JP2015158799A - Space-time clustering processing device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform space-time clustering processing with high accuracy and at high speed.SOLUTION: When extracting a stay point from a positional information sequence while using Mean-Shift clustering, a data point that can be regarded as an error is deleted from the positional information sequence. Thereafter, in a stay point extraction section 12, data points in the positional information sequence are defined as clusters, respectively, discovery processing of a cluster included in a spatial kernel width is successively performed for each cluster regardless of a time difference between clusters and in accordance with a time sequence. At a time point when a cluster that is not included in the kernel width is discovered, subsequence discovery processing is stopped.

Description

  The present invention relates to a spatio-temporal clustering processing apparatus, method, and program for extracting and combining clusters using spatio-temporal mean-shift clustering, for example, in order to extract a stay point from a position information series.

  Analysis of user's interest / preference, behavior history, etc. is important in order to realize recommendations and behavior support technology tailored to users, which have been attracting attention in recent years. To analyze this behavior history and interest / preference It is considered important to acquire the user's place of stay.

  As a method for extracting a place of stay, a method has been proposed in which a position information series of a user is acquired using GPS or a mobile communication system, and a point where the user stays is extracted from the acquired position information series. In order to extract a stay point from a position information series, there is a method of mapping not only the latitude and longitude included in the position information series but also the time axis, and performing mean-shift clustering in time and space (for example, Non-Patent Document 1). reference).

  In addition, a method of performing processing sequentially considering only temporal continuity has been proposed (see, for example, Non-Patent Document 2).

Kyosuke Nishida, Hiroyuki Toda, Ken Kurashima, Atsushi Uchiyama, “Probabilistic Visit POI Analysis: User Modeling from Spatiotemporal Behavior Trajectory”, Multimedia, Distributed, Collaboration and Mobile Symposium (DICOMO2013), July 2013. Kang, Jong Hee, et al. "Extracting places from traces of locations." Proceedings of the 2nd ACM international workshop on Wireless mobile applications and services on WLAN hotspots. ACM, 2004.

  However, in the method described in Non-Patent Document 1, the processing time increases as the number of positioning points increases. In addition, because it is determined whether to belong to the cluster using the spatio-temporal distance between the clusters, the location extraction accuracy when the location information data is missing for a long time such as entering the room is very high There is a problem that it is low.

  Further, in the method described in Non-Patent Document 2, the stay location is determined from the temporally previous cluster in consideration of only the time series. For this reason, it is possible to perform the stay place determination sequentially at a high speed, but there is a problem that an error in the position of the extracted stay place becomes very large.

  The present invention has been made paying attention to the above circumstances, and an object of the present invention is to provide a spatio-temporal clustering processing apparatus, method, and program capable of performing spatio-temporal clustering processing with high accuracy and high speed. There is.

  In order to achieve the above object, a first aspect of the present invention includes means for storing an information sequence in which a plurality of clusters are arranged in time series, cluster discovery processing means, mean shift means, and cluster combining means. Prepare. Then, for each cluster of the stored information series, a process for discovering other clusters included in the preset spatial kernel width of the cluster of interest by the cluster discovery processing means is included in the kernel width. By sequentially performing according to the above time series until another non-existing cluster is found, a cluster group in which the spatial distance between the clusters is within a preset distance and is temporally continuous is found. Then, mean-shift clustering is used to mean shift the center position and time for the discovered cluster group, and a plurality of clusters having the same center position among the cluster groups after the mean shift process are combined. It is a thing.

  According to a second aspect of the present invention, in the cluster discovery processing means, a first parameter representing a spatial distance tolerance is set, and other clusters not included in the kernel width of the cluster of interest are discovered. And when another cluster whose time position is farther from the cluster of interest than the cluster is rediscovered, the time position of the rediscovered cluster is within the range of the first parameter. In this case, the process of including the rediscovered cluster within the kernel width of the cluster of interest when the value is within the range of the first parameter is further performed.

