CN108108766A - Driving behavior recognition methods and system based on Fusion - Google Patents
Driving behavior recognition methods and system based on Fusion Download PDFInfo
- Publication number
- CN108108766A CN108108766A CN201711470875.8A CN201711470875A CN108108766A CN 108108766 A CN108108766 A CN 108108766A CN 201711470875 A CN201711470875 A CN 201711470875A CN 108108766 A CN108108766 A CN 108108766A
- Authority
- CN
- China
- Prior art keywords
- obd
- acceleration
- data
- information
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
Abstract
A kind of driving behavior recognition methods and system based on Fusion, pass through OBD heat transfer agents collector and the acceleration transducer being installed on vehicle, obtain the OBD sensing data acceleration informations of vehicle, and pass through two machine learning models and classify to characteristic information therein, it is independent to calculate the classification confidence of each classification results and classification accuracy in two models, classification results in two models are weighted fusion according to respective classification confidence and classification accuracy, finally with the classification results of Weighted Fusion numerical value maximum, as the recognition result to driving behavior.The present invention by Data fusion technique, can a variety of sensing datas of comprehensive consideration simultaneously, thus, accuracy of identification higher is more advantageous to assessing vehicle drive behavior.Moreover, identify that required most of sensing data can be directly obtained by electric vehicle own bus in the present invention, it is only necessary to acceleration transducer is added on vehicle, it is of low cost, suitable for promoting.
Description
Technical Field
The invention relates to the field of sensing data processing, in particular to a technology for identifying driving behaviors based on multi-sensor data.
Background
The driving behavior habit of drivers has been widely paid attention as an important reference for judging the safety condition of vehicles used by drivers and evaluating users in the automobile leasing and automobile insurance industries. The driving behavior-based recognition technology can reflect the specific driving habits of the driver in the driving process from one side, and further rank the driving risk of the driver according to the driving habits.
At present, the existing driving behavior recognition technology is generally only limited to fuel locomotives, and the research on the driving behavior recognition technology of electric automobiles is less. Also, currently, research on driving behavior recognition technology is mainly directed only to two directions: based on an analysis of the driving behaviour of the driver, or based on an analysis of the driving behaviour of the vehicle.
The former, such as Chiyomi Miyajima, yoshihiro Nishiwaki and Koji Ozawa, models the actuation of the accelerator brake pedal by different drivers, and proposes a hybrid gaussian algorithm by which the driving behavior of the driver is analyzed with an accuracy of approximately 80%.
Some scholars want to use the latter to reflect the driving condition of the vehicle through the sensing signal of the smart phone and recognize the driving behavior. Under the thought, liweijian and the like can identify whether the vehicle has an accident or not through the change degree of the acceleration sensor in the mobile phone.
Although both of the above-described research directions can classify driving behaviors, they have a common problem in that recognition accuracy is not high.
Considering that the number of sensors that can be mounted on an electric vehicle is large, a considerable number of driving data during driving behavior can be obtained actually by various sensors mounted on the vehicle. This means that the identification system input information source is sufficient, which with the use of calendars will make it possible to greatly improve the accuracy of the identification of the driving behaviour. However, the driving behavior recognition technology based on vehicle sensing data is almost blank at present.
Therefore, a driving behavior recognition technology based on sensor data is urgently needed to improve the recognition accuracy of the driving behavior of the electric vehicle.
Disclosure of Invention
In order to solve the defects in the prior art, the invention aims to provide a driving behavior identification method and system based on multi-sensor data fusion.
Firstly, in order to achieve the above purpose, a driving behavior recognition method based on multi-sensor data fusion is provided, which comprises the following steps:
firstly, acquiring OBD (On Board Diagnostics) sensing information and acceleration information of a vehicle, and preprocessing the OBD sensing information and the acceleration information to acquire OBD sensing data and acceleration data of the vehicle;
secondly, respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data;
thirdly, inputting the characteristic information of the OBD sensing data into a first machine learning model trained in advance for classification, and calculating a first classification confidence h1 and a first classification accuracy W1 of each classification result according to the classification data of the first machine learning model; inputting the characteristic information of the acceleration data into a second pre-trained machine learning model and a pre-trained learning model thereof for classification, and calculating a second classification confidence h2 and a second classification accuracy W2 of each classification result according to the classification data of the second machine learning model;
and fourthly, according to the first classification confidence coefficient h1, the first classification accuracy W1, the second classification confidence coefficient h2 and the second classification accuracy W2 obtained in the third step, performing weighted fusion on each classification result of the first machine learning model and the second machine learning model according to a formula W1 x h1+ W2 x h2, and outputting a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior.
Specifically, in the above method, in the first step, the pretreatment is a normalization treatment. The denoising processing adopts a wavelet denoising method.
Specifically, in the above method, in the second step, the feature information of the OBD sensing data includes a mean value, a variance, a maximum value, and a minimum value of the OBD sensing information; the characteristic information of the acceleration data is the time domain fluctuation characteristic and the stability of the acceleration data.
Specifically, in the above method, the time-domain fluctuation characteristic and the stability of the acceleration data are calculated by using an MMA (multiple finite analysis) method; the MMA method comprises the following specific steps:
step M1, calculating K pieces of acceleration data x k The profile signal sequence Y (i) of the formed acceleration sequence is more than or equal to 1 and less than or equal to K:
wherein K represents the acceleration data series x 1 ...x K The length of (a) is greater than (b),<x> represents a set of acceleration sequences intercepted with a fixed time window, i represents the ith data in the profile signal sequence, x k Representing the kth group of acceleration sequences; step M2, dividing the contour signal sequence Y (i) into N s Int (N/s) contiguous and non-overlapping subintervals V j (j＝1，2，...N s ) (ii) a Wherein s represents a scale parameter and N represents the number of data in a set of contour signal sequences Y (i);
step M3, calculating each subinterval V through a least square estimation algorithm j (j＝1，2，...N s ) Each subinterval V is calculated according to the following formula j (j＝1，2，...N s ) Variance G of 2 (s，v)：
Wherein v represents the sequence number of the subinterval; s represents a scale parameter representing the length of the subsequence;
y v (i) Is an n-th order (n is a positive integer range) fitting polynomial for each subinterval:
y v (i)＝a j0 +a j1 i+...a jn-1 i n-1 +a jn i n ，n＝1，2...
wherein n represents the regression order of the fitting polynomial (n =1,2,3 \8230;); a is jn A coefficient representing each order obtained by the above method;
step M4, calculating each subinterval V j (j＝1，2，...N s ) Q-order ripple function ofWherein q represents the fluctuation order (q is generally set to [ -5,5 [)](ii) a When q is a positive value, analysis on large fluctuation of data is indicated; when q is negative, it means that small fluctuations in data are analyzed);
step M5, calculating a Hurst surface function h (q, s) to obtain the time domain fluctuation characteristic and the stability of the acceleration data, wherein,wherein Ri represents a fitting window R i (i =1,2,3, \8230;, i takes the range of positive integers), the fitting window introduced here is a moving fitting window, F (q, s) Ri And s Ri Respectively corresponding values under the fitting serial port Ri; Δ F (q, s) Ri Representing the variation of F (q, s) values under different windows; Δ s Ri Representing the amount of change in the s-value for different windows.
