CN109410586B - Traffic state detection method based on multi-metadata fusion - Google Patents
Traffic state detection method based on multi-metadata fusion Download PDFInfo
- Publication number
- CN109410586B CN109410586B CN201811524358.9A CN201811524358A CN109410586B CN 109410586 B CN109410586 B CN 109410586B CN 201811524358 A CN201811524358 A CN 201811524358A CN 109410586 B CN109410586 B CN 109410586B
- Authority
- CN
- China
- Prior art keywords
- track
- gps
- data
- taxi
- road section
- 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
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0133—Traffic data processing for classifying traffic situation
Abstract
The invention discloses a traffic state detection method based on multi-element data fusion, which comprises the following steps of firstly, utilizing a clustering algorithm to respectively segment taxi GPS data and mobile phone GPS data according to GPS tracks of users to obtain each taxi GPS segmented track and mobile phone GPS segmented track; then, taking taxi GPS segmented tracks as training samples, constructing an identification model, and extracting taxi-like GPS segmented tracks in the mobile phone GPS segmented tracks; matching the taxi GPS sectional track and the taxi-like GPS sectional track to road sections in the urban road network by using a map matching algorithm; respectively calculating the vehicle running speed of the corresponding road section by using the matching results of the two types of tracks; and finally, fusing the vehicle running speeds of the road section respectively calculated by the two types of track matching results by using an evidence theory, and estimating the vehicle running speed of the road section. The invention can accurately detect the traffic state information of the urban road.
Description
Technical Field
The invention belongs to the technical field of traffic, and particularly relates to a traffic state detection method based on multi-metadata fusion.
Background
With the continuous development of urbanization in China, urban roads are increasingly complex, and the difficulty of management and control of urban traffic by administrative departments is also increased, so that the method has very important significance in real-time detection and wide and accurate evaluation of urban road traffic conditions. In recent years, with the rapid development of sensing technology and communication technology, traffic condition information recorded by various traffic sensing devices is widely applied to research in traffic management and control, such as recording the number of vehicles passing by loop detector data, recording snapshot pictures of roads at different times by video monitoring data, recording position information of vehicles at different times by floating car GPS data, and the like. At present, most urban taxis are provided with GPS positioning devices, and taxi GPS data is widely applied to the field of traffic research. Along with the rise and the popularization of various map APP software, a large number of map APP users including private cars, buses and the like can be widely and effectively acquired, and a new direction is provided for the research of urban traffic conditions by utilizing traffic big data. However, the conventional method still has the following problems:
1) although taxi GPS data can effectively record position information of taxies, taxies tend to run on urban main roads, taxi coverage rates of roads with lower urban levels, suburban areas and other areas are still low, and comprehensive detection of urban road traffic conditions has certain limitations. Meanwhile, the recording frequency is low, and certain limitation is also caused to the accuracy of the detected traffic condition.
2) The mobile phone GPS data records the information of traffic users including various traffic modes, including buses, taxis, private cars, bicycles, pedestrians and the like, and the research on urban traffic conditions mainly aims at motor vehicles such as buses, taxis and the like, and the information of bicycles, pedestrians and the like recorded by the mobile phone GPS data can cause serious interference and influence on the traffic condition detection result.
3) The two traffic states sensed by the GPS data may have inconsistency or even contradiction, and a method is needed for measuring the inconsistency, so that fusion is carried out on the basis to obtain the accurate traffic state.
Disclosure of Invention
The invention solves the technical problem that aiming at the defects of the prior art, the invention provides a traffic state detection method based on the fusion of the multivariate data, which fuses taxi GPS data and mobile phone GPS data and can accurately detect the vehicle running speed of each road section in the urban road network, namely the traffic state information.
The technical scheme provided by the invention is as follows:
a traffic state detection method based on multi-metadata fusion comprises the following steps:
step one, constructing an urban road network according to urban road information;
step two, taxi GPS data in a passenger carrying state and mobile phone GPS data are obtained;
thirdly, carrying out GPS track segmentation on taxi GPS data and mobile phone GPS data (continuous track points) according to users by using a clustering algorithm to obtain each taxi GPS segmented track and mobile phone GPS segmented track; taxi GPS data and mobile phone GPS data in a passenger carrying state can be firstly sorted according to time of a user, and then GPS track segmentation is carried out on the sorted GPS data by utilizing a clustering algorithm;
step four, constructing an identification model by taking the taxi GPS sectional track as a training sample, and extracting a taxi (motor vehicle) like GPS sectional track in the mobile phone GPS sectional track;
step five, matching the taxi GPS sectional track and the taxi-like GPS sectional track to road sections in an urban road network by using a map matching algorithm;
sixthly, respectively calculating the vehicle running speed density distribution function of the corresponding road section by using the matching result of the two types of tracks (the taxi GPS sectional track and the taxi-like GPS sectional track);
step seven, for the road section with the two types of track matching results, fusing vehicle running speed density distribution functions of the road section obtained by respectively calculating the two types of track matching results by utilizing an evidence theory to obtain the final estimation of the traffic state (vehicle running speed) of the road section;
and for the road section with only one type of track matching result, taking the vehicle running speed density distribution function of the road section calculated by the matching result of the single type of track as the final estimation of the traffic state of the road section.