  According to a third aspect of the present invention, in the cluster discovery processing means, a second parameter representing an allowable amount of time length is set, and other clusters not included in the kernel width of the cluster of interest are discovered. And when another cluster whose arrangement position in time is far from the cluster of interest is rediscovered, the time length of the rediscovered cluster from the cluster of interest is within the range of the second parameter. If it is within the range of the second parameter, the rediscovered cluster is included in the kernel width of the cluster of interest.

  According to the first aspect of the present invention, since the subsequent discovery process is terminated in response to the discovery of a cluster not included in the kernel width, the amount of calculation required for the discovery process of the cluster included in the kernel width As a result, the time required for the stay location extraction process can be shortened. In addition, since the process of determining whether or not each cluster is included in the spatial kernel width is sequentially performed according to the time series regardless of the time difference between the clusters, even when information cannot be acquired at a certain time interval, Clusters can be extracted with high accuracy.

  According to the second aspect of the present invention, the temporal discontinuity of the cluster is allowed within the range of the first spatial parameter. For this reason, for example, it is possible to reduce the influence of an error in the position information of each cluster and perform a highly stable stay point extraction process.

  According to the third aspect of the present invention, the temporal discontinuity between clusters is allowed if it is within the range of the allowable condition considering the second parameter in time. For this reason, for example, it is possible to reduce the influence of an error in position information of each cluster and an error in measurement time, and to perform a more stable stay point extraction process.

  That is, according to each aspect of the present invention, it is possible to provide a spatiotemporal clustering processing apparatus, method, and program capable of performing spatiotemporal clustering with high accuracy and high speed.

The block diagram which shows the function structure of the stay point extraction apparatus provided with the spatiotemporal clustering processing apparatus which concerns on one Embodiment of this invention. The flowchart which shows the procedure and process content of the stay point extraction process by the stay point extraction apparatus provided with the spatiotemporal clustering processing apparatus which concerns on one Embodiment of this invention. The flowchart which shows the 1st Example of the cluster discovery process in a kernel in the stay point extraction process shown in FIG. The flowchart which shows the 2nd Example of the cluster discovery process in a kernel in the stay point extraction process shown in FIG. The flowchart which shows the 3rd Example of the cluster discovery process in a kernel in the stay point extraction process shown in FIG. The figure for demonstrating the difference with the stay point extraction method in one Embodiment of this invention, and the existing method. The figure for demonstrating the difference of the effect of the stay point extraction method in one Embodiment of this invention, and the existing method.

Embodiments according to the present invention will be described below with reference to the drawings.
[One Embodiment]
(Constitution)
FIG. 1 is a block diagram showing a functional configuration of a stay point extraction apparatus including a spatiotemporal clustering processing apparatus according to an embodiment of the present invention. This stay point extraction device is composed of, for example, a server computer or a personal computer, and includes at least a control unit including a CPU (Central Processing Unit) and a storage unit.

  The storage unit includes a GPS series storage unit 21, a spot name database (point name DB) 22, and a stay spot name series storage unit 23. The GPS series storage unit 21 is used, for example, to store the position information series when the position information series measured by the GPS positioning unit is transmitted from the portable terminal in the user's portable terminal (not shown). The point name database (point name DB) 22 stores information representing the point name in association with the latitude and longitude information of the service target area in advance. Note that the information indicating the location name includes a building name, a facility name, a store name, and the like. The staying point name series storage unit 23 is used for storing information representing a staying point name extracted by the control unit 1 described later.

  The control unit 1 includes an accuracy improvement unit 11, a stay point extraction unit 12, and a point name reference unit 13. Each of these processing units is realized by causing the CPU to execute a program stored in a program memory in the storage unit.

  The accuracy improving unit 11 performs a process of deleting data points that can be clearly regarded as errors among the data points in the position information series. Data points that can be regarded as errors include, for example, data points that would be over 2000 km / h when the moving speed is calculated, that is, data points that are actually not moving at that speed. Is mentioned.