Further, in order to fully utilize data, in the step M2 of the above method, N is respectively set at the head and tail ends of the sequence formed from the contour signal Y (i) s Step cutting of the sub-interval of the equivalent to int (N/s) to obtain 2N s A plurality of contiguous and non-overlapping subintervals;
then to the 2N s And calculating the time domain fluctuation characteristic and the stability of the acceleration data in the subintervals according to the steps from M3 to M5.
Specifically, in the above method, in the third step, the first machine learning model is a random forest classifier;
the first machine learning model is obtained by training the following steps:
step A1, creating a labeled training sample set { (O) of OBD sensing information 1 ，y 1 )，(O 2 ，y 2 )，…，(O k ，y k ) In which k is a sample number, k is not less than 2 k Represents the kth OBD sensory information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 OBD sensing information training sample; carrying out noise reduction and normalization processing on the OBD sensing information training sample to obtain OBD training sample sensing data of the vehicle;
step A2, extracting characteristic information of the OBD training sample sensing data;
step A3, inputting the characteristic information of the OBD training sample sensing data into the random forest classifier, and repeatedly extracting L times from the characteristic information of the OBD training sample sensing data through a self-help (bootstrap) resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples put into the classifier; then randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally larger than 1/3 of the attribute number, so that a decision tree is obtained, and each decision tree obtains a classification result;
step A4, the proportion h of the number of the decision trees of each classification result of the random forest classifier to the total number of all the decision trees 1 (p | x) as the classification data, obtaining a first classification confidence h1= h for each classification result 1 (p|x)；
The first classification accuracy W1 of each classification result isWherein N is acc1 Number, N, of labeled training sample sets representing OBD sensory information that are correctly classified p1 The total number of samples in the labeled training sample set representing the OBD sensory information.
Similarly, in the third step of the above method, the second machine learning model is a random forest classifier with the same parameters as the first machine learning model;
the second machine learning model is obtained by training the following steps:
step B1, creating a labeled training sample set { (x) of acceleration information 1 ，y 1 )，(x 2 ，y 2 )，…，(xk，y k ) H, where k is the sample index, k ≧ 2 k Represents the kth acceleration information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 acceleration information training sample; carrying out noise reduction processing and normalization processing on the acceleration information training samples to obtain acceleration training sample sensing data of the vehicle;
step B2; extracting characteristic information of the acceleration training sample sensing data;
step B3, inputting the characteristic information of the acceleration training sample sensing data into the random forest classifier, and repeatedly extracting the characteristic information of the acceleration training sample sensing data for L times through a self-help (bootstrap) resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples; then, randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally greater than 1/3 of the attribute quantity, so that a decision tree is obtained, and each decision tree obtains a classification result;
b4, the proportion h of the number of the decision trees of each classification result of the random forest classifier to the total number of all the decision trees 2 (p | x) as the classification data, obtaining a second classification confidence h2= h for each classification result 2 (p|x)；
In the second machine learning model, the second classification accuracy W2 of each classification result isWhere N is acc2 Number of correctly classified samples, N, representing acceleration information p2 Representing the total number of test samples.
Specifically, in the method of the present invention, in the first step, the OBD sensor information includes vehicle speed, motor torque, output current information of the motor, and motor power.
Secondly, in order to achieve the purpose, a driving behavior recognition system based on multi-sensor data fusion is also provided, and the driving behavior recognition system comprises an acceleration sensor and an OBD sensing information collector which are arranged on a vehicle, and a server; acceleration sensor with OBD sensing information collector all with the server is connected:
the acceleration sensor is used for acquiring acceleration information of the vehicle and uploading the acceleration information to the server;
the OBD sensing information collector is used for obtaining OBD sensing information of the vehicle and uploading the OBD sensing information to the server; the server is used for receiving the OBD sensing information and the acceleration information, and obtaining OBD sensing data and acceleration data of the vehicle after normalization processing is carried out on the OBD sensing information and the acceleration information; respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data; classifying the characteristic information of the OBD sensing data to obtain a first classification confidence h1 and a first classification accuracy W1; classifying the characteristic information of the acceleration data to obtain a second classification confidence h2 and a second classification accuracy W2; and finally, performing weighted fusion on each classification result according to a formula W1 h1+ W2 h2, calculating to obtain a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior, and outputting the recognition result.
Specifically, in the system, the acceleration sensor is a six-axis accelerometer.
Further, in the system, the server is a vehicle-mounted server or a remote server;
when the server selects a remote server, the driving behavior recognition system based on the multi-sensor data fusion also comprises a storage module and a data transceiver module;
the storage module is simultaneously connected with the acceleration sensor and the OBD sensing information collector, and is used for storing the acceleration information of the vehicle acquired by the acceleration sensor and the OBD sensing information of the vehicle acquired by the OBD sensing information collector and outputting the acceleration information and the OBD sensing information to the data transceiving module;
the data transceiver module is connected with the storage module and the remote server at the same time, and the data transceiver module is used for uploading the acceleration information and the OBD sensing information of the vehicle stored by the storage module to the remote server.
Advantageous effects
According to the invention, OBD sensing data acceleration data of a vehicle are obtained through an OBD sensing information collector and an acceleration sensor which are installed on the vehicle, characteristic information in the OBD sensing data acceleration data is classified through two machine learning models, the classification confidence coefficient and the classification accuracy of each classification result in the two models are independently calculated, the classification results in the two models are weighted and fused according to the classification confidence coefficient and the classification accuracy of each classification result, and finally the classification result with the maximum weighted fusion value is used as a recognition result of driving behaviors. The invention can comprehensively consider various sensing data at the same time through the data fusion technology, thereby having higher identification precision and being more beneficial to evaluating the driving behavior of the vehicle. Moreover, most of sensing data required by identification can be directly obtained through the bus of the electric vehicle, and only an acceleration sensor needs to be additionally arranged on the vehicle, so that the method is low in cost and suitable for popularization.
Further, the time-domain fluctuation characteristic and the stability of the acceleration data are calculated by an MMA (Multiscale multi-fractional analysis) method. In particular, the head and tail ends of the sequence of contour signals Y (i) can be respectively determined according to N s And (2) carrying out step cutting on the sub-intervals of the equivalent to int (N/s) to obtain 2Ns continuous and non-overlapping sub-intervals so as to improve the utilization rate of data and further improve the calculation precision. Of course, there are other methods for the processing of the acceleration signal, such as wavelet packet energy and SVD (singular value decomposition). But considering the problem of device transmission cost, the two methods for time domain require a large amount of time domainData, obviously requiring higher sampling costs. In the present invention, for cost, the acquisition frequency of the acceleration signal is 10 points per second, and each set of data has 50 points. The amount of such data limits the wavelet packet energy and the processing effect of SVD, unlike for the MMA algorithm. The MMA algorithm can efficiently extract the data valid information through the fluctuation and stability of the signal. A comparison experiment also proves that the MMA method has higher accuracy.
