CN111055849B - Intersection intelligent driving method and system based on support vector machine - Google Patents

Intersection intelligent driving method and system based on support vector machine Download PDF

Info

Publication number
CN111055849B
CN111055849B CN201811206747.7A CN201811206747A CN111055849B CN 111055849 B CN111055849 B CN 111055849B CN 201811206747 A CN201811206747 A CN 201811206747A CN 111055849 B CN111055849 B CN 111055849B
Authority
CN
China
Prior art keywords
support vector
vector machine
vehicle
time
sequence
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
CN201811206747.7A
Other languages
Chinese (zh)
Other versions
CN111055849A (en
Inventor
许琮明
王正贤
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.)
Automotive Research and Testing Center
Original Assignee
Automotive Research and Testing Center
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 Automotive Research and Testing Center filed Critical Automotive Research and Testing Center
Priority to CN201811206747.7A priority Critical patent/CN111055849B/en
Publication of CN111055849A publication Critical patent/CN111055849A/en
Application granted granted Critical
Publication of CN111055849B publication Critical patent/CN111055849B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • 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
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W2050/0001—Details of the control system
    • B60W2050/0019—Control system elements or transfer functions
    • B60W2050/0028—Mathematical models, e.g. for simulation

Abstract

The invention provides an intelligent intersection driving method and system based on a support vector machine. In the support vector machine providing step, providing a support vector machine, wherein the support vector machine is subjected to a training process in advance, and in the training process, training data are provided for the support vector machine, and the training data are obtained by processing original data through a dimensionality reduction module and a time compensation module; in the data processing step, the p characteristics acquired by the environment sensing unit are provided to a support vector machine for classification after being processed by a dimensionality reduction module and a time compensation module; in the decision step, the driving behavior of the vehicle is determined according to the classification result of the support vector machine. Therefore, the decision accuracy of the support vector machine can be effectively improved.

