CN109886304B - HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions - Google Patents
HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions Download PDFInfo
- Publication number
- CN109886304B CN109886304B CN201910058190.5A CN201910058190A CN109886304B CN 109886304 B CN109886304 B CN 109886304B CN 201910058190 A CN201910058190 A CN 201910058190A CN 109886304 B CN109886304 B CN 109886304B
- Authority
- CN
- China
- Prior art keywords
- svm
- vehicle
- hmm
- model
- behavior
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a method for identifying surrounding vehicle behaviors under complex road conditions based on an HMM-SVM double-layer improved model, which comprises the following steps of 1) offline training: dividing typical peripheral vehicle behaviors, extracting state characteristic information of peripheral vehicles, transmitting the state characteristic information to a host vehicle through the Internet of vehicles, combining the host vehicle information to convert the state characteristic information into state characteristic information under a road joint coordinate system, generating a characteristic vector X, and respectively inputting the X into HMM and SVM parameter learning; 2) Model improvement: a threshold processor is arranged between the HMM and the SVM, and an NSGA-II algorithm is used for optimizing to obtain an optimal difference factor, so that an HMM-SVM double-layer improved model is obtained; 3) And (3) online testing: the vehicle utilizes an HMM-SVM double-layer improved model to distinguish the behavior mode of the tracked vehicle. The invention expands the application occasions of the surrounding vehicle behavior system by utilizing the road joint coordinate system constructed by the high-precision map; the excellent time sequence modeling capability of the HMM and the extremely strong two-classification capability of the SVM are organically combined, the double-layer model is improved, and the accuracy and the recognition speed of vehicle behavior recognition are improved.
Description
Technical Field
The invention belongs to the technical field of intelligent driving of vehicles, and particularly relates to a method for identifying surrounding vehicle behaviors under complex road conditions based on an HMM-SVM double-layer improved model.
Background
Road scene understanding is one of the important components of a vehicle driving assistance system and is also a fundamental requirement for future vehicle autonomous driving. The real traffic environment is a complex system of mutual influence and dynamic change of multiple traffic participants, and in the complex system, the intelligent automobile not only has visual perception, detection and tracking capability, but also needs to recognize and even predict the behaviors of surrounding vehicles through vehicle-road state information, so that the perception depth is improved.
In the behavior recognition of a nearby vehicle, it is indispensable to acquire accurate and practical nearby vehicle state characteristics. When the existing vehicle behavior recognition method obtains the vehicle state characteristics, the road is defaulted to be a straight road section, and the existing vehicle behavior recognition method is not suitable for recognition of vehicle behaviors under complex road conditions (arc-shaped bend, U-shaped bend and the like). The high-precision map is a map with high precision and definition, and covers data such as road network data, lane line data, lane lines, types of road edges, traffic signs and the like, and the precision of the high-precision map can reach the centimeter level. In addition, with the rapid development of the autopilot industry in recent years, the high-precision map industry enters a developed expressway, and many mainstream electronic map enterprises such as germany and hundred degrees have gradually opened the data of the high-precision map for free to the outside, so that the high-precision map is widely applied. The road joint coordinate system is established by using the high-precision map, and the state characteristic information of the vehicle under the coordinate system is acquired, so that the limitation of the vehicle behavior recognition only on the straight road is broken through, and the application occasions of the surrounding vehicle behavior systems are expanded.
Machine learning algorithms currently related to vehicle behavior recognition can be broadly classified into bayesian networks, hidden markov models, support vector machines, correlation vector machines, and the like. The hidden markov (HMM) model is most widely used because of its strong timing modeling capability, but ignores the influence of the negative sample, and classifies only with the maximum likelihood value, especially when the difference between the output multidimensional probabilities is small, a high false recognition rate is easily caused. The support vector machine algorithm (SVM) has significant advantages in terms of classification problems, can reflect the inter-class variability to a greater extent, and can separate similar samples well at as large euclidean distances as possible by mapping samples that are linearly inseparable in a low-dimensional space into a high-dimensional space. Therefore, in order to fully utilize the excellent time sequence modeling capability of the HMM and the extremely strong two-classification capability of the SVM, and simultaneously consider the real-time performance and the reliability of the recognition result, the invention provides an HMM-SVM double-layer improved model, introduces a difference factor, plays the respective advantages of the two classifiers, and thereby remarkably improves the recognition capability of the surrounding vehicle behavior system.