Further, in the second step, the taxi GPS data is low-frequency recording data, and generally there is a user location recording point every 25 seconds on average, and the latitude and longitude and time information of the user location are recorded. Various map APP software records are used for acquiring mobile phone GPS data, the mobile phone GPS data are high-frequency recorded data, generally, one user position recording point is formed every 1 second, and longitude and latitude and time information of the user position are recorded.
Further, in the third step, the clustering algorithm adopts a DBSCAN clustering algorithm. The DBSCAN clustering algorithm is a density-based clustering algorithm, a data set is divided into different clusters according to density, and each cluster is classified into one type. The density of the data set is measured according to different attributes of the data, the different attributes are different dimensions, and the compactness between any two data points is measured by calculating the Euclidean distance between different data points. The DBSCAN clustering algorithm divides a data set into different clusters, and has two key parameters, wherein the neighborhood distance epsilon and the minimum neighborhood sample number MinPts of the core points in the clusters, the neighborhood distance epsilon is a threshold value of Euclidean distance between the two data points, MinPts is a threshold value of neighborhood sample number of the core points in the clusters (the neighborhood sample number of a certain point is larger than MinPts and is the core point), and the neighborhood of the epsilon-at any point in one cluster at least comprises one core point. In the invention, the data dimensionality comprises longitude, latitude and time, standard division processing is respectively carried out on each data dimensionality, Euclidean distance between data points after the standard division processing is calculated, the neighborhood distance belongs to the minimum neighborhood sample number MinPts of a core point, taxi GPS data and mobile phone GPS data of each user are respectively gathered into different clusters through DBSCAN clustering, and each cluster represents one class, namely a section of researched GPS track (travel track).
Further, a fourth neighborhood distance method is adopted to determine the neighborhood distance parameter belonging to the DBSCAN clustering algorithm, and the specific method is as follows: 1) taking taxi GPS data of each user as a data set, calculating Euclidean distance between any two points in each data set, and finding the fourth largest Euclidean distance value, namely the fourth adjacent domain distance, of each data set from large to small; 2) and (4) making a probability distribution map of the fourth neighborhood distance and determining a distance value corresponding to the inflection point of the probability distribution map as a neighborhood distance parameter epsilon.
Further, the euclidean distance between any two recording points at any position in a data set is calculated as follows:
wherein: x'1And x'2Respectively represents longitude values y 'of two position recording points after standard sub-processing'1And y'2Respectively represents latitude values t 'of two position recording points after standard division processing'1And t'2Respectively representing the time values of the two position recording points after standard division processing; for any dimension data x of any user, the standard processing formula of the data x is as follows:
wherein x' is data after x is subjected to standard division processing,
and sigma is the standard deviation of all the original data of the dimension of the corresponding user.
Further, in the fourth step, an isolated forest model is adopted as the identification model; an isolated forest model (isolation forest) is an unsupervised outlier detection method and is an anomaly detection model constructed based on various statistical characteristics of samples. In the invention, a taxi GPS segmented track is used as a normal sample and is used as a training set to construct a binary tree, and statistical characteristics comprise the running speed, acceleration, turning angle, track length, time and the like of a taxi; and testing the mobile phone GPS segmented track as a test sample through a binary tree, thereby filtering an abnormal track, and extracting a taxi (motor vehicle) like track in the mobile phone GPS segmented track as a detection sample of the road traffic condition. The method specifically comprises the following steps:
step 4.1, taking the taxi GPS sectional track obtained in the step three as a training sample, and taking the mobile phone GPS sectional track as a test sample;
4.2, constructing a plurality of binary trees (iTrees) of the isolated forest model based on the training samples;
4.3, calculating the path length of each test sample in each binary tree to obtain the abnormal score of each test sample;
and 4.4, determining a threshold value of the abnormal score, wherein the test sample with the abnormal score smaller than the threshold value is determined as a normal sample, namely the test sample is a taxi-like GPS segmented track.
Further, in the fifth step, the map matching algorithm adopts an S-T map matching algorithm. The S-T map matching algorithm based on the space-time analysis is a GPS track matching algorithm based on the space-time analysis, and the algorithm comprehensively considers the maximum probability of time and space to match tracks to paths.
Further, in the sixth step, according to a matching result of a certain type of track, the method for calculating the vehicle driving speed density distribution function of the corresponding road section includes:
for a certain road section, obtaining all track sets of the type matched with the road section, respectively calculating each track in the track sets to obtain a speed value, using the speed value as a speed observation sample, and forming a speed observation sample set of the road section; calculating a vehicle running speed density distribution function pos (v) of the road section according to the speed observation sample set:
wherein v is the velocity, n is the number of velocity observation samples in the velocity observation sample set, viObserving the sample for the ith velocity; rhoK(z) is a kernel function, and a Gaussian kernel function is adopted:
wherein h represents a smoothing parameter, the larger h is, the smoother the density function is, and the value of h can be determined by using a Silverman rule, that is, the value of h is
d is the sample dimension (in the present invention, the sample is velocity, and the dimension is 1), and it can be seen that the value of h is larger when the number of samples n is smaller, and vice versa, because when the number of samples is small, sufficient distribution information is often not provided, and a larger value of h may represent such an inaccurate estimation.