Description

Intersection intelligent driving method and system based on support vector machine
Technical Field
The present invention relates to an intelligent driving method and system for an intersection, and more particularly, to an intelligent driving method and system for an intersection based on a support vector machine.
Background
Generally, a crossroad or a road intersection has a multi-directional vehicle turning or straight intersection, and therefore, when passing through the intersection, it is necessary to perform acceleration, deceleration or constant speed driving according to driving judgment, and once the driving judgment is wrong, a traffic accident occurs. According to the statistics of the American statistical agency, the proportion of traffic accidents at crossroads or road intersections in 2008 is as high as 40%; according to the statistics of the German Federal statistical office, the proportion of traffic accidents occurring at the crossroad or road intersection of 2013 is up to 47.5%, and in some countries, the proportion of traffic accidents is even up to 98%.
To assist decision making when driving through intersections, Highly Automated Vehicles (HAV) have been developed by the industry/scholars, which include artificial intelligence for machine learning to assist driving decisions, and support vector machines are one of the machine learning methods that can be predicted or estimated by building models for decision making. Such as when to accelerate, decelerate, or settle at an intersection.
In practical situations, driving can make driving behaviors suitable for the current situation according to surrounding information in the past several units of time, that is, the real driving situation has a time dependence relationship. However, in the training process of the conventional support vector machine, although the data for training is recorded according to the continuous time points, the data observed at each time is regarded as an independent data, and the time dependency of the variables is not considered, so that the determined decision accuracy still has room for improvement.
Therefore, how to effectively improve the decision accuracy of the support vector machine is a goal of the related manufacturers' efforts.
Disclosure of Invention
The invention provides an intelligent intersection driving method and system based on a support vector machine, which can effectively improve the decision accuracy of the support vector machine after dimension reduction processing and time value supplementing processing.
According to an embodiment of an aspect of the present invention, a support vector machine-based intelligent driving method for an intersection is provided, which is applied to a vehicle and includes a support vector machine providing step, a data processing step and a decision step. Providing a support vector machine in a support vector machine providing step, wherein the provided support vector machine is subjected to a training process in advance, and in the training process, training data is provided for the support vector machine, the training data is obtained by processing original data through a one-dimensional reduction module and a time compensation module, wherein the original data comprises a plurality of training samples, each training sample comprises a time total value passing through an intersection, p characteristics corresponding to each sampling time point in a plurality of sampling time points in the time total value and a current decision; the dimensionality reduction module integrates p features into k new features, the time complementation module provides preset time, the time complementation module respectively considers the new features corresponding to any sampling time point in any training sample and other sampling time points before any sampling time point as a to-be-expanded sequence, when the length of the to-be-expanded sequence is less than the number of sampling time points in the preset time, a new to-be-expanded sequence is formed after a pre-estimated value is complemented into the to-be-expanded sequence, the combined distribution of the to-be-expanded sequence which is not complemented into the pre-estimated value and the conditional distribution of the pre-estimated value obtained by the data of all new features subject to Gaussian distribution are adopted, the time complementation module reforms the to-be-expanded sequence into an expanded sequence, the length of the expanded sequence is equal to the number of sampling time points in the preset time, wherein p and k are positive integers, and p is greater than k; in the data processing step, p characteristics acquired by an environment sensing unit are processed by a dimensionality reduction module and a time compensation module and then are provided for a support vector machine to be classified; in the decision step, the driving behavior of the vehicle is determined according to the classification result of the support vector machine.
Therefore, after the training data and the characteristics acquired currently by driving are processed by the dimensionality reduction module and the time compensation module, time dependence can be achieved, and the accuracy of the prediction result can be improved.
According to various embodiments of the foregoing intersection intelligent driving method based on the support vector machine, the dimension reduction module may employ a principal component analysis method. Or the time compensation module may employ a uniform scaling method. Or the preset time may be equal to the maximum of the total value of time.
According to various embodiments of the foregoing intelligent driving method for an intersection based on a support vector machine, the p features may include a lateral speed of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal speed of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance between the vehicle and the intersection, and a speed of the incoming vehicle. The plurality of features in the raw data may be obtained by an environment sensing unit comprising at least one of a radar, a camera and a GPS positioning device.
According to another aspect of the present invention, an intelligent driving system for a vehicle comprises a processing unit and an environment sensing unit, wherein the processing unit is used for processing the environment sensed by the environment sensing unit; the processing unit is arranged on the vehicle and comprises a dimension reduction module, a time compensation module and a support vector machine. The dimensionality reduction module integrates p features corresponding to each sampling time point in the multiple sampling time points into k new features, wherein p and k are positive integers, and p > k; the time compensation module provides a preset time, the time compensation module takes a new characteristic corresponding to any sampling time point and other sampling time points before the sampling time point as a to-be-expanded sequence, and when the length of the to-be-expanded sequence is less than the number of sampling time points in the preset time, a pre-estimated value is added to the to-be-expanded sequence to form a new to-be-expanded sequence, wherein the condition distribution of the pre-estimated value is solved by the joint distribution of the to-be-expanded sequence which is not added with the pre-estimated value and the data of the new characteristics obeying Gaussian distribution, and the time compensation module reforms the new to-be-expanded sequence into an expanded sequence, and the length of the expanded sequence is equal to the number of sampling time points in the preset time; the support vector machine is trained by training data, the training data is obtained by processing original data by a dimensionality reduction module and a time complementation module, the original data comprises a plurality of training samples, each training sample comprises a time total value passing through an intersection, p characteristics corresponding to each sampling time point in the time total value and a current decision. The environment sensing unit is arranged on the vehicle and is connected with the processing unit through signals, and the environment sensing unit is used for acquiring p characteristics; the p features acquired by the environment sensing unit are processed by the dimension reduction module and the time compensation module of the processing unit and then provided to the support vector machine for classification, and the classification result of the support vector machine is used for determining the driving behavior of the vehicle.
