CN108256560B

CN108256560B - Parking identification method based on space-time clustering

Info

Publication number: CN108256560B
Application number: CN201711448160.2A
Authority: CN
Inventors: 周洋; 杨超; 季彦婕
Original assignee: Tongji University
Current assignee: Tongji University
Priority date: 2017-12-27
Filing date: 2017-12-27
Publication date: 2021-05-04
Anticipated expiration: 2037-12-27
Also published as: CN108256560A

Abstract

The invention discloses a parking identification method based on space-time clustering, which comprises the following steps: collecting GPS track data of individual activities based on the smart phone, and extracting spatiotemporal information; searching the nearest k points of any track point on a time axis, and determining a core point in the track through a distance parameter Eps and a minimum number threshold MinPts; the method comprises the steps that core points which are continuous in time form an initial cluster, the inspection is started from a cluster with the maximum density, and adjacent clusters which are adjacent to each other in a time-space mode are combined to obtain parking; and (3) forming an initial trip by the non-core points which are continuous in time, starting the check from the trip with the earliest time, combining the two trips if the time interval between the two trips is smaller than the minimum parking duration threshold value, and correcting the pseudo parking as the trip. The invention can quickly and accurately identify the parking in the individual trip GPS track, lays a foundation for further identifying trip modes and trip purposes, and provides technical support for long-term, large-scale and passive urban resident trip investigation.

Description

Parking identification method based on space-time clustering

Technical Field

The invention belongs to the field of traffic data mining, and particularly relates to clustering analysis of time series data and parking identification in individual travel tracks.

Background

With the rapid popularization and development of smart phones, accurate positioning function and abundant sensor modules provide hardware conditions for real-time collection of individual travel tracks. In the face of a large amount of travel track data, analyzing individual behavior characteristics and identifying activity modes become main problems of urban traffic serving data. The recognition of parking based on individual GPS trajectory data is a precondition for judging OD, estimating a trip mode and a trip purpose. The current related research is mainly based on the speed characteristics and the moving direction characteristics under static state, and is judged by combining with the road network. And identifying parking from the angle of density clustering according to the track point clustering characteristics, the prior art mainly focuses on the DBSCAN method.

The DBSCAN sets two parameters of an Eps neighborhood and a minimum point number MinPts, and the two parameters are continuously expanded and connected outwards to form a point cluster by taking the density as a classification mode on the basis of core points. However, the time sequence of the points is not considered, the clusters of the points with close spatial distance and far time interval are easily and wrongly merged into a class, and the parking cannot be accurately identified under the conditions of coincident paths, short-time travel and the like. Specifically, when the algorithm is used for processing the smartphone GPS mobile data, the following disadvantages are specifically included: when a large number of trace points are processed in the same day, the occupied distance matrix memory is large, so that the program cannot respond or the operation speed is low; the sensitivity of the parameter Eps and MinPts is high, and the generalization capability is poor; the travel times are multiple, the path overlapping degree is high, the identification is not easy, and particularly, a large number of track points are gathered on a part of road sections or intersections with overlapped tracks, so that the parking is judged by mistake.

The method for identifying the origin-destination of the travel of the patent application No. 201611195129.3 based on the space-time clustering analysis algorithm has the core idea that the space distance in DBSCAN is expanded to be the space-time distance, control parameters delta T, Eps and MinPts are set, and the core point is searched from the unmarked point to the periphery, so that an initial cluster is obtained; and merging clusters through a threshold value of 600 seconds or 500m to obtain a travel origin-destination. But for the case of complex trip or poor signal quality, the merging threshold is not necessarily adapted; the merging sequence has randomness and influences the recognition result; the short-time stopping pseudo-parking without eliminating the false recognition results in higher traveling times.

Therefore, the clustering method considering the space-time dimension is explored, the method is suitable for complex travel or data, parking in travel tracks is efficiently and accurately identified, and the method has important significance for urban traffic data mining, travel mode identification and the like.