Further, the method for calculating a speed value from a certain track comprises the following steps: and dividing the length of the real driving path matched with the track on the road section by the travel time of the track to obtain a corresponding speed value.
Further, in the seventh step, for a specific road segment, the concrete steps of using an evidence theory to fuse the vehicle driving speed density distribution functions of the road segment respectively calculated by the matching results of the two types of tracks to obtain the final estimation of the traffic state of the road segment are as follows:
step 7.1, setting taxi GPS sectional railThe vehicle running speed density distribution function of the road section obtained by respectively calculating the GPS sectional tracks of the track and the taxi-like is passt(v) And passm(v) (ii) a Respectively to passt(v) And passm(v) And (3) carrying out normalization treatment:
posstn(v) and passmn(v) Respectively represent post(v) And passm(v) Normalized value, postn(v)∈[0，1]，possmn(v)∈[0，1]；
And 7.2, calculating a fusion result:
possF(v)＝posstn(v)*possmn(v)；
7.3, calculating a fusion effect index;
in a specified speed interval v1，v2]Using likelihood functions in evidence theory
As a fusion effect index, the fusion effect is measured, wherein a and norm are intermediate variables,
PlsDempster([v1，v2]) Has a value range of [0, 1 ]]When the fusion effect is 1, the fusion effect is best, two groups of data have no contradiction, and on the contrary, the contradiction is larger;
and 7.4, setting a threshold, if the fusion index is lower than the threshold, not adopting the corresponding fusion result, and combining historical data and data of the upstream and downstream road sections to adopt passt(v) And passm(v) The one closer to the historical data and the data of the upstream and downstream road sections is used as the traffic state of the road sectionFinal estimation of the state; otherwise, taking the fusion result as the final estimation of the traffic state of the road section.
Further, a velocity interval [ v ] is specified1，v2]Is taken as [ v ]15th，v85th]Wherein v is15thAnd v85thRespectively calculating a 15-quantile value and an 85-quantile value in all speed values obtained by matching the taxi GPS subsection track of the road section; setting the number of all speed values obtained by calculating the taxi GPS sectional track matched to the road section as N, and sequencing the N speed values from small to large; if 0.15N/0.85N is an integer, the speed value ranked at 0.15N/0.85N is a 15-quantile value/85-quantile value; if 0.15N/0.85N is not an integer, the average of the two velocity values that are adjacent to 0.15N/0.85N is taken as 15 quantile/85 quantile.
Has the advantages that:
the invention provides a traffic state detection method based on multi-element data fusion, which is characterized in that the traffic state of an urban road is obtained by fusing taxi GPS data and mobile phone GPS data (map APP data), so that data samples can be greatly increased, the road conditions of areas such as urban secondary roads with few taxis, suburbs and the like can be effectively obtained, and the road coverage rate of data can be greatly improved. Meanwhile, the data volume of the mobile phone GPS data is large, the recording frequency is high, the detection error can be effectively reduced, and the accuracy and the effectiveness of the road section state estimation are improved.
The taxi GPS data and the mobile phone GPS data are easy to acquire in real time, so that the taxi GPS data and the mobile phone GPS data can be used for effectively sensing the urban road traffic state in real time in a large range, important reference values are provided for real-time monitoring and management of traffic and various traffic administrative departments on urban traffic, and a new way is provided for research of scientific researchers on urban traffic by using traffic data by using the mobile phone GPS data
According to the invention, GPS track segmentation is carried out on taxi GPS data and mobile phone GPS data of a user by utilizing a DBSCAN clustering algorithm, a vehicle GPS segmented track is taken as a research object, an isolated forest model (isolation forest) is constructed based on the taxi GPS track capable of reflecting the taxi driving state, and taxi (motor vehicle) -like GPS track in the mobile phone GPS data is extracted to supplement road vehicle coverage information. Carrying out map matching on the taxi GPS track and the extracted taxi-like GPS track by using an S-T map matching algorithm; respectively calculating the vehicle running speed distribution of the corresponding road sections by using the matching result of the taxi GPS track and the taxi-like GPS track; and the taxi GPS track and the taxi-like GPS track are used for acquiring the vehicle running speed distribution of the road section and fusing the vehicle running speed distribution through an evidence theory, so that the final estimation of the road section vehicle running speed, namely the traffic state is obtained.
The method fuses the two types of GPS data by introducing an evidence theory and ensures the validity of the fused result.
