JP7147293B2 - MOVEMENT ESTIMATION APPARATUS, METHOD, AND PROGRAM - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act February 23, 2018 The 10th Forum on Data Engineering and Information Management (The 16th Annual Conference of the Database Society of Japan) Online Proceedings Publication Site DEIM Forum 2018 H3-4 (http http://db-event.jpn.org/deim2018/post/proceedings/, http://db-event.jpn.org/deim2018/data/papers/170.pdf)

The present invention relates to a means of transportation estimation device, method, and program, and more particularly to a means of transportation estimation device, method, and program for estimating means of transportation from a movement trajectory.

With the spread of smart devices with GPS and the collection of communication history information, movement trajectory information is being collected. A movement trajectory is a set of positioning points defined as a triplet of time, latitude, and longitude. There is a need for estimating a user's means of transportation using this. The means of transportation includes, for example, walking, driving, and bicycling. Conventional means of transportation estimation techniques are broadly classified into two types.

The first is a method of estimating one means of transportation corresponding to the whole set of several tens of continuous positioning points (Non-Patent Documents 1 and 4). This involves extracting feature values such as the average speed, variance, and average directional conversion rate from a set of positioning points, and inputting them into a classifier to output one means of transportation for that set. be. In addition, as a method developed from this, there is also a method that considers geographical information such as distances to nearby stations and bus stops as feature amounts (Non-Patent Document 2).

The second is a method of estimating the means of movement of each segment, which is a portion between two consecutive positioning points called a segment (Non-Patent Document 3). In this method, feature values such as speed and the amount of direction change from the previous segment are extracted from each segment, and the mode of transportation is estimated by a classifier such as SVM using the feature values of the segment to be estimated and the segments before and after it.

Yu Zheng et al, “Understanding transportation modes based on GPS data for web applications”, in Proc. of WWW, 2010. Stenneth et al, “Transportation mode detection using mobile phones and GIS information”, in Proc. of SIGSPATIAL, 2010. Bolbol et al, “Inferring hybrid transportation modes from sparse GPS data using a moving window SVM classification”, in Computers, Environment and Urban Systems, 2012. Yuki Endo et al, "Classifying spatial trajectories using representation learning", in International Journal of Data Science and Analytics, 2016.

It is known that the movement trajectory that actually exists has a positioning frequency of once every five minutes or a random positioning frequency. The reasons for these problems are that the frequency of positioning must be sparse due to the problem of battery consumption, and that the timing at which positioning can be performed, such as when communication starts, is limited when the movement trajectory is obtained from communication history information instead of GPS. is the cause. The conventional technology described above cannot correctly handle such movement trajectories.

The first method above originally assumes a movement trajectory with a positioning interval of several seconds, so only one means of transportation is estimated for a set of three or more positioning points. As a result, there is a problem that the accuracy drops significantly when the interval between the positioning points increases. For example, for a movement trajectory that is positioned at 5-minute intervals, when 5 positioning points are processed as one set, 20 minutes have passed within the set, and movement within 20 minutes If the means has changed, it cannot be estimated correctly.

It has been confirmed that the second method, which assumes that the positioning frequency is sparse, can also be estimated with a certain degree of accuracy when positioning is performed once per minute. Accuracy decreases as the interval increases. This is because when the positioning interval widens, it becomes difficult to detect changes due to means of transportation. For example, a segment in which the user moves on foot and a segment in which the user waits at a station for a while before boarding a train have similar movement speeds, so it is difficult to distinguish them by analyzing only the relevant segments.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a moving means estimation apparatus, method, and program capable of accurately estimating moving means in a moving trajectory segment.

In order to achieve the above object, a movement means estimation apparatus according to the present invention is a movement means estimation apparatus for estimating a person's means of movement using a person's movement trajectory, wherein a plurality of An input movement trajectory extraction unit for extracting a movement trajectory for a certain period of time consisting of segments, and a characteristic amount for calculating a feature amount for each of the plurality of segments included in the movement trajectory for the certain period extracted by the input movement trajectory extraction unit. A quantity calculation unit, a short-term trajectory analysis unit that analyzes the feature of each of the plurality of segments based on the feature amount, and a combination of analysis results of each of the plurality of segments by the short-term trajectory analysis unit. and a long-term trajectory analysis unit for estimating the means of transportation.

A method of estimating means of transportation according to the present invention is a method of estimating means of transportation in a device for estimating means of transportation of a person using a trajectory of movement of a person, wherein an input trajectory extraction unit extracts a given trajectory from the input movement trajectory extraction unit extracts a movement trajectory for a certain period of time consisting of a plurality of segments, and a feature amount calculation unit calculates a feature amount for each of the plurality of segments included in the movement trajectory for a certain period of time extracted by the input movement trajectory extraction unit is calculated, a short-term trajectory analysis unit analyzes the feature of each of the plurality of segments based on the feature amount, and a long-term trajectory analysis unit analyzes each of the plurality of segments by the short-term trajectory analysis unit Combining the analysis results for , estimate the means of transportation.

