the feature matrix to eliminate the influence of environmental factors such as illumination and the like.
And S2-8, repeating the steps S2-4-S2-7 within set time, and accurately identifying the training model by machine learning to obtain the identification model of the lower bridge. The machine learning training model comprises an Adaboost iterative algorithm and a harr classifier training method.
Step S3, establishing an optimization model of the ideal vehicle speed when the automobile is off-axle:
and step S3-1, detecting an object in front of the automobile by using an image sensor to obtain an object picture, and detecting an object in front of the automobile by using a gradient sensor to obtain gradient data.
And step S3-2, extracting the image of the object picture according to the feature points set in the identification model of the upper bridge and the identification model of the lower bridge to obtain a feature point image. Wherein, the characteristics of water, railings, concave pavement and the like are represented by '1', and the absence of the characteristics is represented by '0'; the gradient range of the road, the width of the road, the height of the bridge, the land change, the railing change and other characteristics are represented by one-hot codes.
And step S3-3, extracting the characteristic of the characteristic points of the characteristic point image to obtain image data. Wherein, the process of carrying out the feature extraction is as follows: extracting Histogram of Oriented Gradient (HOG) features around the feature points, using a support vector machine (SVM classifier) to form a feature matrix by using the extracted features as matrix elements, and finally performing normalization processing on the feature matrix to eliminate the influence of environmental factors such as illumination and the like.
And step S3-4, when the automobile gets off the bridge, controlling the automobile to decelerate and recording the target speed when the passengers do not have weightlessness feeling.
And step S3-5, when the automobile gets off the bridge, controlling the automobile to decelerate to the target speed, obtaining the weight of the image data and the gradient data at the moment and updating the target speed according to the weight.
And S3-6, repeating the step 3-5 and carrying out model training on the image data and the gradient data through an artificial neural network (BP) algorithm to obtain the optimal weight of the image data and the gradient data.
Two inputs (image data X1 and gradient data X2) of the BP neural network get one output through the three-layer neural network.
FIG. 2 is a schematic diagram of model training of image data and slope data by an artificial neural network in an embodiment of the invention.
As shown in fig. 2, in the process of training the neural networkThe middle inputs are X1 and X2. The first hidden layer function is set to f1(e)、f2(e)、f3(e) The second hidden layer function is set to f4(e)、f5(e) The third hidden layer function is set to f6(e)。
f1(e) The weight corresponding to a node is w(X1)1、w(X2)1Then there is y1＝f1(w(X1)1X1+w(X2)1X2)；f2(e) The weight corresponding to a node is w(X1)2、w(X2)2Then there is y2＝f2(w(X1)2X1+w(X2)2X2)；f3(e) The weight corresponding to a node is w(X1)3、w(X2)3Then there is y3＝f3(w(X1)3X1+w(X2)3X2). Wherein, y1、y2、y3Representing the three outputs of the first hidden layer, respectively.
f4(e) The weight corresponding to a node is w14、w24、w34Then there is y4＝f4(w14y1+w24y2+w34y3)；f5(e) The weight corresponding to a node is w15、w25、w35Then there is y5＝f5(w15y1+w25y2+w35y3). Wherein, y4、y5Respectively representing the two outputs of the second hidden layer.
f6(e) The weight corresponding to a node is w46、w56Then there is y6＝f6(w46y4+w56y5). To obtain an output y6。
Step S4, controlling the speed of the automobile when passing through the bridge to be tested:
and step S4-1, detecting an object in front of the automobile by using the image sensor to obtain a picture of the object to be detected, and detecting the object in front of the automobile by using the gradient sensor to obtain gradient data to be detected.
And step S4-2, extracting the image of the object picture according to the feature points set in the upper bridge identification model and the lower bridge identification model to obtain the feature point image to be detected. Wherein, the characteristics of water, railings, concave pavement and the like are represented by '1', and the absence of the characteristics is represented by '0'; the gradient range of the road, the width of the road, the height of the bridge, the land change, the railing change and other characteristics are represented by one-hot codes.
And step S4-3, performing feature extraction on the feature points in the picture of the object to be detected to obtain image data to be detected. Wherein, the process of carrying out the feature extraction is as follows: extracting Histogram of Oriented Gradient (HOG) features around the feature points, using a support vector machine (SVM classifier) to form a feature matrix by using the extracted features as matrix elements, and finally performing normalization processing on the feature matrix to eliminate the influence of environmental factors such as illumination and the like.
And step S4-4, comparing the image data to be detected with the identification model of the upper bridge and the identification model of the lower bridge respectively to judge whether the automobile is at the lower bridge. Meanwhile, whether the automobile is on the lower axle or not is judged by combining the change condition of the gradient data to be detected.
And step S4-5, obtaining the ideal vehicle speed required by the automobile when getting off the bridge according to the judgment result of yes judgment, the gradient data to be measured, the image data to be measured and the optimal weight.
And step S4-6, reducing the vehicle speed to the ideal vehicle speed in a braking deceleration mode.
Effects and effects of the embodiments
According to the automobile weightlessness detection and control method based on machine learning related to the embodiment, because the method firstly establishes an upper axle identification model and a lower axle identification model, then establishes an ideal speed optimization model when the automobile is at the lower axle, and finally controls the speed when the automobile passes through the bridge to be detected, the method can enable the automobile to identify whether the automobile is at the lower axle or not by comparing the actually detected image of the object to be detected and the gradient data to be detected with the upper axle identification model and the lower axle identification model, obtains the ideal speed required by the actual lower axle according to the ideal speed optimization model when the automobile is at the lower axle, and finally automatically decelerates to the ideal speed to avoid discomfort of passengers when the passengers get off the axle, so that the speed adjustment is more accurate compared with manual deceleration.
In addition, machine learning training models are used for recognizing the upper axle and the lower axle in the process of establishing the identification model of the upper axle and the identification model of the lower axle and in the process of controlling the speed of the automobile when the automobile passes through the axle to be detected, so that the obtained recognition result is more accurate. In the process of establishing an optimal model of ideal vehicle speed when the vehicle is off-axle, the artificial neural network algorithm is used for carrying out model training on the image data and the gradient data to obtain the optimal weight, and the calculation result is more accurate.
In addition, because the identification model of the upper bridge and the identification model of the lower bridge are established in advance, whether the automobile is in the bridge passing state or not can be judged only by comparing the characteristics of the automobile and the pre-established models when the automobile passes through the bridge to be tested, so that the operation amount is greatly simplified, the operation time is shortened, and the burden of an automobile processor is reduced.
In addition, because the method establishes the identification model of the lower bridge, the possibility of mistakenly judging objects such as an overhead bridge and the like as the bridge is avoided, and the accuracy rate of identifying the bridge is improved.
Further, whether the automobile is off-bridge or not can be judged by combining the change condition of the gradient data to be detected, and when the automobile is on-bridge, the gradient data is a positive value; when the automobile is off the axle, the gradient data is a negative value. The combined discrimination method can improve the discrimination accuracy.
The above embodiments are preferred examples of the present invention, and are not intended to limit the scope of the present invention.