Drawings
FIG. 1 is a flow chart of a method for detecting traffic conditions based on multi-metadata fusion according to the present invention;
FIG. 2 is a schematic diagram of an evidence theory fusion process of taxi GPS data and mobile phone GPS data;
FIG. 3 is a schematic diagram of the integration effect of the speeds of the taxi track of the taxi GPS data and the motor vehicle track of the mobile phone GPS data on the early peak (8: 00-8: 10) urban road sections after calculating the road section speeds through S-T map matching and integrating the road section speeds obtained by the two types of track data through an evidence theory.
Detailed Description
The invention provides a traffic state detection method based on multi-metadata fusion, which is shown in fig. 1, takes a Shenzhen city road network as an example, and is practically applied based on Shenzhen city taxi GPS data and mobile phone GPS data recorded by map APP software. Firstly, time sequencing is carried out on data records of each user, segmented tracks of the users are obtained through DBSCAN clustering, then an isolated forest model (isolation forest) is used for building a training model for a taxi GPS track, taxi (motor vehicle) like tracks are extracted from a mobile phone GPS track, finally an S-T map matching algorithm is used for map matching the taxi GPS track and the motor vehicle tracks extracted from the mobile phone GPS track, vehicle running speed distribution of road sections of two types of tracks at different moments is obtained, the vehicle running speed distribution is fitted, and road section speeds obtained by the two types of tracks are fused by an evidence theory.
The Shenzhen city road network is a Shenzhen city simplified version road network and comprises 21115 road sections and 73415 intersections.
The taxi GPS data are collected GPS record data of the Shenzhen taxi 2016 in the week from 9 month 15 to 9 month 22, wherein 13731 taxi users are recorded in the week with one data record point every 25 seconds on average, and 13509 taxi passenger-carrying state users are screened.
The mobile phone GPS data is travel GPS data recorded by the map APP software of Shenzhen city residents in 2016, from 9 month 15 to 9 month 22. The frequency of recording is one GPS recording point per second, during which a total of about 900,000 map APP users are recorded.
Calculating the neighborhood distance of the DBSCAN cluster by considering three dimensionalities of longitude, latitude and time of GPS recorded data, and firstly performing standard division processing on each dimensionality data; for any dimension data of any user, the standard processing formula is as follows:
wherein: x is the original data of the dimension of the user without standard division processing, x' is the data of x after standard division processing,
the mean value of all the original data of the dimension of the user is used, and sigma is the standard deviation of all the original data of the dimension of the user;
then calculating Euclidean distances between data points after standard division processing, and determining a neighborhood distance parameter belonging to the same category;
the euclidean distance is calculated as follows:
wherein: x'1And x'2Respectively representing longitude values y 'of two GPS track points after standard division processing'1And y'2Respectively representing latitude values, t 'of two GPS track points subjected to standard sub-processing'1And t'2And respectively representing the time values of the two GPS track points after standard division processing.
The neighborhood distance parameter e is determined by a typical fourth neighborhood distance method, which comprises the following steps: 1) the taxi GPS data after sequencing of each user is taken as a data set, the Euclidean distance between any two points in each data set is calculated, and the Euclidean distance value, namely the fourth adjacent domain distance, of the fourth large value of each data set is found by arranging the data sets from large to small; 2) making a probability distribution map of the fourth neighborhood distance and determining a distance value corresponding to the inflection point of the probability distribution map as a neighborhood distance parameter belonging to the same category; according to the method, the value of the epsilon is finally determined to be 0.1.
Then, the minimum neighborhood sample number MinPts of the core point in the DBSCAN algorithm is determined to be 4 (the general value of MinPts), and the GPS recorded data in the data set is clustered to obtain the segmentation track of the user.
The isolated forest model (isolation forest) is an unsupervised outlier detection model, a training sample is a taxi GPS segmented track clustered by DBSCAN, a test sample is a mobile phone GPS segmented track clustered by DBSCAN, samples are randomly extracted from the training sample (taking a user track segment as a unit), various statistical characteristics of the samples are extracted to construct a binary tree, the training model is constructed, the test sample is tested, the abnormal score of the test sample is calculated, and the higher the abnormal score is, the higher the possibility of indicating the sample is abnormal is. And determining an abnormal threshold value of the isolated forest test model, regarding the test sample with the abnormal score smaller than the abnormal threshold value as the test sample without obvious difference from the taxi track, and determining the test sample as a taxi (motor vehicle) like GPS (global positioning system) segmented track. The selected data statistical characteristics comprise the average speed, the median speed, the maximum speed, the track length, the average acceleration, the staying rate, the average rotation angle, the rotation angle conversion rate, the speed conversion rate, the starting and ending point linear distance of the track, the track length and whether the track occurs in the traffic flow peak period (7: 00-10:00 early and 17:00-20:00 late).
The S-T map matching algorithm considers the combination of a data track point and a road network and matches the track to the path by considering the maximum probability of time and space, so that the vehicle running speed of the path is calculated according to the path length and the track time and is corresponding to the corresponding road section. The specific algorithm is as follows:
1) for a segment of track (p)1→p2→…→pn) Finding track candidate points of each track point
(i is the ith track point in the track, j is the jth track candidate point of the ith track point), wherein the track candidate point is a foot hanging position where a perpendicular line is drawn from the track point to the road section, and the length of the perpendicular line needs to be less than 35 meters;
2) calculating trajectory candidate points
Observation probability of
In the formula:
candidate points for the trajectory
And the locus point piStraight geographic distance therebetween.