Further, the invention takes a random forest classifier as a machine learning model. There are many kinds of machine learning models, and in order to embody different classification effects of different classifiers on each sample, the present invention introduces the confidence h (p | x). The machine learning model may generally select random forests, naive Bayes and libsvm. For the three classifiers, a naive Bayes classifier is simpler, belongs to an early classifier, and has no advantage for the classification with excessive categories; whereas libsvm requires a higher choice of parameters. Because a plurality of decision trees are established in the random forest classification process, each decision tree obtains a classification result, and the classification result of each decision tree is voted to be determined as a final classification result. Therefore, the classification data can be directly the proportion h (p | x) of the number of decision trees of each classification result to the total number of all decision trees, and the classification correct proportion of each classification result can be used as the classification dataAs the classification accuracy. Through comparison experiments, the random forest has higher accuracy compared with other two forests, does not need too many parameter selections, and is more suitable for the method.
Particularly, in the structure of the device, as the CAN bus of the electric automobile is provided with most OBD sensing data required by the invention, the invention CAN directly acquire the information by a 0BD sensing information acquisition device. In addition, because the data on the CAN bus is updated in real time along with the running condition of the vehicle, the invention is more sensitive to the driving condition of the vehicle and has higher precision compared with the prior information obtained by an intermediate sensor or a data processing unit. Particularly, after various types of sensing information of the vehicle is obtained, the method also comprises preprocessing the information so as to extract characteristic information of the information. The characteristic information only retains relatively sensitive data characteristics, so that the deviation of the identification accuracy caused by the sensitivity of the sensor can be effectively corrected, and the identification accuracy of the invention is further improved. By increasing the number of training samples, the method can also effectively inhibit overfitting and further improve the identification precision. The remote server can greatly improve the computing capability of the system and is more convenient for the extension and the update of the training samples.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, are incorporated in and constitute a part of this specification, illustrate embodiments of the invention, and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart of a driving behavior recognition method based on multi-sensor data fusion according to the present invention;
FIG. 2 is a block diagram of a driving behavior recognition system based on multi-sensor data fusion in accordance with the present invention;
fig. 3 is a schematic flow chart of processing training samples or sensing data in the server according to the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it should be understood that they are presented herein only to illustrate and explain the present invention and not to limit the present invention.
Fig. 1 is a flow chart of a driving behavior recognition method based on multi-sensor data fusion according to the present invention, which includes the following steps:
the method comprises the steps that firstly, an On Board Diagnostics (OBD) of a vehicle is obtained, wherein OBD sensing information is actually data interaction is realized through a CAN bus of an electric vehicle, and some data under the running state of the electric vehicle are obtained;
secondly, respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data;
thirdly, inputting the characteristic information of the OBD sensing data into a first machine learning model trained in advance for classification, and calculating a first classification confidence h1 and a first classification accuracy W1 of each classification result according to the classification data of the first machine learning model; inputting the characteristic information of the acceleration data into a pre-trained second machine learning model and a pre-trained learning model thereof for classification, and calculating a second classification confidence h2 and a second classification accuracy W2 of each classification result according to the classification data of the second machine learning model; ( The specific calculation modes of the classification confidence coefficient are different for different machine learning models: taking a Bayesian model as an example, the posterior probability of the Bayesian model can be used as a classification confidence; for the classification accuracy, the accuracy of different classifiers can be effectively represented, that is, the classification capability of the classifier for the data sample is represented )
And fourthly, according to the first classification confidence coefficient h1, the first classification accuracy W1, the second classification confidence coefficient h2 and the second classification accuracy W2 obtained in the third step, performing weighted fusion on each classification result of the first machine learning model and the second machine learning model according to a formula W1 x h1+ W2 x h2, and outputting a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior.
Specifically, in the method, in the first step, the pretreatment is a normalization treatment. The denoising processing adopts a wavelet denoising method.
Specifically, in the above method, in the second step, the feature information of the OBD sensing data includes a mean value, a variance, a maximum value, and a minimum value of the OBD sensing information; the characteristic information of the acceleration data is the time domain fluctuation characteristic and the stability of the acceleration data.
Specifically, in the above method, the time-domain fluctuation characteristic and the stability of the acceleration data are calculated by using an MMA (multiple finite analysis) method; the MMA algorithm is a method of processing time domain signals. In effect, the standard multiple analysis algorithm incorporates a scaling factor. The fluctuation behavior of the signal and its stability can be evaluated by MMA algorithm analysis. The result of this algorithm is a Hurst surface describing the amplitude of variation of the Hurst index in the case of fluctuations with different parameters q (fluctuation order) and s (scale) [7]. However, in practice, the method has not been used for feature extraction, and the method is adopted herein, and the Hurst exponential sequence thereof is selected as the extracted feature. The MMA method comprises the following specific steps:
step M1, calculating K pieces of acceleration data x k The profile signal sequence Y (i) of the formed acceleration sequence is more than or equal to 1 and less than or equal to K:
wherein K represents the acceleration data series x 1 ...x K The length of (a) is greater than (b),<x&represents that the data collected during the period is intercepted by a fixed time window (for example, a time window of 5 seconds is selected in the embodiment) to be used as a group of acceleration sequences, i represents the ith data in the profile signal sequence, and x k Representing the kth group of acceleration sequences;
step M2, dividing the contour signal sequence Y (i) into N s Int (N/s) consecutive and non-overlapping subintervals V j (j＝1，2，...N s ) (ii) a Wherein s represents a scale parameter and N represents the number of data in a set of contour signal sequences Y (i);
step M3, passing the minimumCalculating each subinterval V by using a quadratic estimation algorithm j (j＝1，2，...N s ) Each subinterval V is calculated according to the following formula j (j＝1，2，...N s ) Variance G of 2 (s，v)：
Wherein v represents the sequence number of the subinterval; s represents a scale parameter representing the length of the subsequence; y is v (i) Denotes each subinterval V j (j＝1，2，...N s ) The fitting polynomial of order n (in this embodiment, the order n may be 5, and n may be a value within a positive integer range):
y v (i)＝a j0 +a j1 i+...a jn-1 i n-1 +a jn i n ，n＝1，2...
wherein n represents the regression order of the fitting polynomial (n =1,2,3 \8230;); a is a jn Coefficients representing each order of the general solution (as will be understood by those skilled in the art, this n-th order fitting polynomial is obtained by a toolbox, and the fitting polynomial can be obtained by knowing the subintervals and setting n);
step M4, calculating each subinterval V j (j＝1，2，...N s ) Of order q of a ripple functionWhere q represents the fluctuation order (q is an artificially set parameter, typically set to [ -5,5,5 [)]When q is a positive value, it means that large fluctuation of data is analyzed, and when q is a negative value, it means that small fluctuation of data is analyzed);
step M5, calculating a Hurst surface function h (q, s) to obtain the time domain fluctuation characteristic and the stability of the acceleration data, wherein,wherein Ri represents a fitting window R i (i =1,2,3, \8230; selected in this examplei to 5, the value range of i includes the whole positive integer range), ri is introduced as a moving fitting window, and the s value with quasi-continuous change can be brought into the q-order fluctuation function F (q, s) operation, F (q, s) Ri And s Ri Respectively representing a q-order fluctuation function F (q, s) under a Ri fitting window Ri The corresponding value and the value corresponding to the scale parameter s; Δ F (q, s) Ri Representing the variation of F (q, s) values under different windows; Δ s Ri Representing the amount of change in the s-value for different windows.