Claims (5)

1. A vehicle weightlessness detection and control method based on machine learning is used for controlling the deceleration of a vehicle to avoid passengers from generating weightlessness feeling when the vehicle passes a bridge, and is characterized by comprising the following steps:
step 1, establishing an identification model of an upper bridge:
step 1-1, collecting a predetermined number of positive and negative examples of pictures about a bridge to obtain a collected data set,
step 1-2, setting the type of the characteristic points of the bridge,
step 1-3, extracting images of the collected data set according to the set characteristic points to obtain a first characteristic point image,
step 1-4, processing the first feature point image by a salient region detection method to obtain a first salient image,
step 1-5, detecting an object in front of the automobile in the driving process of the automobile to obtain an object picture,
step 1-6, extracting the image of the object picture according to the set characteristic point to obtain an object characteristic point image,
step 1-7, processing the object characteristic point image by a salient region detection method to obtain an object salient image,
step 1-8, respectively carrying out feature extraction on the object saliency image and the first saliency image and comparing and identifying extraction results,
1-9, repeating the steps 1-5-1-8 within a set time, and accurately identifying by adopting a machine learning training model to obtain an identification model of the upper bridge;
step 2, establishing an identification model of the lower bridge:
step 2-1, setting the type of the characteristic points of the bridge,
step 2-2, extracting images of the collected data set according to the set characteristic points to obtain a second characteristic point image,
step 2-3, processing the second characteristic point image by a salient region detection method to obtain a second salient image,
step 2-4, detecting the object in front of the automobile in the driving process of the automobile to obtain an object picture,
step 2-5, extracting the image of the object picture according to the set characteristic points to obtain an object characteristic point image,
step 2-6, processing the object characteristic point image by a salient region detection method to obtain an object salient image,
step 2-7, respectively carrying out feature extraction on the object saliency image and the second saliency image and comparing and identifying extraction results,
step 2-8, repeating step 2-4 to step 2-7 within a set time, and accurately identifying the training model by machine learning to obtain an identification model of the lower bridge;
step 3, establishing an optimization model of the ideal speed when the automobile is off-axle:
step 3-1, detecting the object in front of the automobile to obtain an object picture and gradient data,
step 3-2, extracting the characteristic points in the object picture to obtain image data,
step 3-3, when the automobile gets off the bridge, controlling the automobile to decelerate and recording the target speed when the passenger has no weightlessness feeling,
step 3-4, when the automobile gets off the bridge, the automobile is controlled to be decelerated to the target speed, the weight of the image data and the gradient data at the moment is obtained, the target speed is updated according to the weight,
3-5, repeating the step 3-4, and performing model training on the image data and the gradient data through an artificial neural network algorithm to obtain the optimal weight of the image data and the gradient data;
step 4, controlling the speed of the automobile when passing through the bridge to be measured,
step 4-1, detecting the object in front of the automobile to obtain the image of the object to be detected and the data of the gradient to be detected,
step 4-2, extracting the characteristic points in the picture of the object to be detected to obtain the image data to be detected,
step 4-3, comparing the image data to be detected with the identification model of the upper bridge and the identification model of the lower bridge respectively to judge whether the automobile is at the lower bridge or not,
4-4, obtaining the ideal vehicle speed required by the automobile to get off the axle according to the judgment result of yes judgment, the gradient data to be detected, the image data to be detected and the optimal weight;
step 4-5, reducing the vehicle speed to the ideal vehicle speed in a braking and decelerating way,
wherein, in the step 1-2, the characteristic points of the bridge at least comprise water, a railing, the gradient range of the road, the width of the road and the height of the bridge when the identification model of the upper bridge is established,
in the step 2-1, the characteristic points of the bridge at least comprise land change, railing change and road surface recess when the identification model of the lower bridge is established.
2. The machine learning-based automobile weight loss detection and control method according to claim 1, characterized in that:
and 4-3, judging whether the automobile is on the lower axle or not according to the change condition of the gradient data to be detected.
3. The machine learning-based automobile weight loss detection and control method according to claim 1, characterized in that:
wherein the object image and the object image to be measured are both obtained by detecting an object in front of the automobile using an image sensor,
the slope data and the slope data that awaits measuring all obtain through using the slope sensor to detect the object in front of the car.
4. The machine learning-based automobile weight loss detection and control method according to claim 1, characterized in that:
in steps 1-8, 2-7, 3-2 and 4-2, the method for performing feature extraction includes histogram feature extraction and SVM classifier processing.
5. The machine learning-based automobile weight loss detection and control method according to claim 1, characterized in that:
in steps 1-9 and steps 2-8, the machine learning method includes an Adaboost iterative algorithm and a harr classifier training method.
CN201810789221.XA 2018-07-18 2018-07-18 Automobile weightlessness detection and control method based on machine learning Active CN109101980B (en)