3) Calculating the previous track candidate point
To the next track candidate point
Transition probability of
In the formula: di-1→ i is the straight-line geographic distance from the ith-1 track point to the ith track point, w(i-1，t)→(i，s)The candidate points of the road section of the ith-1 track point
Candidate points of road section to ith track point
Shortest path distance in a road network.
4) Combining the observation probability and the transition probability to obtain track candidate points
To the track candidate point
Integrated probability of
And for each section of track, selecting a series of track candidate points with the highest comprehensive probability (one track candidate point is selected for each track) as the real driving track corresponding to the track on the road section.
5) Calculating the path driving speed v of the section of track according to the travel time of the section of track and the real driving path corresponding to the section of track1→nAnd corresponds the speed to the road segment.
In the formula: w is a1→nFor the actual path length, Δ t, of the section of track corresponding to the section of road1→nThe travel time for that segment of the trajectory.
Since various vehicles equipped with GPS devices travel on a road section, it is a dynamic coverage problem itself. For a certain road, in real-time application, the speed observation sample number is uncertain, and the prior information of the speed distribution is uncertain, so that the speed density is estimated by adopting a nonparametric method-kernel function method, and then the speed density is converted into a possible function.
For a specific road section, obtaining all track sets matched to the road section after map matching, and respectively calculating each track in the track sets to obtain a speed value serving as a speed observation sample to form a speed observation sample set of the road section; and calculating the speed of the road section according to the speed observation sample set:
wherein v is velocity, viThe speed observation sample for the ith speed on the road section is obtained, and n is the number of the speed observation samples on the road section; rhoK(z) is a kernel function, and the invention selects a Gaussian kernel function commonly used in the engineering field:
wherein h represents a smoothing parameter, the larger h is, the smoother the density function is, and the value of h can be determined by using a Silverman rule, that is to say
d is the sample dimension (in the present invention, the sample is velocity, and the dimension is 1), and it can be seen that the smaller the number of samples n, the larger the value of h, and vice versa, because the smaller the number of samples n, the insufficient number of samples is providedSufficient distribution information, a larger value of h, may represent such an inaccurate estimate.
The evidence theory is an effective method capable of fusing uncertain data, and comprises a support function and a plausibility function, wherein the lower limit and the upper limit of new data after fusing the two data can be obtained by calculating the support function and the plausibility function, and meanwhile, the fused effect can be measured. The concrete steps of fusing the road speed obtained by the two types of tracks by adopting the evidence theory are as follows:
step 1: setting the vehicle running speed density distribution function of the road section obtained by respectively calculating the taxi GPS sectional track and the taxi-like GPS sectional track as passt(v) And passm(v) (ii) a Respectively to passt(v) And passm(v) And (3) carrying out normalization treatment:
posstn(v) and passmn(v) Respectively represent post(v) And passm(v) Normalized value, postn(v)，possmn(v)∈[0，1](ii) a In this example v ∈ [0, 120 ]]；
Step 2: calculating a fusion result:
possF(v)＝posstn(v)*possmn(v)；
and step 3: calculating a fusion effect index;
in a specified speed interval v1，v2]Using likelihood functions in evidence theory
As a fusion effect index, the fusion effect is measured, wherein a and norm are intermediate variables,
PlsDempster([v1，v2]) Has a value range of [0, 1 ]]When the fusion effect is 1, the fusion effect is best, two groups of data have no contradiction, and on the contrary, the contradiction is larger;
and 4, step 4: setting a threshold (0.5 in the embodiment), if the fusion index is lower than the threshold, the corresponding fusion result is not adopted, and the pass is adopted by combining the historical data and the data of the upstream and downstream road sectionst(v) And passm(v) The other one which is closer to the historical data and the data of the upstream and downstream road sections is used as the final estimation of the traffic state of the road section; otherwise, taking the fusion result as the final estimation of the traffic state of the road section.
The evidence theory fusion process of taxi GPS data and mobile phone GPS data is shown in FIG. 2.
The taxi track of the taxi GPS data and the motor vehicle track extracted by the isolated forest model in the mobile phone GPS data are subjected to S-T map matching, and the calculated urban road speed distribution result is shown in figure 3 (early peak result, 8: 00-8: 10 in the morning).
