CN113159105B - Driving behavior unsupervised mode identification method and data acquisition monitoring system - Google Patents
Driving behavior unsupervised mode identification method and data acquisition monitoring system Download PDFInfo
- Publication number
- CN113159105B CN113159105B CN202110216457.6A CN202110216457A CN113159105B CN 113159105 B CN113159105 B CN 113159105B CN 202110216457 A CN202110216457 A CN 202110216457A CN 113159105 B CN113159105 B CN 113159105B
- Authority
- CN
- China
- Prior art keywords
- driving behavior
- driving
- acceleration
- data
- speed
- 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
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Abstract
The invention discloses a driving behavior unsupervised mode identification method and a data acquisition monitoring system, wherein the method comprises the following steps: collecting driving data fragments of a driver in the driving process and extracting features; clustering the extracted features by using an improved Wasserstein distance-based self-adaptive K-Means clustering algorithm to obtain a driving behavior mode; and establishing a driving behavior theme description, and matching the driving behavior theme description with driving behavior patterns obtained by all clusters by using an LDA model to obtain an identification result of the unsupervised driving behavior. In addition, the invention also constructs a corresponding data acquisition monitoring system, and a complete driving behavior unsupervised mode monitoring system is formed by a mode identification algorithm, a hardware architecture and a monitoring APP, so that the driving behavior can be more effectively identified, the real-time monitoring of the driving behavior is realized, and the driving safety of a driver is improved.
Description
Technical Field
The invention relates to the technical field of driving behavior unsupervised mode recognition and predictive monitoring, in particular to a driving behavior unsupervised mode recognition method and a data acquisition monitoring system.
Background
In recent years, the motor vehicle industry in China is rapidly developed. In 2020, the domestic automobile has 2.7 hundred million vehicles, the driver has 4.1 hundred million vehicles, and the total amount and increment of the motor vehicle and the driver are all the first in the world. However, with the increase in the amount of maintenance of automobiles, a deep contradiction in traffic management work is gradually revealed. China is one of countries with the most road traffic accidents in the world. According to the data published by the national statistical bureau, the death rate of every ten thousand vehicles in China has reached incredibly 6.2, and the data has reached 4 to 8 times that of developed countries, so that the supervision task of the China on bad driving behaviors is far from the priority. Therefore, a high degree of attention should be paid to road traffic safety issues, whether it be a transportation enterprise or a government related department.
Analysis of a large number of accident cases shows that although the causes of traffic accidents are various, artificial reasons are always the main factors of the accidents. Abnormal driving behavior of a driver tends to be a cause of a major traffic accident. Therefore, if abnormal driving behaviors can be effectively identified and an alarm can be timely given to a driver, road traffic accidents can be avoided to a certain extent. On the other hand, the abnormal driving behavior can be effectively identified, so that the traffic department and the operation enterprises can be helped to realize dynamic supervision and evaluation of the safety situation, and the safety management level is improved.
In order to save cost, monitoring and early warning by using information such as vehicle GPS, speed, acceleration and the like become a hot spot of current research. By analyzing related data at home and abroad, the identification method of abnormal driving behaviors is effectively applied to a series of safety supervision fields such as driver behavior management, so that the supervision level is improved, and the occurrence probability of accidents is reduced. Therefore, in order to effectively improve the safety supervision of the traffic industry, it is imperative to perform deep analysis and application on the vehicle monitoring data. Nevertheless, the current abnormal driving behavior pattern recognition algorithm is still in a stage of comparison, and still has the following problems: (1) The aim of the research is still to detect isolated events at the traffic running level, and other factors causing traffic mode changes are ignored; (2) At present, a supervised algorithm is mostly adopted for abnormal driving behavior identification, however, the marked data are expensive and rare, and the loss of related information and the reduction of generalization performance are caused. Therefore, it is imperative to explore an unsupervised abnormal driving behavior recognition algorithm.
Disclosure of Invention
The invention aims to provide a driving behavior unsupervised mode identification method and a data acquisition monitoring system, which are used for detecting and identifying abnormal driving behaviors by adopting a bottom-up unsupervised method, providing timely feedback for a driver and further improving the driving safety of the driver.
In order to solve the technical problems, the embodiment of the invention provides the following scheme:
in one aspect, a driving behavior unsupervised mode recognition method is provided, including the following steps:
s1, collecting driving data fragments of a driver in a driving process and extracting features;
s2, clustering the extracted features by using an improved self-adaptive K-Means clustering algorithm based on Wasserstein distance to obtain a driving behavior mode;
s3, establishing driving behavior theme description, and matching the driving behavior theme description with driving behavior modes obtained by all clusters by using an LDA model to obtain an identification result of the unsupervised driving behavior.
Preferably, in the step S1, the driving data segment includes four variables of track curvature, acceleration, speed, steering angle; the extracted features are divided into two categories: statistical features and temporal features; the statistical features include: average, variance, ACF1, remainder autocorrelation, trend, curvature, and entropy; the time profile includes: maximum mean difference, maximum variance difference, maximum displacement of Kullback-Leibler divergence, and remainder variance.
Preferably, the step S2 specifically includes:
s200, inputting a characteristic sample data set Q= { x s |s＝1,...,N}；
S201, let k be from 1 to k max Calculating clusters when the number of clusters is k, wherein k max Is the maximum number of clusters;
s202, randomly selecting k sample points in the characteristic sample data set Q, and respectively assigning the k sample points to an initial clustering center mu 1 ，μ 2 ，…，μ b ，…，μ k ；
S203, making S be 1-N circulation, wherein N represents the number of samples to be clustered, and calculating sample x b With each cluster center mu b Distance d (x) s ，μ b )；
S204, finding x s About cluster center μ b Will x s Are classified as mu b Cluster C with minimum distance b In (a) and (b);
s205, updating the cluster centers of all the clusters, and if the new cluster centers are not equal to the original cluster centers, updating the cluster centers;
s206, repeating the steps S203-S205 until the clustering center is not updated;
s207, calculating the current cluster error square sum SSE k ；
S208, calculating the difference between the current cluster error square sum and the last cluster error square sum, namely the decreasing amplitude delta SSE k ＝SSE k-1 -SSE k ；
S209, calculating the difference of the decrease amplitude variation, i.e. ΔΔSSE k ＝ΔSSE k-1 -ΔSSE k ；
S210, repeating step S201-step S209 until k=k max ；
S211, the optimal cluster number is delta SSE k The value of k is the largest, and the clustering result is the optimal clustering result;
s212, outputting a clustering result.
Preferably, the wasperstein distance is defined as follows:
Assuming that θ and t represent two distributions, the distance between them is calculated by the following formula:
wherein the method comprises the steps ofIs->The set of all possible joint distributions combined with the t distribution, pi representing one of the joint distributions represented by θ and t, E (m，n)～Π [||m-n||]Calculating the distance expected value of the two distributions under the condition of the current joint distribution pi; after calculation of all joint distributions, use +.>Obtaining all expected lower bounds, namely Wasserstein distance values;
let the set of subsequences describe as x= (X 1 ，...，x s ，…，x N ) WhereinN is the total number of data set samples, < >>For the dimension of the subsequence feature, k is the number of clusters, and the cluster center vector is { μ } 1 ，μ 2 ，...，μ b ，...，μ k -a }; sample x s And cluster center mu b The distance of (2) is:
Π(x s ，μ b ) Is x s Sum mu b Distribution of a set of all possible joint distributions combined, pi representing x s Sum mu b One of the joint distributions represented; wherein x is sf ，μ bf Respectively represent x s And mu b Is the f-th feature of (2).
