KR100923723B1 - Method for clustering similar trajectories of moving objects in road network databases - Google Patents

Abstract

The present invention relates to a method for clustering similar trajectories in a road network space, comprising: (a) storing trajectory data of a moving object in a road network space in a database; (b) determining whether there is a similarity measure function value matched with the trajectory data stored in the database; (c) measuring and storing the similarity according to the similarity measuring function when it does not match with the similarity measuring function in step (b); (d) searching for any two trajectories of which the similarity measurement function has the largest value; (e) mapping the trajectories in k dimensions based on the two trajectories retrieved in step (d); (f) performing clustering on the result mapped in step (e); And (g) storing the cluster configured in the step (f) in a database, thereby enabling effective similar trajectory search and clustering for moving objects in the road network.

Road network, moving object, similar trajectory, similarity function, DSL function, summary information

Description

Method for clustering similar trajectories of moving objects in road network databases}

1 is a schematic diagram of a segment representing a trajectory of a moving object according to an embodiment of the present invention.

2 is a flowchart for measuring similarity of a moving object trajectory in a road network space according to the present invention.

3 is a flowchart of a process for clustering similar trajectories of moving objects in a road network space according to the present invention.

4 is an exemplary diagram of clustering similar trajectories of a moving object on a road network according to the present invention.

5 illustrates a summary process of cluster information in a clustering process of a similar trajectory of a moving object according to the present invention.

6 is a flowchart of searching for a trajectory similar to a trajectory of a query of a moving object in a road network space according to the present invention.

The present invention relates to a similar trajectory clustering method in a road network space, and more particularly, to efficiently track a trajectory cluster similar to a given query trajectory for a large trajectory information of moving objects moving over time in a road network trajectory space. Similar trajectory clustering method in road network space for searching.

Recently, due to the development of technologies such as mobile communication, telematics, and GPS, there is increasing interest in a method for effectively utilizing the location information of a moving object.

The moving object periodically transmits its location to the server, which has the property of spatio-temporal data in which spatial location information changes over time.

The user's query for the moving object database can be broadly classified into two types: a past time query that retrieves the past movement path history of the moving object and a future time query that predicts and searches for the future movement of the moving object. The future time query can be used for various services based on forecasting future conditions, such as location-based services, traffic information systems, and aircraft control systems.

The service is based on trajectory information of a moving object measured using GPS in the satellite at regular time intervals.

Attempts have been made to search or cluster trajectories similar to the trajectories of a given moving object among the techniques for efficiently storing and managing the trajectory information, and to analyze them in connection with road information and user information.

Most of these attempts are based on the trajectory information of a moving object represented as a series of two-dimensional spatial coordinates (x, y) in Euclidean space, and the spatial coordinates as a similarity measure to search for similar trajectories. Use Euclidean distance, which is a distance measurement method.

To measure the similarity between moving object trajectories in Euclidean space, the EU (Euclidean distance), dynamic time warping distance (DTW), edit distance with real penalty (ERP), longest common sub-sequences (LCSS), and EDR ( Techniques using distance functions such as edit distance in real sequence have been proposed.

In the EU method, when two trajectories having the same length are given, the Euclidean distance between k-dimensional space-time coordinates constituting the trajectory is obtained.

This method has the constraint that the lengths of the two trajectories being compared must be the same, so in practice, this method is not suitable because the lengths of the trajectories are not the same.

The DTW method is a method that allows to repeat an arbitrary number of specific space-time coordinate values in the trajectory to measure the similarity between trajectories having different lengths. The ERP method allows gaps in similar trajectories, and can measure similarity between two trajectories of different lengths, but it is very sensitive to noise. EU, DTW, and ERP functions are all noise-sensitive similarity measurement methods, and thus have a disadvantage of low accuracy when targeting actual trajectories that are likely to generate noise during the acquisition or representation of the trajectory.

The LCSS method and the EDR method are solved. These methods consider the two coordinates to be matched if the difference in the spatiotemporal coordinate values in the pseudo trajectory is less than a given allowance to reduce the sensitivity due to noise. However, the LCSS method has a disadvantage in that accuracy is low as a similarity measurement method that does not allow a gap in the trajectory. The EDR method uses an edit distance for measuring similarity and, unlike LCSS, allows a gap in a trajectory.