Disclosure of Invention

In order to solve the technical problems, the invention aims to provide a density clustering method considering a time sequence, which is used for processing individual trip GPS data acquired by a smart phone and providing a parking identification operation interface and result visualization for a user.

And a visual interface display.

The invention is realized by the following technical scheme:

1. a parking identification method based on spatio-temporal clustering comprises the following steps:

s1, collecting GPS track data of individual activities based on the smart phone, extracting space-time information from the GPS track data, and obtaining a time sequence data set;

s2, identifying the parking in the GPS track of the individual trip by adopting clustering algorithm processing to the time sequence data set, and outputting a parking identification result;

s3, evaluating the parking recognition result;

and S4, displaying the parking recognition user interface in a visual mode.

Further, in step S1, the GPS track data are collected at the same sampling frequency; the spatiotemporal information comprises time, longitude and latitude information; the time-series data set is a data set D ═ P { P } obtained by converting the longitude and latitude into

plane coordinate values

_i1, 2, …, M, where M is the number of trace points, D is the data set of M trace points, and subset P_iFor data sets of the ith trace point, P_i＝(t_i，x_i，y_i) Wherein, t_i，x_i，y_iThe time, the abscissa value and the ordinate value of the ith track point are respectively.

Further, in step S2, the clustering algorithm process further includes the following steps:

s21, setting clustering algorithm parameters including a search length k, a time adjacent threshold I, a space adjacent threshold Eps, a core point minimum discrimination number MinPts and a parking minimum duration DU;

s22, initializing labels, wherein all track points in the data set D are initially marked as Label ═ infinity, and the initial number of parking is ClustID ═ 1;

s23, searching core points, and aiming at each track point P in the data set D_iDetermining said trace point P_iK points which are most adjacent in time are respectively calculated, and each point in the k points and the track point P are respectively calculated_iIf the spatial distance is less than the number of points N of the spatially adjacent threshold Eps_epsIf the minimum number MinPts is judged to be larger than the core point, the point P is judged_iMarking as a core point, changing the Label Label-1, and otherwise keeping the Label Label-infinity;

s24, stopMerging the temporally successive core points into a set of core points { C_j}; searching core point set C with maximum core point set density DS_mCalculating the core point set C_mWith neighboring point set { C_m-1,C_m+1The space-time distance of }; if the adjacent parking time interval L_TA distance L from the adjacent parking space_DBoth are within the threshold range, then the two point sets are merged, otherwise the point set C is used_mMarking the core point set as parking, updating the Label Label (ClustID) and the cluster number (ClustID + 1), and searching the next core point set with the maximum density until all the core point sets complete labeling;

s25, row check, and merging points at the Label ∞ in the data set D into a row set { TR) if they are temporally continuous_n}; if the time interval of the adjacent trips is smaller than the minimum parking duration DU, merging the trips, and updating the Label of the pseudo-parking inner point to be 0;

s26, outputting a recognition result, merging the points which are continuous in time, adjacent and have the same mark number into the same class, wherein the parking mark number is larger than 0, and the travel mark number is equal to 0; and calculating the starting and stopping moments of parking and traveling, calculating the position coordinates of the parking center, and outputting the identification result.

Further, in step S21, the search length k refers to any trace point P_iK points P which are temporally closest to each other_Si,...,P_TiIn which P is_SiAs a starting point, P_TiAs an end point, the starting point and the end point are marked with the following symbols (S, T):

the temporal proximity, refers to a certain parking A_jTime-adjacent parking { A }_j-1,A_j+1If an adjacent parking time interval L between a next parking start time and a previous parking end time is met_TLess than the temporally adjacent threshold I;

the spatial proximity, meaning to a certain parking A_jTime-adjacent parking { A }_j-1,A_j+1If the adjacent parking space distance L between the next parking center position and the previous parking center position is satisfied_DLess than the spatially neighboring threshold value Eps.

Further, the adjacent parking time interval L_TExpressed as:

wherein, T_jsIs the start time of the jth parking, T_j,EIs the end time of the jth parking;

the adjacent parking space distance L_DExpressed as:

wherein, X_j,Y_jIs the position coordinate of the j-th parking center.