Preferably, the elbow method is introduced to calculate the objective function to obtain the optimal number of clusters:
and obtaining the optimal clustering number when the SSE value reaches the minimum.
Preferably, in the step S3, establishing a driving behavior theme description refers to semantic expression of the driving behavior by extracting driving features; the driving behavior topics comprise 5 normal driving behavior topics and 6 abnormal driving behavior topics; the 5 normal driving behavior topics include: straight running at a constant speed, mild starting, curve stable running, normal lane changing and mild stopping; the 6 abnormal driving behavior subjects include: quick lane change, sudden braking, rapid overtaking, curve rapid running, sudden acceleration and curve overtaking.
Preferably, in the step S3, the unsupervised driving behavior recognition process based on the LDA model is as follows:
the LDA model is essentially a three-layer Bayesian network, and unsupervised driving behavior matching is performed by calculating feature generation probability; set Ω= { Ω 1 ，...，Ω d ，...，Ω k The driving behavior pattern is represented, wherein the number of the driving behavior patterns is the number k of the extracted clusters, and O is the number of the driving behavior subjects; the local driving behavior characteristic parameter observed in the driving behavior pattern Ω is expressed asWherein U is d The number of local features is 8 features, namely a track curvature mean value, a track curvature variance, a speed mean value, a speed variance, an acceleration mean value, an acceleration variance, a steering angle mean value and a steering angle variance;
LDA assumes a third feature w observed in the third driving behavior pattern d，u Is based on potential driving behavior z d，u Generated, i.e. assumed, features w d，u The actual value w of (2) is according toGenerated, then the LDA is assigned by θ o|d And Dirichlet parameter alpha, wherein phi 0 Is a feature profile with a potential driving topic with Dirichlet parameter β; thus, the generation model of LDA is assumed to be:
θ d ～Dir(θ；α)
φ o ～Dir(φ；β)
z d，u ～Mult(z；θ d )
when the feature w= { w is observed d，u The probability of generation of the feature is expressed as:
And generating a driving behavior theme with the highest probability as a behavior matching result of the driving data fragment.
In one aspect, a data acquisition monitoring system based on the driving behavior unsupervised mode identification method is provided, where the data acquisition monitoring system includes:
the data acquisition subsystem is used for acquiring driving data fragments of a driver in the driving process and extracting characteristics to realize local data monitoring, historical data storage and key real-time data uploading;
the cloud platform centralized monitoring subsystem is used for clustering the extracted features by using an improved Wasserstein distance-based self-adaptive K-Means clustering algorithm to obtain a driving behavior mode; and establishing a driving behavior theme description, and matching the driving behavior theme description with driving behavior patterns obtained by all clusters by using the LDA model to obtain an identification result of the unsupervised driving behavior.
Preferably, the data acquisition subsystem comprises a data acquisition front end, a serial port server, a computer, a display and data acquisition software, and realizes interactive control through a touch screen;
the data acquisition subsystem provides two modes of automatic operation and manual operation, and in the automatic operation mode, the system is started or stopped by automatically receiving a trigger signal; in the manual operation mode, an operator starts or stops the system work by pressing an operation button;
The cloud platform centralized monitoring subsystem comprises a cloud data center and is used for monitoring, storing, retrieving, analyzing and publishing data.
Preferably, the data acquisition monitoring system further comprises a vehicle abnormal driving behavior monitoring APP, wherein the vehicle abnormal driving behavior monitoring APP comprises a user registration and login module, an online inquiry module, an area display module and a modification module, and is used for inquiring, displaying, online updating and modifying vehicle driving information so as to be monitored by management personnel in real time.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
the driving behavior unsupervised mode identification method provided by the invention does not need to set a threshold value for abnormal driving behaviors, and aims at the characteristics of the abnormal driving behaviors, and the driving behavior unsupervised identification is realized through the following two steps: firstly, extracting statistical features and space-time related features as feature parameters for driving behavior recognition, introducing Wasserstein distance as a clustering measurement method for improving clustering accuracy and fully considering structural data characteristics, providing a self-adaptive K-Means algorithm based on the Wasserstein distance, introducing feature distribution concepts, aggregating driving data fragments, and generating a driving behavior mode; secondly, setting a driving behavior theme description, matching the behavior pattern represented by the driving characteristics of the actual driving behavior by using the LDA model, and performing unsupervised marking on the driving behavior by generating probability matching driving behavior. The invention also constructs a corresponding data acquisition monitoring system, and a complete driving behavior unsupervised mode monitoring system is formed by a mode recognition algorithm, a hardware architecture and a monitoring APP, so that the driving behavior can be more effectively recognized, the real-time monitoring of the driving behavior is realized, and the driving safety of a driver is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a driving behavior unsupervised pattern recognition method provided by an embodiment of the present invention;
FIG. 2 is a schematic diagram of a feature extraction process according to an embodiment of the present invention;
fig. 3 is a flowchart of an AKCWD algorithm provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram of an LDA model according to an embodiment of the present invention;
FIG. 5 is a flow chart of driving behavior recognition provided by an embodiment of the present invention;
FIG. 6 is a schematic view of an LDA model driving theme visualization effect provided by an embodiment of the present invention;
FIG. 7 is a visual effect diagram of a theme of rapid acceleration driving provided by an embodiment of the present invention;
FIG. 8 is a schematic diagram of a data acquisition subsystem provided by an embodiment of the present invention;
fig. 9 is a data link and access network structure diagram of a centralized monitoring subsystem of a cloud platform according to an embodiment of the present invention;
FIG. 10 is a schematic diagram of vehicle monitoring data provided by an embodiment of the present invention;
FIG. 11 is a trace segment feature semantic graph provided by an embodiment of the present invention;
FIG. 12 is a schematic diagram of normal lane change driving behavior theme identification provided by an embodiment of the present invention;
fig. 13 is a schematic diagram of an LDA-based driving behavior theme recognition effect provided by an embodiment of the present invention;
FIGS. 14 a-14 d are graphs illustrating speed change in four driving behavior themes provided by embodiments of the invention;
15 a-15 d are schematic diagrams of acceleration changes in four driving behavior topics provided by an embodiment of the present invention;
FIGS. 16 a-16 d are schematic views of steering angle changes in four driving behavior topics provided by embodiments of the present invention;
fig. 17 is an abnormal driving behavior feature analysis chart provided by an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the embodiments of the present invention will be described in further detail with reference to the accompanying drawings.