Further, a program of the present invention is a program for functioning as each unit constituting the transportation estimation apparatus.

As described above, according to the transportation estimating device, method, and program of the present invention, a movement trajectory for a certain period of time consisting of a plurality of segments is extracted, and for each of the plurality of segments included in the movement trajectory for a certain period of time, Segments of a movement trajectory are calculated by calculating feature amounts, analyzing segment features based on the feature amounts for each of a plurality of segments, combining the analysis results for each of the plurality of segments, and estimating means of transportation. It is possible to obtain an effect that the means of transportation in can be estimated with high accuracy.

It is a figure for demonstrating the method to estimate a means of transportation in 1st Embodiment. 1 is a block diagram showing a functional configuration of a transportation estimation device according to a first embodiment; FIG. FIG. 10 is a diagram showing an example of data in a movement trajectory accumulation unit; (A) A diagram showing an example of data that is a set of a segment and a positioning point, and (B) A diagram that shows an example of data that is a set of an estimation target segment and an input segment. It is a figure which shows an example of the output data of a map information conversion part. It is a figure which shows an example of the output data of a feature-value calculation part. It is a figure which shows an example of the output data of a short-term locus|trajectory analysis part. FIG. 4 is a diagram for explaining a data input/output relationship between a short-term trajectory analysis unit and a long-term trajectory analysis unit; It is a figure which shows an example of the output data of a long-term locus|trajectory analysis part. 6 is a flowchart showing a transportation estimation processing routine by the transportation estimation device according to the first embodiment; It is a figure for demonstrating the method to estimate a means of transportation in 2nd Embodiment. FIG. 11 is a block diagram showing a functional configuration of a transportation estimation device according to a second embodiment; FIG. It is a figure which shows an example of the output data of a feature-value calculation part. It is a figure which shows an example of the output data of a short-term locus|trajectory analysis part. It is a figure which shows an example of the output data of a transportation means change possibility calculation part. FIG. 10 is a diagram for explaining a method of estimating a mode of transportation sequence by a long-term trajectory analysis unit; FIG. 10 is a diagram showing an example of data in a moving means labeled moving trajectory storage unit; 9 is a flow chart showing a transportation means change probability learning processing routine by the transportation estimation device according to the second embodiment. 9 is a flow chart showing a transportation estimation processing routine by the transportation estimation device according to the second embodiment.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[First Embodiment]
<Overview of the first embodiment of the present invention>
First, an overview of the first embodiment of the present invention will be described. In the embodiment of the present invention, the mode of transportation of the estimation target segment is estimated while considering long-term transition details. Consideration of long-term transition content means that, for example, when traveling by train for a long time before and after a certain segment, even if the distance traveled in the target segment is short, it is because the train stopped at the station for a while. As a means, it is to consider that it is still a train. In order to make this possible, the embodiment of the present invention analyzes each part of the movement trajectory in the short term, and then considers the result in the long term to estimate the means of movement of the estimation target segment.

A procedure for estimating means of transportation in the device according to claim 1 will be described. First, from the trajectory of movement, a trajectory of movement over a certain period of time around the segment to be estimated is extracted. For each segment in the extracted movement trajectory, a feature amount vector representing information of the segment is calculated. A small set of contiguous segments is then extracted and each analyzed. This analysis result is called a short-term analysis result. The short-term analysis results may be output in the form of a moving means of the set, or alternatively in the form of a vector representing the features. Next, the means of transportation of the estimation target segment is estimated using the short-term analysis result as an input. The estimation method is a class classifier using logistic regression, neural network, or the like.

An image diagram of the estimation process in the invention shown in claim 1 is shown in FIG. The estimation flow in FIG. 1 will be described. Movement trajectory information is shown at the bottom of FIG. Positioning points are indicated by circles, and arrows between the circles correspond to segments. The movement trajectory in FIG. 1 indicates that the user has been on the train for a long time, but the train stopped at a station before and after the target of estimation, so the speed temporarily slows down. Next, we describe the estimation process. A total of seven segments, three before and after the segment to be estimated, are used as inputs to estimate the mode of transportation. First, a feature quantity such as speed is extracted from each segment as needed. Next, short-term trajectory analysis is performed. Here, the segments are taken three by three and analyzed for information related to the mode of transportation. Long-term trajectory analysis is then performed.

Here, from the information obtained as a result of the short-term trajectory analysis, it is determined that there is a high possibility that the transportation means of the estimation target segment is a train, taking into consideration that the estimation target segment is a train for a long period of time. In the conventional technology, only immediately before and after the segment is considered, so it is not possible to appropriately determine whether an estimation target whose speed is temporarily slowing down is walking or a train. Appropriate estimation is possible by considering the transition relation.