According to the foregoing embodiments of the intersection intelligent driving system based on the support vector machine, the p features include: a lateral velocity of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal velocity of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance of the vehicle from the intersection, and a velocity of the incoming vehicle. Or the environment sensing unit may comprise at least one of a radar, a camera and a GPS positioning device. Or the current decision may comprise at least one of acceleration, deceleration or pacing.
Drawings
FIG. 1 is a flow chart of a method for intelligent driving at an intersection based on a support vector machine according to a first embodiment of the invention;
FIG. 2 illustrates a first simulated training of the support vector machine-based intelligent driving method of an intersection according to FIG. 1;
FIG. 3 illustrates a second simulated training of the support vector machine based intelligent driving method for intersections according to FIG. 1;
FIG. 4 illustrates a third simulated training of the support vector machine based intelligent driving method for intersections according to FIG. 1;
FIG. 5 illustrates a first accumulation rate for the first simulated training of FIG. 2;
FIG. 6 illustrates a first accumulation rate for the second simulated training of FIG. 3;
FIG. 7 depicts a first accumulation rate of the third simulated training of FIG. 4; and
fig. 8 is a block diagram of an intelligent driving system at an intersection based on a support vector machine according to a third embodiment of the invention.
Detailed Description
Embodiments of the present invention will be described below with reference to the accompanying drawings. For the purpose of clarity, numerous implementation details are set forth in the following description. However, the reader should understand that these implementation details should not be used to limit the invention. That is, in some embodiments of the invention, these implementation details are not necessary. In addition, for the sake of simplicity, some conventional structures and elements are shown in the drawings in a simplified schematic manner; and repeated elements will likely be referred to using the same reference number or similar reference numbers.
Referring to fig. 1, fig. 1 is a flowchart illustrating a crossing intelligent driving method 100 based on a support vector machine according to a first embodiment of the invention. The intersection intelligent driving method 100 based on the support vector machine is applied to a vehicle and comprises a support vector machine providing step 110, a data processing step 120 and a decision step 130.
In the support vector machine providing step 110, a support vector machine is provided, the support vector machine is previously subjected to a training process, in the training process, training data is provided to the support vector machine, the training data is obtained by processing an original data through a one-dimensional reduction module and a time compensation module, wherein the original data comprises a plurality of training samples, each training sample comprises a time total value passing through an intersection, p characteristics corresponding to each sampling time point in a plurality of sampling time points in the time total value and a current decision, the dimension reduction module integrates the p characteristics into k new characteristics, the time compensation module provides a preset time, the time compensation module considers the new characteristics corresponding to any sampling time point in any training sample and other sampling time points before any sampling time point as an expansion sequence, and the time compensation module reforms the expansion sequence into an expansion sequence, the length of the expansion sequence is equal to the number of sampling time points in a preset time, wherein p and k are positive integers, and p > k.
In the data processing step 120, the p features obtained by an environment sensing unit are processed by the dimensionality reduction module and the time interpolation module, and then provided to the support vector machine for classification.
In the decision step 130, the driving behavior of the vehicle is determined according to the classification result of the support vector machine.
Therefore, after the training data and the characteristics acquired currently by driving are processed by the dimensionality reduction module and the time compensation module, time dependence can be achieved, and the accuracy of the prediction result can be improved. Details of the intersection intelligent driving method 100 based on the support vector machine will be described later.
A support vector machine is a supervised machine learning classifier that can be used to assist in determining the behavior of a vehicle. In the support vector machine providing step 110, the provided support vector machine is trained through a training process, and can determine the deceleration, acceleration or constant speed behavior of the vehicle when passing through the intersection.
In the training process, a plurality of training samples can be formed by simulating the condition that the vehicle passes through the intersection in a simulation mode. In a first embodiment, the simulation platform may be a Prescan Advanced Driver Assistance Systems (ADAS) developed by Tass International, Inc. that can build relevant intersection information to simulate the passage of vehicles through an intersection. In other embodiments, training samples may be obtained on the actual road, or other simulation software may be used, but not limited thereto.
The data obtained by the vehicle passing the intersection once can be regarded as a training sample. That is, ten training samples may be taken when the vehicle passes the intersection ten times. Each training sample includes a total time value for passing through the intersection, p features corresponding to each of a plurality of sampling points within the total time value, and a current decision. For example, assuming that in the 1 st training sample, the total time passing through the intersection is 2 seconds, and the sampling is performed every 0.4 seconds, there are 5 sampling time points, and each sampling time point will collect p features and a next decision, where the next decision can be acceleration, deceleration or constant speed, and the p features can include a lateral speed of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal speed of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance between the vehicle and the intersection, and a speed of the incoming vehicle.
The data for the 1 st training sample may be as shown in table 1. In a single training sample, p features obtained from q sampling points can form an original feature matrix X (X ═ X)1,…,xp) Wherein x isi＝(xi1,..,xiq)TIt should be understood by the reader that when there are n training samples, there are n primitive feature matrices XlCorresponding to different sampling time point numbers qlN and q are positive integers, l is a positive integer from 1 to n, and i is a positive integer from 1 to p. All current decisions in the n training samples form a current decision matrix ZZ, ZZ ═ z (z)1,…,zn). Hereinafter, TlwRepresents the w sampling time point in the l training sample, w is 1 to qlPositive integer of (1), xlwiRepresenting the time T of samplinglwThe ith feature obtained, zlwRepresenting the time T of samplinglwThe current decision obtained. Thus, T in Table 111The 1 st sampling time point in the 1 st training sample is 0.4 seconds, T in the first embodiment12The 2 nd sampling point in the 1 st training sample is 0.8 second, x in the first embodiment111Represents the 1 st sampling time point T in the 1 st training sample 111 st feature, x, obtained122Is represented in the 1 st training sampleMiddle 2 sampling time point T12The obtained 2 nd feature, z13Represents the 3 rd sampling time point T in the 1 st training sample13The obtained current decision is analogized, and the description is not repeated.