The embodiment of the invention firstly provides a driving behavior unsupervised mode identification method, as shown in fig. 1, comprising the following steps:
s1, collecting driving data fragments of a driver in a driving process and extracting features;
S2, clustering the extracted features by using an improved self-adaptive K-Means clustering algorithm based on Wasserstein distance to obtain a driving behavior mode;
s3, establishing driving behavior theme description, and matching the driving behavior theme description with driving behavior modes obtained by all clusters by using an LDA model to obtain an identification result of the unsupervised driving behavior.
Specifically, in the step S1, the driving data segment includes four variables of track curvature, acceleration, speed, and steering angle.
When the driving behavior is identified, a large amount of actually measured driving data needs to be referenced, and the degree of independence between the data is high. According to the invention, heuristic characteristic parameter selection criteria are obtained through analysis of the existing driving behavior recognition method. For example, australian RTA finds that an increase in vehicle speed by studying abnormal driving behaviour tends to be accompanied by an increase in risk of traffic accidents. Therefore, the vehicle speed is an important consideration for the recognition of abnormal driving behavior; in addition, the driver often steps on the accelerator pedal or the brake pedal severely, which is accompanied by abnormal driving behavior, has a great influence on driving safety.
Therefore, the invention can effectively identify the phenomenon of four urgency (sudden acceleration, sudden deceleration, sudden braking and sudden turning) by analyzing the external expression of the driving behavior and combining the actual demand of the driving behavior analysis, and the selected driving data segment mainly comprises four variables of track curvature, acceleration, speed and steering angle, thereby establishing an abnormal driving behavior identification model.
The feature extraction process converts the raw vehicle driving data segment into a set of features (Fl) that capture both contextual information around the point in time and temporal information around the point in time. These features can be divided into two classes: statistical features and temporal features. The feature extraction flow chart is shown in fig. 2.
The invention uses 11 feature extractors to perform feature conversion segment by segment aiming at four variable data of track curvature, speed, acceleration and steering angle. Statistical features describe some basic features around each data point in the time series where the mean, variance, ACF1, remainder autocorrelation, trend, curvature and entropy are extracted, as shown in table 1; the time characteristics describe the change of the sequence data over time, and the time characteristics are determined by comparing the data in two consecutive windows, where the maximum mean difference, the maximum variance difference, the maximum displacement of Kullback-Leibler divergence, and the remainder variance are extracted as shown in table 2.
Table 1 table of statistical characteristics
Table 2 time profile
In summary, each driving data segment includes four variables of track curvature, acceleration, speed, and steering angle, and 44 behavior features can be captured. All of these features are normalized to be comparable in different sequences.
Further, in order to study the occurrence of abnormal driving behaviors, the segmented data needs to be clustered, driving behavior patterns are mined from a large number of unlabeled tracks, and driving behavior scenes are provided for unsupervised pattern recognition. In order to fully embody the characteristics of multi-feature clustering data and measure the distribution difference of each index feature of different clusters, the invention adopts Wasserstein distance measurement to replace Euclidean distance in the traditional K-Means algorithm, and provides a self-adaptive K-Means clustering Algorithm (AKCWD) based on the Wasserstein distance to improve the degree of feature discrimination.
The Wasserstein distance is first defined as follows:
assume thatAnd t represents two distributions, the distance between which is calculated by the following formula:
wherein the method comprises the steps ofIs->All possible sets of joint distributions combined with the t distribution, pi representing +.>And t, E (m，n)～Π [||m-n||]Calculating the distance expected value of the two distributions under the condition of the current joint distribution pi; after calculation of all joint distributions, use +.>Obtaining all expected lower bounds, namely Wasserstein distance values;
let the set of subsequences describe as x= (X 1 ，...，x s ，…，x N ) WhereinN is the total number of data set samples, < > >For the dimension of the subsequence feature, k is the number of clusters, and the cluster center vector is { μ } 1 ，μ 2 ，...，μ b ，...，μ k -a }; sample x s And cluster center mu b The distance of (2) is:
Π(x s ，μ b ) Is x s Sum mu b Distribution of a set of all possible joint distributions combined, pi representing x s Sum mu b One of the joint distributions represented; wherein x is sf ，μ bf Respectively represent x s And mu b Is the f-th feature of (2).
To obtain the best number of clusters, an objective function is calculated by introducing an elbow method to obtain the best number of clusters:
and obtaining the optimal clustering number when the SSE value reaches the minimum.
Specifically, the AKCWD algorithm in step S2 includes the steps of:
s200, inputting a characteristic sample data set Q= { x s |s＝1，...，N}；
S201, let k be from 1 to k max Calculating clusters when the number of clusters is k, wherein k max Is the maximum number of clusters;
s202, randomly selecting k sample points in the characteristic sample data set Q, and respectively assigning the k sample points to an initial clustering center mu 1 ，μ 2 ，…，μ b ，…，μ k ；
S203, making S be 1-N circulation, wherein N represents the number of samples to be clustered, and calculating sample x b With each cluster center mu b Distance d (x) s ，μ b )；
S204, finding x s About cluster center μ b Will x s Are classified as mu b Cluster C with minimum distance b In (a) and (b);
s205, updating the cluster centers of all the clusters, and if the new cluster centers are not equal to the original cluster centers, updating the cluster centers;
S206, repeating the steps S203-S205 until the clustering center is not updated;
s207, calculating the timeSum of squares of pre-cluster errors SSE k ；
S208, calculating the difference between the current cluster error square sum and the last cluster error square sum, namely the decreasing amplitude delta SSE k ＝SSE k-1 -SSE k ；
S209, calculating the difference of the decrease amplitude variation, i.e. ΔΔSSE k ＝ΔSSE k-1 -ΔSSE k ；
S210, repeating step S201-step S209 until k=k max ；
S211, the optimal cluster number is delta SSE k The value of k is the largest, and the clustering result is the optimal clustering result;
s212, outputting a clustering result.
The algorithm flow chart is shown in fig. 3.
Further, in step S3, a driving behavior theme is first set, that is, a behavior description of a driving behavior scene is given according to characteristics of a trajectory curvature, a speed, an acceleration, and a steering angle. And then matching the driving behavior theme with driving behavior patterns obtained by all clusters, and selecting the most similar patterns as the representative of each class cluster. And finally, marking the class cluster with the highest matching degree as the driving behavior theme, and taking the class cluster as the identification result of the unsupervised driving behavior.
The essence of the driving behavior unsupervised pattern recognition is to convert continuous driving behavior data into a sequence of driving characteristics in natural language; for example, "rapid acceleration from low speed", "high speed cruising", and "rapid braking and stopping", thereby identifying driver behavior. Therefore, in the pattern recognition process, a large number of driving features can be extracted to realize semantic expression of driving behaviors, and such a recognition manner is effective for long-term prediction of driving behaviors, which does not require specifying a specific detection threshold, and can be analyzed by using behavior semantics.
The invention provides some common driving behavior theme descriptions, and total 11 driving behavior themes comprise 5 normal driving behavior themes and 6 abnormal driving behavior themes. Wherein, 5 normal driving behavior topics include: straight running at a constant speed, mild starting, curve stable running, normal lane changing and mild stopping; the 6 abnormal driving behavior subjects include: quick lane change, sudden braking, rapid overtaking, curve rapid running, sudden acceleration and curve overtaking. The theme basically comprises frequent driving behaviors of a driver in the driving process, has high representativeness and can reflect the driving style of the driver in the driving process. The driving behavior theme description is shown in table 3.