Further, according to claim 3, map information is taken into consideration when calculating the feature amount of each segment. This can further improve the estimation accuracy. As for how to incorporate it into the feature amount, it is necessary to devise a method when the positioning interval of the movement trajectory is sparse. Conventionally, the distance from each positioning point to its nearest railroad track was used. This is not a problem if the positioning interval is about several seconds, but if the positioning frequency is about 5 minutes, even if you are moving by car, you may be near the railroad track at the timing of positioning. In order to ascertain whether it was continuously near the railroad during the movement, it is necessary to consider whether the railroad type of the nearest railroad track from the positioning point is consistent.

<Configuration of transportation estimation device according to first embodiment of the present invention>
Next, the configuration of the transportation estimation device according to the first embodiment of the present invention will be described. As shown in FIG. 2, the transportation estimating device 100 according to the first embodiment of the present invention includes a CPU, a RAM, and a ROM storing programs for executing a transportation estimating processing routine described later and various data. and a computer including: The transportation estimation apparatus 100 functionally includes an input unit 10, a calculation unit 20, and an output unit 90, as shown in FIG. The means of transportation estimation device 100 is a device for estimating means of transportation from a locus of movement.

The input unit 10 receives each of the movement trajectories of a person from the outside as an input.

The calculation unit 20 includes a movement trajectory accumulation unit 22, a map information accumulation unit 24, an input movement trajectory extraction unit 26, a map information conversion unit 28, a feature amount calculation unit 30, a short-term trajectory analysis unit 32, and a long-term trajectory analysis unit 34. consists of

The trajectory storage unit 22 records each trajectory input by the input unit 10 . Specifically, the time and latitude/longitude are recorded for the positioning points belonging to each trajectory. A data example is shown in FIG.

The input movement trajectory extraction unit 26 reads the movement trajectory from the movement trajectory storage unit 22, and extracts a movement trajectory for a certain period of time, which consists of a plurality of segments centering on the estimation target segment, from the read movement trajectory. Specifically, one segment ID is assigned to every two consecutive positioning points, and an input segment ID set consisting of an estimation target segment ID and a fixed number of segments before and after the corresponding segment ID are output together.

However, in the vicinity of the start point and the end point of the movement trajectory, there may not be the required number of segments before and after. In that case, the input segment ID set is obtained by supplementing the missing part with temporary IDs that do not have corresponding segments. An output example is shown in FIG. In this example, the input segment set is a set with a length of 5, which is obtained by adding two before and after the estimation target segment. In addition, when the estimation target segment ID is 1 or 2, the number of segments before and after is insufficient, so the input segment is supplemented with an ID that has no corresponding segment, ie, 0.

The map information conversion unit 28 processes and outputs map information for each of a plurality of segments included in the movement trajectory for a certain period of time extracted by the input movement trajectory extraction unit. Specifically, for each segment included in the output of the input movement trajectory extraction section 26, the position information of that segment is read from the movement trajectory accumulation section 22. FIG. Using the positional information, information relating to estimation of the means of transportation is read out from the map information storage unit 24, converted into a form that can be easily used as a feature amount, and output. Specific examples of output include the distance to the nearest station, the distance to the nearest railroad track, whether the routes of the nearest railroad tracks at both ends are the same, the distance to the nearest bus stop, the density of the railroad tracks and roads in the vicinity, Other factors include the population density of the surrounding area. An output example is shown in FIG.

The feature quantity calculation unit 30 calculates a feature quantity for each of a plurality of segments included in the movement trajectory extracted by the input movement trajectory extraction unit 26 for a certain period of time. Specifically, for each estimation target segment ID included in the output of the input movement trajectory extraction unit 26, for each segment in the corresponding input segment ID set, the position information of the corresponding positioning point is read from the movement trajectory accumulation unit 22. . Using the position information and the output content of the map information conversion unit 28, a feature amount useful for estimating the means of transportation is calculated and output for each segment.

Specific examples of feature amounts include speed, length of time, moving distance, angle change amount from the previous segment, and the like. At this time, the feature amount is also calculated for the segment ID used for complementing the missing part in the input movement trajectory extraction unit 26 . This feature amount is set to a value that does not become noise in subsequent processing. For example, it is conceivable to use the average of each feature amount of all other segments. An output example is shown in FIG. Also, as shown in FIG. 4, an input segment ID set corresponding to the estimation target segment ID is also output together.

The short-term trajectory analysis unit 32 analyzes the feature of each segment based on the feature amount. Specifically, using the output of the feature amount calculation unit 30, a set of feature vectors obtained by analyzing the short-term features of each part of one input segment ID set is output. An output example is shown in FIG. Gated Convolutional Neural Network (Gated CNN) can be used as an analysis method (see Non-Patent Document 5).