TABLE 1 st training sample
Figure GDA0002801333490000061
In the 2 nd training sample, the total time taken to pass through the intersection is 2.4 seconds, and samples are taken at intervals of 0.4 seconds, for a total of 6 sampling points. The data for the 2 nd training sample may be as shown in table 2.
TABLE 2 nd training sample
Figure GDA0002801333490000062
Assuming there are only 2 training samples, the raw data includes the data of tables 1 and 2.
The original data can be converted into training data after being processed by the dimensionality reduction module and the time compensation module. The dimension reduction module may use Principal Component Analysis (PCA), Partial Least Squares Regression (PLSR), Multidimensional Scaling (MDS), Projection Pursuit (Projection Pursuit method), Principal Component Regression (PCR), Quadratic Discriminant Analysis (QDA), normalized Discriminant Analysis (RDA), Linear Discriminant Analysis (LDA), or the like. Preferably, the time compensation module uses a principal component analysis method, and the correlation formulas of the principal component analysis method are shown in formula (1), formula (2) and formula (3).
Y＝aTX (1)。
Figure GDA0002801333490000071
Figure GDA0002801333490000072
The above formula is based on a sampling time point in a training sample, so the variable l representing the number of training samples and the variable w representing the sampling time point are not included. Wherein Y represents the integrated new feature matrix, which contains k new features, i.e., Y ═ Y (Y)1,....,yk) Wherein y isjRepresents the jth new feature, j being a positive integer from 1 to k. When considering n training samples and the corresponding sampling time point number qlWhen, Yl＝(y11,....,ylk)，
Figure GDA0002801333490000074
a is a coefficient matrix, ajiDenotes the ith feature xiThe corresponding coefficient. In the first embodiment, the dimensionality reduction module integrates p features in each training sample into 1 new feature, that is, k is 1. Therefore, the 1 st training sample processed by the dimensionality reduction module is shown in table 3, and the 2 nd training sample processed by the dimensionality reduction module is shown in table 4. Wherein, yljwRepresenting the corresponding sampling time T after reforminglwThe jth new feature of (a), the data processed by the dimensionality reduction module is (Y)l,zl)。
TABLE 3 1 st training sample processed by dimensionality reduction Module
Figure GDA0002801333490000073
Figure GDA0002801333490000081
TABLE 4 2 training samples processed by the dimensionality reduction module
Sampling time point Y2 z2
T21 y211 z21
T22 y212 z22
T23 y213 z23
T24 y214 z24
T25 y215 z25
T26 y216 z26
Then, the data is processed by the time padding module to perform padding. Since the classification data inputted by the supervised classifier must be of the same length and array data, the accumulated data per unit time can be pulled to the same time length by the time complementation module.
The value complementing method of the time value complementing module can adopt a Dynamic Time Warping (DTW) method or a Uniform scaling (Uniform scaling) method. Preferably, the time compensation module adopts a uniform scaling method.
When the uniform scaling method is used, the time padding module may provide a preset time, wherein the preset time may be equal to the maximum of the total time value. That is, in the first embodiment, the total time value of the 1 st training sample is 2 seconds, the total time value of the 2 nd training sample is 2.4 seconds, and the maximum value is 2.4, so the predetermined time can be 2.4 seconds, and the number of sampling time points in the predetermined time is 6.
Before the value is complemented, the time complementing module considers new characteristics corresponding to any sampling time point in any training sample and other sampling time points before the sampling time point as an extended sequence. Table 5 is a list of numbers to be expanded, where LljwRepresenting the sequence of numbers to be expanded, Lljw＝(ylj1,…,yljw). If p features in each training sample are integrated into 1 new feature after passing through the dimensionality reduction module, j in table 7 is 1.
TABLE 5 List of numbers to be expanded
Figure GDA0002801333490000082
Figure GDA0002801333490000091
In more detail, in Table 5, the number to be expanded LljwIncluding the 1 st to w th samples of the l training samplePoint Tl1～TlwCorresponding all jth new features ylj1～yljw. For example, for the extension sequence L111Since w is 1 and j is 1, the sequence L is extended111Having a 1 st sampling time point T 111 st new feature y111And the sequence L to be expanded111Is 1, i.e. it consists of one number. Array L to be expanded213Since w is 3 and j is 1, the sequence L is extended213Having the 1 st to 3 rd sampling time points T in the 2 nd training sample21～T23Corresponding 3 1 st new features y211～y213And the sequence L to be expanded213Is 3, i.e. it consists of three values, and so on. Therefore, the sequence to be expanded can be reformed into an expanded sequence through the complementary value, and the length of the expanded sequence is equal to the number of sampling time points in the preset time. Since the number of sampling time points in the predetermined time is 6 in the first embodiment, the extended sequence is composed of 6 values. Thus, after being complemented by the time-complementing module, all the expansion series consist of 6 values, as shown in Table 6, where L* ljwRepresenting an expansion sequence. If p features in each training sample are integrated into 1 new feature after passing through the dimensionality reduction module, j in table 7 is 1.
TABLE 6 expansion number List
Figure GDA0002801333490000092
Figure GDA0002801333490000101
Extension sequence L in Table 6* ljw＝(L* ljw1,…,L* ljwr) Which is the sequence L to be expandedljwThrough the transformation of formula (4) and formula (5).
Figure GDA0002801333490000102
Figure GDA0002801333490000103
Wherein r1s，qsRepresents the number of sampling time points in a predetermined time, qsQ is not more than qlWhen the preset time is equal to the maximum of the total time value, qs＝max(ql)。
Figure GDA0002801333490000104
Represents a floor formula, and
Figure GDA0002801333490000105
z is an integer set, i.e., r × w/qsUnconditionally truncating only the integer.
For example, L113＝(y111,y112,y113) Is extended to L* 113＝(L* 1131,L* 1132,L* 1133,L* 1134,L* 1135,L* 1136) And L is* 1131＝y111，L* 1132＝y111(since 2x3/6 is 1, the choice will be at L113Y of position 1111Supplement L* 113Second position) of L* 1133＝y111(since unconditional truncation leaves only the integer 1 because 3x3/6 ═ 1.5, the choice will be at L113Y of position 1111Supplement L* 113Third position) of (c), L* 1134＝y112(since 4x3/6 is 2, the choice will be at L113Y of 2 nd position112Supplement L* 113Fourth position) of (d), L* 1135＝Y112(since unconditional truncation leaves only the integer 2, since 5x3/6 ═ 2.5, the choice will be toAt L113Position y of the 2 nd position112Is supplemented with L* 113Fifth position) of L), L* 1136＝y113(since 6x3/6 is 3, the choice will be at L113Position of the 3 rd position y113Is supplemented with L* 113The sixth position of (1). It should be noted that, when the length of the sequence to be expanded is greater than the number of sampling time points in the predetermined time, that is, the length of the sequence to be expanded is greater than the length of the expanded sequence after the reforming is expected, the reduction can still be performed according to the equations (4) and (5), so as to achieve the object of the present invention.
The data of the original data processed by the dimensionality reduction module is (Y)l,zl)，(Yl,zl) Processed by a time compensation module and converted into training data (L)* l,zl)，L* l＝(L* l1,…,L* lk)，
Figure GDA0002801333490000112
Figure GDA0002801333490000113
The training data is shown in Table 7, where the last column can maintain the original value because it does not need to be complemented, but for convenience of representing the relationship with the SVM, it is still expressed by L* ljwrAnd (4) showing. If p features in each training sample are integrated into 1 new feature after passing through the dimensionality reduction module, j in table 7 is 1.