TABLE 3 Driving behavior topic description
Further, the invention uses a document topic generation model (LDA) to complete an unsupervised driving behavior recognition algorithm. The LDA model is essentially a three-layer Bayesian network, and unsupervised driving behavior matching is performed by calculating feature generation probability; set Ω= { Ω 1 ，...，Ω d ，...，Ω k And (3) representing driving behavior patterns, wherein the number of the driving behavior patterns is the number k of the extracted clusters, and O is the number of driving behavior topics, namely the behavior topics given in the table 3. The local driving behavior characteristic parameter observed in the driving behavior pattern Ω is expressed as Wherein U is d The number of local features is 8 features, namely a track curvature mean value, a track curvature variance, a speed mean value, a speed variance, an acceleration mean value, an acceleration variance, a steering angle mean value and a steering angle variance.
Here, LDA assumes a planchette feature w observed in the d-th driving behavior pattern d，u Is based on potential driving behavior z d，u Generated, i.e. assumed, features w d，u The actual value w of (2) is according toGenerated, then the LDA is assigned by θ o|d And Dirichlet parameter alpha, wherein phi 0 Is a feature profile with a potential driving topic with Dirichlet parameter β; thus, the generation model of LDA is assumed to be:
θ d ～Dir(θ；α) (4)
φ o ～Dir(φ；β) (5)
z d，u ～Mult(z；θ d ) (6)
when the feature w= { w is observed d，u The probability of generation of the feature is expressed as:
the algorithm model schematic diagram is shown in fig. 4, and the driving behavior subject with the highest probability is generated as the behavior matching result of the driving data segment.
The driving behavior recognition flowchart is shown in fig. 5. The driving behavior scene comprises physical characteristics such as acceleration mean value, acceleration variance, speed mean value, speed variance, steering angle mean value, steering angle variance, track curvature mean value, track curvature variance and the like, and the characteristics are sequentially matched with the characteristics of the driving behavior theme, and the generation probability is calculated. Finally, the semantic meaning of the driving behavior theme with the highest probability of matching with the actual driving behavior scene is searched and is given to the driving behavior scene to be used as a label of the driving behavior scene.
In order to obtain a more refined matching effect, each feature of the driving behavior scene is normalized to a numerical value between [0,1] and discretized into 20 intervals, the width of each discrete interval is set to be 0.05, and the theme tag of the driving behavior scene is determined by estimating the salient feature of the driving behavior scene. The 20 discrete sections such as the speed average are divided into five groups of "stationary" (section 0), "very slow running" (section 1 to section 5), "low-speed running" (section 6 to section 10), "medium-speed running" (section 11 to section 15), "high-speed running" (section 16 to section 20).
And translating 8 driving behavior physical characteristics of the driving behavior scene to obtain semantic tags corresponding to different discrete intervals. And matching the semantic tags with the semantics of the driving behavior theme description of the table 3 to finish the conversion from the unknown behavior to the known behavior. The feature semantic translations are shown in table 4.
TABLE 4 feature semantic translation Table
By integrating the behavior descriptions of the salient features, a behavior description of each driving topic is generated, the visual effect of which in a parallel coordinate system is shown in fig. 6. The vertical axis represents eight coordinate systems, namely a speed mean value, a speed variance, an acceleration mean value, an acceleration variance, a track rotation angle mean value, a track rotation angle variance, a track curvature mean value and a track curvature variance, wherein the coordinate systems have a value range of [0, 20], and represent 20 intervals after being discretized. As can be seen from the figure, the invention realizes the unsupervised recognition of driving behaviors by utilizing the LDA model and combining the characteristic semantic translation table without specific threshold setting.
To better represent a specific driving behavior, fig. 7 shows the visual effect of the rapid acceleration driving behavior in a parallel coordinate system. As can be seen from the figure, the speed average of the rapid acceleration driving behavior is distributed in the interval [5, 15], the speed variance is distributed in the interval [10, 15], the acceleration average is distributed in the interval [15, 20], the acceleration variance is distributed in the interval [15, 20], the track corner average is distributed in the interval [1,0], the track corner variance is distributed in the interval [1, 10], the track curvature average is distributed in the interval [1, 20], the track curvature variance is distributed in the interval [1, 20], and the rapid acceleration behavior is identified according to the combined action of 8 driving behavior characteristics.
Correspondingly, the embodiment of the invention also provides a data acquisition monitoring system based on the driving behavior unsupervised mode identification method, which comprises the following steps:
the data acquisition subsystem is used for acquiring driving data fragments of a driver in the driving process and extracting characteristics to realize local data monitoring, historical data storage and key real-time data uploading;
the cloud platform centralized monitoring subsystem is used for clustering the extracted features by using an improved Wasserstein distance-based self-adaptive K-Means clustering algorithm to obtain a driving behavior mode; and establishing a driving behavior theme description, and matching the driving behavior theme description with driving behavior patterns obtained by all clusters by using the LDA model to obtain an identification result of the unsupervised driving behavior.
The data acquisition subsystem is responsible for acquiring the data of abnormal driving behaviors of a multi-element time sequence, takes C++ as a development language, and embeds various data communication protocols such as 101, 102, 103, 104, modbus, CDT, DISA and the like of various IEC 60870-5; modeling accords with the requirements of an interface reference model, a Common Information Model (CIM) and a Component Interface Specification (CIS) in IEC 61970, accords with international standards, and can be used as middleware to be seamlessly integrated with each system.
The cloud platform centralized monitoring subsystem acquires real-time monitoring data from the data acquisition subsystem, the communication protocol between the two subsystems can adopt standard IEC104 protocol or other protocols, the real-time data acquisition frequency can support second level according to the protocol requirement, and modes such as variable quantity uploading, cyclic uploading and calling can be supported.
Fig. 8 is a schematic diagram of a data acquisition subsystem provided by an embodiment of the present invention, and fig. 9 is a data link and access network structure diagram of a cloud platform centralized monitoring subsystem provided by an embodiment of the present invention. The data acquisition subsystem comprises a data acquisition front end, a serial server, a computer, a display and data acquisition software, and realizes interactive control through a touch screen; the data acquisition subsystem provides two modes of automatic operation and manual operation, and in the automatic operation mode, the system is started or stopped by automatically receiving a trigger signal; in the manual operation mode, the operator starts or stops the system operation by pressing the operation button. The cloud platform centralized monitoring subsystem comprises a cloud data center and is used for monitoring, storing, retrieving, analyzing and publishing data.
The cloud data center stores real-time data of a monitoring process by using a real-time database, provides retrieval service, stores static data of a business process by using a business SQL database (Oracle or MYSQL), provides retrieval service, and uses an application mode of offline analysis data of a big data analysis platform, so that the real-time monitoring requirement of the running state of the vehicle running monitoring data can be supported, and the analysis requirements of various application-oriented and theme-oriented objects can be met. The database design organizes the management of the database in an object-oriented mode conforming to the natural mode of human thinking, and the speed and efficiency of data retrieval and search are improved.