[Non-Patent Document 5] YN Dauphin et al, “Language Modeling with Gated Convolutional Networks”, in ArXiv, 2016.

The analysis procedure using Gated CNN is described. For a number of continuous segments called the mask width, a vector obtained by combining these feature vectors is used as input, and two layers of neural networks are used to output two vectors of the same number of dimensions. After applying the sigmoid function to one of these, we compute the Hadamard product of the two vectors and take that as the output vector. This series of operations is called a Gated Unit. The output vector of the Gated Unit can be regarded as a feature vector representing the content of movement of the segment set that is the input of the Gated Unit. By repeatedly applying this to the segments in the set of input segment IDs while shifting the segments to be used by a number called the stride width, a set of feature vectors of a certain number is obtained. The lower half of FIG. 8 shows the input/output relationship of data when analyzed by Gated CNN with a mask width of 3 and a stride width of 1. In FIG. In this case, if the length of the input segment is 5, the process with the gated CNN with the mask width of 3 is shifted 1 each and performed 3 times, so the number of output feature vectors is 3.

As an analysis method here, a simple CNN may be used, or a method such as a random forest using the same input as the above may be used to perform class classification.

The long-term trajectory analysis unit 34 combines the analysis results for each of the plurality of segments by the short-term trajectory analysis unit 32 to estimate the means of transportation of the estimation target segment, and outputs the results from the output unit 90 . Specifically, using a set of analysis results output by the short-term trajectory analysis unit 32 from a certain number of segments before and after the segment to be estimated, the transportation means of the target segment is estimated and output by the output unit 90 . FIG. 9 shows an output example.

As an analysis method, a vector with the same number of dimensions as the number of candidates for the means of transportation is output using a one-layer fully-connected neural network, and the result of applying a softmax function to it is regarded as the probability of each means of transportation. etc. can be considered. The data input/output relationship in this case is shown in the upper half of FIG. As an estimation method here, after repeating Gated CNN several times, it may be input to a layer neural network, or it may be repeated until the output vector of Gated CNN becomes one. Classification may also be performed using a random forest or the like.

<Operation of the transportation estimation device according to the first embodiment of the present invention>
Next, operation of the transportation estimation device 100 according to the first embodiment of the present invention will be described. First, upon receiving each movement trajectory, the transportation means estimation device 100 stores it in the movement trajectory storage unit 22 . Further, the transportation estimation device 100 executes a transportation estimation processing routine shown in FIG.

First, in step S100, the input movement trajectory extracting unit 26 reads out the movement trajectory from the movement trajectory storage unit 22, assigns one segment ID to every two consecutive positioning points, and estimates target segment IDs and their corresponding segment IDs. It outputs together with an input segment ID set consisting of a certain number of segments before and after.

Next, in step S102, the map information conversion unit 28 reads the position information of each segment included in the output of the input movement trajectory extraction unit 26 from the movement trajectory storage unit 22, and uses the position information to map the map. Information related to estimation of means of transportation is read out from the information storage unit 24, converted into a form that can be easily used as a feature amount, and output.

Next, in step S<b>104 , the feature amount calculation unit 30 extracts from the movement trajectory accumulation unit 22 for each segment in the set of corresponding input segment IDs for each estimation target segment ID included in the output of the input movement trajectory extraction unit 26 . Read the position information of the corresponding positioning point. Using the position information and the output content of the map information conversion unit 28, a feature amount useful for estimating the means of transportation is calculated and output for each segment.

Next, in step S106, the short-term trajectory analysis unit 32 uses the output of the feature amount calculation unit 30 to output a set of feature vectors obtained by analyzing short-term features of each part of one input segment ID set. .

Next, in step S108, the long-term trajectory analysis unit 34 estimates the means of transportation of the target segment using a set of analysis results output by the short-term trajectory analysis unit 32 from a fixed number of segments before and after the segment to be estimated. and output by the output unit 90, and the transportation means estimation processing routine ends.

As described above, according to the transportation estimating apparatus according to the first embodiment of the present invention, a movement trajectory for a certain period of time, which is composed of a plurality of segments centered on an estimation target segment, is extracted, and a movement trajectory for a certain period of time is extracted. Calculate the feature value for each of the multiple segments included in , analyze the segment features based on the feature value for each of the multiple segments, and combine the analysis results for each of the multiple segments to determine the target segment for estimation. By estimating the means of transportation, the means of transportation in the segment of the movement trajectory can be estimated with high accuracy.

In addition, it is possible to highly accurately estimate the means of transportation of the user from a movement trajectory whose positioning interval is long or random. This makes it possible to understand the behavior of the user with high accuracy, which can be used for recommendations in services related to various map information.