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 CN109101980A (en) 2018-12-28
CN109101980B true 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)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN114412988B (en) * 2022-02-11 2022-11-29 吉林大学 Uphill predictive gear shifting control method based on road information and machine learning

Citations (3)

* Cited by examiner, † Cited by third party
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
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

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1662989B1 (en) * 2000-06-16 2014-09-03 BodyMedia, Inc. System for monitoring and managing body weight and other physiological conditions including iterative and personalized planning, intervention and reporting capability

Patent Citations (3)

* Cited by examiner, † Cited by third party
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
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)

* Cited by examiner, † Cited by third party
Title
考虑道路几何特征的车速自适应控制方法研究;张蕊;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20150415;全文 *

Also Published As

Publication number Publication date
CN109101980A (en) 2018-12-28

Similar Documents

Publication Publication Date Title
CN109101980B (en) Automobile weightlessness detection and control method based on machine learning
CN108537197A (en) A kind of lane detection prior-warning device and method for early warning based on deep learning
CN104573646B (en) Chinese herbaceous peony pedestrian detection method and system based on laser radar and binocular camera
CN108609019B (en) Automatic electric vehicle driving method based on artificial intelligence platform
CN105745131B (en) Travel controlling system, server, car-mounted device
CN105206109B (en) A kind of vehicle greasy weather identification early warning system and method based on infrared CCD
CN106503636B (en) A kind of road sighting distance detection method and device of view-based access control model image
CN109334563B (en) Anti-collision early warning method based on pedestrians and riders in front of road
CN109941288A (en) Safe driving auxiliary device and method
CN107463890B (en) A kind of Foregut fermenters and tracking based on monocular forward sight camera
CN103839264A (en) Detection method of lane line
CN103714660A (en) System for achieving fatigue driving judgment on basis of image processing and fusion between heart rate characteristic and expression characteristic
CN104182729B (en) Pedestrian detection method based on ARM embedded platforms
CN110458050B (en) Vehicle cut-in detection method and device based on vehicle-mounted video
CN112802356B (en) Vehicle automatic driving method and terminal based on Internet of things
CN112070823A (en) Video identification-based automobile intelligent cabin adjusting method, device and system
CN110097783A (en) Vehicle early warning method and system
WO2019213982A1 (en) Driver control behavior quantification method and device employing principle of least action
Karaduman et al. Deep learning based traffic direction sign detection and determining driving style
CN110991264A (en) Front vehicle detection method and device
CN108108703A (en) Deceleration strip missing detection method, device and electronic equipment
CN114708567A (en) Road surface depression detection and avoidance method and system based on binocular camera
CN114559961A (en) Method and device for avoiding obstacles on road surface of automatic driving vehicle and readable storage medium
CN109815836A (en) A kind of urban road surfaces guiding arrow detection recognition method
Karungaru et al. Road traffic signs recognition using genetic algorithms and neural networks

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