Disclosure of Invention
Aiming at the requirements of real-time performance and reliability of surrounding vehicle behavior recognition, the invention provides a surrounding vehicle behavior recognition method under complex road conditions based on an HMM-SVM double-layer improved model, which can accurately recognize the behaviors of surrounding vehicles under complex road conditions in real time and provides a reference basis for decision planning of intelligent vehicles.
The invention aims to realize the following technical scheme, namely a method for identifying the behavior of surrounding vehicles under complex road conditions based on an HMM-SVM double-layer improved model, which specifically comprises the following steps:
step (1): offline training phase
Dividing N typical peripheral vehicle behaviors, extracting state characteristic information of peripheral vehicles, transmitting the state characteristic information to a host vehicle through the Internet of vehicles, converting the state characteristic information of the host vehicle and the state characteristic information of the peripheral vehicles into state characteristic information under a road joint coordinate system, and generating new characteristic vectors X (delta X, delta Y and delta V) X ，△V Y ，△a X ，△a Y ) Inputting feature vectors X of the vehicle behaviors around the same class into the HMM model as an observation sequence to obtain HMMs of all the vehicle behavior classes; the peripheral vehicle behaviors are in groups of two pairs, feature vectors corresponding to two behavior categories in each group are input to each SVM model for parameter learning, and each group of SVM models is obtained;
step (2): model improvement stage
Forming a double-layer model by the trained HMM model and the SVM model, arranging a threshold processor between the two layers of models, and directly outputting an HMM layer recognition result when the threshold processor obtains that the difference between probability values is large; otherwise, extracting the behaviors corresponding to the two HMM models with the maximum probability values, selecting the corresponding SVM models for two-classification, outputting an SVM layer identification result, optimizing by using an NSGA-II algorithm to obtain an optimal difference factor sigma, and finally obtaining an HMM-SVM double-layer improved model;
step (3): on-line testing stage
The tracked target vehicle transmits the acquired self-vehicle driving information to the vehicle in real time through the vehicle network, and the vehicle utilizes the trained HMM-SVM double-layer improved model to distinguish the behavior mode of the tracked vehicle.
Further, the typical surrounding vehicle behavior includes following, left lane change, right lane change, overtaking.
Further, the joint coordinate system of the road uses the center line of the lane as a coordinate axis X and uses a line perpendicular to a tangent line of the coordinate axis X as a coordinate axis Y.
Further, the feature vector in the road joint coordinate system includes the position (Xego, yego), speed (V Xego ，V Yego ) Acceleration (a) Xego ，a Yego ) And the position (Xaro, yaro), speed (V) of the surrounding vehicle Xaro ，V Yaro ) Acceleration (a) Xaro ，a Yaro )。
Still further, in the feature vector X, Δx=xaro-Xego, Δy=yaro-ygo, and Δv X ＝V Xaro -V Xego 、△V Y ＝V Yaro -V Yego 、△a X ＝a Xaro -a Xego 、△a Y ＝a Yaro -a Yego 。
Further, the NSGA-II algorithm takes recognition accuracy and recognition time as optimization targets, and the optimization target function is f=T 0 -u 1 -u 2 -…u N Wherein u is 1 ，u 2 ，…，u N Training the recognition accuracy of each behavior of the sample, T 0 To identify the total time.
Further, the classifier equation of the two classifications is f (x) =sign (ω) T X+b), where ω is an adjustable weight vector and b is the offset.