[Second embodiment]
<Overview of Second Embodiment of the Present Invention>
First, the outline of the second embodiment of the present invention will be described.

Since people move continuously, there is naturally consistency. For example, a continuous section without a train station will always be either a train or not a train at all. In the conventional technology, this point is not taken into consideration, and estimation is performed independently for each part, and the most probable means of transportation is output for each, resulting in inconsistent estimation results.

Although there are techniques that consider the above consistency, they have the problem that they are applicable only in limited situations. In Non-Patent Document 1, when the mode of transportation is switched, even in cases such as a transfer from a train to a bus, a short period of time is spent moving on foot. are below the threshold as candidate points for changing means of transportation, and the means of transportation is estimated for each of the parts between the candidate points. However, this method is based on the premise that the movement trajectory is closely measured at intervals of several seconds. It is known that there are cases in which the frequency of positioning is once every few minutes or that the frequency of positioning is random in real-world trajectories, and this technology cannot be applied to such trajectories. . This is because it is not possible to detect a short-time walking movement from such a trajectory with low positioning frequency. Non-Patent Document 1 also proposes to learn the transfer possibility in each area on the map from the learning data, and to correct the transportation means series based on it. However, the method of Non-Patent Document 1 has a problem that it is not possible to obtain the possibility of changing means of transportation in a region that is not included in the learning data.

In order to solve the above problem, after independently estimating the movement means of the partial movement trajectory extracted from the movement trajectory, a consistent long-term movement means sequence is estimated and output using this as an input. A consistent mode of transport series means that no mode of transport has changed from non-train to train in areas other than train stations, and no mode of transportation has changed to buses in areas other than bus stops. (See FIG. 11). Also, in order to calculate the possibility of changing means of transportation even in areas where there is no past data, the relationship between geographic information such as railway stations and bus stops and the possibility of changing means of transportation is learned in advance.

The apparatus according to claim 4 estimates and outputs a long-term movement means sequence considering consistency. First, from the trajectory of movement, partial trajectories are extracted by dividing the trajectory into short segments that can be handled by the conventional method, and the probability distribution of the means of transportation is calculated for each segment. At the same time, the possibility of changing the means of transportation at each part of the movement trajectory is calculated in consideration of map information such as surrounding railway stations and bus stops. Next, these two are used simultaneously as inputs to estimate and output a long-term consistent transportation sequence without unnatural transportation changes.

Claim 4 is a device that performs supervised learning on the relationship between geographic information and the possibility of changing the means of transportation by inputting a movement trajectory with a means of transportation label before performing the estimation according to the invention of claim 2 . By performing this in advance, the estimation in claim 2 can be performed with higher accuracy.

<Configuration of transportation estimation apparatus according to the second embodiment of the present invention>
Next, the configuration of the transportation estimation device according to the second embodiment of the present invention will be described. In addition, the same reference numerals are given to the parts having the same configuration as in the first embodiment, and the description thereof is omitted.

As shown in FIG. 12, a transportation estimating apparatus 200 according to the second embodiment of the present invention includes a CPU, a RAM, and a CPU for executing a transportation change probability learning processing routine and a transportation estimation processing routine, which will be described later. It can be configured by a computer including a ROM storing programs and various data. The transportation estimation apparatus 200 functionally includes an input unit 10, a calculation unit 220, and an output unit 90 as shown in FIG. The means of transportation estimation device 200 is a device for estimating means of transportation from a locus of movement.

The input unit 10 receives each of the movement trajectories of a person from the outside as an input.

The calculation unit 220 includes the movement trajectory accumulation unit 22, the map information accumulation unit 24, the input movement trajectory extraction unit 26, the map information conversion unit 28, the feature amount calculation unit 30, the transportation means change possibility calculation unit 231, and the short-term trajectory analysis unit 32. , a long-term trajectory analysis unit 34, a movement means labeled movement trajectory accumulation unit 240, a movement means change possibility model estimation unit 242, a movement means change possibility model accumulation unit 244, and a movement means estimation result accumulation unit 246. ing.

The trajectory storage unit 22 records each trajectory input by the input unit 10 . Specifically, the time and latitude/longitude are recorded for the positioning points belonging to each trajectory. A data example is shown in FIG.

The input movement trajectory extraction unit 26 reads the movement trajectory from the movement trajectory storage unit 22 and extracts a movement trajectory for a certain period of time consisting of a plurality of segments from the read movement trajectory. Specifically, one segment ID assigned to every two consecutive positioning points is output. In this embodiment, this segment is a unit for allocating means of transportation. An output example is shown in FIG.