The stay point extraction unit 12 performs the following processing.
(1) The position information series after the error is deleted is received from the accuracy improving unit 11, and each data point of the position information series is set as a cluster. Then, for each cluster, processing for extracting clusters that are included in the spatial kernel width and that are temporally continuous.
(2) A process of mean-shifting the center position and time for each extracted cluster using Mean-Shift clustering.
(3) Processing for combining clusters having the same center position among the clusters after the mean shift processing.
(4) Processing for deleting a cluster having a minimum stay time or less from the clusters after the joining processing.

  The spot name reference unit 13 accesses the spot name DB 22 based on the latitude / longitude information of the clusters obtained by the stay spot extracting unit 12 and reads information representing the corresponding spot name from the spot name DB 22. And the process which stores the information showing this read point name in the stay point name series memory | storage part 23 is performed.

(Operation)
Next, the stay point extraction processing operation of the apparatus configured as described above will be described. In the following description, it is assumed that the position information series acquired from a user's portable terminal (not shown) is already stored in the GPS series storage unit 21. FIG. 1 illustrates an example of the GPS series.

  First, the control unit 1 inputs the position information sequence read from the GPS sequence storage unit 21 to the accuracy improving unit 11, and extracts and deletes data points that can be clearly regarded as errors for each data point of the position information sequence. . This process is performed, for example, by calculating a moving speed based on the measurement position and the time between the previous data point and comparing the calculated moving speed with a threshold value. As a result, for example, “2013-12-03 00:00:34, 11.76, 149.52” is deleted as error data in the GPS series shown in FIG.

Next, the control unit 1 performs the stay point extraction process by the stay point extraction unit 12 as follows. FIG. 2 is a flowchart showing the processing procedure and processing contents.
That is, first, in step S11, each data point of the position information series from which the error has been deleted by the accuracy improving unit 11 is initialized as a cluster, and until it is determined in step S12 that the mean shift by Mean Shift clustering has converged. The processes in steps S13 to S15 are repeatedly executed.

  First of all, in step S13, a process for finding a cluster in the spatiotemporal kernel is performed for each cluster. The following three embodiments can be considered as the discovery processing method for clusters in the space-time kernel.

(1) First Example In the first example, when performing mean-shift clustering, a spatial kernel is used, and only temporally continuous clusters are determined to be in the same kernel. This is a technique for extracting stay points. For example, when extracting a cluster included in the kernel width of the cluster cls_i, it is determined whether or not the clusters i + 1, i + 2,... Following the cluster cls_i are sequentially included in the kernel width of the cluster cls_i. As a result of this determination, even if the cluster cls_i + 2 is included in the kernel width, if the previous cluster cls_i + 1 is not included in the kernel width, the cluster cls_i + 2 is also included in the kernel width. Is not included. Similarly, it is determined whether the clusters i-1, i-2,... Following the cluster cls_i are sequentially included in the kernel width of the cluster cls_i, and the cluster cls_i-2 is included in the kernel width. However, if the next cls_i-1 is not included in the kernel width, the cluster cls_i-2 is not included in the kernel width.

FIG. 3 is a flowchart showing the processing procedure and processing contents of the in-kernel cluster finding method according to the first embodiment.
The clusters are arranged in the order of measurement time according to the time series, and the i-th cluster is set as cls_i, and the cluster array is set as {cls}. First, in step S21, one cluster cls_i is selected, and in this state, each cluster cls_j: {cls | i <j} whose order of measurement time is later than the cluster cls_i in step S22 is i + 1, i + 2,. Is selected. In step S23, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i + 1, i + 2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i + 1, i + 2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S24. The

  Then, in the above determination process, it is determined that the cluster cls_j (j = i + k1) is not within the kernel width of the cluster cls_i. In this case, it is considered that all the clusters after the order i + k1 are not included in the kernel width of the cluster cls_i. At this time, the discovery process of the in-kernel clusters in the advance direction of the order on the time axis is finished, Control goes to step S26.