Since these methods are all based on Euclidean space, they are not suitable for measuring similarity between trajectories on the road network space targeted by the present invention.

The trajectory on the road network space is represented by a series of road segments passed by the moving object, in which case the same road segment does not appear in the trajectory repeatedly in succession. Therefore, DTW or ERP method that allows repetition is not applicable. In addition, the LCSS or the EDR method can measure the similarity with respect to the identifier of the road segment constituting the trajectory, but the similarity measurement process for the sub trajectory may be repeatedly performed, thereby causing a performance degradation problem.

The present invention is proposed to solve the above problems, and to provide a clustering method through the similarity measurement function for the trajectory information of the moving object stored in the database, not in the existing Euclidean space.

Another object of the present invention is to provide a query processing method for quickly searching a corresponding cluster for a query trajectory of a moving object.

In addition, it is to provide a clustering search method that is utilized by providing the user information and road information associated with the cluster found by the similarity to the trajectory of the moving object to the user.

In the road network according to the present invention for achieving the above object, the method for measuring the similarity of the trajectory of the moving object, (a) the trajectory of the moving object in the road network space as the trajectory data including the segment identifier and the length of the segment in the database; Storing; (b) determining whether there is a measured value calculated by a similarity measurement function for the stored specific trajectory and the trajectory already stored in the database; (c) calculating a similarity measure function value by the similarity measure function when there is no value calculated by the similarity measure function for the stored specific trajectory; (d) storing the function value measured in step (c) in the database as a similarity value between the specific trajectory in the database and the trajectory stored in the database; And (e) if there is no value calculated by the similarity measurement function for the specific trajectory stored in the step (b), storing the similarity measurement value for which the similarity measurement function value has already been calculated for the specific trajectory. It is characterized by.

In addition, in the method for measuring the similarity of the moving object trajectory in the road network according to the present invention, the similarity measurement function, when any two trajectories are respectively T _i , T _j , the following equation DSN (T _i , T _j )

It is characterized by being given.

In addition, in the method for measuring the similarity of the moving object trajectory in the road network according to the present invention, when the similarity measuring function is any two trajectories, respectively, T _i , T _j , the following equation DSL (T _i , T _j )

It is characterized by being given.

Meanwhile, a method for clustering similar trajectories of moving objects in a road network space may include: (a) storing trajectory data of the moving objects in a database in a road network space; (b) determining whether there is a similarity measure function value matched with the trajectory data stored in the database; (c) measuring and storing the similarity according to the similarity measuring function when it does not match with the similarity measuring function in step (b); (d) searching for any two trajectories of which the similarity measurement function has the largest value; (e) mapping the trajectories in k dimensions based on the two trajectories retrieved in step (d); (f) performing clustering on the result mapped in step (e); And (g) storing the cluster configured in the step (f) in a database.

In addition, in the method of clustering similar trajectories of moving objects in a road network space according to the present invention, the similarity measurement function is assuming that any two trajectories are T _i and T _j , respectively, the following equation DSL (T _i , T _j )

It is characterized by being given.

In addition, the method for clustering similar trajectories of moving objects in a road network space according to the present invention, the method comprising: (i) determining whether information of clusters is summarized; (j) if it is determined in step (i) that the information of the cluster is not summarized, summarizing the cluster information by a frequency calculation formula; And (k) storing the cluster summary information in a database.

In addition, in the method of clustering similar trajectories of moving objects in a road network space according to the present invention, the cluster summary information is a segment list of each trajectory included in the cluster as segment summary information for each cluster, and the corresponding segment in each cluster. And a weight of a segment indicating a frequency of occurrence of the list.

Further, in the method for clustering similar trajectories of moving objects in a road network space according to the present invention, the frequency calculation equation is

It is characterized by being given.