The real-time database system is novel database management system software, and based on a high-speed database engine developed by a 64-bit system and an advanced distributed cluster architecture, the real-time database system is suitable for collecting, storing, retrieving and publishing massive real-time/historical data, has good horizontal expansion capability and high availability, and can process dynamic data which changes rapidly along with time.
The technical indexes of the real-time database system are as follows: 1) Scale of: support the label count scale of more than 100 ten thousand. 2) Speed of: high-speed real-time and historical data retrieval capability; real-time data millisecond-level response; historical data for the month span retrieves the second order response. 3) Storage type: support flexible and diverse multiple data types: boolean type; integer (8 bits/16 bits/32 bits/64 bits); floating point (32 bits/64 bits); date-type data (time stamp); others. 4) Efficient data compression: the method supports multiple lossless and lossy compression modes, greatly improves the storage efficiency, and improves the analysis and retrieval speeds in massive historical data; the two-stage compression capability is supported, the utilization rate of network resources can be effectively improved, the requirement on hardware is reduced, multi-stage buffering is provided, and the high availability of the system is improved; the configuration of the granularity of the points is supported, and the compression algorithm can be flexibly selected according to the characteristics of different data; up to tens of times the compression ratio; advanced distributed architecture: the cluster-based hot backup mechanism enables the system to have high availability of data; the distributed redundant storage architecture enables the system to have high elastic expansion capability; the disaster recovery mechanism fully ensures the high safety of the vehicle driving monitoring data. 5) Flexible data access interface: a C/c++/JAVA/JSON interface is provided for third party calls and writing of data.
Further, the data acquisition monitoring system further comprises a vehicle abnormal driving behavior monitoring APP, wherein the vehicle abnormal driving behavior monitoring APP comprises a user registration and login module, an online query module, an area display module and a modification module and is used for achieving query, display, online update and modification of vehicle driving information so as to enable management personnel to monitor in real time.
The APP is developed by using HBuilderX as a development tool and using HTML5+CSS+JavaScript, and is built by using a MUI front end framework. The APP realizes the registration and login functions, acquires real-time vehicle running monitoring data, has the functions of time sequence feature extraction, cluster analysis, pattern recognition, theme matching and the like, and displays the result in a visual interface.
The composition of the APP comprises:
the client uses MUI front end framework to develop and design, and uses HTML5, CSS and JavaScript language to develop the front end.
The server uses the ThinkJS server framework to develop, and is matched with a MySQL database, so that functions of registration, login verification, data transmission, addition, modification and deletion can be realized.
The system management background is developed by using HTML5, CSS and JavaScript language and is used for managing the database.
The MUI front-end framework (based on HTML5, CSS and JavaScript) is used for designing and developing an Android client and is also used for developing a system management background. The ThinkJS server end framework (based on NodeJS) is used for designing a logic interface to provide services for a client and a system management background, and corresponding functions are realized. The database MySQL is used for storing vehicle travel monitoring data and user information.
Further, the invention uses the driving data of partial vehicles in the vehicle monitoring data to verify the self-adaptive K-Means algorithm based on the Wasserstein distance and the abnormal behavior recognition algorithm based on the LDA algorithm.
The vehicle monitoring data is from experimental data of a certain automobile company, and records types of abnormal driving behaviors of drivers in different traffic conditions (including different road conditions and weather) of different driving styles. As shown in table 5:
table 5 driving behavior monitoring table
The raw vehicle monitoring data includes a plurality of dimensions such as GPS track point, acceleration, speed, accelerator pedal pressure, steering angle, battery voltage, etc. Because the intention of the invention is to learn and identify the abnormal driving behavior of the driver through the vehicle monitoring data, only data parameters with the driver control intention such as GPS track points, speed, acceleration, steering angle and the like are adopted. The data sampling period is 0.1 seconds, wherein each set of samples contains 10000 to 20000 sampling points. Fig. 10 shows a schematic diagram of vehicle monitoring data, which shows recorded data of a vehicle during driving, and is marked with data serial numbers, vehicle numbers, driver numbers, positioning time, vehicle speed, acceleration, position and corner information, wherein the vehicle speed, acceleration, position and corner information can be analyzed as main basis for abnormal driving behavior identification.
The invention adopts SD validity index (SD) index as the evaluation index of the final clustering effect. The SD evaluation algorithm relies on several metrics, the variance of the dataset and the variance of each cluster, respectively.
Let k represent the number of clusters, mu b Represents the center of the cluster, 1 < b < k, N represents the number of the cluster samples, and all the cluster samples are collected as Q= { x s S=1,.. b ，N b Is C b Number of elements in a matrix.
First, calculating a sample mean value:
dividing according to dimensions, wherein the data set variance of the f dimension is defined as follows:
wherein the method comprises the steps ofIs the mean +.>
According to the definition, the f-th dimension cluster C b Variance of (2)The definition is as follows:
for all clusters, the index for measuring the compactness of the clusters is defined as follows:
if the compactness of the clusters obtained by clustering is good, this result should be smaller than the variance of the dataset.
In addition, the degree of dispersion among the clusters as a whole is defined as follows:
wherein D is max Represents the maximum distance between the centers of any two clusters, D min Representing the minimum distance between the centers of any two clusters.
Finally, SD is defined as follows:
SD(k)＝αScat(k)+Dist(k) (14)
where α is a scaling factor.
The smaller the SD value result, the better the clustering effect is represented.
In consideration of the fact that drivers tend to show different driving behavior trends under different driving environments, three driving data fragment sets with large driving environment differences are selected for clustering analysis, and the three driving data fragment sets are respectively the driving environments of one type of weather and expressway, the driving environments of one type of weather and arterial road and the driving environments of three types of weather and expressway. One type of weather is good weather, and three types of weather are heavy rain, heavy snow and the like.
Tables 6, 7 and 8 are the clustering experiment results of one type of weather and expressway, one type of weather and arterial road, and three types of weather and expressway respectively.
TABLE 6 clustering experiments on weather and expressways
TABLE 7 clustering experiments on weather and arterial roads
Table 8 clustering experiments of three types of weather and expressways
From tables 6 to 8, it can be found that the adaptive K-Means clustering algorithm based on Wasserstein distance provided by the invention obtains better clustering effect under the evaluation index of SD. In addition, in different weather and road environments, the optimal cluster number often presents different distributions, and in combination with actual information, it can be speculated that when the weather is heavy, due to weather limitation, drivers may also present different driving styles, resulting in different driving behavior patterns.