The beneficial effects of the invention are as follows:
(1) In the stage of acquiring the state characteristics of the surrounding vehicles and the own vehicles, a road joint coordinate system constructed by utilizing a high-precision map is introduced, so that the limitation of identifying the vehicle behavior on the straight road can be broken through, and the application occasions of the surrounding vehicle behavior system are expanded;
(2) On the basis of identifying the peripheral vehicle behaviors by the HMM single model, an SVM model is added to form a double-layer model to identify the peripheral vehicle behaviors, and the excellent time sequence modeling capability of the HMM and the extremely strong two classification capability of the SVM are organically combined, so that the respective advantages of the two classifiers are fully exerted;
(3) And establishing a mathematical model by taking the recognition accuracy and the recognition time as optimization targets, performing multi-target optimization on a threshold processor by using an NSGA-II algorithm to obtain a difference factor sigma, improving an HMM-SVM double-layer model, and improving the accuracy and the recognition speed of vehicle behavior recognition.
Drawings
FIG. 1 is a schematic diagram of the establishment of a joint coordinate system of a road, FIG. 1 (a) is a schematic diagram of the establishment of a straight road in the joint coordinate system, and FIG. 1 (b) is a schematic diagram of the establishment of a curve in the joint coordinate system;
FIG. 2 is a block diagram of a method for identifying surrounding vehicle behavior under complex road conditions based on an HMM-SVM double-layer improved model.
Detailed Description
The technical scheme of the present invention will be further described with reference to the accompanying drawings, but the scope of the present invention is not limited thereto.
Step1: state feature acquisition and data processing
Firstly, N typical peripheral vehicle behaviors, namely following, left lane changing, right lane changing, overtaking and the like, are respectively summarized and divided. Each experimental vehicle is set to be provided with an interface connected with a high-precision map, the high-precision map contains a large amount of road information, one important characteristic different from the traditional map is precision, the precision of centimeter level can be achieved, and the experimental vehicle can be helped to realize high-precision positioning. Each experimental vehicle is provided with an independent ID, and is connected with the Internet of vehicles through an OBD (On-Board Diagnostics) interface, data extracted from the vehicle is uploaded to the cloud, and the experimental vehicle interacts with other vehicles in real time through the background. In consideration of the real-time performance and the robustness of data transmission, the sampling frequency is set to be 50Hz, namely the time length between the data before and after acquisition is 0.02s. One of the test vehicles is selected as the host vehicle (Ego), and the vehicles Around the host vehicle, namely, the surrounding vehicles (Around 1,2 …). The surrounding vehicles herein include not only vehicles in the field of view of the driver of the host vehicle or in the sensing area of equipment such as a vehicle-mounted camera, a laser radar, a millimeter wave radar, etc., but also vehicles in blind areas such as vehicles in which the turning opening of a mountain road is blocked by a mountain, etc.
And aiming at the target vehicles to be identified in the surrounding vehicles, the vehicle downloads the state characteristic information of the surrounding vehicles on line from the Internet of vehicles. By using the accessed high-precision map, referring to a natural system representation method of motion particles in theoretical mechanics, taking the central line of a lane as a coordinate axis X, taking a line perpendicular to a tangent line of the coordinate axis X as a coordinate axis Y, and establishing a road joint coordinate system, as shown in FIG. 1, FIG. 1 (a) is a schematic diagram of establishing a straight road in the joint coordinate system, and FIG. 1 (b) is a schematic diagram of establishing a curve in the joint coordinate system. The state characteristic information of the vehicle and the downloaded surrounding vehicles is converted into characteristic information under a road joint coordinate system, namely the position (Xygo, yego) and the speed (V) of the vehicle at the moment t Xego ，V Yego ) Acceleration (a) Xego ，a Yego ) Peripheral vehicle position (Xaro, yaro), speed (V Xaro ，V Yaro ) Acceleration (a) Xaro ，a Yaro ) The characteristic information is converted into a new characteristic vector X (delta X, delta Y, delta V) X ，△V Y ，△a X ，△a Y ) Wherein Δx=xaro-Xego, Δy=yaro-ygo, Δv X ＝V Xaro -V Xego ，△V Y ＝V Yaro -V Yego ，△a X ＝a Xaro -a Xego ，△a Y ＝a Yaro -a Yego . The feature vector sets corresponding to the behavior categories of each typical surrounding vehicle are equally obtained as training sample sets according to the steps, wherein the training sample sets can be set as B according to the behavior categories 1 ，B 2 ，…，B N 。