In addition, the method for clustering similar trajectories of moving objects in a road network space according to the present invention, the method comprising: (m-1) introducing a query trajectory into a database; (m-2) extracting segments constituting the query trajectory; (m-3) determining whether the segment of the query trajectory and the segment of the trajectory cluster summary information stored in the database match each other; (m-4) obtaining a sum of segment weights of the matching clusters when the matching clusters are present in the step (m-3); (m-5) searching for a cluster having the largest sum of the weights; And (m-6) determining the searched cluster as a cluster to which the query trajectory belongs.

In addition, the method for clustering similar trajectories of a moving object in a road network space according to the present invention, the method further comprises the step of presenting to the user a cluster found to be similar to the query trajectory, user information and road information associated with the cluster. It is to be done.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

1 is a schematic diagram of a segment representing a trajectory of a moving object according to an embodiment of the present invention.

As shown in FIG. 1, in the present invention, the trajectory of the moving object is expressed as Ti = {(S ₁ , L ₁ ), ..., (S _n , L _n )}. Here, T _i is an identifier of a trajectory, S _j is an identifier of a segment, and L _j is a length of a segment.

In addition to the trajectory information, user information, road information, and the like are stored and managed together as additional information.

In the scheme proposed by the present invention, the trajectory is composed of a list of segment identifiers represented by a string.

2 is a flowchart for measuring similarity of a moving object trajectory in a road network space according to the present invention.

Referring to FIG. 2, first, a new trajectory of a moving object on a road network space is stored in a database as trajectory data including a segment identifier and a length of a segment (S21).

Next, it is determined whether there is already a measured value calculated by the similarity measurement function with respect to the stored specific trajectory and the trajectory already stored in the database (S22).

As a result of the determination in step S22, when there is no measurement value calculated by the similarity measurement function for the stored specific trajectory, the similarity measurement function value is determined for the stored specific trajectory and the trajectories already stored in the database by the similarity measurement function. Calculate (S23).

The function value measured in step S23 is stored in the database as a similarity value between the specific trajectory and the trajectories already stored in the database (S24).

As a result of the determination in step S22, when there is a value already calculated by the similarity measurement function for the stored specific trajectory, the similarity measurement function value already calculated for the stored specific trajectory and the trajectory already stored in the database is used as the similarity value. Save (S25).

In general, the ED (edit distance) function, which is widely used as a distance function between strings, can be used to measure similarity between two given trajectories.

However, when the number of segments of two compared trajectories are different from each other, the similarity is affected by the trajectories having a large number of segments.

The present invention proposes a trace T _i and the similarity measure function DSN (disimilarity with number) of the formula (1) between T _j.

(1)

(One)

For example, the similarity value obtained by the DSN function for the trajectories of FIG. 1 is obtained as follows.

DSN (T1, T2) = 0.56, DSN (T1, T3) = 0.4, DSN (T1, T4) = 0.56, DSN (T2, T3) = 0.64, DSN (T2, T4) = 0.6, DSN (T3, T4) = 0.64

The similarity by the ED function for T2, T3, and T4 compared to T1 yields the same value of 3.

On the other hand, the value of the proposed DSN function shows the most similar value of T3 to T1, as suggested above.

In fact, since there are more common segments of T1 and T3, the value of DSN is more reliable than the value of ED. This is because the DSN function calculates the distance between two trajectories using the sum of the number of segments of the two trajectories.

That is, when calculating the similarity, the influence of any one trajectory having a large number of segments among the two trajectories as in the ED function can be reduced.

However, since the similarity is measured only by the match and the number of segment identifiers, the degree of proximity of the two trajectories cannot be known.

For example, the similarity of the trajectories T1 and T2 and the similarity of the trajectories T1 and T4 have the same value by the DSN function. This is because the number of non-common segments of T1 and T2 and the number of non-common segments of T1 and T4 are the same.

However, as shown in FIG. 1, since the trajectory of T2 is closer to T1 than the trajectory of T4, it can be said that T1 and T2 are more similar.

Even in the same road, the number of segments included in the trajectory may vary depending on the criteria for dividing the segments. Since the proposed DSN function is affected by the number of segments, the similarity value has a completely different value when one segment is divided into a plurality of segments.