In order to mark an actual driving scene as a specific driving behavior theme, taking a expressway as an example, 11 driving behavior scenes obtained by clustering in a driving environment under a weather type, and analyzing a driving behavior theme matching process based on an LDA model. Here, 4 typical driving behavior scenes are intercepted for feature semantic translation, as shown in FIG. 11, eight behavior features are extracted from each driving behavior scene, and the eight behavior features are respectively speed average valuesVelocity variance δv, acceleration mean +.>Acceleration variance δa, steering angle mean +.>Steering angle variance delta theta and track curvature mean +.>The track curvature variance deltac is discretized into 20 discrete values with the width of 0.05 and further divided into 5 groups of intervals 0, and intervals 1E [1,5 ]]Interval 2E [6, 10]Interval 3 epsilon [11, 15]Interval 4E [16, 20]。
According to the feature semantic translation table in table 4, the behavior feature interval value is converted into behavior description, and the discrete value is the behavior feature interval value. And translates into a behavioral description based on the magnitude of its characteristic value. Table 9 gives the results of the translation of 11 driving behavior scenarios from features to behavior descriptions.
Table 9 expressway, driving behavior translation under a class of weather
FIG. 12 shows driving scenario C 1 Representing the driving behavior semantic translation process of the driving behavior sample 1. Firstly calculating behavior feature values of a sample, secondly translating each behavior feature value into a driving behavior description, integrating 8 features to obtain a final driving behavior description, matching the final driving behavior description with a driving behavior theme, wherein the driving behavior theme with the highest matching degree is a normal lane change, and the final behavior recognition result is obtained.
FIG. 13 more intuitively shows recognition effects of different driving behavior topics based on LDA by adopting parallel coordinate systems, wherein the vertical axis has nine coordinate systems, and the first eight coordinate systems are respectively a speed mean value, a speed variance value, an acceleration mean value, an acceleration variance value, a track rotation angle mean value, a track rotation angle variance value, a track curvature mean value and a track curvature variance value, and each coordinate system takes a value of [0, 20] to represent 20 discrete intervals; the ninth coordinate system corresponds to the recognition result of the different driving behavior subjects.
In order to better show that different driving behaviors have different trend conditions on different parameters so as to verify the recognition effect of the driving subjects, the trend changes of three characteristics of speed, acceleration and steering angle in the four driving behavior subjects of rapid lane change, rapid acceleration, acceleration and deceleration and rapid left turn are respectively shown in fig. 14 a-14 d, fig. 15 a-15 d and fig. 16 a-16 d. Such as jerk behavior, acceleration followed by deceleration are presented in the figure, and there is a dramatic change in steering angle halfway, which is consistent with behavior expectations, consistent with the driving behavior theme description of table 4.
After the driving behavior subject is matched, the total driving behavior scene fragments of 10093 are combined, wherein the driving behavior scene fragments have 626 sudden braking/sudden acceleration behaviors, 516 rapid lane change behaviors, 501 rapid overtaking behaviors, 313 overspeed behaviors and 278 curved rapid running behaviors. Fig. 17 shows the distribution of four driving behaviors of rapid acceleration and deceleration, rapid turning, rapid overtaking, and rapid lane change, which are presented in the 3d scene, of acceleration, trajectory and steering angle characteristics. It can be seen that for the rapid lane change behavior, the acceleration is low, the rotation angle value is high, the track value is extremely high, and the behavior theme description given by priori knowledge is very met. For the sharp turning behavior and the sharp acceleration/deceleration behavior, the acceleration value is high, the steering angle value is low, the track characteristic distribution is wide, and the behavior theme description given by priori knowledge is also met.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (7)
1. An unsupervised mode recognition method for driving behavior is characterized by comprising the following steps:
s1, collecting driving data fragments of a driver in a driving process and extracting features;
The driving data segment comprises four variables of track curvature, acceleration, speed and steering angle; the extracted features are divided into two categories: statistical features and temporal features; the statistical features include: average, variance, ACF1, remainder autocorrelation, trend, curvature, and entropy; the time profile includes: maximum mean difference, maximum variance difference, maximum displacement of Kullback-Leibler divergence, and remainder variance;
s2, clustering the extracted features by using an improved self-adaptive K-Means clustering algorithm based on Wasserstein distance to obtain a driving behavior mode;
s3, establishing driving behavior theme description, and matching the driving behavior theme description with driving behavior modes obtained by all clusters by using an LDA model to obtain an identification result of the unsupervised driving behavior;
establishing a driving behavior theme description refers to extracting driving characteristics according to track curvature, speed, acceleration and steering angle to carry out semantic expression on driving behavior; the driving behavior topics comprise 5 normal driving behavior topics and 6 abnormal driving behavior topics;
the 5 normal driving behavior topics include:
the vehicle runs at a normal speed, the acceleration is very low, and the track is gentle; the corresponding driving behavior pattern is: straight running at a constant speed;
The speed is accelerated from 0, the acceleration is very low, and the track is gentle; the corresponding driving behavior pattern is: mild starting;
the vehicle runs at a normal speed, the acceleration is very low, and the track curvature is large; the corresponding driving behavior pattern is: the curve stably runs;
the vehicle runs at a normal speed, the acceleration is low, and the track change is obvious; the corresponding driving behavior pattern is: normal lane change;
the speed is reduced to 0, the acceleration is very low, and the track is gentle; the corresponding driving behavior pattern is: gently stopping;
the 6 abnormal driving behavior subjects include:
the running speed is high, and the track inclination angle acceleration is high; the corresponding driving behavior pattern is: fast lane changing;
the speed is rapidly reduced to 0, the acceleration is changed severely, and the track is gentle; the corresponding driving behavior pattern is: sudden braking;
the speed change is severe, the acceleration change is severe, the track change is severe and rapid, and the track change is expressed as two rapid lane changes; the corresponding driving behavior pattern is: rapidly overtaking;
the vehicle runs at a faster speed, the acceleration is low, and the track curvature is large; the corresponding driving behavior pattern is: the curve runs rapidly;
the speed is rapidly increased from 0, the acceleration is changed vigorously, and the track is gentle; the corresponding driving behavior pattern is: rapid acceleration;
The vehicle runs at a normal speed, the acceleration is higher, the track curvature is large, and the variance is large; the corresponding driving behavior pattern is: curve overtaking;
the unsupervised driving behavior recognition process based on the LDA model is as follows:
the LDA model is essentially a three-layer Bayesian network, and unsupervised driving behavior is performed by calculating feature generation probabilityMatching; set Ω= { Ω 1 ,...,Ω d ,...,Ω k The driving behavior pattern is represented, wherein the number of the driving behavior patterns is the number k of the extracted clusters, and O is the number of the driving behavior subjects; the local driving behavior characteristic parameter observed in the driving behavior pattern Ω is expressed asWherein U is d The number of local features is 8 features, namely a track curvature mean value, a track curvature variance, a speed mean value, a speed variance, an acceleration mean value, an acceleration variance, a steering angle mean value and a steering angle variance;
LDA assumes the u-th feature w observed in the d-th driving behavior pattern d,u Is based on potential driving behavior z d,u Generated, i.e. assumed, features w d,u The actual value w of (2) is according toGenerated, then the LDA is assigned by θ o|d And Dirichlet parameter alpha, wherein phi o Is a feature profile with a potential driving topic with Dirichlet parameter β; thus, the generation model of LDA is assumed to be:
θ d ～Dir(θ；α)
φ o ～Dir(φ；β)
z d,u ～Mult(z；θ d )
When the feature w= { w is observed d,u The probability of generation of the feature is expressed as:
and generating a driving behavior theme with the highest probability as a behavior matching result of the driving data fragment.