Claims (10)
1. A traffic state detection method based on multi-metadata fusion is characterized by comprising the following steps:
step one, constructing an urban road network according to urban road information;
step two, taxi GPS data in a passenger carrying state and mobile phone GPS data are obtained;
thirdly, carrying out GPS track segmentation on taxi GPS data and mobile phone GPS data according to users by using a clustering algorithm to obtain each taxi GPS segmented track and mobile phone GPS segmented track;
step four, constructing an identification model by taking taxi GPS sectional tracks as training samples, and extracting taxi-like GPS sectional tracks in the mobile phone GPS sectional tracks;
step five, matching the taxi GPS sectional track and the taxi-like GPS sectional track to road sections in an urban road network by using a map matching algorithm;
sixthly, respectively calculating the vehicle running speed density distribution function of the corresponding road section by using the matching results of the two types of tracks;
step seven, for the road section with the two types of track matching results, fusing vehicle running speed density distribution functions of the road section obtained by respectively calculating the two types of track matching results by utilizing an evidence theory to obtain the vehicle running speed of the road section, namely the final estimation of the traffic state;
and for the road section with only one type of track matching result, taking the vehicle running speed density distribution function of the road section calculated by the matching result of the single type of track as the final estimation of the traffic state of the road section.
2. The method for detecting traffic states based on multi-data fusion of claim 1, wherein in the second step, each position recording point in the taxi GPS data and the mobile phone GPS data comprises information of three dimensions of longitude, latitude and time.
3. The method for detecting traffic status based on multi-metadata fusion of claim 2, wherein in the third step, the clustering algorithm employs a DBSCAN clustering algorithm.
4. The method for detecting the traffic state based on the multivariate data fusion as claimed in claim 3, wherein a fourth neighborhood distance method is adopted to determine a neighborhood distance parameter e in the DBSCAN clustering algorithm, and the specific method is as follows: 1) taking taxi GPS data of each user as a data set, calculating Euclidean distance between any two points in each data set, and finding the fourth largest Euclidean distance value, namely the fourth adjacent domain distance, of each data set from large to small; 2) and (4) making a probability distribution map of the fourth neighborhood distance and determining a distance value corresponding to the inflection point of the probability distribution map as a neighborhood distance parameter epsilon.
5. The method of claim 4, wherein the Euclidean distance between any two recording points at any position in a data set is calculated according to the following formula:
wherein: x'1And x'2Respectively represents longitude values y 'of two position recording points after standard sub-processing'1And y'2Respectively represents latitude values t 'of two position recording points after standard division processing'1And t'2Respectively representing the time values of the two position recording points after standard division processing; for any dimension data x of any user, the standard processing formula of the data x is as follows:
wherein x' is data after x is subjected to standard division processing,
and sigma is the standard deviation of all the original data of the dimension of the corresponding user.
6. The method for detecting the traffic state based on the multi-metadata fusion as claimed in claim 1, wherein in the fourth step, an isolated forest model is adopted as the identification model; the method specifically comprises the following steps:
step 4.1, taking the taxi GPS sectional track obtained in the step three as a training sample, and taking the mobile phone GPS sectional track as a test sample;
4.2, constructing a plurality of binary trees of the isolated forest model based on the training samples;
4.3, calculating the path length of each test sample in each binary tree to obtain the abnormal score of each test sample;
and 4.4, determining a threshold value of the abnormal score, wherein the test sample with the abnormal score smaller than the threshold value is determined as a normal sample, namely the test sample is a taxi-like GPS segmented track.
7. The method for detecting traffic states based on multi-data fusion of claim 1, wherein in the sixth step, the method for calculating the vehicle driving speed density distribution function of the corresponding road section according to the matching result of a certain type of track comprises the following steps:
for a certain road section, obtaining all track sets of the type matched with the road section, respectively calculating each track in the track sets to obtain a speed value, using the speed value as a speed observation sample, and forming a speed observation sample set of the road section; calculating a vehicle running speed density distribution function pos (v) of the road section according to the speed observation sample set:
wherein v is the velocity, n is the number of velocity observation samples in the velocity observation sample set, viObserving the sample for the ith velocity; rhoK(z) is a kernel function, and a Gaussian kernel function is adopted:
wherein h represents a smoothing parameter, the larger h is, the smoother the density function is, and the value of h can be determined by using a Silverman rule, that is, the value of h is
d is the sample dimension, d 1.
8. The method of claim 7, wherein the calculating a speed value from a track comprises: and dividing the length of the real driving path matched with the track on the road section by the travel time of the track to obtain a corresponding speed value.
9. The method for detecting traffic states based on multi-data fusion of claim 7, wherein in the seventh step, for a specific road segment, the vehicle driving speed density distribution functions of the road segment, which are respectively calculated from the matching results of the two types of tracks, are fused by using an evidence theory, and the specific step of obtaining the final estimation of the traffic state of the road segment is as follows:
step 7.1, setting the vehicle running speed density distribution function of the road section, which is obtained by respectively calculating the taxi GPS sectional track and the taxi-like GPS sectional track, as passt(v) And passm(v) (ii) a Respectively to passt(v) And passm(v) And (3) carrying out normalization treatment:
posstn(v) and passmn(v) Respectively represent post(v) And passm(v) Normalized value, postn(v)∈[0，1]，possmn(v)∈[0，1]；
And 7.2, calculating a fusion result:
possF(v)＝posstn(v)*possmn(v)；
7.3, calculating a fusion effect index;
in a specified speed interval v1，v2]Using likelihood functions in evidence theory
As a fusion effect index, the fusion effect is measured, wherein a and norm are intermediate variables,
norm＝max(possF(v))
PlsDempster([v1，v2]) Has a value range of [0, 1 ]]When the fusion effect is 1, the fusion effect is best, two groups of data have no contradiction, and on the contrary, the contradiction is larger;
and 7.4, setting a threshold, if the fusion index is lower than the threshold, not adopting the corresponding fusion result, and combining historical data and data of the upstream and downstream road sections to adopt passt(v) And passm(v) The other one which is closer to the historical data and the data of the upstream and downstream road sections is used as the final estimation of the traffic state of the road section; otherwise, taking the fusion result as the final estimation of the traffic state of the road section.