For example, in FIG. 1, DSN (T1, T2) = 0.56 of two trajectories T1 and T2. If segment s6 of trajectory T2 is divided into three segments, DSN (T1, T2) = 0.64.

Therefore, the similarity of two trajectories cannot be determined using only the number of segments. In order to solve this problem, the present invention proposes a similarity measurement function DSL (dissimilarity with length) of Equation (2) using the length of a segment.

(2)

(2)

A similarity value obtained by the DSL function with respect to the trajectory of FIG. 1 is obtained as follows.

DSL (T1, T2) = 0.56, DSL (T1, T3) = 0.4, DSL (T1, T4) = 0.65, DSL (T2, T3) = 0.59, DSL (T2, T4) = 0.68, DSL (T3, T4) = 0.7

The similarity values of DSN (T1, T2) and DSN (T1, T4) by DSN function are the same, while the similarity values of DSL (T1, T2) and DSL (T1, T4) by DSL function are different. have. That is, it can be seen that the trajectory T2 is more similar to the trajectory T1 than the trajectory T4 by the DSL function. The proposed DSL function judges that the trajectories having a large number of common segments and a small difference in the length of the segments are more similar between the two trajectories.

In addition, the proposed scheme is not affected by the change in the number of segments according to the criteria for dividing the segment. For example, in the case of measuring the similarity of the trajectories T1 and T2, when dividing the segment of s6 into three, the total length of the three segments divided into four is unchanged before and after dividing. Therefore, the value of DSL (T1, T2) before and after the division is the same as 0.56.

By using the similarity measurement function proposed in the present invention, it is possible to cluster similar trajectories of a moving object.

Similar trajectories can also be clustered to provide users with more useful information.

Trajectory clustering refers to grouping using the similarity between trajectories. In a general clustering method, clusters are formed by repetitive change of a center point, which is a center of gravity of a cluster.

3 is a flowchart of a process for clustering similar trajectories of moving objects in a road network space according to the present invention.

Referring to FIG. 3, first, new trajectory data of a moving object on a road network space is stored in a database (S31).

It is determined whether there is a matching similarity measure function value for the trajectory data stored in the database (S32).

Next, as a result of the determination in step S32, if there is no similarity measurement function value matched with the trajectory data in the database, a new similarity is measured with the similarity measurement function with respect to the trajectory data, and the similarity with respect to the trajectory data is measured. The measurement function value is stored in the database (S33).

Among the similarity measurement function values stored in the database, two random trajectories having the highest value are retrieved (S34).

The trajectories are mapped in k-dimensions based on the two trajectories found in step S34 (S35).

Next, clustering is performed on the result mapped in step S35 (S36).

In addition, the cluster configured in the step S36 is stored in the database (S37).

If the cluster is composed of the relative distance between the trajectories, the criterion of the center point becomes ambiguous. In order to solve this problem, the present invention expresses each trajectory as a point in k-dimensional space using FastMap, and then clusters the points corresponding to the entire trajectories.

In this case, in order to map trajectories having different lengths into one dimension, the present invention uses a similarity value between two trajectories measured by the proposed similarity measurement function DSL. Here, the measured value of DSL means the distance between two trajectories.

FastMap is a method of mapping these objects to k-dimensional points given n objects and the distance-to-object distance function.

Using FastMap, the trajectory can be easily mapped to a point in k-dimensional space by using the measured distance between two trajectories.

How to convert trajectory to k-dimensional point using FastMap is as follows.

First, find the two trajectories with the largest value among the DSL values. Referred to these two trajectories T _maxa, T _maxb.

점으로 매핑한다.Second, the trajectory are T _'i, which is based on the T _{maxa, maxb} T using the equation

Map to a point.

Third, the new distance difference between the T ' _i points mapped by the second process is recalculated by the following equation.

Finally, we repeat the process and expand the dimensions to map.

After converting the trajectory to a k-dimensional point using FastMap, the present invention performs clustering on the k-dimensional points transformed using the k-medoids method. The k-medoids method finds k clusters among n objects by randomly finding representative objects in each cluster.