Step2: model training learning
(1) Hidden Markov model training learning
The hidden Markov model (Hidden Markov Model, HMM) is a probabilistic model about time sequences, generallyThe observable variables were passed to study the unobservable variables. The hidden Markov model can be set as a five-tuple (Q, V, A, B, pi), wherein the hidden state Q= { Q 1 ，Q 2 ，…，Q N N is the number of hidden states; observable state v= { V 1 ，V 2 ，…，V M M is the number of observation states; hidden state transition probability matrix a= [ α ] ij ] N×N Elements of (a) represent transition probabilities, alpha, between hidden states in the HMM model ij Is hidden state Q at time t i Hidden state at time t+1 is Q j Probability of a) ij ＝P(I t+1 ＝Q j |I t ＝Q i ) I=1, 2 …, N; j=1, 2 …, N, I is a state sequence of length T, and i= { I 1 ,I 2 ,…,I T -a }; confusion matrix b= [ B ] j (k)] N×M The elements of (a) represent the transition probabilities between the respective hidden states and the observation states in the HMM model, b j (k) Indicating at time t that the hidden state is Q j The observation state is O t Probability of b j (k)＝P(O t ＝V k |I t ＝Q j ) K=1, 2 …, M; j=1, 2 …, N, O is the corresponding observation sequence; initial state probability matrix pi= (pi) i ) Wherein pi is i ＝P(I 1 ＝Q i ) I=1, 2, …, N, represents the respective hidden states Q at the initial time t=1 i Is a probability of (2).
Firstly, initializing each vehicle behavior recognition HMM model to obtain initial parameters N, M, A, B and pi; and taking the feature vector X of the vehicle behaviors (following, left lane changing, right lane changing and overtaking) around the same class obtained in Step1 as the input of an HMM model, adjusting the parameters of a model lambda= (A, B, pi) by adopting a Baum-Welch iterative algorithm according to the parameters after the model initialization, maximizing a probability function, gradually updating the model parameters, finally obtaining the HMM corresponding to each vehicle behavior class, and completing the learning of the first layer model.
(2) Support vector machine model training learning
The support vector machine (Support Vector Machine, SVM) model is the currentThe most widely used classifier shows many unique advantages in addressing small sample, non-linear and high dimensional pattern recognition. The basic principle is to find an optimal hyperplane meeting the data classification requirement, so that the hyperplane has the largest distance from two types of sample points under the condition of ensuring the classification accuracy. The hyperplane should satisfy ω T X+b=0, where ω is an adjustable weight vector, b is an offset, X is a feature vector, and T is a transposed symbol of the matrix. The optimal hyperplane requires maximization of the classification interval, and two parallel hyperplanes the distance is 2/||omega|, that is, a minimization of ω is required, i.e., there is a minimization equation when solving: phi (omega) =1/2|| ω|=1/2 (ω, ω), in order for all samples to be out of the hyperplane, the above constraint Y should also be satisfied i ω T ·X i +b＞1，Y i E { -1,1}, i=1, 2, … l, where Y i The sample class is represented, and l is the number of samples.
The N typical peripheral vehicle behaviors were grouped in pairs, and were grouped together into h=n (N-1)/2 groups, each group including two behavior categories i, j, one as positive samples and the other as negative samples. Training sample set B corresponding to positive and negative samples obtained in Step1 i 、B j Training sample sets (B) i ，B j ) For training the SVM model. Based on MATLAB environment, each group of training sample set (B i ，B j ) The feature vector X of each group is input to each SVM model for parameter learning, and the optimal hyperplane result f of each group is found k (x)＝ω T X+b, k=1, 2, … h, thereby obtaining SVM models of two classification for each group.