2. The driving behavior unsupervised pattern recognition method according to claim 1, wherein the step S2 specifically comprises:
s200, inputting a characteristic sample data set Q= { x s |s＝1,...,N}；
S201, let k be from 1 to k max Calculating clusters when the number of clusters is k, wherein k max Is the maximum number of clusters;
s202, randomly selecting k sample points in the characteristic sample data set Q, and respectively assigning the k sample points to an initial clustering center mu 1 ,μ 2 ,...,μ b ,...,μ k ；
S203, making S be 1-N circulation, wherein N represents the number of samples to be clustered, and calculating sample x b With each cluster center mu b Distance d (x) s ,μ b )；
S204, finding x s About cluster center μ b Will x s Are classified as mu b Cluster C with minimum distance b In (a) and (b);
s205, updating the cluster centers of all the clusters, and if the new cluster centers are not equal to the original cluster centers, updating the cluster centers;
s206, repeating the steps S203-S205 until the clustering center is not updated;
s207, calculating the current cluster error square sum SSE k ；
S208, calculating the difference between the current cluster error square sum and the last cluster error square sum, namely the decreasing amplitude delta SSE k ＝SSE k-1 -SSE k ；
S209, calculating the difference of the decrease amplitude variation, i.e. ΔΔSSE k ＝ΔSSE k-1 -ΔSSE k ；
S210, repeating step S201-step S209 until k=k max ；
S211, the optimal cluster number is delta SSE k The value of k is the largest, and the clustering result is the optimal clustering result;
s212, outputting a clustering result.
3. The driving behavior unsupervised pattern recognition method according to claim 2, wherein the wasperstein distance is defined as follows:
assuming that θ and iota represent two distributions, the distance between them is calculated by the following formula:
where pi (θ, iota) is the set of all possible joint distributions that combine the θ and iota distributions, pi representing one of the joint distributions represented by θ and iota, E (m,n)～Π [||m-n||]Calculating the distance expected value of the two distributions under the condition of the current joint distribution pi; after all the joint distributions are calculated, the method usesObtaining all expected lower bounds, namely Wasserstein distance values;
let the set of subsequences describe as x= (X 1 ，...，x s ，...，x N ) Wherein x is s ＝(x s1 ，x s2 ，...，x sf ，...，x sl ) T N is the total number of data set samples, l is the dimension of the subsequence feature, k is the number of clusters, and the cluster center vector is { mu } 1 ,μ 2 ,...,μ b ,...,μ k -a }; sample x s And cluster center mu b The distance of (2) is:
Π(x s ,μ b ) Is x s Sum mu b Distribution of a set of all possible joint distributions combined, pi representing x s Sum mu b One of the joint distributions represented; wherein x is sf ,μ bf Respectively represent x s And mu b Is the f-th feature of (2).
4. A driving behavior unsupervised pattern recognition method according to claim 3, wherein an elbow method is introduced to calculate an objective function to obtain an optimal number of clusters:
and obtaining the optimal clustering number when the SSE value reaches the minimum.
5. A data acquisition monitoring system based on the driving behavior unsupervised pattern recognition method according to any one of claims 1 to 4, characterized in that the data acquisition monitoring system comprises:
the data acquisition subsystem is used for acquiring driving data fragments of a driver in the driving process and extracting characteristics to realize local data monitoring, historical data storage and key real-time data uploading;
the cloud platform centralized monitoring subsystem is used for clustering the extracted features by using an improved Wasserstein distance-based self-adaptive K-Means clustering algorithm to obtain a driving behavior mode; and establishing a driving behavior theme description, and matching the driving behavior theme description with driving behavior patterns obtained by all clusters by using the LDA model to obtain an identification result of the unsupervised driving behavior.
6. The data acquisition and monitoring system according to claim 5, wherein the data acquisition subsystem comprises a data acquisition front end, a serial server, a computer, a display and data acquisition software, and interaction control is realized through a touch screen;
The data acquisition subsystem provides two modes of automatic operation and manual operation, and in the automatic operation mode, the system is started or stopped by automatically receiving a trigger signal; in the manual operation mode, an operator starts or stops the system work by pressing an operation button;
the cloud platform centralized monitoring subsystem comprises a cloud data center and is used for monitoring, storing, retrieving, analyzing and publishing data.
7. The data acquisition and monitoring system according to claim 5, further comprising a vehicle abnormal driving behavior monitoring APP, wherein the vehicle abnormal driving behavior monitoring APP comprises a user registration and login module, an online query module, a region display module and a modification module, and is used for implementing query, display, online update and modification of vehicle driving information for real-time monitoring by management personnel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110216457.6A CN113159105B (en) | 2021-02-26 | 2021-02-26 | Driving behavior unsupervised mode identification method and data acquisition monitoring system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110216457.6A CN113159105B (en) | 2021-02-26 | 2021-02-26 | Driving behavior unsupervised mode identification method and data acquisition monitoring system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113159105A CN113159105A (en) | 2021-07-23 |
| CN113159105B true CN113159105B (en) | 2023-08-08 |
Family
ID=76883511
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110216457.6A Active CN113159105B (en) | 2021-02-26 | 2021-02-26 | Driving behavior unsupervised mode identification method and data acquisition monitoring system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159105B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487936A (en) * | 2021-07-26 | 2021-10-08 | 广州星锐传媒信息科技有限公司 | Simulation driving judgment method based on multi-dimensional continuous signal analysis |
| CN113505852B (en) * | 2021-07-28 | 2023-07-28 | 云南电网有限责任公司电力科学研究院 | Low-voltage user phase identification method based on principal component analysis and EM distance |
| CN115204323B (en) * | 2022-09-16 | 2022-12-02 | 华智生物技术有限公司 | Seed multi-feature based clustering and synthesis method, system, device and medium |
| CN116975731B (en) * | 2023-08-08 | 2024-02-20 | 山东大学 | Cross-domain cutter damage monitoring method and system based on transfer learning |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015217707A (en) * | 2014-05-14 | 2015-12-07 | 株式会社豊田中央研究所 | Vehicle control device and program |
| WO2016009266A1 (en) * | 2014-07-16 | 2016-01-21 | Schlumberger Canada Limited | Systems and methods for estimating time of flight for an acoustic wave |
| CN107284442A (en) * | 2017-05-15 | 2017-10-24 | 北京理工大学 | A kind of longitudinally controlled method of negotiation of bends for automatic driving vehicle |