In the MMA method, the selection of a scale value s is crucial, and a moving fitting window is adopted in the MMA, and the quasi-continuous changing s value in the window is substituted into the F (q, s) operation. Assume a fitting window Ri (i =1, 2.., n), and hRi corresponds to the calculated h value in Ri. The value of h, h(s) = { h } can then be calculated for each fitting window R1 ，h R2 ，...，h Rn For a fixed value of q, values of h (q) can be obtained that vary quasi-continuously over a range of s. Repeating the above operations for different q values can obtain a generalized Hurst surface.
The h (q, s) obtained by the above method also has a certain physical significance: h e (0, 0.5) indicates that the time domain signal has certain reverse persistence, h =0.5 indicates that the signal is uncorrelated noise, h e (0.5, 1) indicates that the signal is persistent, h =1.5 indicates that the object is making brownian motion, and h > 2 indicates that the object is black noise.
Further, in order to fully utilize data, in the step M2 of the above method, N is respectively set from the head and tail ends of the sequence of the contour signal Y (i) s Step cutting of the sub-interval of equivalent to int (N/s) to obtain 2N s A plurality of contiguous and non-overlapping subintervals;
then to the 2N s And calculating the time domain fluctuation characteristic and the stability of the acceleration data in the subintervals according to the steps from M3 to M5.
Specifically, in the above method, in the third step, the first machine learning model is a random forest classifier. Random forests are decision tree-based integrated methods that can be used for classification, regression, and other ensemble learning. The minimum decision unit of the random forest is a random decision tree, a plurality of decision trees are constructed during training, a plurality of output classification result modes are obtained according to the decision trees, and finally, the decision results are comprehensively voted by the decision trees. In the embodiment, a plurality of decision trees are established by utilizing a random forest, each decision tree obtains a classification result, the classification result of each decision tree is voted, and the class with the highest vote number is used as the final classification result;
the first machine learning model is obtained by training the following steps:
step A1, creating a labeled training sample set { (O) of OBD sensing information 1 ，y 1 )，(O 2 ，y 2 )，…，(O k ，y k ) In which k is a sample number, k is not less than 2 k Represents the kth OBD sensory information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 OBD sensing information training sample; carrying out normalization processing on the OBD sensing information training sample to obtain the OBD training sample sensing data of the vehicle;
a2, extracting characteristic information of the OBD training sample sensing data;
step A3, inputting the characteristic information of the OBD training sample sensing data into the random forest classifier, and repeatedly extracting L times from the characteristic information of the OBD training sample sensing data through a self-help (bootstrap) resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples put into the classifier; then randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally larger than 1/3 of the attribute number, so that a decision tree is obtained, and each decision tree obtains a classification result;
step A4, the proportion h of the number of the decision trees of each classification result of the random forest classifier to the total number of all the decision trees 1 (p | x) as the classification data, obtaining a first classification confidence h1= h for each classification result 1 (p|x)；
The first classification accuracy W1 of each classification result isWherein N is acc1 Number, N, of labeled training sample sets representing OBD sensory information that are correctly classified p1 The total number of samples in the labeled training sample set representing the OBD sensory information.
Similarly, in the third step of the method, the second machine learning model is a random forest classifier with the same parameters as the first machine learning model;
the second machine learning model is obtained by training the following steps:
step B1, creating a labeled training sample set { (x) of acceleration information 1 ，y 1 )，(x 2 ，y 2 )，…，(x k ，y k ) Where k is the sample number, k is 2,x k Represents the kth acceleration information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 acceleration information training sample; carrying out normalization processing on the acceleration information training sample to obtain the acceleration training sample sensing data of the vehicle;
step B2; extracting characteristic information of the acceleration training sample sensing data;
step B3, inputting the characteristic information of the acceleration training sample sensing data into the random forest classifier, and repeatedly extracting L times from the characteristic information of the acceleration training sample sensing data through a self-help (bootstrap) resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples; then, randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally greater than 1/3 of the attribute quantity, so that a decision tree is obtained, and each decision tree obtains a classification result; here exactly in line with the classifier step employed in step 6;
step B4The proportion h of the number of decision trees of each classification result of the random forest classifier to the total number of all decision trees 2 (p | x) as the classification data, obtaining a second classification confidence h2= h for each classification result 2 (p|x)；
In the second machine learning model, the second classification accuracy W2 of each classification result isWhere N is acc2 Number of correctly classified samples, N, representing acceleration information p2 Representing the total number of test samples.
Specifically, in the method of the present invention, in the first step, the OBD sensor information includes vehicle speed, motor torque, output current information of the motor, and motor power.
To better plan the driving behavior, the model built here is mainly used to identify the following driving behaviors: accelerating, decelerating, turning left, turning right, changing left lane, changing right lane, and driving normally.
Secondly, referring to fig. 2, in order to achieve the above object, a driving behavior recognition system based on multi-sensor data fusion is further provided, which includes an acceleration sensor and an OBD sensing information collector mounted on a vehicle, and further includes a server; acceleration sensor with OBD sensing information collector all with the server is connected:
the acceleration sensor is used for acquiring acceleration information of a vehicle and uploading the acceleration information to the server;
the OBD sensing information collector is used for obtaining OBD sensing information of the vehicle and uploading the OBD sensing information to the server; the server is used for receiving the OBD sensing information and the acceleration information, and obtaining OBD sensing data and acceleration data of the vehicle after denoising and normalizing the OBD sensing information and the acceleration information; respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data; classifying the characteristic information of the OBD sensing data to obtain a first classification confidence h1 and a first classification accuracy W1; classifying the characteristic information of the acceleration data to obtain a second classification confidence h2 and a second classification accuracy W2; and finally, performing weighted fusion on each classification result according to a formula W1 h1+ W2 h2, calculating to obtain a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior, and outputting the recognition result.
The system only needs to be provided with an accelerometer and an OBD acquisition instrument on the electric automobile. The OBD information is provided by the electric automobile, the accuracy rate of the OBD information is quite reliable, the needed cost is low, and the OBD information is suitable for mass production. The result obtained by the method can be used as a basis for judging the safe driving condition of the driver, and provides a scientific basis for providing differentiated products and services for automobile leasing and vehicle insurance industries.