Step3: improved model for forming double layers
The trained HMM model and SVM model are combined into a double-layer model, a threshold processor is arranged between the two-layer model, training sample sets corresponding to the typical peripheral vehicle behaviors are combined into a total training sample set (B1, B2, …, BN), the characteristic vector X of each sample is used as an observation sequence to be input into the HMM model, and the multidimensional output probability (P 1 ，P 2 ，…，P n ) Acting asIs the input to the threshold processor. The threshold processor compares the output probabilities to extract the maximum two probabilities P max-i 、P max-ii The method comprises the steps of carrying out a first treatment on the surface of the Let sigma be the difference factor in the threshold processor, when P max-ii /P max-i When sigma is less than or equal to sigma, the difference between the output results of the HMM models is larger, the identification result of the first layer model is directly output, namely, the output probability of each model is the largest (P max-i ) The corresponding behavior. When P max-ii /P max-i If the output probability is greater than or equal to sigma, the difference between the output results of the HMM model is smaller, and if the direct output probability is maximum (P max-i ) The corresponding behavior is easy to generate higher recognition error rate, and the recognition of the second layer model is needed. Will P max-i 、P max-ii Corresponding behavior Q i 、Q ii Extracting, inputting the feature vector X into the behavior Q i 、Q ii In the set of SVM models, the SVM model classifier equation f (x) =sign (ω) T X+b) outputs the behavior class of the sample, and takes the result as the final recognition result. In view of the important influence of the threshold sigma in the threshold processor on the real-time performance and reliability of double-layer model identification, the invention further analyzes the identification result of the training sample set, and takes the identification accuracy of each behavior of the training sample as u 1 ，u 2 ，…，u N Identifying the total time as T 0 。
In order to improve the HMM-SVM double-layer model, a mathematical model is established by taking the recognition accuracy and recognition time as optimization targets, and a NSGA-II algorithm is used for carrying out multi-target optimization on a threshold processor. The NSGA-II algorithm is one of the most popular multi-objective genetic algorithms at present, and has the advantages of high running speed and good convergence of solution sets. Defining an optimization objective function as f=t 0 -u 1 -u 2 -…u N And optimizing a difference factor sigma aiming at the improvement of the recognition accuracy and recognition time, and finally obtaining the HMM-SVM double-layer improved model.
Step4: on-line testing stage
The tracked target vehicle transmits the acquired self-vehicle driving information to the host vehicle in real time through the vehicle network, the host vehicle combines the characteristics of the two vehicles under the road joint coordinate system to form a new characteristic vector, and the behavior mode of the tracked vehicle is distinguished by using a trained HMM-SVM double-layer improved model.
The examples are preferred embodiments of the present invention, but the present invention is not limited to the above-described embodiments, and any obvious modifications and substitutions which can be made by one skilled in the art without departing from the spirit of the present invention are within the scope of the present invention.
Claims (5)
1. The method for identifying the behavior of the surrounding vehicles under the complex road conditions based on the HMM-SVM double-layer improved model is characterized by comprising the following steps of:
step (1): offline training phase
Dividing N typical peripheral vehicle behaviors, extracting state characteristic information of peripheral vehicles, transmitting the state characteristic information to a host vehicle through the Internet of vehicles, converting the state characteristic information of the host vehicle and the state characteristic information of the peripheral vehicles into state characteristic information under a road joint coordinate system, and generating new characteristic vectors X (delta X, delta Y and delta V) X ，△V Y ，△a X ，△a Y ) Inputting feature vectors X of the vehicle behaviors around the same class into the HMM model as an observation sequence to obtain HMMs of all the vehicle behavior classes; the peripheral vehicle behaviors are in groups of two pairs, feature vectors corresponding to two behavior categories in each group are input to each SVM model for parameter learning, and each group of SVM models is obtained;