  In step S26, the clusters cls_j: {cls | i> j} whose measurement time is earlier than the cluster cls_i are sequentially selected as i-1, i-2,. In step S27, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i-1, i-2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i−1, i−2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S28. The

  Then, in the above determination process, it is determined that the cluster cls_j (j = i-k2) is not within the kernel width of the cluster cls_i. In this case, it is considered that all the clusters whose order is i-k2 or later are not included in the kernel width of the cluster cls_i, and at this point, the discovery process of the in-kernel clusters in the return direction of the order on the time axis ends.

  When the in-kernel cluster discovery process for the cluster cls_i is completed, the process returns from step S30 to step S21 to select the next cluster cls_i + 1. This time, the space between this cluster cls_i + 1 and another cluster cls_j is selected. In-kernel cluster discovery processing based on the distance is performed in steps S22 to S29. Thereafter, similarly, every time the next cluster cls_i is sequentially selected, the in-kernel cluster discovery process based on the spatial distance between the selected cluster cls_i and another cluster cls_j is repeatedly performed, and all the cluster cls_i When the in-kernel cluster discovery process for is completed, the control unit 1 proceeds to step S14.

(2) Second Embodiment The second embodiment is an extension of the first embodiment, and allows temporal discontinuity as long as it is within a preset spatial parameter range. It is a technique. In other words, when the parameter is n, when it is sequentially determined whether or not each cluster cls_j is included in the spatial kernel width of the cluster cls_i, it will not be included in the kernel width after a certain cluster. However, if the cluster ranks n1 more after the cluster and enters the kernel width again, the cluster after n1 is a cluster only if n1 is within the range of parameter n (n1 ≤ n). It is determined to be included in the kernel width of cls_i.

  FIG. 4 is a flowchart showing the processing procedure and processing contents of the in-kernel cluster finding method according to the second embodiment. In the figure, the same parts as those in FIG.

  The clusters are arranged in the order of measurement time according to the time series, and the i-th cluster is set as cls_i, and the cluster array is set as {cls}. First, in step S21, one cluster cls_i is selected, and in this state, each cluster cls_j: {cls | i <j} whose order of measurement time is later than the cluster cls_i in step S22 is i + 1, i + 2,. Is selected. In step S23, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i + 1, i + 2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i + 1, i + 2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S24. The

  On the other hand, assume that it is determined that the cluster cls_j (j = i + k1) does not fall within the kernel width of the cluster cls_i in the above determination process. In this case, the determination is further continued for each of the k1 + 1th and subsequent clusters. When the i + k1 + n1th cluster enters the kernel width of the cluster cls_i again, the rank i + k1 + n1 of the cluster is within the preset parameter n range (n1 ≦ n). Is determined in step S31, and if it is within the range of the parameter n, it is determined that the i + k1 + n1-th cluster is included in the kernel width of the cluster cls_i. Saved in memory.

  On the other hand, if the cluster order i + k1 + n1 is not within the range of parameter n (n1 ≦ n), all clusters after the order i + k1 are within the kernel width of the cluster cls_i. At this time, the discovery process of the cluster in the kernel in the order of advance in the time axis is terminated, and the process proceeds to step S26.

  In step S26, the clusters cls_j: {cls | i> j} whose measurement time is earlier than the cluster cls_i are sequentially selected as i-1, i-2,. In step S27, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i-1, i-2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i−1, i−2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S28. The

  On the other hand, in the above determination process, it is determined that the cluster cls_j (j = i-k2) is not within the kernel width of the cluster cls_i. In this case, the determination is continued for each of the k2-1 and subsequent clusters. Then, when the i-k2-n2th cluster enters the kernel width of the cluster cls_i again, whether or not the rank i-k2-n2 of the cluster is within the range of the parameter n (n2 ≦ n) Is determined in step S32 and if it is within the range of the parameter n, it is determined that the i-k2-n2 th cluster is included in the kernel width of the cluster cls_i, and is stored in the memory in the control unit 1 in step S29. Is done.

  On the other hand, if the cluster order i-k2-n2 is not within the range of parameter n (n1 ≤ n), all clusters in the order prior to i-k2 are within the kernel width of cluster cls_i. At this time, the discovery process of the cluster in the kernel in the return direction in the order on the time axis is finished.