10. The method of claim 8, wherein the speed interval [ v ] is specified1，v2]Is taken as [ v ]15th，v85th]Wherein v is15thAnd v85thRespectively calculating a 15-quantile value and an 85-quantile value in all speed values obtained by matching the taxi GPS subsection track of the road section; setting the number of all speed values obtained by calculating the taxi GPS sectional track matched to the road section as N, and sequencing the N speed values from small to large; if 0.15N/0.85N is an integer, the speed value ranked at 0.15N/0.85N is a 15-quantile value/85-quantile value; if 0.15N/0.85N is not an integer, the average of the two velocity values that are adjacent to 0.15N/0.85N is taken as 15 quantile/85 quantile.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811524358.9A CN109410586B (en) | 2018-12-13 | 2018-12-13 | Traffic state detection method based on multi-metadata fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811524358.9A CN109410586B (en) | 2018-12-13 | 2018-12-13 | Traffic state detection method based on multi-metadata fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109410586A CN109410586A (en) | 2019-03-01 |
| CN109410586B true CN109410586B (en) | 2020-06-05 |
Family
ID=65458932
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811524358.9A Active CN109410586B (en) | 2018-12-13 | 2018-12-13 | Traffic state detection method based on multi-metadata fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109410586B (en) |
Families Citing this family (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110113716B (en) * | 2019-04-22 | 2021-08-31 | 腾讯大地通途（北京）科技有限公司 | Path state information acquisition method and device and storage medium |
| GB2611225B (en) * | 2019-05-29 | 2023-10-18 | Motional Ad Llc | Estimating speed profiles |
| CN110264740A (en) * | 2019-06-24 | 2019-09-20 | 长沙理工大学 | The inhuman real-time track detector of traffic machine and detection method based on video |
| CN110285817B (en) * | 2019-07-12 | 2023-03-21 | 东北电力大学 | Complex road network map matching method based on self-adaptive D-S evidence theory |
| CN110555544B (en) * | 2019-07-16 | 2022-04-08 | 广东工业大学 | Traffic demand estimation method based on GPS navigation data |
| CN110766940A (en) * | 2019-09-24 | 2020-02-07 | 重庆交通大学 | Method for evaluating running condition of road signalized intersection |
| CN110751196B (en) * | 2019-10-12 | 2020-09-18 | 东北石油大学 | Oil-like drop attachment identification method in oil-water two-phase flow transparent pipe wall |
| CN110570658B (en) * | 2019-10-23 | 2022-02-01 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
| CN110634299B (en) * | 2019-10-25 | 2022-03-25 | 福州大学 | Urban traffic state fine division and identification method based on multi-source track data |
| CN112712729B (en) * | 2019-10-26 | 2023-01-06 | 华为技术有限公司 | Method and system for predicting motion trajectory |
| CN111739283B (en) * | 2019-10-30 | 2022-05-20 | 腾讯科技（深圳）有限公司 | Road condition calculation method, device, equipment and medium based on clustering |
| CN112837393B (en) * | 2019-11-22 | 2024-04-09 | 中国航天系统工程有限公司 | Method and system for generating oversized city vector road network based on vehicle position data |
| CN111081016B (en) * | 2019-12-18 | 2021-07-06 | 北京航空航天大学 | Urban traffic abnormity identification method based on complex network theory |
| CN111190891B (en) * | 2019-12-27 | 2023-07-25 | 武汉长江通信产业集团股份有限公司 | Multi-semantic track data segment storage method |
| CN111123333B (en) * | 2019-12-30 | 2022-05-03 | 公安部交通管理科学研究所 | Vehicle track positioning method fusing bayonet and GPS data |
| CN111179592B (en) * | 2019-12-31 | 2021-06-11 | 合肥工业大学 | Urban traffic prediction method and system based on spatio-temporal data flow fusion analysis |
| CN111311910B (en) * | 2020-02-19 | 2020-12-22 | 中南大学 | Abnormal track detection method for multi-level road-level floating vehicle |
| CN111368881B (en) * | 2020-02-19 | 2023-10-03 | 浙江工业大学 | Low-frequency GPS track road network matching method based on multidimensional data fusion analysis |
| CN111666971A (en) * | 2020-04-27 | 2020-09-15 | 深圳壹账通智能科技有限公司 | Event identification method, device and equipment based on position location and storage medium |
| CN111862605B (en) * | 2020-07-20 | 2022-03-08 | 腾讯科技（深圳）有限公司 | Road condition detection method and device, electronic equipment and readable storage medium |
| CN111949896B (en) * | 2020-07-30 | 2023-09-15 | 安徽师范大学 | Track abnormality detection method based on track big data |
| CN112183589B (en) * | 2020-09-14 | 2022-04-22 | 西北工业大学 | Real-time vehicle K neighbor query method under low sampling rate |