Furthermore, different signal data are acquired through each sensor, characteristics of the signal data are extracted and used for training of a machine learning model, different sensor data have different sensitivities to different driving behaviors, and then information of the sensors is fused through the information fusion method, so that the identification accuracy can be effectively improved. The driving behavior of the driver can be effectively distinguished. By adding the data enhancement processing step, the amount of training sample data is increased, and overfitting is inhibited.
Specifically, in the system, the acceleration sensor is a six-axis accelerometer in the MEMS acquisition system, and the acceleration information acquired by the acceleration sensor includes acceleration information in three directions of X, Y, and Z axes.
Further, in the system, the server is a vehicle-mounted server or a remote server;
when the server selects a remote server, the driving behavior recognition system based on multi-sensor data fusion also comprises a storage module and a data transceiving module;
the storage module is simultaneously connected with the acceleration sensor and the OBD sensing information collector, and is used for storing the acceleration information of the vehicle acquired by the acceleration sensor and the OBD sensing information of the vehicle acquired by the OBD sensing information collector and outputting the acceleration information and the OBD sensing information to the data transceiver module;
the data transceiver module is connected with the storage module and the remote server at the same time, and the data transceiver module is used for uploading the acceleration information and the OBD sensing information of the vehicle stored by the storage module to the remote server.
The method applies the machine learning method to the electric vehicle driving behavior identification for the first time, and analyzes the collected related sensor data through the server to obtain the correct driving behavior of the driver. The invention CAN directly analyze the vehicle CAN bus data, thereby avoiding increasing the number of sensors meaninglessly, effectively controlling the cost and obviously improving the identification result of the driving behavior. The result obtained by the method can be used as a basis for judging the safe driving condition of the driver, and provides a scientific basis for providing differentiated products and services for automobile leasing and vehicle insurance industries.
By adopting the method for identifying the driving behaviors of the electric vehicle based on the information fusion, higher accuracy can be obtained. Reference is made to the following comparative data:
1) The characteristic value of the acceleration information is extracted through an MMA algorithm, a classifier is trained through a random forest algorithm, and the obtained classification result is as follows
TABLE 1 acceleration information recognition results based on MMA algorithm (based on random forest)
2) By carrying out the feature extraction on the OBD information data and training a classifier by adopting a random forest algorithm, the obtained classification result is as follows:
TABLE 2 recognition results based on ODB information (based on random forest)
3) The classification result obtained by the information fusion algorithm is as follows:
TABLE 3 identification results based on information fusion method (based on random forest)
Through comparing the three experiments, when the driving behavior is analyzed by independently using the acceleration information and the OBD information, although the driving behavior can be identified to a certain degree, the accuracy of the result is not high, and the accuracy is greatly improved by adopting the information fusion method, so that the information fusion method has considerable feasibility.
The technical scheme of the invention has the advantages that:
the invention collects different sensor information in the driving process of the electric automobile through a plurality of sensors of the automobile, processes original data through an effective characteristic processing method, extracts the characteristics of signals, adopts a machine learning method to obtain a training model, classifies the collected data by using the training model, and simultaneously adopts an information fusion method to fuse the information of the plurality of sensors to obtain a final classification result. The invention can automatically identify different driving behaviors of the driver of the electric automobile, thereby judging the safe driving condition of the driver. The method provides a basis for providing differentiated products and services for the electric automobile leasing and vehicle insurance industries. Compared with a traditional machine learning method, the method has the advantages that through information fusion, the identification result has higher identification precision, and the identification of the driving behaviors of the electric automobile is facilitated.
Those of ordinary skill in the art will understand that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A driving behavior identification method based on multi-sensor data fusion is characterized by comprising the following steps:
the method comprises the steps of firstly, obtaining OBD sensing information and acceleration information of a vehicle, preprocessing the OBD sensing information and the acceleration information, and obtaining OBD sensing data and acceleration data of the vehicle;
secondly, respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data;
thirdly, inputting the characteristic information of the OBD sensing data into a first machine learning model trained in advance for classification, and calculating a first classification confidence h1 and a first classification accuracy W1 of each classification result according to the classification data of the first machine learning model; inputting the characteristic information of the acceleration data into a pre-trained second machine learning model and a pre-trained learning model thereof for classification, and calculating a second classification confidence h2 and a second classification accuracy W2 of each classification result according to the classification data of the second machine learning model;
and fourthly, according to the first classification confidence coefficient h1, the first classification accuracy W1, the second classification confidence coefficient h2 and the second classification accuracy W2 obtained in the third step, performing weighted fusion on each classification result of the first machine learning model and the second machine learning model according to a formula W1 x h1+ W2 x h2, and outputting a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior.
2. The driving behavior recognition method based on multi-sensor data fusion of claim 1, wherein in the second step, the characteristic information of the OBD sensing data comprises a mean value, a variance, a maximum value and a minimum value of the OBD sensing data; the characteristic information of the acceleration data is the time domain fluctuation characteristic and the stability of the acceleration data.
3. The multi-sensor data fusion-based driving behavior recognition method according to claim 2, wherein the time-domain fluctuation characteristics and stability of the acceleration data are calculated by an MMA (moving asymptote) method; the MMA method comprises the following specific steps:
step M1, calculating K pieces of acceleration data x k The profile signal sequence Y (i) of the formed acceleration sequence is more than or equal to 1 and less than or equal to K:
wherein K represents the acceleration data sequence x 1 ...x K The length of (a) is greater than (b),<x> represents a set of acceleration sequences intercepted with a fixed time window, i represents the ith data in the profile signal sequence, x k Representing the kth group of acceleration sequences;
step M2, dividing the contour signal sequence Y (i) into N s Int (N/s) contiguous and non-overlapping subintervals V j (j＝1，2，...N s ) (ii) a Wherein s represents a scale parameter and N represents the number of data in a set of contour signal sequences Y (i);
step M3, calculating each subinterval V through a least square estimation algorithm j (j＝1，2，...N s ) Each subinterval V is calculated according to the following formula j (j＝1，2，...N s ) Variance G of 2 (s，v)：
Wherein v denotes the sequence number of the subinterval, s denotes the scale parameter,
y v (i) Is each sub-areaN-th order fit polynomial between:
y v (i)＝a j0 +a j1 i+...aj n-1 i n-1 +a jn i n ，n＝1，2...
n represents the regression order of the fitting polynomial (n =1,2,3 \ 8230;); ajn represents a coefficient of each order obtained by the above method;
step M4, calculating each subinterval V j (j＝1，2，...N s ) Of order q of a ripple functionWherein q represents the fluctuation order;
step M5, calculating Hutst surface functions h (q, s) to obtain the time domain fluctuation characteristic and the stability of the acceleration data, wherein,wherein R is i Represents a fitting window R i (i =1,2,3, \ 8230;, positive integer range); f (q, s) Ri And s Ri Are all the corresponding values under the Ri window, Δ F (q, s) Ri Representing the variation of F (q, s) values under different windows; Δ s Ri And representing the variation of the s value under different windows.