step (2): model improvement stage
Forming a double-layer model by the trained HMM model and the SVM model, arranging a threshold processor between the two layers of models, and directly outputting an HMM layer recognition result when the threshold processor obtains that the difference between probability values is large; otherwise, extracting the behaviors corresponding to the two HMM models with the maximum probability values, selecting the corresponding SVM models for two-classification, outputting an SVM layer identification result, optimizing by using an NSGA-II algorithm to obtain an optimal difference factor sigma, and finally obtaining an HMM-SVM double-layer improved model; the NSGA-II algorithm takes recognition accuracy and recognition time as optimization targets; the optimization objective function is f=t 0 -u 1 -u 2 -…u N Wherein u is 1 ，u 2 ，…，u N Training the recognition accuracy of each behavior of the sample, T 0 To identify a total time;
specifically, the multidimensional output probabilities (P 1 ，P 2 ，…，P n ) As input to the threshold processor, the threshold processor compares the output probabilities to extract the maximum two probabilities P max-i 、P max-ii The method comprises the steps of carrying out a first treatment on the surface of the When P max-ii /P max-i When sigma is less than or equal to sigma, the maximum probability P of each model output is directly output max-i The corresponding behavior; when P max-ii /P max-i When not less than sigma, P is max-i 、P max-ii Corresponding behavior Q i 、Q ii Extracting, inputting the feature vector X into the behavior Q i 、Q ii In the set of SVM models, the SVM model classifier equation f (x) =sign (ω) T X+b) outputting the behavior category of the training sample, and taking the result as a final recognition result; wherein ω is an adjustable weight vector, b is an offset, and T is the transposed symbol of the matrix;
step (3): on-line testing stage
The tracked target vehicle transmits the acquired self-vehicle driving information to the vehicle in real time through the vehicle network, and the vehicle utilizes the trained HMM-SVM double-layer improved model to distinguish the behavior mode of the tracked vehicle.
2. The method for identifying surrounding vehicle behavior under complex road conditions based on the HMM-SVM dual-layer improved model according to claim 1, wherein the typical surrounding vehicle behavior comprises following, left lane change, right lane change, overtaking.
3. The method for identifying the behavior of surrounding vehicles under the complex road conditions based on the HMM-SVM double-layer improved model according to claim 1, wherein the road joint coordinate system takes the central line of a lane as a coordinate axis X and a line perpendicular to a tangent line of the coordinate axis X as a coordinate axis Y.
4. A method for identifying behavior of surrounding vehicles under complex road conditions based on HMM-SVM double-layer improved model as claimed in claim 1 or 3, wherein the feature vector under the combined road coordinate system comprises position (Xego, ygo), speed (V) of the vehicle at time t Xego ，V Yego ) Acceleration (a) Xego ，a Yego ) And the position (Xaro, yaro), speed (V) of the surrounding vehicle Xaro ，V Yaro ) Acceleration (a) Xaro ，a Yaro )。
5. The method for identifying the behavior of a surrounding vehicle under a complex road condition based on a HMM-SVM double-layer improved model according to claim 4, wherein in the feature vector X, Δx=xaro-Xego, Δy=yaro-ygo, and Δv X ＝V Xaro -V Xego 、△V Y ＝V Yaro -V Yego 、△a X ＝a Xaro -a Xego 、△a Y ＝a Yaro -a Yego 。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910058190.5A CN109886304B (en) | 2019-01-22 | 2019-01-22 | HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions |
| PCT/CN2019/077584 WO2020151059A1 (en) | 2019-01-22 | 2019-03-11 | Method for recognizing surrounding vehicle behaviors in complex road conditions based on hmm-svm two-layer improved model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910058190.5A CN109886304B (en) | 2019-01-22 | 2019-01-22 | HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886304A CN109886304A (en) | 2019-06-14 |
| CN109886304B true CN109886304B (en) | 2023-09-29 |
Family
ID=66926458
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910058190.5A Active CN109886304B (en) | 2019-01-22 | 2019-01-22 | HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109886304B (en) |
| WO (1) | WO2020151059A1 (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288835B (en) * | 2019-06-28 | 2021-08-17 | 江苏大学 | Surrounding vehicle behavior real-time identification method based on kinematic prediction compensation mechanism |