The representative object of the cluster is an object whose average (or total) distance from other objects among the objects belonging to the cluster is minimum. The k-medoids clustering method divides objects into a given number of clusters, which minimizes the sum of the distances between each object in the cluster and the representative object. This method has the advantage of being less sensitive to noise than the k-means method.

4 is an exemplary diagram of clustering similar trajectories of a moving object on a road network according to the present invention.

4 shows an example of mapping the trajectory of FIG. 1 to a two-dimensional point using FastMap, and then making it into three clusters by k-medoids.

In the clustered results, it can be seen that T1 and T3 having the highest similarity are grouped in the same cluster.

5 illustrates a summary process of cluster information in a clustering process of a similar trajectory of a moving object according to the present invention.

Referring to FIG. 5, when there is a configured cluster, it is first determined whether information of the cluster is summarized (S51).

If it is determined in step S51 that the information of the cluster is not summarized, the cluster information is summarized by a frequency calculation formula (S52).

Next, the cluster summary information is stored in a database (S53).

If cluster information is already summarized in step S51, it is stored in the database as cluster information (S54).

In the present invention, the information on the cluster obtained through the clustering process is provided to each cluster C with segment summary information S = { <(SegList ₁ ), Weight ₁ >, ..., <(SegList _n ), Weight _n > } and the user. Summary Information U = { User ₁ , .., User _m } is saved and managed together.

Here, SegList _j means a segment list of each trajectory included in the corresponding cluster, and Weight _j means the frequency of occurrence of the corresponding segment list in the cluster and is obtained by Equation (3).

(3)

(3)

For example, if C1 = {T1, T3} obtained by performing a clustering process on each trajectory of FIG. 1 and each trajectory user is user1 and user2, segment summary information S = {<(s1, s3, s4), 1>, <(s2, s8, s9, s10), 0.5>}, and user summary information U = {user1, user2}.

Each process of clustering using the above similarity function is performed as in the processes of FIGS. 3 and 5.

We propose an efficient query processing technique using segment summary information and user summary information of a cluster configured as described above.

6 is a flowchart of searching for a trajectory similar to a trajectory of a query of a moving object in a road network space according to the present invention.

Referring to FIG. 6, a user's query trajectory for a moving object in a road network space enters a database (S61).

Next, a segment constituting the query trace introduced to the database is extracted (S62).

It is determined whether there is a match between the extracted segment of the query trajectory and the segment of the trajectory cluster summary information stored in the database (S63).

In this case, when there is a matching cluster in step S63, a sum of segment weights of the matching clusters is obtained (S64).

Next, the cluster having the largest sum of the weights is searched, and the searched cluster is determined as the cluster to which the query trajectory belongs (S65 and S66).

If there is no matching cluster in step S63, since the query trajectory does not match the cluster stored in the database, it is determined that no cluster belongs to the query trajectory (S67).

For example, if the query trajectory Q = { s1, s2, s10, s9, s8, s5 } is given to the database for the segment summary information of the cluster constructed from FIG. 1, the segment summary information of the cluster for each segment constituting Q. Search for.

The sum of the weight values of the corresponding segments in the segment summary information of each cluster compared with each segment is obtained, and the cluster having the highest weight value is searched.

In the above example, since the segments s1, s2, s10, s9, and s8 in Q exist in the cluster C1, the weights of s1, s2, s10, s9, and s8 at this time add up to 1 + 0.5 + 0.5 + 0.5 + 0.5 = 3. . Since the segments s1 and s5 in the next Q exist in the cluster C2, the sum of the weights is 1 + 1 = 2, and since the segment s1 exists in the cluster C3, the sum of the weights is 1. The cluster mapped to the query trajectory Q becomes the cluster C1 having the largest weight value.

Therefore, a cluster matching the query trajectory can be searched without the additional overhead of reconfiguring the cluster for a given query trajectory.

This query processing is performed in the same manner as the process of FIG. 6 described above, and then presents the user information and the road information associated with the searched cluster to the user.

For example, suppose cluster C1 has users A, B, and C who have moved in similar trajectories in Seoul, and cluster C2 has users A and B who have moved in similar trajectories in Busan. Visits the Busan area, the user A and B belonging to the cluster C1 may recommend to the user C the route traveled from the Busan area.