The map information conversion unit 28 processes and outputs map information for each of a plurality of segments included in the movement trajectory extracted by the input movement trajectory extraction unit 26 for a certain period of time. Specifically, for each segment included in the output of the input movement trajectory extraction section 26, the position information of that segment is read from the movement trajectory accumulation section 22. FIG. Using the position information, information on estimation of means of transportation is read out from the map information accumulation unit 24, converted into a form that is easy to use as a feature value, and output, and information on estimation of possibility of changing means of transportation is read out, Output. Specific examples of output include the distance to the nearest station, the distance to the nearest bus stop, the distance to the nearest train line, and the real-time congestion degree of the nearest road. An output example is shown in FIG.

The feature quantity calculation unit 30 calculates a feature quantity for each of a plurality of segments included in the movement trajectory extracted by the input movement trajectory extraction unit 26 for a certain period of time. Specifically, for each estimation target segment ID included in the output of the input movement trajectory extraction unit 26, for each segment in the corresponding input segment ID set, the position information of the corresponding positioning point is read from the movement trajectory accumulation unit 22. . Using the position information and the output content of the map information conversion unit 28, a feature amount useful for estimating the means of transportation is calculated and output for each segment.

Specific examples of feature amounts include speed, length of time, distance traveled, distance from the nearest station, angle change amount from the previous segment, and the like. In addition, it is conceivable to also use the feature amounts of the segments before and after the estimation target in some cases. An output example is shown in FIG.

The short-term trajectory analysis unit 32 analyzes the feature of each of the plurality of segments based on the feature amount, and calculates a probability distribution representing the probability of each means of transportation. Specifically, a method in which a softmax function is used as the final layer using a neural network in which the feature amount of each segment output by the feature amount calculation unit 30 is used as an input, or an ensemble of multiple classifiers such as a random forest and consider the ratio of the number of votes in the ensemble as a probability distribution. An output example is shown in FIG.

For each pair of continuous segments, the means of transportation change possibility calculation unit 231 calculates, for each combination of means of transportation, the probability that the means of transportation in the pair will change from one of the combinations to the other. Specifically, using the output of the map information conversion unit 28, an inquiry is made to the means of transportation change possibility model storage unit 244, and the probability that the means of transportation will change in that part for a pair of continuous segments is calculated. Output for each combination. An output example is shown in FIG.

The long-term trajectory analysis unit 34 calculates the probability distribution calculated for each of the plurality of segments by the short-term trajectory analysis unit 32, and the combination of transportation means calculated for each pair of segments by the transportation means change possibility calculation unit 231. and the probability that the transportation means change, a transportation means sequence consisting of transportation means for each of the plurality of segments, which has long-term consistency, is estimated and recorded in the transportation means estimation result accumulation unit 246 .

As an implementation method, for example, a method using a Markov random field can be considered. A Markov random field is a technique for estimating the true value of each node based on the observed data present at each node of the graph. At this time, it is possible to set a similarity constraint between the true data of adjacent nodes and use it at the time of estimation. When a Markov random field is used as a processing method, the segments in the movement trajectory are regarded as nodes in the Markov random field, and the means of movement of each segment, which is the object of estimation, is regarded as the true data of the node. Also, the analysis result of the short-term trajectory analysis unit 32 for each segment is regarded as observation data for each node. The relationship of means of transportation between adjacent nodes is expressed as the degree of similarity between true data of adjacent nodes.

A schematic diagram showing this modeling is shown in FIG.

An example cost function is shown below. This shows the cost function for a certain segment s with a means of transportation m, and the cost of the entire series is calculated by calculating this for all segments and taking the sum. By assigning a transport m to each segment s in such a way as to minimize the cost of the entire sequence, a consistent transport sequence can be obtained.

とする方法が一例として考えられる。βはセグメントそれぞれの移動手段確率の項（p_s(m)）と変化可能性の項

の重視する度合いを調整するための重み項である。 p _s (m) is the probability that the means of transportation of segment s is m, and is obtained as the output of short-term trajectory analysis unit 32 . M _s is a set of segments temporally continuous with segment s, that is, segments before and after segment s. w(m _s , m _s′ ) is the cost for the means of transportation ms and m _{s ′} of segments s and s′ to continue, and is calculated using the output result of the means of transportation change possibility calculation unit 231 . can do Various specific calculation formulas are conceivable. For example, when the probability of changing from m _s to m _s' is p,

As an example, a method of β is the term of transportation probability (p _s (m)) and the term of possibility of change for each segment

is a weighting term for adjusting the degree of emphasis on

When estimating, iterative conditional models (ICM) can be used. First of all, assign one transportation means appropriately for every segment. Next, one segment is selected, and after fixing the moving means of the other segments, the moving means of that segment is changed so that the cost function is minimized. By repeating this process, it is possible to obtain a locally optimal moving means sequence for the cost function. Alternatively, a max-sum algorithm or the like may be used.