TABLE 7 training data
Figure GDA0002801333490000111
The above training data can be provided to the support vector machine to find out the hyperplane, the relevant formulas of the support vector machine are shown as formulas (6) to (9), wherein the starting target of the support vector machine is shown as formula (6), formula (6) is substituted for formula (7), and formulas (6) and (7) are rewritten into formula (8) according to the calculus and the if principle.
Figure GDA0002801333490000114
Figure GDA0002801333490000115
Figure GDA0002801333490000121
Figure GDA0002801333490000122
Wherein C is penalty coefficient (cost variable) and is greater than 0, and W is element coefficient (entries parameter), ξlIs a relaxation parameter (slack variable), b is an intercept term parameter (intercept term), alphad、αeLagrange multipliers (lagrange multipliers), Φ radial basis function kernels (radial bias functions), which extend the hyperplane to nonlinear cuts. L is* l、L* dAnd L* eRepresents the extension value L mentioned above* ljwrFor simplicity, other variables are omitted, and d and e are both variables.
After the support vector machine passes through the training data, the hyperplane can be found to assist in decision making.
In the data processing step 120, when the vehicle passes through the intersection, the environment sensing unit may collect p features in real time, the environment sensing unit may include a plurality of sensing devices such as radar, camera, GPS positioning device, etc., the sensing devices may be used to detect distance, vehicle speed, etc., and may sense p features, but the types and number of the sensing devices are not limited thereto. The p features collected at each sampling time point of the vehicle are processed by the dimensionality reduction module and the time compensation module in the manner described above. Then, in the decision step 130, the processed data enters the support vector machine, since the support vector machine has previously passed through the training process and found the hyperplane, the p features obtained at the moment are processed and then enter the support vector machine, so as to generate the classification result, and the classification result is used to determine the driving behavior of the vehicle, such as deceleration, acceleration or constant speed.
Referring to fig. 2, fig. 3 and fig. 4, fig. 2 shows a first simulation training of the intelligent intersection driving method based on the support vector machine 100 according to fig. 1, fig. 3 shows a second simulation training of the intelligent intersection driving method based on the support vector machine 100 according to fig. 1, and fig. 4 shows a third simulation training of the intelligent intersection driving method based on the support vector machine 100 according to fig. 1. In the first simulation training, the second simulation training and the third simulation training, the vehicle V1 passes through a T-shaped intersection, and during the first simulation training, no vehicle comes on the transverse road R1; in the second simulation training, the left side of the transverse road R1 has an incoming vehicle V2; in the third simulation training, there is an oncoming vehicle V2 to the right of the lateral road R1.
The speed of the vehicle V1 is 40 kilometers per hour, and the speed of the vehicle V2 is between 15 kilometers per hour and 40 kilometers per hour. The 7 features gathered are a lateral velocity of vehicle V1 relative to incoming vehicle V2, a lateral acceleration of vehicle V1 relative to incoming vehicle V2, a longitudinal velocity of vehicle V1 relative to incoming vehicle V2, a longitudinal acceleration of vehicle V1 relative to incoming vehicle V2, a distance of vehicle V1 from incoming vehicle V2, a distance of vehicle V1 from the intersection, and a velocity of incoming vehicle V2. And the first simulation training, the second simulation training and the third simulation training respectively comprise 20 training samples, the current decision comprises the behaviors of deceleration, constant speed and acceleration, and the preset time of the first simulation training is 16.9 seconds, the preset time of the second simulation training is 28.8 seconds, and the preset time of the third simulation training is 21.7 seconds.
Referring to fig. 5, 6 and 7, fig. 5 shows a first accumulation rate of the first simulated training of fig. 2, fig. 6 shows a first accumulation rate of the second simulated training of fig. 3, and fig. 7 shows a first accumulation rate of the third simulated training of fig. 4. The average first accumulation rate of the first simulation training is 0.9619, the average first accumulation rate of the second simulation training is 0.7588, and the average first accumulation rate of the third simulation training is 0.8014, wherein the first accumulation rates are all above 0.7, which indicates that the information after dimensionality reduction has explained the original data as much as possible and meets the requirements.
Table 8 shows that the decision result Accuracy (AC) of the first comparative example is compared with that of the intersection intelligent driving method 100 based on the support vector machine in the present application under the same condition as the first simulation training, and it can be seen from table 8 that the decision accuracy of the intersection intelligent driving method 100 based on the support vector machine is higher. The first comparative example also uses a support vector machine for classification, but the difference is that the support vector machine of the first comparative example is trained only by raw data.
TABLE 8 decision accuracy comparison in the first simulation training scenario
Figure GDA0002801333490000131
Table 9 shows that, in the same situation as the second simulation training, the decision result Accuracy (AC) of the second comparative example is compared with that of the intersection intelligent driving method 100 based on the support vector machine in the present application, and it can be seen from table 9 that the decision accuracy of the intersection intelligent driving method 100 based on the support vector machine is higher. The second comparative example also uses the support vector machine for classification, but the difference is that the support vector machine of the second comparative example is trained only by the original data.
TABLE 9 decision accuracy comparison in the second simulated training scenario
Figure GDA0002801333490000141
Table 10 shows that the decision result Accuracy (AC) of the third comparative example is compared with that of the intersection intelligent driving method 100 based on the support vector machine in the present application under the same condition as the third simulation training, and it can be seen from table 10 that the decision accuracy of the intersection intelligent driving method 100 based on the support vector machine is higher. The third comparative example also uses a support vector machine for classification, but the difference is that the support vector machine of the third comparative example is trained only by the original data.
TABLE 10 decision accuracy comparison in the third simulation training scenario
Figure GDA0002801333490000142
Figure GDA0002801333490000151
It should be noted that all the above tests are simulation results of PreScan, but the tests can also be performed using actual roads.
In a second embodiment of the present invention, in the support vector machine providing step, the time padding module provides a predetermined time, the time padding module is a to-be-extended sequence according to new characteristics respectively corresponding to any sampling time point in any training sample and other sampling time points before the any sampling time point, and when the length of the to-be-extended sequence is less than the number of sampling time points within the predetermined time, a new to-be-extended sequence is formed after the to-be-extended sequence is padded into a pre-estimated value of a next sampling time point of the sampling time point, and the time padding module rearranges the new to-be-extended sequence into an extended sequence whose length is equal to the number of sampling time points within the predetermined time.