| CN108985488A (en) * | 2018-06-06 | 2018-12-11 | 福建工程学院 | The method predicted to individual trip purpose |
| CN109086794A (en) * | 2018-06-27 | 2018-12-25 | 武汉理工大学 | A kind of driving behavior mode knowledge method based on T-LDA topic model |
| CN109558482A (en) * | 2018-07-27 | 2019-04-02 | 中山大学 | A kind of parallel method of the text cluster model PW-LDA based on Spark frame |
| CN109849926A (en) * | 2019-01-29 | 2019-06-07 | 北京市交通信息中心 | A kind of method and system for distinguishing taxi and whether giving other people drive |
| CN109918476A (en) * | 2019-01-26 | 2019-06-21 | 北京工业大学 | A kind of subject retrieval method based on topic model |
| CN110175557A (en) * | 2019-05-24 | 2019-08-27 | 北京交通大学 | Driver's driving behavior method for identifying and classifying and system based on driving characteristics group |
| CN110171426A (en) * | 2019-05-23 | 2019-08-27 | 重庆大学 | A kind of recognition methods of driver's dangerous driving behavior and system based on smart machine |
| CN110471411A (en) * | 2019-07-26 | 2019-11-19 | 华为技术有限公司 | Automatic Pilot method and servomechanism |
| CN111126438A (en) * | 2019-11-22 | 2020-05-08 | 北京理工大学 | Driving behavior recognition method and system |
| CN111461185A (en) * | 2020-03-19 | 2020-07-28 | 哈尔滨工程大学 | Driving behavior analysis method based on improved K-means |
| CN111724423A (en) * | 2020-06-03 | 2020-09-29 | 西安交通大学 | Non-rigid body registration method of differential homoembryo based on fluid divergence loss |
| CN111752254A (en) * | 2019-03-28 | 2020-10-09 | 深圳市商汤科技有限公司 | Test method, device, medium and equipment for trajectory tracking controller |
| CN112339747A (en) * | 2020-10-30 | 2021-02-09 | 上海欧菲智能车联科技有限公司 | Automatic parking track generation method and device, electronic equipment and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8676937B2 (en) * | 2011-05-12 | 2014-03-18 | Jeffrey Alan Rapaport | Social-topical adaptive networking (STAN) system allowing for group based contextual transaction offers and acceptances and hot topic watchdogging |
| US10300922B2 (en) * | 2017-09-29 | 2019-05-28 | Denso International America, Inc. | Risk assessment system for assessing current driver behavior relative to past behavior and behaviors of other drivers |
| US11194331B2 (en) * | 2018-10-30 | 2021-12-07 | The Regents Of The University Of Michigan | Unsupervised classification of encountering scenarios using connected vehicle datasets |
-
2021
- 2021-02-26 CN CN202110216457.6A patent/CN113159105B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015217707A (en) * | 2014-05-14 | 2015-12-07 | 株式会社豊田中央研究所 | Vehicle control device and program |
| WO2016009266A1 (en) * | 2014-07-16 | 2016-01-21 | Schlumberger Canada Limited | Systems and methods for estimating time of flight for an acoustic wave |
| CN107284442A (en) * | 2017-05-15 | 2017-10-24 | 北京理工大学 | A kind of longitudinally controlled method of negotiation of bends for automatic driving vehicle |
| CN108985488A (en) * | 2018-06-06 | 2018-12-11 | 福建工程学院 | The method predicted to individual trip purpose |
| CN109086794A (en) * | 2018-06-27 | 2018-12-25 | 武汉理工大学 | A kind of driving behavior mode knowledge method based on T-LDA topic model |
| CN109558482A (en) * | 2018-07-27 | 2019-04-02 | 中山大学 | A kind of parallel method of the text cluster model PW-LDA based on Spark frame |
| CN109918476A (en) * | 2019-01-26 | 2019-06-21 | 北京工业大学 | A kind of subject retrieval method based on topic model |
| CN109849926A (en) * | 2019-01-29 | 2019-06-07 | 北京市交通信息中心 | A kind of method and system for distinguishing taxi and whether giving other people drive |
| CN111752254A (en) * | 2019-03-28 | 2020-10-09 | 深圳市商汤科技有限公司 | Test method, device, medium and equipment for trajectory tracking controller |
| CN110171426A (en) * | 2019-05-23 | 2019-08-27 | 重庆大学 | A kind of recognition methods of driver's dangerous driving behavior and system based on smart machine |
| CN110175557A (en) * | 2019-05-24 | 2019-08-27 | 北京交通大学 | Driver's driving behavior method for identifying and classifying and system based on driving characteristics group |
| CN110471411A (en) * | 2019-07-26 | 2019-11-19 | 华为技术有限公司 | Automatic Pilot method and servomechanism |
| CN111126438A (en) * | 2019-11-22 | 2020-05-08 | 北京理工大学 | Driving behavior recognition method and system |
| CN111461185A (en) * | 2020-03-19 | 2020-07-28 | 哈尔滨工程大学 | Driving behavior analysis method based on improved K-means |
| CN111724423A (en) * | 2020-06-03 | 2020-09-29 | 西安交通大学 | Non-rigid body registration method of differential homoembryo based on fluid divergence loss |
| CN112339747A (en) * | 2020-10-30 | 2021-02-09 | 上海欧菲智能车联科技有限公司 | Automatic parking track generation method and device, electronic equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| "Large-scale unusual time series detection";R.J.Hyndman等;《2015 IEEE international conference on data mining workshop》;第1616-1619页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113159105A (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113159105B (en) | Driving behavior unsupervised mode identification method and data acquisition monitoring system | |
| JP6026959B2 (en) | Driving scene label estimation device | |
| US9639521B2 (en) | Cognitive neuro-linguistic behavior recognition system for multi-sensor data fusion | |
| CN110428091A (en) | Risk Identification Method and relevant device based on data analysis | |
| CN109410588B (en) | Traffic accident evolution analysis method based on traffic big data | |
| JP2018502386A (en) | Mapper element for neuro-language behavior recognition system | |
| CN110727706A (en) | Method for rapidly extracting and grading risk driving scene for intelligent networking automobile test | |
| CN110633729A (en) | Driving risk hierarchical clustering method for intelligent networking vehicle group test | |
| CN109408578A (en) | One kind being directed to isomerous environment monitoring data fusion method | |
| JP2017073021A (en) | Driving support device | |
| JP2018503183A (en) | Vocabulary analyzer for neuro-language behavior recognition system | |
| CN111242387A (en) | Talent departure prediction method and device, electronic equipment and storage medium | |
| KR102028930B1 (en) | method of providing categorized video processing for moving objects based on AI learning using moving information of objects | |
| JP2017010206A (en) | Driving operation determination device | |
| CN110533094B (en) | Evaluation method and system for driver | |
| JP2017071333A (en) | Drive assisting device | |
| CN112084339B (en) | Traffic knowledge graph construction method based on cross-media data | |
| JP2019095878A (en) | Retrieval device for vehicle | |
| CN113919544B (en) | Crime early warning method and device, computer equipment and storage medium | |
| CN115457982A (en) | Pre-training optimization method, device, equipment and medium of emotion prediction model | |
| CN114169247A (en) | Method, device and equipment for generating simulated traffic flow and computer readable storage medium | |
| CN112686327A (en) | Method for clustering based on trajectory triple features | |
| CN111914138A (en) | Working condition analysis method, device and system based on Internet of vehicles big data | |
| Elnekave et al. | Discovering regular groups of mobile objects using incremental clustering | |
| Han et al. | A pattern accumulated compression method for trajectories constrained by urban road networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |