CN108122186B - Job and live position estimation method based on checkpoint data - Google Patents

Abstract

The patent discloses a vehicle position estimation method, which includes: step (1) obtaining card port data and preprocessing the card port data; dividing the vehicle journey; extracting important recording points; and (4) clustering the sample points by using an improved k-means clustering algorithm. The home position of the vehicle can be effectively estimated, and a foundation is laid for further analyzing the travel characteristics of the vehicle and the like.

Description

Job and live position estimation method based on checkpoint data

Technical Field

The invention relates to the field of traffic, in particular to a position estimation method based on checkpoint data.

Background

The gate device is an intelligent control device which is arranged on a traffic junction road and is used for monitoring the road by combining software and hardware which are supported by a camera, can record passing vehicle information including license plate numbers, passing time, driving directions and the like in 24 hours all day, and provides reliable data support for vehicle travel characteristics and road network state analysis. In the travel characteristic, if the position of the vehicle is known, the method has great significance for analyzing and controlling the related traffic conditions such as estimation of traffic flow and the like. The job location is the location where the job (job) and the residence (house) are located.

At present, there are many methods for estimating the OD (starting point and end point) of a traveler, wherein a patent with application number 201110163206.2- "dynamic OD matrix estimation method based on vehicle automatic identification device" mainly uses the microscopic path restoration of a vehicle as a main means for dynamic OD estimation and adopts a bayesian estimation algorithm to operate, and the technical scheme has large data amount and large calculation amount, is not applicable to the environment where hardware conditions cannot be met, and has a space for improving the prediction result.

The patent with application number 201310213953.1- 'public transport card passenger commuting OD distribution estimation method based on intelligent public transport system data' is mainly based on three resident travel characteristic assumptions, a riding frequency statistical method and a clustering analysis method are used for calculating resident OD, and the method is limited by OD data of public transport IC cards.

Disclosure of Invention

The vehicle position estimation method based on the checkpoint data can overcome the defects and realize vehicle position estimation by combining the characteristics of the checkpoint data.

In order to solve the technical problem, the technical scheme provided by the patent comprises: a vehicle occupancy location estimation method, the method comprising: step (1) obtaining card port data and preprocessing the card port data; the pretreatment comprises the following steps: (1.1) extracting a gate number K, a license plate number M, a passing time S and a form direction F from original gate data by data dimension reduction, and deleting the rest columns of data; (1.2) grouping the data according to the license plate number; all data are grouped according to the license plate number M, namely the data with the same license plate number are put into a set M_i＝{M_i1，M_i2，…，M_inIn which M is_inIndicates that the number plate is M_iThe nth piece of data of the vehicle; secondly, sorting according to the passing time; then, the data of the same license plate number are sequenced according to the passing time S to obtain a vehicle data set M arranged according to the time sequence_is＝{M_is1，M_is2，…，M_isnIn which M is_isnIndicates that the number plate is M_iThe nth piece of data of the vehicle; (1.3) data cleaning and deleting unidentified data; deleting the duplicate records; step (2) dividing vehicle travel (2.1) dividing each vehicle recording point every day into independent travel based on data recording time intervals; first, data set M is paired by date_isThe track data set M of each vehicle is obtained by grouping the data in the step (1)_id＝{M_id1，M_id2，…，M_idnIn which M is_idnIndicates that the number plate is M_iThe nth data of the vehicle on day d; (2.2) setting a threshold value B, and dividing recording points; according to a vehicle daily track data set M_idCalculating the time difference Deltat between two adjacent recording points_d＝{Δt_d1，Δt_d2，…，Δt_dn-1And setting a threshold value B based on the checkpoint data and the actual road network, and if delta t is obtained_di<B, the (i + 1) th recording point belongs to the next travel, otherwise, the recording point belongs to the current track set, and all travel track sets, M, of each vehicle per day are obtained_id＝{M_id1，M_id2，…，M_idn}; extracting important recording points, selecting the first and last journey of the vehicle every day, and extracting two types of recording points: the first type is a first recording point of a first stroke and a last recording point of a last stroke and is marked as type A; the second type is the last recording point of the first journey and the first recording point of the last journey, and is marked as type B; and according to the bayonet number and the actual position latitude and longitude coordinates of the bayonet, putting the extracted latitude and longitude coordinates of the recording point into a new sample set P, wherein P is_i＝{PA_i，PB_iIn which PA is_iA class a record point data set representing all the ith vehicle; step (4) clustering sample points by using the improved K-means clustering algorithm (4.1) improving the measurement method of the K-means clustering algorithm, and measuring the similarity between samples by adopting a spherical distance to replace an Euclidean distance in the K-means; the calculation formula of the spherical distance is as follows: d (P)_i，P_j) R × arccos (a + b); wherein D (P)_i，P_j) Represents P_iPoint and point P_jThe shortest spherical distance therebetween; r is the radius of the earth; a sin (P)_iLon×π/180)×sin(P_jLon×π/180)，P_iLonIs a point P_iLongitude, P of_jLonIs a point P_jLongitude of (d); co (P)_iLat×π/180)×cos(P_jLat×π/180)×cos(P_jLon-P_iLon)，P_iLatIs a point P_iLatitude of (P)_jLatIs a point P_jThe latitude of (d); (4.2) setting a clustering center K, performing clustering analysis, setting the number K of data subsets to be generated, and performing clustering analysis on the data subsetsInput data set PA_iDividing into k classes to obtain a data set C ═ C₁，…，c_kIn which c is_kRepresenting a set of data divided into a kth class; the initial value of k is randomly selected from the samples to obtain k classes, the class max { C } with the largest number of samples is selected from the k classes, the clustering center of the k classes is used as an estimated home position, and the k classes are also selected from the data set PB_iAnd the clustering center obtained by the middle clustering is used as an estimated working position.

Preferably, the threshold B is calculated specifically as follows: (1) acquiring length information L ═ L of all road sections of m roads₁，…，l_mAnd the maximum allowable speed V ═ V for each road₁，…，v_m}; (2) calculating travel time T ═ T for each road segment₁，…，t_mTherein of

(3) And B is max { T }, and the maximum value is found from the travel time of all the road sections in the road network and is used as the threshold B.

The invention provides a trip travel identification method based on vehicle track recording time intervals around an exploratory algorithm and based on bayonet data; in the calculation of the time interval threshold B, the actual road network information is fully considered, and road travel time is calculated by combining the checkpoint data, so that the reliability of the threshold B is ensured; according to the method, the commuting characteristics of the vehicle are combined, a series of 'important' recording points in the driving process are selected from the single-stroke track points of the vehicle to serve as a position estimation sample set, the Euclidean distance in a K-means algorithm is improved to be a spherical distance according to the characteristics of bayonet data, the important recording points are clustered, and the complexity of the method is reduced; the home position of the vehicle can be effectively estimated, and a foundation is laid for further analyzing the travel characteristics of the vehicle and the like.

Drawings

FIG. 1 is a schematic diagram of a road network according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of vehicle trip division based on an exploratory algorithm;

FIG. 3 is a flow chart of vehicle trip travel track point division;

fig. 4 is a schematic diagram of important recording points of the vehicle track.

Detailed Description

The present invention is further described in detail below with reference to the drawings and examples so that those skilled in the art can practice the invention with reference to the description. The specific embodiment provides a bayonet data-based home location estimation method, which comprises the following specific steps:

(1) preprocessing the data of the bayonet;

and the preprocessing of the buckle data comprises data dimension reduction, data grouping and data cleaning. The data content with the maximum value can be obtained through data preprocessing, the implementation of the method is facilitated, and the accuracy and the efficiency of the method are improved.

(1.1) data dimension reduction

The amount of raw bayonet data information is large and not all information is useful for this embodiment. In order to improve the analysis efficiency, the key information is extracted from the checkpoint data and the data dimension is reduced in the specific embodiment. Specifically, in the present embodiment, the gate number K, the license plate number M, the passing time S, the traveling direction F, and the remaining row data are extracted from the original gate data.

(1.2) data reconstitution

1. Grouping according to license plate numbers; all data are grouped according to the license plate number M, namely the data with the same license plate number are put into a set M_i＝{M_i1，M_i2，…，M_inIn which M is_inIndicates that the number plate is M_iThe nth piece of data of the vehicle.

2. Sorting according to the passing time; then, the data of the same license plate number are sequenced according to the passing time S to obtain a vehicle data set M arranged according to the time sequence_is＝{M_is1，M_is2，…，M_isnIn which M is_isnIndicates that the number plate is M_iThe nth piece of data of the vehicle.

(1.3) data cleansing

Deleting unidentified data; deleting the record of which the license plate number in the card port data cannot be successfully identified; deleting the duplicate records; and performing deletion operation aiming at repeated records of which the vehicle is continuously detected for many times.

Data that is generally considered to be the word "unrecognized" for the license plate number and data that is a wrong identification of the license plate number, such as a scrambling code, are records that the license plate number has not been successfully recognized and that need to be deleted. The repeated data contains data with the same bayonet number K, license plate number M, passing time S and driving direction F, but the data with overlapped contents can not be regarded as the repeated data, only one piece of data in the repeated data is reserved for the repeated data, and the rest of data is deleted.

(2) Dividing vehicle range

(2.1) dividing each vehicle recording point per day into independent trips based on the data recording time interval,

first, data set M is paired by date_isThe track data set M of each vehicle is obtained by grouping the data in the step (1)_id＝{M_id1，M_id2，…，M_idnIn which M is_idnIndicates that the number plate is M_iThe nth data of the vehicle on day d.

(2.2) setting a threshold B, and dividing recording points

According to a vehicle daily track data set M_idCalculating the time difference Deltat between two adjacent recording points_d＝{Δt_d1，Δt_d2，…，Δt_dn-1And setting a threshold value B based on the checkpoint data and the actual road network, and if delta t is obtained_di<B, the (i + 1) th recording point belongs to the next travel, otherwise, the recording point belongs to the current track set, and all travel track sets, M, of each vehicle per day are obtained_id＝{M_id1，M_id2，…，M_idn}。

For example, in the road network shown in fig. 1, the length L ═ L of all the roads in the road network is obtained₁，…，l_mAnd maximum allowable speed information V ═ V₁，…，v_mCalculating the travel time T of all road sections as T₁，…，t_mSelecting the maximum travel time max { T } as the value of the threshold B。

Based on the track data of each vehicle which is obtained in the step (1) and sorted according to time every day, calculating the time intervals of the adjacent recording points, comparing the time intervals with B one by one, and if the time intervals are smaller than B, enabling the two adjacent track points corresponding to the time intervals to belong to a track set; otherwise, if the distance is larger than B, the two track points belong to two different track sets, and the track points at the back in the time sequence are divided into the next track set, as shown in FIG. 2, t₁，t₂，t₃，t₄All are greater than threshold B, the flow chart is shown in fig. 3;

(3) extracting important recording points

According to the commuting rule of the vehicle, selecting the first and the last journey of the vehicle every day, and extracting two types of recording points:

the first type is a first recording point of a first stroke and a last recording point of a last stroke and is marked as type A;

the second type is the last recording point of the first journey and the first recording point of the last journey, and is marked as type B;

and according to the bayonet number and the actual position latitude and longitude coordinates of the bayonet, putting the extracted latitude and longitude coordinates of the recording point into a new sample set P, wherein P is_i＝{PA_i，PB_iIn which PA is_iAll the class a record point data sets of the i-th vehicle are represented.

: extracting A-type recording points and B-type recording points of each vehicle based on all the travel track recording points of the vehicles obtained in the step (2), and respectively putting the A-type recording points and the B-type recording points into sample sets A and B as shown in fig. 4 to obtain two types of sample sets of the ith vehicle, wherein the two types of sample sets are respectively PA_i，PB_i；

(4) Improving a k-means clustering algorithm to cluster sample points

(4.1) method for progress rate modification

In K-means, Euclidean distances are typically used to measure the similarity between samples. And track points in the bayonet data are longitude and latitude coordinates, so that the Euclidean distance is replaced by the spherical distance. The calculation formula of the spherical distance is as follows:

D(P_i，P_j)＝R×arccos(a+b)

wherein D (P)_i，P_j) Represents P_iPoint and point P_jThe shortest spherical distance therebetween; r is the radius of the earth;

a＝sin(P_iLon×π/180)×sin(P_jLon×π/180)，P_iLonis a point P_iLongitude, P of_jLonIs a point P_jLongitude of (d);

b＝cos(P_iLat×π/180)×cos(P_jLat×π/180)×cos(P_jLon-P_iLon)，P_iLatis a point P_iLatitude of (P)_jLatIs a point P_jThe latitude of (c).

(4.2) setting a clustering center K, and carrying out clustering analysis

Setting the number k of data subsets to be generated, the data set PA to be input_iDividing into k classes to obtain a data set C ═ C₁，…，c_kIn which c is_kRepresenting a set of data divided into the kth class. Where the initial value of k is randomly chosen from the samples. Obtaining k classes, selecting the class max { C } containing the largest number of samples from the k classes, taking the cluster center of the class max { C } as the estimated home position, and similarly, obtaining the data set PB from the data set_iAnd the clustering center obtained by the middle clustering is used as an estimated working position.