  When the in-kernel cluster discovery process for the cluster cls_i is completed, the process returns from step S30 to step S21 to select the next cluster cls_i + 1. This time, the space between this cluster cls_i + 1 and another cluster cls_j is selected. In-kernel cluster discovery processing based on the distance is performed in steps S22 to S29 (including steps S31 and S32). Thereafter, similarly, every time the next cluster cls_i is sequentially selected, the in-kernel cluster discovery process based on the spatial distance between the selected cluster cls_i and another cluster cls_j is repeatedly performed, and all the cluster cls_i When the intra-kernel cluster discovery process for is completed, the control unit 1 proceeds to step S14 described above.

(3) Third Embodiment The third embodiment is a further expansion of the second embodiment, and the temporal discontinuity is determined by the range of the spatial parameters described in the second embodiment. If it is within the range of parameters within a preset time, the method is allowed. In other words, in addition to the spatial parameter n, a permissible condition that further considers a temporal parameter is further defined. When it is sequentially determined whether or not each cluster cls_j is included in the spatial kernel width of the cluster cls_i, even if the cluster cls_i is not included in the kernel width after a certain cluster, the ranking is higher than that cluster. If the rank n1 of the cluster falls within the range of the parameter n (n1 ≦ n) and the above-mentioned allowable condition is satisfied Is determined to be included in the kernel width of the cluster cls_i.

  FIG. 5 is a flowchart showing the processing procedure and processing contents of the in-kernel cluster finding method according to the third embodiment. In the figure, the same parts as those in FIG.

  The clusters are arranged in the order of measurement time according to the time series, and the i-th cluster is set as cls_i, and the cluster array is set as {cls}. First, in step S21, one cluster cls_i is selected, and in this state, each cluster cls_j: {cls | i <j} whose order of measurement time is later than the cluster cls_i in step S22 is i + 1, i + 2,. Is selected. In step S23, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i + 1, i + 2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i + 1, i + 2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S24. The

On the other hand, assume that it is determined that the cluster cls_j (j = i + k1) does not fall within the kernel width of the cluster cls_i in the above determination process. In this case, the determination is further continued for each of the k1 + 1th and subsequent clusters. When the i + k1 + n1-th cluster enters the kernel width of the cluster cls_i again, first in step S41, the rank i + k1 + n1 of the cluster is within the range of the preset spatial parameter n. It is determined whether (n1 ≦ n). Next, the following permissible conditions considering the preset time parameter t
n1
Σ | cls_i + k1 + x | <t
x = 0
Is satisfied, and if the permissible condition is satisfied, it is determined that the i + k1 + n1-th cluster is included in the kernel width of the cluster cls_i. Saved in memory.

  On the other hand, if the rank i + k1 + n1 of the cluster is not included in the range of the parameter n (n1 ≦ n) or if it is included, the order is All the clusters after i + k1 are considered not to be within the kernel width of the cluster cls_i. At this time, the discovery process of the in-kernel clusters in the forward progression direction on the time axis ends, and the process proceeds to step S26. .

  Next, in step S26, each cluster cls_j: {cls | i> j} whose measurement time is earlier than the cluster cls_i is sequentially selected as i-1, i-2,. In step S27, a spatial distance from the cluster cls_i is obtained for each of the selected clusters cls_j (j = i-1, i-2,...), And this spatial distance falls within the kernel width of the cluster cls_i. It is determined whether or not there is. As a result of this determination, the cluster cls_j (j = i−1, i−2,...) Determined that the spatial distance is within the kernel width of the cluster cls_i is stored in the memory in the control unit 1 in step S28. The

On the other hand, in the above determination process, it is determined that the cluster cls_j (j = i-k2) is not within the kernel width of the cluster cls_i. In this case, the determination is continued for each of the k2-1 and subsequent clusters. When the i-k2-n2th cluster enters the kernel width of the cluster cls_i again, first, in step S41, the rank i-k2-n2 of the cluster is within the range of the preset spatial parameter n. It is determined whether (n1 ≦ n). Next, the following permissible conditions considering the preset time parameter t
n2
Σ | cls_i-k2-x | <t
x = 0
Is satisfied, and if the allowable condition is satisfied, it is determined that the i-k2-n2 th cluster is included in the kernel width of the cluster cls_i. Saved in memory.