Further, in step S24, the method further includes the following steps:

s241, merging the core points which are subject to Label-1 and are continuous in time into a core point set C_jCalculating each of the core point sets C_jAt a starting time T_jSEnd time T_jECenter position coordinate (X)_j,Y_j) Wherein X is_j，Y_jRespectively as the core point set C_jAverage value of horizontal and vertical coordinates of all track points;

s242, defining density DS of core point sets, and setting the maximum space distance d between the number NP of core points in a core point set and each point of the core points_maxRatio of (D) to (NP/d)_max；

The core point set C_jMiddle, DS_j＝NP_j/d_jmax；

S243, searching to obtain the core point set C with the maximum core point set density DS_mWherein the core point set C_mStart of (2)Time T_mSEnd time T_mECenter position coordinates, (X)_m,Y_m) Wherein X is_m，Y_mRespectively as the core point set C_mAverage value of horizontal and vertical coordinates of all track points;

computing the set of core points C_mWith neighboring point set { C_m-1,C_m+1The space-time distance of }; if the point set C_m-1And the core point set C_mSaid adjacent parking time interval L_T＝T_js-T_j-1,E< I and the adjacent parking space distance

Then merge point set C_m-1,C_mAnd change C_mAt a starting time T_mS＝T_m,S-1(ii) a If C_m+1Satisfying the adjacent parking time interval L_T＝T_j+1,s-T_j,E< I and the adjacent parking space distance

Then merge point set C_m,C_m+1And change C_mEnd time T of_mE＝T_m,E+1(ii) a If { C_m-1,C_m+1All the results satisfy the merging condition, then determine C_mIn order to park, and update the Label Label as ClustID;

step S244: and updating the cluster number of ClustID +1, searching the next core point set with the maximum density and Label of-1, and repeating the step S243 until all the core point sets complete the labeling.

Further, the step S25 further includes the following steps:

s251, forming a trip set { TR for points of the data set D having Label ∞ and continuous time_nCalculating the starting time T of the trip_nSAnd an end time T_nE；

S252 for the trip TR_nCalculating and next time interval trip TR_n+1Time interval L of_T＝T_n+1,S-T_n,EIf L is_T＜DU，Description of TR_nAnd TR_n+1The duration of the parking between the two does not meet the requirement of the minimum duration, and the travel { TR is combined_n,TR_n+1And change the termination time T_nE＝T_n,E+1；

S253: the merged travel labels and the pseudo-parking labels in between are changed to Label 0, and the step S252 is repeated until all the travel tests are finished.

Further, in step S3, the parking recognition result evaluation index includes,

(1) travel time consistency TripNumFit

If predicted number of trips N_PTWith the true number of trips N_TTEqual, then 1, otherwise 0;

(2) travel start-stop time difference TimeDiff

Predicted starting time t of all trips_psAnd end time t_peRespectively with the start time t of the log record_tsAnd end time t_teAverage difference of (d);

(3) dwell duration difference DUDiff

Predicted duration DU of all parking_pkDuration of parking DU with logging_tkAverage difference of (d);

(4) active position deviation distance DistP

Predicted center position P for all parking_pcAnd the true position P_tcAverage deviation distance of (d);

wherein N is_TAThe number of stops.

Furthermore, a parking recognition user interface is displayed visually, the clustering algorithm parameters are input into the interface facing the user, and a parking recognition graph, the stop-start position and stop position stay time of the parking in the GPS data and the evaluation index of the parking recognition result are displayed.

Furthermore, the sampling frequency of the GPS track data is 1 Hz.

And the minimum parking duration DU is used for judging whether the identified cluster is parked or not, and the threshold value is defaulted to 120 s.

The method identifies a core point, searches k points which are nearest on a time axis of the core point from a first point of a track, and judges that the point is the core point when the number of points in the range of Eps reaches MinPts, wherein the number of the points mainly depends on k and Eps.

The invention is combined by calculating the difference between the previous parking ending time and the next parking starting time and the distance between the previous parking center position and the next parking center position, and combining if the difference is less than the parameters I and Eps respectively.

According to the false parking filtering method, due to the fact that accidental parking possibly exists in the track, such as red light at signalized intersections, traffic jam at rush hour road sections, bus waiting at bus stations and the like, the false parking filtering method can obtain a more accurate identification result by setting the minimum parking time length to filter the false parking filtering.

Compared with the prior art, the invention has the following beneficial effects:

the invention makes up the defect of processing time sequence data by DBSCAN, can accurately identify the conditions of a coincident path, short trip and the like by setting the threshold values of time proximity and space proximity, and has better generalization capability.

The method comprises the steps of searching k nearby points to judge core points, and generating an initial cluster from the perspective of space-time aggregation; defining the density of clusters to reflect the space aggregation degree and determining the merging order of the clusters, thereby avoiding randomness and reducing merging errors; the method is suitable for the conditions of complex travel or poor signal quality by setting a merging threshold; and meanwhile, considering the minimum parking time length, and filtering the false parking.

The invention can quickly and accurately identify the parking in the individual trip GPS track, lays a foundation for further identifying trip modes and trip purposes, and provides technical support for long-term, large-scale and passive urban resident trip investigation.

Drawings

FIG. 1 is a flow chart of a spatio-temporal clustering-based parking identification method of the present invention.

Fig. 2 is a schematic diagram of an effect track I recognition result of the parking recognition example of the present invention, where a thick solid line indicates parking and a thin solid line indicates travel.

Fig. 3 is a schematic diagram of an effect track II recognition result of the parking recognition example of the present invention, where a thick solid line indicates parking and a thin solid line indicates travel.

FIG. 4 is a graphical illustration of a docking recognition visualization in accordance with the present invention.

Detailed Description

The following describes embodiments of the present invention in detail, and the embodiments are developed based on the technical solutions of the present invention, and provide detailed implementation manners and specific operation procedures.

The implementation case is as follows:

selecting two full-day travel GPS tracks collected by a smart phone, wherein the track I is characterized by long parking time and divergent points in space, and the recording time is 6:37: 02-20: 06:53, which comprises 3 times of parking and 2 times of traveling; the track II is characterized by more travel times, high path overlapping degree and recording time of 7:04: 13-18: 44:05, wherein 6 times of parking and 5 times of travel are included.

Setting parameters: the search length k is 61, the temporal proximity threshold I is 30s, the spatial proximity threshold Eps is 30m, the core point minimum number MinPts is 30, and the parking minimum duration DU is 120 s.

Table 1 trip log

And (3) processing and identifying the parking in the track I and the track II according to the travel log GPS track data in the table 1 and the parking identification method flow chart based on the space-time clustering shown in the figure 1, and outputting a parking identification result.

The track I recognition result is shown in fig. 2, the thick solid line is parking, and the thin solid line is travel.

The recognition result of the trajectory II is shown in FIG. 3, wherein the thick solid line is parking and the thin solid line is traveling.

And evaluating the parking recognition results of the track I and the track II according to the evaluation indexes, as shown in the table 2 for the parking recognition evaluation.

TABLE 2 parking identification evaluation

The parking identification can be visually displayed, for example, as shown in fig. 4, clustering algorithm parameters are input into a display interface, and a parking identification map, the stop-start position dwell time of the parking in the GPS data and the evaluation index of the parking identification result are displayed.

The above embodiments are preferred embodiments of the present application, and those skilled in the art can make various changes or modifications without departing from the general concept of the present application, and such changes or modifications should fall within the scope of the claims of the present application.