In more detail, assuming that the original data includes the data in tables 1 and 2, the predetermined time is 2.4 seconds, and the sequence of the number to be expanded is shown in table 5, because the cumulative sampling time points 0-T in table 511The length of the corresponding number sequence to be expanded is 1, which is less than the number (equal to 6) of sampling time points in the preset time, and the sampling time points are accumulated from 0 to T11Has not obtained the next sampling time T12Can be complemented by an estimated value y'1j2Corresponding to the sampling time point T12. Similarly, in table 5, the length of the sequence to be expanded corresponding to all the accumulated sampling time points is less than 6, so that the estimated values need to be respectively supplemented, and new values are to be obtainedExtended column L'ljwIt is shown in Table 11, wherein y'ljwRepresenting the estimated value. It should be noted that in the new expansion sequence of Table 11, the sampling time points 0-T are accumulated11、0～T25The length of the corresponding sequence to be expanded is 6, and no complementary value is needed, but for clearly showing the difference between the new sequence to be expanded and the sequence to be expanded, the sequence is still listed, and the invention is not limited thereby. In the second embodiment, the method for reforming the new to-be-expanded sequence into the to-be-expanded sequence and the training method of the support vector machine in the following are the same as those in the first embodiment, and are not repeated.
Table 11, new to be expanded number list
Figure GDA0002801333490000152
Figure GDA0002801333490000161
In a second embodiment, a joint assignment of the sequence to be expanded and all new features y are usedljwThe data of the data are subjected to Gaussian distribution, so that the conditional distribution of the estimated value can be obtained, and finally the estimated value can be obtained. Due to the above-mentioned new feature y after dimensionality reductionljwThe data can be adapted to comply with gaussian distribution (or gaussian random process), i.e. yljw～GP(μjw，∑jw) Wherein (mu)jw，∑jw) Can be estimated from the equations (10) and (11). Furthermore, other new features y before the estimated value can be determined from equation (12)ljwJoint distribution (equivalent to a sequence of numbers to be extended), where w in equation (12) is greater than 2, c in equation (14) refers to the c-th training sample, and m refers to the m-th training sample.
Figure GDA0002801333490000162
Figure GDA0002801333490000163
yij[1：(w-1)]≡(yij1，...，yij(w-1))～GP(μj[1：(w-1)]，∑j[1：(w-1)]) (12)。
(μj[1：(w-1)]，∑j[1：(w-1)]) Can be estimated from the equations (13), (14) and (15).
Figure GDA0002801333490000164
Figure GDA0002801333490000165
Figure GDA0002801333490000166
Finally, a conditional distribution (conditional distribution) for predicting the next unit time at the past time point, that is, from 0 to T at the cumulative sampling time point, can be obtained11Predicting sampling time T12The condition distribution of (3) is as shown in the formulas (16) and (17).
Figure GDA0002801333490000173
Figure GDA0002801333490000174
While
Figure GDA0002801333490000175
Can be composed of
Figure GDA0002801333490000176
Estimating, with conditional assignments, the assignment according to the conditionAn estimate of the next sampling time point is predicted (which may be generated according to the Monte Carlo concept).
Wherein the content of the first and second substances,
Figure GDA0002801333490000177
is yljw，ylj[1：w-1]The variance matrix (Covariance matrix) of (g,), GP (·,) is the abbreviation of gaussian random process, μ is the mean of gaussian random process, Σ is the variance matrix of gaussian random process (Covariance-variance matrix),
Figure GDA0002801333490000178
the quantities are estimated for the elements (elements) in the variance matrix.
Table 12 shows the decision result accuracy in the first simulated training scenario after the addition of the pre-estimate, table 13 shows the decision result accuracy in the second simulated training scenario after the addition of the pre-estimate, and table 14 shows the decision result accuracy in the third simulated training scenario after the addition of the pre-estimate. As can be seen from the results of tables 12, 13 and 14, the decision accuracy can be improved by adding the prediction value.
TABLE 12 decision accuracy in first simulation training scenario
Figure GDA0002801333490000171
TABLE 13 decision accuracy in second simulation training scenario
Figure GDA0002801333490000172
Figure GDA0002801333490000181
TABLE 14 decision accuracy in the third simulation training scenario
Figure GDA0002801333490000182
Referring to fig. 8, fig. 8 is a block diagram illustrating a support vector machine-based intersection intelligent driving system 200 according to a third embodiment of the invention. The intersection intelligent driving system 200 based on the support vector machine is applied to a vehicle and comprises a processing unit 210 and an environment sensing unit 220, wherein the processing unit comprises a dimension reduction module 211, a time compensation module 212 and a support vector machine 213, the dimension reduction module 211 integrates p features corresponding to each sampling time point in a plurality of sampling time points into k new features, wherein p and k are positive integers, and p > k; the time interpolation module 212 provides a preset time, the time interpolation module 212 regards new characteristics respectively corresponding to any sampling time point and other sampling time points before the sampling time point as a to-be-expanded sequence, and when the length of the to-be-expanded sequence is less than the number of sampling time points in the preset time, a new to-be-expanded sequence is formed after a pre-estimated value is added to the to-be-expanded sequence, and the time interpolation module 212 reforms the new to-be-expanded sequence into an expanded sequence, wherein the length of the expanded sequence is equal to the number of sampling time points in the preset time; the support vector machine 213 is trained with a training data, which is obtained by processing an original data through the dimensionality reduction module 211 and the time interpolation module 212, wherein the original data comprises a plurality of training samples, each training sample comprises a time total value passing through an intersection, p features corresponding to each sampling time point in the time total value and a current decision.
The environment sensing unit 220 is disposed on the vehicle and connected to the processing unit 210, and the environment sensing unit 220 is configured to obtain p features; the p features acquired by the environment sensing unit 220 are processed by the dimension reduction module 211 and the time interpolation module 212 of the processing unit 210, and then provided to the support vector machine 213 for classification, and the classification result of the support vector machine 213 is used to determine the driving behavior of the vehicle.
Therefore, the vehicle can be assisted to determine the acceleration, deceleration or constant speed behaviors when passing through the intersection. The processing details of the dimension reduction module 211 and the time padding module 212 and the relationship between the processing details and the support vector machine 213 are as described above, and are not described herein again. The p features may include a lateral velocity of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal velocity of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance of the vehicle from the intersection, and a velocity of the incoming vehicle. The environment sensing unit 220 may comprise at least one of a radar, a camera, and a GPS positioning device. Or the current decision may comprise at least one of acceleration, deceleration or pacing.
Although the present invention has been described with reference to the above embodiments, it should be understood that various changes and modifications can be made therein by those skilled in the art without departing from the spirit and scope of the invention.