| CN110986994B (en) * | 2019-11-14 | 2021-08-03 | 苏州智加科技有限公司 | Automatic lane change intention marking method based on high-noise vehicle track data |
| CN111354193B (en) * | 2020-02-26 | 2021-09-10 | 江苏大学 | Highway vehicle abnormal behavior early warning system based on 5G communication |
| CN112069599B (en) * | 2020-08-28 | 2023-08-22 | 北京科技大学 | Method for predicting reliability of running performance of tracked vehicle |
| CN114299742B (en) * | 2022-01-20 | 2023-10-13 | 福建工程学院 | Speed limit information dynamic identification and update recommendation method for expressway |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106971194A (en) * | 2017-02-16 | 2017-07-21 | 江苏大学 | A kind of driving intention recognition methods based on the double-deck algorithms of improvement HMM and SVM |
| CN107967486A (en) * | 2017-11-17 | 2018-04-27 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9784592B2 (en) * | 2015-07-17 | 2017-10-10 | Honda Motor Co., Ltd. | Turn predictions |
| US20180190377A1 (en) * | 2016-12-30 | 2018-07-05 | Dirk Schneemann, LLC | Modeling and learning character traits and medical condition based on 3d facial features |
| CN108470460B (en) * | 2018-04-11 | 2020-08-28 | 江苏大学 | Peripheral vehicle behavior identification method based on smart phone and RNN |
-
2019
- 2019-01-22 CN CN201910058190.5A patent/CN109886304B/en active Active
- 2019-03-11 WO PCT/CN2019/077584 patent/WO2020151059A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106971194A (en) * | 2017-02-16 | 2017-07-21 | 江苏大学 | A kind of driving intention recognition methods based on the double-deck algorithms of improvement HMM and SVM |
| CN107967486A (en) * | 2017-11-17 | 2018-04-27 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886304A (en) | 2019-06-14 |
| WO2020151059A1 (en) | 2020-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886304B (en) | HMM-SVM double-layer improved model-based surrounding vehicle behavior recognition method under complex road conditions | |
| US10733755B2 (en) | Learning geometric differentials for matching 3D models to objects in a 2D image | |
| Jain et al. | Discrete residual flow for probabilistic pedestrian behavior prediction | |
| CN111310583B (en) | Vehicle abnormal behavior identification method based on improved long-term and short-term memory network | |
| CN106971194B (en) | Driving intention recognition method based on improved HMM and SVM double-layer algorithm | |
| CN112347993B (en) | Expressway vehicle behavior and track prediction method based on vehicle-unmanned aerial vehicle cooperation | |
| Gadepally et al. | A framework for estimating driver decisions near intersections | |
| CN103605362B (en) | Based on motor pattern study and the method for detecting abnormality of track of vehicle multiple features | |
| CN108877267B (en) | Intersection detection method based on vehicle-mounted monocular camera | |
| CN110778714B (en) | Fuel vehicle gear identification method and system | |
| US20210166085A1 (en) | Object Classification Method, Object Classification Circuit, Motor Vehicle | |
| Moore et al. | Data-driven extraction of vehicle states from CAN bus traffic for cyberprotection and safety | |
| Moosavi et al. | Driving style representation in convolutional recurrent neural network model of driver identification | |
| Ranjbar et al. | Safety monitoring of neural networks using unsupervised feature learning and novelty estimation | |
| CN113406623A (en) | Target identification method, device and medium based on radar high-resolution range profile | |
| xi Wang et al. | Detection of MMW radar target based on doppler characteristics and deep learning | |
| Yu et al. | Track-before-detect labeled multi-Bernoulli smoothing for multiple extended objects | |
| CN115294449A (en) | Ground identification method for crawler robot | |
| Gao et al. | Deep learning‐based hybrid model for the behaviour prediction of surrounding vehicles over long‐time periods | |
| CN114445456A (en) | Data-driven intelligent maneuvering target tracking method and device based on partial model | |
| CN112559968A (en) | Driving style representation learning method based on multi-situation data | |
| De Candido et al. | An analysis of distributional shifts in automated driving functions in highway scenarios | |
| CN113534131B (en) | GM-PHD-based motion parameter estimation combined smooth filtering method | |
| Nguyen et al. | Reliability-aware and robust multi-sensor fusion toward ego-lane estimation using artificial neural networks | |
| Cai et al. | CAEV-Deep Learning-Based Hybrid Model for the Behavior Prediction of Surrounding Vehicles Over Long-time Periods |
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 |