Claims

1. A parking identification method based on spatio-temporal clustering is characterized by comprising the following steps:

the GPS track data are collected at the same sampling frequency; the spatiotemporal information comprises time, longitude and latitude information; the time-series data set is a data set D ═ P { P } obtained by converting the longitude and latitude into plane coordinate values_i1, 2, …, M, where M is the number of trace points, D is the data set of M trace points, and subset P_iFor data sets of the ith trace point, P_i＝(t_i，x_i，y_i) Wherein, t_i，x_i，y_iRespectively time, an abscissa value and an ordinate value of the ith track point;

s2, identifying the parking in the GPS track of the individual trip by adopting clustering algorithm processing to the time sequence data set, and outputting a parking identification result, wherein the clustering algorithm processing further comprises the following steps:

s24, merging the core points which are continuous in time into a core point set { C_j}; searching core point set C with maximum core point set density DS_mCalculating the core point set C_mWith neighboring point set { C_m-1,C_m+1The space-time distance of }; if adjacent parking time interval L_TA distance L from the adjacent parking space_DBoth are within the threshold range, then the two point sets are merged, otherwise the point set C is used_mMarking as parking, updating the Label Label ClustID and the cluster number ClustID +1, and searching the next core point set with the maximum density until all the core point sets are finishedForming a label;

s26, outputting a recognition result, and merging the points which are continuous and adjacent in time and have the same mark number into the same class, wherein the parking mark number is not 0 and not infinity, and the trip mark number is equal to 0 or infinity; calculating the starting and stopping moments of parking and traveling, calculating the position coordinates of a parking center, and outputting an identification result;

s3, evaluating the parking recognition result;

and S4, displaying the parking recognition user interface in a visual mode.

2. The spatio-temporal clustering-based parking identification method according to claim 1, wherein in step S21, the search length k refers to any locus point P_iK points P which are temporally closest to each other_Si,...,P_TiIn which P is_SiAs a starting point, P_TiAs an end point, the starting point and the end point are marked with the following symbols (S, T):

3. The spatio-temporal clustering-based parking identification method according to claim 2, wherein the adjacent parking time intervals L are_TExpressed as:

the adjacent parking space distance L_DExpressed as:

wherein, X_j,Y_jIs the position coordinate of the j-th parking center.

4. The spatio-temporal clustering-based parking identification method according to claim 1, wherein in step S24, the method further comprises the following steps:

The core point set C_jMiddle, DS_j＝NP_j/d_jmax；

S243, searching to obtain the core point set C with the maximum core point set density DS_mWherein the corePoint set C_mAt a starting time T_mSEnd time T_mECenter position coordinates, (X)_m,Y_m) Wherein X is_m，Y_mRespectively as the core point set C_mAverage value of horizontal and vertical coordinates of all track points;

computing the set of core points C_mWith neighboring point set { C_m-1,C_m+1The space-time distance of }; if the point set C_m-1And the core point set C_mSaid adjacent parking time interval L_T＝T_ms-T_m-1,E< I and the adjacent parking space distance

Then merge point set C_m-1,C_mAnd change C_mAt a starting time T_mS＝T_m,S-1(ii) a If C_m+1Satisfying the adjacent parking time interval L_T＝T_m+1,s-T_m,E< I and the adjacent parking space distance

5. The spatio-temporal clustering-based parking identification method according to claim 1, wherein said step S25 further comprises the steps of:

S252 for the trip TR_nCalculating and next time interval trip TR_n+1Time interval L of_T＝T_n+1,S-T_n,EIf L is_T< DU, indicating TR_nAnd TR_n+1The duration of the parking between the two does not meet the requirement of the minimum duration, and the travel { TR is combined_n,TR_n+1And change the termination time T_nE＝T_n,E+1；

6. The spatio-temporal clustering-based parking recognition method according to claim 1, wherein in step S3, the evaluation index of the parking recognition result comprises,

(1) travel time consistency TripNumFit

(2) travel start-stop time difference TimeDiff

(3) dwell duration difference DUDiff

(4) active position deviation distance DistP

wherein N is_TAThe number of stops.

7. The spatio-temporal clustering-based parking recognition method according to claim 6, wherein the parking recognition user interface is visually displayed, the clustering algorithm parameters are input in the user-oriented interface, and a parking recognition map and the start-stop position dwell time and the parking recognition result evaluation index of the parking recognition in the GPS data are displayed.

8. The spatio-temporal clustering-based parking identification method according to claim 1, wherein the sampling frequency of the GPS trajectory data is 1 Hz.