Claims (10)

1. An intelligent intersection driving method based on a support vector machine is applied to a vehicle, and is characterized by comprising the following steps:
a support vector machine providing step, providing a support vector machine, the support vector machine is pre-trained, in the training process, providing a training data to the support vector machine, the training data is obtained by processing an original data through a dimension reduction module and a time compensation module, wherein the original data comprises a plurality of training samples, each training sample comprises a time total value passing through an intersection, p characteristics corresponding to each sampling time point in a plurality of sampling time points in the time total value and a current decision, the dimension reduction module integrates the p characteristics into k new characteristics, the time compensation module provides a preset time, the time compensation module regards the new characteristics respectively corresponding to any sampling time point in any training sample and other sampling time points before any sampling time point as an expansion number series, when the length of the sequence to be expanded is less than the number of the sampling time points in the preset time, a new sequence to be expanded is formed after a pre-estimated value is added into the sequence to be expanded, wherein the conditional distribution of the pre-estimated value is obtained by the joint distribution of the sequence to be expanded without adding the pre-estimated value and the data of the new characteristics obeying Gaussian distribution, the time value adding module reforms the new sequence to be expanded into an expanded sequence, the length of the expanded sequence is equal to the number of the sampling time points in the preset time, wherein p and k are positive integers, and p is greater than k;
a data processing step, wherein the p characteristics acquired by an environment sensing unit are provided for the support vector machine to be classified after being processed by the dimensionality reduction module and the time complementation module; and
a decision step, deciding the driving behavior of the vehicle according to the classification result of the support vector machine.
2. The intelligent intersection driving method based on the support vector machine as claimed in claim 1, wherein the dimensionality reduction module adopts a principal component analysis method.
3. The intelligent intersection driving method based on the support vector machine as claimed in claim 1, wherein the time compensation module adopts a uniform scaling method.
4. The intelligent intersection driving method based on the support vector machine as claimed in claim 1, wherein the predetermined time is equal to the maximum of the total time values.
5. The intelligent intersection driving method based on a support vector machine as claimed in claim 1, wherein the p features include a lateral velocity of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal velocity of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance between the vehicle and the intersection, and a velocity of the incoming vehicle.
6. The method of claim 1, wherein the features of the raw data are obtained from the environment sensing unit, the environment sensing unit comprising at least one of a radar, a camera, and a GPS device.
7. An intelligent intersection driving system based on a support vector machine is applied to a vehicle, and is characterized by comprising the following components:
a processing unit disposed on the vehicle and comprising:
a dimension reduction module for integrating p features corresponding to each of the plurality of sampling time points into k new features, wherein p and k are positive integers, and p > k;
a time interpolation module, providing a preset time, regarding the new features respectively corresponding to any sampling time point and other sampling time points before the sampling time point as an to-be-expanded sequence, and forming a new to-be-expanded sequence after a pre-estimated value is added to the to-be-expanded sequence when the length of the to-be-expanded sequence is less than the number of the sampling time points in the preset time, wherein the condition distribution of the pre-estimated value is calculated by using the joint distribution of the to-be-expanded sequence which is not added with the pre-estimated value and the data of the new features to obey Gaussian distribution, and the time interpolation module reforms the new to-be-expanded sequence into an expanded sequence, and the length of the expanded sequence is equal to the number of the sampling time points in the preset time; and
a support vector machine, trained by a training data, the training data being obtained by processing an original data through the dimensionality reduction module and the time complementation module, the original data comprising a plurality of training samples, each training sample comprising a time total value passing through an intersection, the p features corresponding to each sampling time point in the time total value and a current decision; and
the environment sensing unit is arranged on the vehicle and is in signal connection with the processing unit;
the p features obtained by the environment sensing unit are processed by the dimension reduction module and the time compensation module of the processing unit and then provided to the support vector machine for classification, and the classification result of the support vector machine is used for determining the driving behavior of the vehicle.
8. The intelligent intersection driving system based on a support vector machine of claim 7, wherein the p features comprise a lateral velocity of the vehicle relative to an incoming vehicle, a lateral acceleration of the vehicle relative to the incoming vehicle, a longitudinal velocity of the vehicle relative to the incoming vehicle, a longitudinal acceleration of the vehicle relative to the incoming vehicle, a distance between the vehicle and the intersection, and a velocity of the incoming vehicle.
9. The support vector machine-based intersection intelligent driving system of claim 7, wherein the environment sensing unit comprises at least one of a radar, a camera and a GPS positioning device.
10. The support vector machine-based intersection intelligent driving system of claim 7, wherein the current decision comprises at least one of acceleration, deceleration or constant speed.
CN201811206747.7A 2018-10-17 2018-10-17 Intersection intelligent driving method and system based on support vector machine Active CN111055849B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811206747.7A CN111055849B (en) 2018-10-17 2018-10-17 Intersection intelligent driving method and system based on support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811206747.7A CN111055849B (en) 2018-10-17 2018-10-17 Intersection intelligent driving method and system based on support vector machine