In addition to this, it can also be realized by a method of modeling using a hidden Markov model and optimizing with a Viterbi algorithm.

The transportation means labeled movement trajectory accumulation unit 240 records the time, latitude and longitude of the positioning points belonging to each movement trajectory, and the means of transportation for each segment that is a set of continuous positioning points. A data example is shown in FIG.

The means of transportation change possibility model estimating unit 242 calculates the probability that the means of transportation will change for each combination of means of transportation and the map information based on the movement trajectory with the label of the means of transportation for each of the plurality of segments and the map information. to learn the relationship between

Specifically, information is read from the moving means labeled movement locus accumulation unit 240, and based on the position information, the map information conversion unit 28 is called to obtain map information of the surrounding area. Then, it learns the relationship between the map information and the change possibility of the means of transportation, and records it in the means of transportation change possibility model accumulation unit 244 .

As a method of implementation, for example, using the output of the map information conversion unit 28 as an explanatory variable and changes in means of transportation as a dependent variable, learning is performed in an ensemble of multiple decision trees such as a random forest, and the parameters are recorded. can be considered. In this case, the calculation of the possibility of changing the means of transportation performed by the calculating unit 231 of the possibility of changing the means of transportation is performed by ensemble of a plurality of classifiers using the parameters of the model accumulation unit 244 of the possibility of changing the means of transportation with map information as an input. It can be realized by regarding the ratio of the number of votes in the ensemble as a probability distribution. In addition to this, a method using a neural network is also conceivable.

<Operation of the transportation estimation device according to the second embodiment of the present invention>
Next, operation of the transportation estimation device 200 according to the second embodiment of the present invention will be described. In advance, transportation estimation apparatus 200 executes a transportation change probability learning processing routine shown in FIG.

In step S200, the mode of transportation change possibility model estimation unit 242 reads information from the mode of transportation labeled movement trajectory accumulation unit 240, and based on the position information, calls the map information conversion unit 28 to obtain map information of the surrounding area. . Based on the movement trajectory with the transportation means label for each of the plurality of segments and the map information, the relationship between the probability that the transportation means will change for each combination of transportation means and the map information is learned. It is recorded in the possibility model accumulation unit 244 .

Further, upon receiving each movement trajectory, the transportation means estimation device 200 stores it in the movement trajectory storage unit 22 . Further, the transportation estimation device 200 executes a transportation estimation processing routine shown in FIG.

First, in step S210, the input movement trajectory extraction unit 26 reads the movement trajectory from the movement trajectory storage unit 22, and outputs one segment ID assigned to every two consecutive positioning points.

Next, in step S212, the map information conversion unit 28 reads the position information of each segment included in the output of the input movement trajectory extraction unit 26 from the movement trajectory accumulation unit 22, and uses the position information to map the map. Information on estimation of means of transportation and information on estimation of possibility of changing means of transportation are read from the information storage unit 24 and output.

Next, in step S<b>214 , the feature amount calculation unit 30 extracts from the movement trajectory accumulation unit 22 for each segment in the set of corresponding input segment IDs for each estimation target segment ID included in the output of the input movement trajectory extraction unit 26 . Read the position information of the corresponding positioning point. Using the position information and the output content of the map information conversion unit 28, a feature amount useful for estimating the means of transportation is calculated and output for each segment.

Next, in step S216, the short-term trajectory analysis unit 32 receives the output of the feature amount calculation unit 30 and outputs the probability that each segment is each means of transportation.

In step S218, the means of transportation change possibility calculation unit 231 uses the output of the map information conversion unit 28 to inquire of the means of transportation change possibility model accumulation unit 244, and determines whether a pair of consecutive segments has a means of transportation at that part. output the probability that changes in for each combination of means of transportation.

In step S<b>220 , the long-term trajectory analysis unit 34 combines the probability distribution calculated for each of the plurality of segments by the short-term trajectory analysis unit 32 and the movement mode calculated for each pair of segments by the transportation means change possibility calculation unit 231 . Based on the probability that the means of transportation will change for each combination of means, a means of transportation sequence consisting of means of transportation for each of a plurality of segments with long-term consistency is estimated, and a means of transportation estimation result accumulation unit 246 , and ends the transportation estimation processing routine.

As described above, according to the movement means estimation apparatus according to the second embodiment of the present invention, a long-term movement means sequence is extracted from a movement trajectory that has large noise and is difficult to estimate a movement means with high accuracy using each part alone. It becomes possible to estimate with high accuracy. As a result, it is possible to understand the user's behavior with high accuracy, and it can be used for recommendations in services related to various geographic information.

The present invention is not limited to the above-described embodiments, and various modifications and applications are possible without departing from the gist of the present invention.

For example, the transportation means estimation apparatus 100 described above has a computer system inside, and the "computer system" includes the homepage provision environment (or display environment) if the WWW system is used. shall be taken.