| CN112116813B (en) * | 2020-11-20 | 2021-04-30 | 深圳市城市交通规划设计研究中心股份有限公司 | State recognition model establishing method, road traffic state recognition method and road traffic state recognition device |
| CN112382095B (en) * | 2020-11-26 | 2021-09-10 | 长沙理工大学 | Urban expressway traffic state estimation method based on multi-source data fusion |
| CN113053107B (en) * | 2020-12-23 | 2022-08-02 | 沈阳世纪高通科技有限公司 | Method and device for calculating road congestion and slowness state by combining free flow |
| CN113375685B (en) * | 2021-03-31 | 2022-09-30 | 福建工程学院 | Urban intersection center recognition and intersection turning rule extraction method based on sub-track intersection |
| CN113259900B (en) * | 2021-05-27 | 2021-10-15 | 华砺智行(武汉)科技有限公司 | Distributed multi-source heterogeneous traffic data fusion method and device |
| CN113486719A (en) * | 2021-06-08 | 2021-10-08 | 南京邮电大学 | Vehicle destination prediction method, vehicle destination prediction device, computer equipment and storage medium |
| CN113743767B (en) * | 2021-08-27 | 2022-11-04 | 广东工业大学 | Vehicle dispatching method, system, computer and medium based on time and safety |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20160119321A (en) * | 2015-04-02 | 2016-10-13 | 김진영 | Method for taxi call service |
| CN107146404A (en) * | 2016-03-01 | 2017-09-08 | 香港生产力促进局 | The information processing system and its method of a kind of real-time collecting and transmission traffic data |
| CN108335482A (en) * | 2017-01-20 | 2018-07-27 | 亚信蓝涛（江苏）数据科技有限公司 | A kind of urban transportation Situation Awareness method and method for visualizing |
| CN108538054B (en) * | 2018-05-17 | 2020-12-01 | 北京中交汇智数据有限公司 | Method and system for acquiring traffic road condition information based on mobile phone signaling data |
| CN108806250B (en) * | 2018-06-08 | 2020-10-13 | 北京航空航天大学 | Regional traffic jam evaluation method based on speed sampling data |
-
2018
- 2018-12-13 CN CN201811524358.9A patent/CN109410586B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109410586A (en) | 2019-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109410586B (en) | Traffic state detection method based on multi-metadata fusion | |
| CN109215347B (en) | Traffic data quality control method based on crowdsourcing trajectory data | |
| CN109544932B (en) | Urban road network flow estimation method based on fusion of taxi GPS data and gate data | |
| Biljecki et al. | Transportation mode-based segmentation and classification of movement trajectories | |
| CN110176139B (en) | DBSCAN + -based road congestion identification visualization method | |
| Zhang et al. | Integration of GPS traces with road map | |
| KR100820467B1 (en) | a traffic estimating system and the method considered road type | |
| CN111949896B (en) | Track abnormality detection method based on track big data | |
| CN109459047A (en) | Vehicle high-precision location data is accurately matched with navigation map and lane change behavior detection method | |
| Tang et al. | Exploring urban travel patterns using density-based clustering with multi-attributes from large-scaled vehicle trajectories | |
| CN112053558A (en) | Traffic jam state identification method, device and equipment | |
| Zhu et al. | A driving cycle detection approach using map service API | |
| CN114882069A (en) | Taxi track abnormity detection method based on LSTM network and attention mechanism | |
| Thiruppathiraj et al. | Automatic pothole classification and segmentation using android smartphone sensors and camera images with machine learning techniques | |
| Sadeghian et al. | A stepwise methodology for transport mode detection in GPS tracking data | |
| Trogh et al. | Map matching and lane detection based on Markovian behavior, GIS, and IMU data | |
| CN111444286B (en) | Long-distance traffic node relevance mining method based on trajectory data | |
| CN113295173A (en) | Map matching method for annular road section | |
| Li et al. | Driving performances assessment based on speed variation using dedicated route truck GPS data | |
| Choudhry et al. | Inferring truck activities using privacy-preserving truck trajectories data | |
| CN114091581A (en) | Vehicle operation behavior type identification method based on sparse track | |
| Li et al. | Route restoration method for sparse taxi GPS trajectory based on Bayesian network | |
| CN116989801A (en) | Map matching method and device for low-frequency long tracks of complex road network | |
| Ng et al. | Urban Travel Pattern Recognition Based on Clustering Techniques Using License Plate Sensing Data | |
| CN114662792B (en) | Traffic flow prediction method of recurrent neural network based on dynamic diffusion graph convolution |
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 |