4. The driving behavior recognition method based on multi-sensor data fusion of claim 3, wherein in the step M2, the head and tail ends of the sequence formed by the contour signals Y (i) are respectively in accordance with N s Step cutting of the sub-interval of equivalent to int (N/s) to obtain 2N s A plurality of contiguous and non-overlapping subintervals; then to the 2N s And calculating the time domain fluctuation characteristic and the stability of the acceleration data in the subintervals according to the steps from M3 to M5.
5. The multi-sensor data fusion-based driving behavior recognition method of claim 1, wherein in the third step, the first machine learning model is a random forest classifier;
the first machine learning model is obtained by training the following steps:
step A1, creating a labeled training sample set { (O) of OBD sensing information 1 ，y 1 )，(O 2 ，y 2 )，…，(O k ，y k ) In which k is a sample number, k is not less than 2 k Represents the kth OBD sensory information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 OBD sensing information training sample; carrying out normalization processing on the OBD sensing information training sample to obtain the OBD training sample sensing data of the vehicle;
a2, extracting characteristic information of the OBD training sample sensing data;
step A3, inputting the characteristic information of the OBD training sample sensing data into the random forest classifier, and repeatedly extracting L times from the characteristic information of the OBD training sample sensing data through a self-help resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples put into the classifier; then randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally larger than 1/3 of the attribute number, so that a decision tree is obtained, and each decision tree obtains a classification result;
step A4, the proportion h of the number of the decision trees of each classification result of the random forest classifier to the total number of all the decision trees 1 (p | x) as the classification data, obtaining a first classification confidence h1= h for each classification result 1 (p|x)；
The first classification accuracy W1 of each classification result is
Where N is acc1 Number, N, of labeled training sample sets representing OBD sensory information that are correctly classified p1 The total number of samples in the labeled training sample set representing the OBD sensory information.
6. The multi-sensor data fusion-based driving behavior recognition method of claim 5, wherein in the third step, the second machine learning model is a random forest classifier with the same parameters as the first machine learning model;
the second machine learning model is obtained by training the following steps:
step B1, creating a labeled training sample set { (x) of acceleration information 1 ，y 1 )，(x 2 ，y 2 )，…，(x k ，y k ) Where k is the sample number, k is 2,x k Represents the kth acceleration information training sample, y k A label value representing a kth training sample; each label in the labeled training sample set at least comprises 1 acceleration information training sample; carrying out normalization processing on the acceleration information training sample to obtain the acceleration training sample sensing data of the vehicle;
step B2; extracting characteristic information of the acceleration training sample sensing data;
step B3, inputting the characteristic information of the acceleration training sample sensing data into the random forest classifier, and repeatedly extracting the characteristic information of the acceleration training sample sensing data for L times through a self-help resampling technology by the random forest classifier to obtain a sample set, wherein L is the total number of samples; then randomly selecting m training sample attribute values to form a new sample set, wherein m is a set parameter and is generally larger than 1/3 of the attribute number, so that a decision tree is obtained, and each decision tree obtains a classification result;
step B4, the proportion h of the number of the decision trees of each classification result of the random forest classifier to the total number of all the decision trees 2 (p | x) as the classification data, obtaining a second classification confidence h2= h for each classification result 2 (p|X)
In the second machine learning model, the second classification accuracy W2 of each classification result is
Where N is acc2 Number of correctly classified samples, N, representing acceleration information p2 Representing the total number of samples tested.
7. The driving behavior recognition method based on multi-sensor data fusion of claim 1, wherein in the first step, the OBD sensing information comprises vehicle speed, motor torque, output current information of the motor, and motor power.
8. A driving behavior recognition system based on multi-sensor data fusion is characterized by comprising an acceleration sensor, an OBD sensing information collector and a server, wherein the acceleration sensor and the OBD sensing information collector are arranged on a vehicle; the acceleration sensor and the OBD sensing information collector are both connected with the server;
the acceleration sensor is used for acquiring acceleration information of a vehicle and uploading the acceleration information to the server;
the OBD sensing information collector is used for obtaining OBD sensing information of the vehicle and uploading the OBD sensing information to the server;
the server is used for receiving the OBD sensing information and the acceleration information, and carrying out normalization processing on the OBD sensing information and the acceleration information to obtain OBD sensing data and acceleration data of the vehicle; respectively extracting characteristic information of the OBD sensing data and characteristic information of the acceleration data; classifying the characteristic information of the OBD sensing data to obtain a first classification confidence h1 and a first classification accuracy W1; classifying the characteristic information of the acceleration data to obtain a second classification confidence h2 and a second classification accuracy W2; and finally, performing weighted fusion on each classification result according to a formula W1 h1+ W2 h2, calculating to obtain a classification result with the maximum weighted fusion value, namely a recognition result of the driving behavior, and outputting the recognition result.
9. The multi-sensor data fusion-based driving behavior recognition system of claim 8, wherein the acceleration sensor is a six-axis accelerometer.
10. The multi-sensor data fusion based driving behavior recognition system of claim 9, wherein the server is an onboard server or a remote server;
if the server is a remote server, the driving behavior recognition system based on multi-sensor data fusion also comprises a storage module and a data transceiving module;
the storage module is simultaneously connected with the acceleration sensor and the OBD sensing information collector, and is used for storing the acceleration information of the vehicle acquired by the acceleration sensor and the OBD sensing information of the vehicle acquired by the OBD sensing information collector and outputting the acceleration information and the OBD sensing information to the data transceiving module;
the data transceiver module is connected with the storage module and the remote server at the same time, and the data transceiver module is used for uploading the acceleration information and the OBD sensing information of the vehicle stored by the storage module to the remote server.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711470875.8A CN108108766B (en) | 2017-12-28 | 2017-12-28 | Driving behavior identification method and system based on multi-sensor data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711470875.8A CN108108766B (en) | 2017-12-28 | 2017-12-28 | Driving behavior identification method and system based on multi-sensor data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108108766A true CN108108766A (en) | 2018-06-01 |
| CN108108766B CN108108766B (en) | 2021-10-29 |
Family
ID=62214485
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711470875.8A Active CN108108766B (en) | 2017-12-28 | 2017-12-28 | Driving behavior identification method and system based on multi-sensor data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108108766B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921096A (en) * | 2018-06-29 | 2018-11-30 | 北京百度网讯科技有限公司 | Time tracking method, apparatus, equipment and computer-readable medium |
| CN109229108A (en) * | 2018-08-07 | 2019-01-18 | 武汉理工大学 | A kind of driving behavior safe evaluation method based on driving fingerprint |
| CN109447127A (en) * | 2018-09-29 | 2019-03-08 | 深圳市元征科技股份有限公司 | Data processing method and device |
| CN109614904A (en) * | 2018-12-03 | 2019-04-12 | 东北大学 | A kind of activity recognition method of the Multi-sensor Fusion based on Shapelet |
| CN109635335A (en) * | 2018-11-12 | 2019-04-16 | 平安科技（深圳）有限公司 | Drive Risk Forecast Method, device, computer equipment and storage medium |
| CN109922439A (en) * | 2019-03-29 | 2019-06-21 | 重庆长安汽车股份有限公司 | The fusion method of multi-sensor data, the object detection method based on V2X and system |
| CN110210513A (en) * | 2019-04-23 | 2019-09-06 | 深圳信息职业技术学院 | Data classification method, device and terminal device |
| CN110365648A (en) * | 2019-06-14 | 2019-10-22 | 东南大学 | A kind of vehicle-mounted CAN bus method for detecting abnormality based on decision tree |
| CN110447214A (en) * | 2018-03-01 | 2019-11-12 | 北京嘀嘀无限科技发展有限公司 | A kind of system, method, apparatus and storage medium identifying driving behavior |
| CN111104959A (en) * | 2018-10-25 | 2020-05-05 | 激发认知有限公司 | Method and apparatus for machine learning classifier generation |
| CN111376910A (en) * | 2018-12-29 | 2020-07-07 | 北京嘀嘀无限科技发展有限公司 | User behavior identification method and system and computer equipment |
| CN111814836A (en) * | 2020-06-12 | 2020-10-23 | 武汉理工大学 | Vehicle driving behavior detection method and device based on class imbalance algorithm |
| CN111923927A (en) * | 2019-05-13 | 2020-11-13 | 长城汽车股份有限公司 | Method and apparatus for interactive perception of traffic scene prediction |
| CN112099471A (en) * | 2020-08-27 | 2020-12-18 | 深圳市元征科技股份有限公司 | Vehicle power type judgment method and device, terminal device and storage medium |
| CN112348039A (en) * | 2019-08-07 | 2021-02-09 | 中国移动通信集团上海有限公司 | Training method of driving behavior analysis model, driving behavior analysis method and equipment |
| CN112567437A (en) * | 2018-08-17 | 2021-03-26 | 索尼公司 | Information processing device, information processing system, information processing method, and program |
| CN112889270A (en) * | 2018-10-22 | 2021-06-01 | 松下电器（美国）知识产权公司 | Information processing method, information processing device, information processing program, and information processing system |
| CN112883190A (en) * | 2021-01-28 | 2021-06-01 | 平安科技（深圳）有限公司 | Text classification method and device, electronic equipment and storage medium |
| CN113112796A (en) * | 2021-04-07 | 2021-07-13 | 吉林大学 | Construction method and system of driving behavior characteristics and driving behavior analysis method |
| WO2022011722A1 (en) * | 2020-07-17 | 2022-01-20 | 华为技术有限公司 | Data processing method and apparatus, and intelligent vehicle |
| CN114169374A (en) * | 2021-12-10 | 2022-03-11 | 湖南工商大学 | Cable-stayed bridge stay cable damage identification method and electronic equipment |
| CN115496978A (en) * | 2022-09-14 | 2022-12-20 | 北京化工大学 | Image and vehicle speed information fused driving behavior classification method and device |
| CN116702000A (en) * | 2023-05-31 | 2023-09-05 | 深圳技术大学 | Road surface quality dynamic monitoring and evaluating method based on multi-layer data fusion |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060204107A1 (en) * | 2005-03-04 | 2006-09-14 | Lockheed Martin Corporation | Object recognition system using dynamic length genetic training |
| CN101633359A (en) * | 2008-07-24 | 2010-01-27 | 通用汽车环球科技运作公司 | Adaptive vehicle control system with driving style recognition |
| CN101872171A (en) * | 2009-04-24 | 2010-10-27 | 中国农业大学 | Driver fatigue state recognition method and system based on information fusion |
| CN102023010A (en) * | 2010-10-26 | 2011-04-20 | 西安理工大学 | MEMS (micro-electromechanical system)-based wavelet field multisensor information fusion system and fusion method |
| CN103310202A (en) * | 2013-06-27 | 2013-09-18 | 西安电子科技大学 | System and method for guaranteeing driving safety |
| CN104463244A (en) * | 2014-12-04 | 2015-03-25 | 上海交通大学 | Aberrant driving behavior monitoring and recognizing method and system based on smart mobile terminal |
| CN105320966A (en) * | 2015-10-30 | 2016-02-10 | 东软集团股份有限公司 | Vehicle driving state recognition method and apparatus |
| CN106203437A (en) * | 2015-05-07 | 2016-12-07 | 平安科技（深圳）有限公司 | Individual driving behavior recognition methods and device |
-
2017
- 2017-12-28 CN CN201711470875.8A patent/CN108108766B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060204107A1 (en) * | 2005-03-04 | 2006-09-14 | Lockheed Martin Corporation | Object recognition system using dynamic length genetic training |
| CN101633359A (en) * | 2008-07-24 | 2010-01-27 | 通用汽车环球科技运作公司 | Adaptive vehicle control system with driving style recognition |
| CN101872171A (en) * | 2009-04-24 | 2010-10-27 | 中国农业大学 | Driver fatigue state recognition method and system based on information fusion |
| CN102023010A (en) * | 2010-10-26 | 2011-04-20 | 西安理工大学 | MEMS (micro-electromechanical system)-based wavelet field multisensor information fusion system and fusion method |
| CN103310202A (en) * | 2013-06-27 | 2013-09-18 | 西安电子科技大学 | System and method for guaranteeing driving safety |
| CN104463244A (en) * | 2014-12-04 | 2015-03-25 | 上海交通大学 | Aberrant driving behavior monitoring and recognizing method and system based on smart mobile terminal |
| CN106203437A (en) * | 2015-05-07 | 2016-12-07 | 平安科技（深圳）有限公司 | Individual driving behavior recognition methods and device |
| CN105320966A (en) * | 2015-10-30 | 2016-02-10 | 东软集团股份有限公司 | Vehicle driving state recognition method and apparatus |
Non-Patent Citations (2)
| Title |
|---|
| CHIYOMI MIYAJIMA 等: "Driver Modeling Based on Driving Behavior and Its Evaluation in Driver Identification", 《PROCEEDINGS OF THE IEEE 》 * |
| 周后飞 等: "智能手机车辆异常驾驶行为检测方法", 《智能系统学报》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110447214A (en) * | 2018-03-01 | 2019-11-12 | 北京嘀嘀无限科技发展有限公司 | A kind of system, method, apparatus and storage medium identifying driving behavior |
| CN110447214B (en) * | 2018-03-01 | 2022-04-15 | 北京嘀嘀无限科技发展有限公司 | System, method and device for identifying driving behavior and storage medium |
| CN108921096A (en) * | 2018-06-29 | 2018-11-30 | 北京百度网讯科技有限公司 | Time tracking method, apparatus, equipment and computer-readable medium |
| CN109229108A (en) * | 2018-08-07 | 2019-01-18 | 武汉理工大学 | A kind of driving behavior safe evaluation method based on driving fingerprint |
| CN112567437A (en) * | 2018-08-17 | 2021-03-26 | 索尼公司 | Information processing device, information processing system, information processing method, and program |
| CN109447127A (en) * | 2018-09-29 | 2019-03-08 | 深圳市元征科技股份有限公司 | Data processing method and device |
| CN112889270A (en) * | 2018-10-22 | 2021-06-01 | 松下电器（美国）知识产权公司 | Information processing method, information processing device, information processing program, and information processing system |
| CN111104959A (en) * | 2018-10-25 | 2020-05-05 | 激发认知有限公司 | Method and apparatus for machine learning classifier generation |
| CN109635335A (en) * | 2018-11-12 | 2019-04-16 | 平安科技（深圳）有限公司 | Drive Risk Forecast Method, device, computer equipment and storage medium |
| CN109614904A (en) * | 2018-12-03 | 2019-04-12 | 东北大学 | A kind of activity recognition method of the Multi-sensor Fusion based on Shapelet |
| CN111376910B (en) * | 2018-12-29 | 2022-04-15 | 北京嘀嘀无限科技发展有限公司 | User behavior identification method and system and computer equipment |
| CN111376910A (en) * | 2018-12-29 | 2020-07-07 | 北京嘀嘀无限科技发展有限公司 | User behavior identification method and system and computer equipment |
| CN109922439B (en) * | 2019-03-29 | 2020-10-16 | 重庆长安汽车股份有限公司 | Fusion method of multi-sensor data, and target detection method and system based on V2X |
| CN109922439A (en) * | 2019-03-29 | 2019-06-21 | 重庆长安汽车股份有限公司 | The fusion method of multi-sensor data, the object detection method based on V2X and system |
| CN110210513B (en) * | 2019-04-23 | 2021-04-09 | 深圳信息职业技术学院 | Data classification method and device and terminal equipment |
| CN110210513A (en) * | 2019-04-23 | 2019-09-06 | 深圳信息职业技术学院 | Data classification method, device and terminal device |
| CN111923927B (en) * | 2019-05-13 | 2022-04-29 | 长城汽车股份有限公司 | Method and apparatus for interactive perception of traffic scene prediction |
| CN111923927A (en) * | 2019-05-13 | 2020-11-13 | 长城汽车股份有限公司 | Method and apparatus for interactive perception of traffic scene prediction |
| CN110365648A (en) * | 2019-06-14 | 2019-10-22 | 东南大学 | A kind of vehicle-mounted CAN bus method for detecting abnormality based on decision tree |
| CN112348039A (en) * | 2019-08-07 | 2021-02-09 | 中国移动通信集团上海有限公司 | Training method of driving behavior analysis model, driving behavior analysis method and equipment |
| CN112348039B (en) * | 2019-08-07 | 2023-04-07 | 中国移动通信集团上海有限公司 | Training method of driving behavior analysis model, driving behavior analysis method and equipment |
| CN111814836A (en) * | 2020-06-12 | 2020-10-23 | 武汉理工大学 | Vehicle driving behavior detection method and device based on class imbalance algorithm |
| WO2022011722A1 (en) * | 2020-07-17 | 2022-01-20 | 华为技术有限公司 | Data processing method and apparatus, and intelligent vehicle |
| CN112099471A (en) * | 2020-08-27 | 2020-12-18 | 深圳市元征科技股份有限公司 | Vehicle power type judgment method and device, terminal device and storage medium |
| CN112883190A (en) * | 2021-01-28 | 2021-06-01 | 平安科技（深圳）有限公司 | Text classification method and device, electronic equipment and storage medium |
| CN113112796A (en) * | 2021-04-07 | 2021-07-13 | 吉林大学 | Construction method and system of driving behavior characteristics and driving behavior analysis method |
| CN113112796B (en) * | 2021-04-07 | 2022-05-27 | 吉林大学 | Construction method and system of driving behavior characteristics and driving behavior analysis method |
| CN114169374A (en) * | 2021-12-10 | 2022-03-11 | 湖南工商大学 | Cable-stayed bridge stay cable damage identification method and electronic equipment |
| CN114169374B (en) * | 2021-12-10 | 2024-02-20 | 湖南工商大学 | Cable-stayed bridge stay cable damage identification method and electronic equipment |
| CN115496978A (en) * | 2022-09-14 | 2022-12-20 | 北京化工大学 | Image and vehicle speed information fused driving behavior classification method and device |
| CN116702000A (en) * | 2023-05-31 | 2023-09-05 | 深圳技术大学 | Road surface quality dynamic monitoring and evaluating method based on multi-layer data fusion |
| CN116702000B (en) * | 2023-05-31 | 2024-03-29 | 深圳技术大学 | Road surface quality dynamic monitoring and evaluating method based on multi-layer data fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108108766B (en) | 2021-10-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108108766B (en) | Driving behavior identification method and system based on multi-sensor data fusion | |
| Yu et al. | A bearing fault diagnosis technique based on singular values of EEMD spatial condition matrix and Gath-Geva clustering | |
| CN103996287B (en) | A kind of vehicle compulsory based on decision-tree model changes decision-making technique | |
| US8972213B2 (en) | Pattern recognition approach to battery diagnosis and prognosis | |
| CN108280484B (en) | Driver acceleration characteristic online classification and identification method | |
| Martínez et al. | Driver identification and impostor detection based on driving behavior signals | |
| CN106971194A (en) | A kind of driving intention recognition methods based on the double-deck algorithms of improvement HMM and SVM | |
| CN111994084B (en) | Method and system for identifying driving style of driver and storage medium | |
| CN112766182B (en) | Broaching tool wear state identification method | |
| CN112699793A (en) | Fatigue driving detection optimization identification method based on random forest | |
| CN104688252A (en) | Method for detecting fatigue status of driver through steering wheel rotation angle information | |
| Xun et al. | An experimental study towards driver identification for intelligent and connected vehicles | |
| Yang et al. | Driver2vec: Driver identification from automotive data | |
| CN114926825A (en) | Vehicle driving behavior detection method based on space-time feature fusion | |
| CN113581188A (en) | Commercial vehicle driver driving style identification method based on Internet of vehicles data | |
| CN115817500A (en) | Driving style determination method and device, vehicle and storage medium | |
| Hamed et al. | Fuel consumption prediction model using machine learning | |
| CN113642114A (en) | Modeling method for humanoid random car following driving behavior capable of making mistakes | |
| CN106056150A (en) | System and method for establishing part division remote damage assessment of different vehicle types based on artificial intelligence random forest method | |
| Li et al. | Classification of adaptive cruise control vehicle type based on car following trajectories | |
| CN112061136A (en) | Automatic analysis method for driver driving behavior based on long-term and short-term memory network | |
| CN115830368A (en) | Vehicle shock absorber defect diagnosis method, system, equipment and medium | |
| CN115774942A (en) | Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM | |
| Nanda et al. | Real-time surface material identification using infrared sensor to control speed of an arduino based car like mobile robot | |
| Wang et al. | A data-driven estimation of driving style using deep clustering |
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 |