According to the method of clustering similar trajectories in the road network according to the present invention, the similar trajectory search and clustering for moving objects in the road network can be performed by applying the similarity measuring method considering the road network space.

In addition, it is possible to efficiently search for a cluster matching the user's query trajectory by using segment summary information on the configured cluster.

Although the present invention has been described in detail only with respect to the specific embodiments described, it will be apparent to those skilled in the art that various modifications and variations are possible within the technical scope of the present invention, and such modifications and modifications belong to the appended claims. .

Claims (10)

(a) storing the trajectory of the moving object on the road network space as a trajectory data including a segment identifier, a length of the segment, user information, and road information in a database; (b) determining whether there is a measured value calculated by a similarity measurement function for the stored specific trajectory and the trajectory already stored in the database; (c) If there is no measured value calculated by the similarity measurement function for the stored specific trajectory as a result of the step (b), the stored specific trajectory and the trajectories already stored in the database by the similarity measurement function. Calculating a similarity measure function; (d) storing the function value measured in step (c) in the database as a similarity value between the specific trajectory and the trajectory already stored in the database; And (e) If the determination result of step (b) indicates that there is a value already calculated by the similarity measurement function for the stored specific trajectory, the similarity measurement function already calculated for the stored specific trajectory and the trajectory already stored in the database. And storing the value as a similarity value. The method of claim 1, The similarity measurement function is assuming that any two trajectories, respectively, T _i , T _j , the following equation DSN (T _i , T _j )

A method for measuring the similarity of a moving object trajectory in a road network space, characterized in that. The method of claim 1, The similarity measurement function is assuming that two arbitrary trajectories T _i and T _j , respectively, the following equation DSL (T _i , T _j )

A method for measuring the similarity of a moving object trajectory in a road network space, characterized in that. (a) storing the trajectory of the moving object on the road network space as a trajectory data including a segment identifier, a length of the segment, user information, and road information in a database; (b) determining whether there is a similarity measure function value in the database that matches the trajectory data; (c) If the similarity measurement function value matched for the trajectory data is not found in the database as a result of the determination of step (b), a new similarity is measured for the trajectory data by a similarity measurement function, and the trajectory data is measured. Storing in the database as a similarity measure function for; (d) retrieving any two trajectories having the largest value of the similarity measurement function value among all similarity measurement function values stored in the database; (e) mapping the trajectories in k dimensions based on the two trajectories retrieved in step (d); (f) performing clustering on the result mapped in step (e); And and (g) storing the cluster configured in the step (f) in a database. The method of claim 4, wherein The similarity measurement function, When two arbitrary trajectories are called T _i and T _j , respectively, the following equation DSL (T _i , T _j )

A method for clustering similar trajectories of moving objects in a road network space. The method of claim 4, wherein (i) determining whether information of the cluster is summarized; (j) if it is determined in step (i) that the information of the cluster is not summarized, summarizing the cluster information by a frequency calculation formula; And (k) storing the cluster summary information in a database, the method of clustering similar trajectories of moving objects in a road network space. The method of claim 6, The cluster summary information is segment segmentation information for each cluster, and includes a segment list of each trajectory included in the cluster, and a weight of a segment indicating a frequency of occurrence of the segment list in each cluster. Similar trajectory clustering method for moving objects. The method of claim 6, The frequency calculation formula

A method for clustering similar trajectories of moving objects in a road network space. The method of claim 6, (m-1) inputting a query trajectory into the database; (m-2) extracting segments constituting the query trajectory; (m-3) determining whether the segment of the query trajectory and the segment of the trajectory cluster summary information stored in the database match each other; (m-4) obtaining a sum of segment weights of the matching clusters when the matching clusters are present in the step (m-3); (m-5) searching for a cluster having the largest sum of the weights; And (m-6) further comprising determining the searched cluster as a cluster to which the query trajectories belong. The method of claim 9, And presenting to the user the cluster found to be similar to the query trajectory, the user information and the road information associated with the cluster to the user.

Images