  On the other hand, if the cluster order i-k2-n2 is not included in the range of the parameter n (n2 ≦ n), or if it is included, the order is All clusters prior to i-k2 are considered not to fall within the kernel width of the cluster cls_i, and at this point, the discovery process of the in-kernel clusters in the return direction in the order on the time axis ends.

  When the in-kernel cluster discovery process for the cluster cls_i is completed, the process returns from step S30 to step S21 to select the next cluster cls_i + 1. This time, the space between this cluster cls_i + 1 and another cluster cls_j is selected. In-kernel cluster discovery processing based on the distance is performed in steps S22 to S29 (including steps S31 and S32).

  Thereafter, similarly, every time the next cluster cls_i is sequentially selected, the in-kernel cluster discovery process based on the spatial distance between the selected cluster cls_i and another cluster cls_j is repeatedly performed, and all the cluster cls_i When the intra-kernel cluster discovery process for is completed, the control unit 1 proceeds to step S14 described above.

  In step S14, the stay point extraction unit 12 of the control unit 1 performs a mean shift (moves to a weighted average) the center position and time for each cluster. This mean shift is described in detail in Non-Patent Document 1 described above. Next, in step S15, a process of combining clusters having the same center position is performed for each cluster after the mean shift.

  If it is determined in step S12 that the mean shift for each cluster has converged, in step S16, a process of deleting a cluster that does not satisfy the minimum stay time from the cluster after the combination process is performed. Then, the cluster group remaining after the deletion process is output to the point name reference unit 13 as a stay point series.

  In the point name reference unit 13, the point name DB 22 is accessed based on the latitude / longitude information of the clusters obtained by the stay point extraction unit 12. And the information showing the corresponding point name is searched from the said point name DB22. Then, the information indicating the searched spot name is stored in the stay spot name series storage unit 23.

  The stay point name series information stored in the stay point name series storage unit 23 is read in response to a request from an operator terminal used by an operator such as a network service provider or an information distribution operator, Sent to the requesting operator terminal. This staying point name series information is used, for example, for a recommendation service or action support for the user.

(Effect of the first embodiment)
As described above in detail, in the first example of the spatio-temporal clustering method according to one embodiment of the present invention, when a stay point is extracted from a position information sequence using Mean-Shift clustering, After deleting the data points that can be regarded as errors, the stay point extraction unit 12 sets each data point of the position information series as a cluster, and performs a cluster discovery process included in the spatial kernel width for each cluster. The processing is performed sequentially according to the time series regardless of the time difference between them, and the subsequent discovery process is terminated when a cluster not included in the kernel width is found.

  FIG. 6 shows the result of the cluster discovery processing according to the first embodiment compared with the processing result by the existing method. (A) shows the result by the existing method, and (b) shows the result by the first embodiment. Respectively. As shown in the figure, in the existing method, since it is always determined whether or not all clusters are included in the kernel width, as shown in FIG. If they return to within the radius within a certain time, they are combined as the same cluster.

  On the other hand, in the first embodiment, as described above, the cluster discovery process is sequentially performed according to the time series regardless of the time difference between the clusters, and after the cluster that is not included in the kernel width is discovered. The discovery process is terminated. For this reason, as shown in (b), once a certain cluster goes out of the cluster radius, even if the cluster subsequently enters the radius, the cluster is not considered to represent the same point.

  Therefore, according to the first embodiment, it is possible to reduce the amount of calculation required for the cluster discovery process included in the kernel width, thereby shortening the time required for the stay location extraction process. .