Claims (1)

1. A method for estimating a position of a vehicle occupancy, the method comprising:

step (1) obtaining card port data and preprocessing the card port data;

the bayonet device is an intelligent control device which is arranged on a traffic junction road and is used for monitoring the road by combining software and hardware which are supported by a camera, and the pretreatment comprises the following steps:

(1.1) data dimension reduction

Extracting a gate number K, a license plate number M, a passing time S and a form direction F from original gate data, and deleting the rest columns of data;

(1.2) data reconstitution

Firstly, grouping according to license plate numbers; all the data are grouped according to the license plate number M, namely the number of the same license plate numberAccording to put into a set M_i＝{M_i1，M_i2，...，M_inIn which M is_inIndicates that the number plate is M_iThe nth piece of data of the vehicle; secondly, sorting according to the passing time; then, the data of the same license plate number are sequenced according to the passing time S to obtain a vehicle data set M arranged according to the time sequence_is＝{M_is1，M_is2，...，M_isnIn which M is_isnIndicates that the number plate is M_iThe nth piece of data of the vehicle;

(1.3) data cleansing

Deleting unidentified data; deleting the duplicate records;

step (2) dividing the vehicle journey

(2.1) dividing each vehicle recording point every day into independent trips based on the data recording time interval; first, data set M is paired by date_isThe track data set M of each vehicle is obtained by grouping the data in the step (1)_id＝{M_id1，M_id2，...，M_idnIn which M is_idnIndicates that the number plate is M_iThe nth data of the vehicle on day d;

(2.2) setting a threshold value B, and dividing recording points; according to a vehicle daily track data set M_idCalculating the time difference Deltat between two adjacent recording points_d＝{Δt_d1，Δt_d2，...，Δt_dn-1And setting a threshold value B based on the checkpoint data and the actual road network, and if delta t is obtained_diIf the number of the recording points is less than B, the (i + 1) th recording point belongs to the next travel, otherwise, the recording points belong to the current track set, and all travel track sets, M, of each vehicle per day are obtained according to the current track set_id＝{M_id1，M_id2，...，M_idn}；

Extracting important record points in step (3)

The first and last journey of the vehicle each day are selected, and two types of recording points are extracted: the first type is a first recording point of a first stroke and a last recording point of a last stroke and is marked as type A; the second type is the last recording point of the first journey and the first recording point of the last journey, and is marked as type B; and according to the cardThe mouth number and the longitude and latitude coordinates of the actual position of the bayonet are taken, and the extracted longitude and latitude coordinates of the recording point are put into a new sample set P_i＝{PA_i，PB_iIn which PA is_iA class a record point data set representing all the ith vehicle;

step (4) clustering sample points by using an improved k-means clustering algorithm

(4.1) measurement method for improving k-means clustering algorithm

The similarity between the samples is measured by adopting the spherical distance to replace the Euclidean distance in the K-means; the calculation formula of the spherical distance is as follows: d (P)_i，P_j) R × arccos (a + b); wherein D (P)_i，P_j) Represents P_iPoint and point P_jThe shortest spherical distance therebetween; r is the radius of the earth; a sin (P)_iLon×π/180)×sin(P_jLon×π/180)，P_iLonIs a point P_iLongitude, P of_jLonIs a point P_jLongitude of (d); co (P)_iLat×π/180)×cos(P_jLat×π/180)×cos(P_jLon-P_iLon)，P_iLatIs a point P_iLatitude of (P)_jLatIs a point P_jThe latitude of (d);

(4.2) setting a clustering center K, and carrying out clustering analysis

Setting the number k of data subsets to be generated, the data set PA to be input_iDividing into k classes to obtain a data set C ═ C₁，...，c_kIn which c is_kRepresenting a set of data divided into a kth class; the initial value of k is randomly selected from the samples to obtain k classes, the class max { C } with the largest number of samples is selected from the k classes, the clustering center of the k classes is used as an estimated home position, and the k classes are also selected from the data set PB_iThe clustering center obtained by the middle clustering is used as an estimated working position;

the threshold B is specifically calculated as follows:

(1) acquiring length information L ═ L of all road sections of m roads₁，...，l_mAnd the maximum allowable speed V ═ V for each road₁，...，v_m}；

(2) Calculating travel time for each road segmentT＝{t₁，...，t_mTherein of

(3) And B is max { T }, and the maximum value is found from the travel time of all the road sections in the road network and is used as the threshold B.

Images