Further, in the specification of the present application, an embodiment in which the program is pre-installed has been described, but it is also possible to store the program in a computer-readable recording medium such as a CD-ROM, memory card, etc. and provide it. is.

10 Input unit 20 Calculation unit 22 Movement trajectory accumulation unit 24 Map information accumulation unit 26 Input movement trajectory extraction unit 28 Map information conversion unit 30 Feature quantity calculation unit 32 Short-term trajectory analysis unit 34 Long-term trajectory analysis unit 90 Output unit 100 Transportation means estimation device 200 means of transportation estimation device 220 calculation unit 231 means of transportation change possibility calculation unit 240 means of transportation labeled movement trajectory accumulation unit 242 means of transportation change possibility model estimation unit 244 means of transportation change possibility model accumulation unit 246 means of transportation estimation result accumulation unit

Claims (5)

A means of transportation estimation device for estimating a means of transportation of a person using a person's movement trajectory,
an input movement trajectory extraction unit that extracts a movement trajectory of a certain period consisting of a plurality of segments from a given movement trajectory;
a feature quantity calculation unit that calculates a feature quantity for each of the plurality of segments included in the movement trajectory for the fixed period extracted by the input movement trajectory extraction unit;
Moving means change possibility calculation for each combination of moving means for each pair of continuous segments among the plurality of segments, calculating a probability that the moving means changes from one of the combinations to the other in the pair of segments. Department and
a short-term trajectory analysis unit that analyzes the feature of each of the plurality of segments based on the feature quantity;
a long-term trajectory analysis unit that combines the analysis results of each of the plurality of segments by the short-term trajectory analysis unit to estimate a means of transportation;
including
The short-term trajectory analysis unit analyzes the characteristics of each segment based on the feature amount for each of the plurality of segments, thereby calculating a probability distribution representing the probability of each means of transportation,
The long-term trajectory analysis unit calculates the probability distribution calculated for each of the plurality of segments by the short-term trajectory analysis unit, and the movement method calculated for each pair of segments by the transportation means change possibility calculation unit. Means of transportation calculated using the sum of cost functions for the plurality of segments that the means of transportation of the segments is the means of transportation m, based on the probability that the means of transportation will change for each combination of means. A mobile means estimator for estimating a mobile means sequence consisting of mobile means m for each of said plurality of segments so as to minimize the cost of the entire sequence . further comprising a map information conversion unit for processing and outputting map information for each of the plurality of segments included in the movement trajectory for the fixed period extracted by the input movement trajectory extraction unit;
The feature amount calculation unit calculates a feature amount using the map information output by the map information conversion unit for each of the plurality of segments included in the movement trajectory for the fixed period extracted by the input movement trajectory extraction unit. 2. The moving means estimating device according to claim 1 , which calculates the . Transportation means for learning the relationship between the probability of change of the transportation means for each combination of the transportation means and the map information, based on the movement trajectory with the transportation means label for each of the plurality of segments and the map information. 3. The transportation estimating device according to claim 2 , further comprising a variability model estimating unit. A method of estimating means of transportation in a means of transportation estimating device for estimating means of transportation of a person using a movement trajectory of a person, comprising:
an input movement trajectory extraction unit extracting a movement trajectory for a certain period of time consisting of a plurality of segments from the given movement trajectory;
a feature quantity calculation unit calculating a feature quantity for each of the plurality of segments included in the movement trajectory of the fixed period extracted by the input movement trajectory extraction unit;
a means of transportation change possibility calculation unit, for each combination of means of transportation for each of the plurality of segments, calculates the probability that the means of transportation changes from one of the combinations to the other in the segment;
A short-term trajectory analysis unit, for each of the plurality of segments, analyzes the feature of the segment based on the feature amount;
a long-term trajectory analysis unit estimating means of transportation by combining analysis results for each of the plurality of segments by the short-term trajectory analysis unit ;
The short-term trajectory analysis unit analyzes, for each of the plurality of segments, by analyzing the characteristics of the segment based on the feature amount, calculating a probability distribution representing the probability of each means of transportation,
In the estimation by the long-term trajectory analysis unit, the probability distribution calculated for each of the plurality of segments by the short-term trajectory analysis unit and the probability distribution calculated for each of the plurality of segments by the transportation means change possibility calculation unit Also, based on the probability that the means of transportation for each combination of the means of transportation will change, the calculation is performed using the sum of the cost functions for the plurality of segments that the means of transportation of the segments is the means of transportation m. a transportation means estimation method for estimating a transportation means m sequence for each of said plurality of segments so as to minimize the cost of the overall transportation sequence taken . A program for causing a computer to function as each unit constituting the moving means estimation apparatus according to any one of claims 1 to 3 .

Images