  In addition, because the process of determining whether or not each cluster is included in the spatial kernel width is performed sequentially according to the time series regardless of the time difference between the clusters, indoor environments where position information cannot be acquired at regular time intervals Also, the place of stay can be extracted with high accuracy. That is, even when the temporal distance between the positional information i and the positional information i + 1 is very large but the distance is very close, the positional information i and the positional information i + 1 are included in the same cluster. It becomes possible.

  FIGS. 7A and 7B are diagrams for explaining this effect. FIG. 7A shows a cluster discovery result by the existing method, and FIG. 7B shows a discovery result by the first embodiment. As shown in FIG. 5A, in the existing method, when the time difference between clusters exceeds a certain value, these clusters are not extracted as information representing the same place of stay. On the other hand, in the first embodiment, as shown in (b), even if the time difference between the clusters is large, these clusters are extracted as information representing the same place of stay if these clusters are continuous. . Therefore, for example, even when a user stays in an underground mall or office building for a long time, the two clusters can be combined as indicating the same place of stay and the place of stay can be reliably extracted.

  In the second embodiment, when it is sequentially determined whether or not each cluster cls_j is included in the spatial kernel width of the cluster cls_i during the cluster discovery process, Even if it is not included in the kernel width in (1), if the rank of the n1th cluster after the cluster is within the kernel width again, the rank n1 of the cluster is within the preset parameter n ( If n1 ≦ n), it is determined that the n1-th cluster is included in the kernel width of the cluster cls_i.

  Therefore, the temporal discontinuity of the cluster can be allowed within the range of the spatial parameter n. For this reason, for example, it is possible to reduce the influence of an error in the position information of each cluster and perform a highly stable stay point extraction process.

  Further, in the third embodiment, an allowable condition that further considers a temporal parameter in addition to the spatial parameter n is further defined. When it is sequentially determined whether or not each cluster cls_j is included in the spatial kernel width of the cluster cls_i, even if the cluster cls_i is not included in the kernel width after a certain cluster, the ranking is higher than that cluster. Is again within the kernel width in the n1th cluster later, the rank n1 of the cluster is within the range of parameter n (n1 ≤ n), and is preset in consideration of the temporal parameter t When the allowable condition is satisfied, it is determined that the n1-th cluster is included in the kernel width of the cluster cls_i.

  Therefore, the temporal discontinuity between clusters is allowed if it is within the range of the spatial parameter n described in the second embodiment and within the allowable condition in consideration of the time parameter. be able to. For this reason, for example, it is possible to reduce the influence of an error in position information of each cluster and an error in measurement time, and to perform a more stable stay point extraction process.

[Other Embodiments]
In the third example of the above embodiment, even if a cluster is out of the kernel width, the cluster is within the range of the spatial parameter n and within the allowable condition range considering the time parameter. For example, a time discontinuity between clusters is allowed. However, the temporal discontinuity between clusters may be allowed without considering the spatial parameter n and using only the temporal parameter as an allowable condition.

  In the embodiment, the GPS is used as the position measurement unit. However, the position information provided by a base station of a mobile communication network system such as a cellular mobile communication system or a wireless LAN, A position measuring means using an acceleration sensor may be used.

  Further, in the embodiment, when a location information series measured by a user's mobile terminal or the like is collected by a server computer, and a spatiotemporal clustering process for extracting a stay point from the location information series is executed in the server computer Was described as an example. However, the present invention is not limited thereto, and the spatio-temporal clustering process may be directly performed on the location information sequence obtained by the mobile terminal in the user's mobile terminal, and the location information sequence obtained by the user's mobile terminal may be It may be transferred to a fixed terminal and the spatio-temporal clustering process may be executed on the transferred position information series in the fixed terminal.

  Furthermore, in the embodiment, the case where the staying point is extracted from the position information series has been described as an example. However, the present invention is not limited to this, for example, when stagnation of movement is extracted from information representing the movement history of the monitoring target detected by the acceleration sensor, or when no change is extracted from the monitoring video of the target obtained by the camera Also, the present invention can be applied.