Publications (2)

Publication Number Publication Date
CN111055849A CN111055849A (en) 2020-04-24
CN111055849B true CN111055849B (en) 2021-04-06

Family

ID=70296833

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811206747.7A Active CN111055849B (en) 2018-10-17 2018-10-17 Intersection intelligent driving method and system based on support vector machine

Country Status (1)

Country Link
CN (1) CN111055849B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI765720B (en) * 2021-05-24 2022-05-21 中國鋼鐵股份有限公司 Method and system for detecting a temperature of conveyor system

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104494600B (en) * 2014-12-16 2016-11-02 电子科技大学 A kind of Driver intention recognition method based on SVM algorithm
CN105260712B (en) * 2015-10-03 2019-02-01 上海大学 A kind of vehicle front pedestrian detection method and system
KR101865766B1 (en) * 2016-10-11 2018-06-11 주식회사 피엘케이 테크놀로지 Moving objects collision warning apparatus and method for large vehicle
CN106650644B (en) * 2016-12-07 2017-12-19 上海交通大学 The recognition methods of driver's hazardous act and system
CN107491720A (en) * 2017-04-01 2017-12-19 江苏移动信息系统集成有限公司 A kind of model recognizing method based on modified convolutional neural networks
CN107284452B (en) * 2017-07-18 2018-04-10 吉林大学 Merge the hybrid vehicle future operating mode forecasting system of intelligent communication information

Also Published As

Publication number Publication date
CN111055849A (en) 2020-04-24

Similar Documents

Publication Publication Date Title
CN109774724B (en) Method and apparatus for evaluating driving performance under a variety of driving scenarios and conditions
Wang et al. How much data are enough? A statistical approach with case study on longitudinal driving behavior
Oh et al. Estimation of rear-end crash potential using vehicle trajectory data
Eluru et al. A note on generalized ordered outcome models
Wang et al. A safety-based approaching behavioural model with various driving characteristics
Hamdar et al. From behavioral psychology to acceleration modeling: Calibration, validation, and exploration of drivers’ cognitive and safety parameters in a risk-taking environment
EP3940487B1 (en) Estimation of probability of collision with increasing severity level for autonomous vehicles
Aghabayk et al. Modelling heavy vehicle car‐following behaviour in congested traffic conditions
Choudhury et al. Modelling acceleration decisions in traffic streams with weak lane discipline: A latent leader approach
TWI690440B (en) Intelligent driving method for passing intersections based on support vector machine and intelligent driving system thereof
DE102021203353A1 (en) Method for determining a trajectory of an at least partially assisted motor vehicle, computer program and assistance system
CN115017742A (en) Automatic driving test scene generation method, device, equipment and storage medium
Platho et al. Predicting velocity profiles of road users at intersections using configurations
CN111055849B (en) Intersection intelligent driving method and system based on support vector machine
Kibalama et al. AV/ADAS Safety-Critical Testing Scenario Generation from Vehicle Crash Data
DE102019217533A1 (en) Method for estimating a coverage of the area of traffic scenarios
Reiterer et al. Beyond-design-basis evaluation of advanced driver assistance systems
CN114906170A (en) Motion planning method and device, electronic equipment and storage medium
Tawfeek Human-like speed modeling for autonomous vehicles during car-following at intersections
Marzbanrad et al. Prediction of driver’s accelerating behavior in the stop and go maneuvers using genetic algorithm-artificial neural network hybrid intelligence
DE102020129451A1 (en) Method for predicting driving interventions, method for training an algorithm and motor vehicle
Li et al. Traffic Modeling Considering Motion Uncertainties
Inoue et al. Validation methodology to establish safe autonomous driving algorithms with a high driver acceptance using a virtual environment
Nassuna et al. Feature selection for abnormal driving behavior recognition based on variance distribution of power spectral density
US20230195977A1 (en) Method and system for classifying scenarios of a virtual test, and training method

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