  The apparatus of the present invention can also be realized by a computer and a program, and the program can be recorded on a recording medium or provided through a network. In addition, the processing procedure and processing contents for extracting clusters included in the kernel can be variously modified and implemented without departing from the gist of the present invention.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  DESCRIPTION OF SYMBOLS 1 ... Control unit, 11 ... Accuracy improvement part, 12 ... Stay spot extraction part, 13 ... Spot name reference part, 21 ... GPS series memory | storage part, 22 ... Spot name DB, 23 ... Stay spot name series memory part.

Claims (7)

Means for storing an information sequence in which a plurality of clusters are arranged in time series;
For each cluster of the stored information series, another cluster that is not included in the kernel width is found in the process of finding another cluster included in the preset spatial kernel width of the cluster of interest. A cluster discovery processing means for discovering a cluster group that is temporally continuous within a preset distance of the spatial distance between the clusters, by sequentially performing according to the time series until
Means for shifting the center position and time for the discovered clusters using Mean-Shift clustering;
A spatio-temporal clustering processing apparatus comprising: means for combining a plurality of clusters having the same center position in the cluster group after the mean shift processing. The cluster discovery processing means includes:
A first parameter representing an allowable amount of spatial distance is set, another cluster that is not included in the kernel width of the cluster of interest is found, and the array position in time from the cluster is far from the cluster of interest When another cluster is rediscovered, it is determined whether the temporal arrangement position of the rediscovered cluster is within the range of the first parameter, and is within the range of the first parameter. 2. The spatiotemporal clustering processing apparatus according to claim 1, further comprising means for including the rediscovered cluster in a kernel width of the cluster of interest in some cases. The cluster discovery processing means includes:
A second parameter representing an allowable amount of time length is set, another cluster not included in the kernel width of the cluster of interest is found, and an array position in time from the cluster is changed from the cluster of interest When another distant cluster is rediscovered, it is determined whether the time length of the rediscovered cluster from the target cluster is within the range of the second parameter, and the range of the second parameter 3. The spatiotemporal clustering processing apparatus according to claim 1, further comprising means for including the rediscovered cluster within the kernel width of the cluster of interest when the cluster is within the range. 4. A spatiotemporal clustering method executed by a spatiotemporal clustering device comprising a computer and a storage unit,
A process of storing an information sequence in which a plurality of clusters are arranged in time series in a storage unit;
For each cluster of the stored information series, another cluster that is not included in the kernel width is found in the process of finding another cluster included in the preset spatial kernel width of the cluster of interest. A cluster discovery process for discovering clusters in which the spatial distance between clusters is within a preset distance and is temporally continuous, by sequentially performing according to the above time series,
Mean-shifting the center position and time using Mean-Shift clustering for the discovered clusters,
A spatio-temporal clustering processing method comprising: combining a plurality of clusters having the same center position among the cluster groups after the mean shift processing. The cluster discovery process includes:
A first parameter representing an allowable amount of spatial distance is set, another cluster that is not included in the kernel width of the cluster of interest is found, and the array position in time from the cluster is far from the cluster of interest When another cluster is rediscovered, it is determined whether the temporal arrangement position of the rediscovered cluster is within the range of the first parameter, and is within the range of the first parameter. 5. The spatiotemporal clustering method according to claim 4, further comprising a step of including the rediscovered cluster in a kernel width of the cluster of interest in some cases. The cluster discovery process includes:
A second parameter representing an allowable amount of time length is set, another cluster not included in the kernel width of the cluster of interest is found, and an array position in time from the cluster is changed from the cluster of interest When another distant cluster is rediscovered, it is determined whether the time length of the rediscovered cluster from the target cluster is within the range of the second parameter, and the range of the second parameter 6. The spatiotemporal clustering processing method according to claim 4, further comprising the step of including the rediscovered cluster within the kernel width of the cluster of interest when the cluster is within the range.   The program which makes the computer with which the said spatiotemporal clustering processing apparatus performs the process in each means with which the spatiotemporal clustering processing apparatus in any one of Claims 1 thru | or 3 is provided.

Images