JP2010250807A - Method and system for detecting state change of object from position data stream - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To detect state change of a position data stream at high speed and with high precision. <P>SOLUTION: The method for detecting the state change of an object includes steps of: recording a state immediately before the object as B<SB>L</SB>; receiving position data one by one from the position data stream; determining accumulated state change tendency from the state B<SB>L</SB>immediately before the object to respective states B<SB>i</SB>(i=1, 2, ... I) whenever the position data is received; and detecting presence/absence of a state change point in the position data stream based on the determined accumulated state change tendency. The method further describes a position of the state change point and a state of an object at a change destination when the state change point is detected. An object state change detection method includes higher speed and higher precision and descriptiveness of the detection result is higher in comparison with the conventional technology. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates generally to position data stream processing, and more particularly to a method and system for detecting state changes of interest from a position data stream.

  The state of a moving object (user or real object) that is the basis of context-aware computing is generally implicitly indicated by real-time object position data and mined from that data. In executing state mining, detection of a change in the state of an object is important and useful. By using the detected state change, it becomes possible to provide various services based on the context in a timely manner to the user. State change detection for these applications usually involves 1) fast (immediate detection), 2) high precision (false alarm probability is almost zero), and 3) high descriptiveness (not only the time of occurrence of state change, However, the existing technology cannot fully satisfy these requirements.

  The purpose of state change detection is to identify the state change point and state change pattern of interest by processing a position data stream based on observation. Several related techniques already exist for detecting underlying pattern changes through real-time data stream processing. For example, Non-Patent Document 1 (Kenji Yamanishi and Junichi Takeuchi's paper “A unifying framework for detecting outliers and change points from non-stationary time series data (Japanese paper title: change and detection value using forgetting learning algorithm) Unified handling of output))) proposes an AR (autoregressive) model for expressing the statistical behavior of time-series data. This paper further proposes a SDAR (sequentially counting AR model learning) method for online learning of an AR model to describe the underlying statistical behavior. A scoring function based on a learned AR model for detecting outliers and change points has also been proposed.

  In addition to the above, Non-Patent Document 2 (the paper “Paper: Detecting Changes in Data Streams” by Shai Ben-David et al.) Provides non-parametric detection for detecting changes in the data stream. A law has been proposed. This scheme performs a parallel difference test using sliding windows and feature windows of different lengths, and converts the change detection problem into a test of whether the two samples were generated by different distributions.

  FIG. 1 is a structural block diagram showing a state change detection system proposed by Shai Ben-David et al. In this solution, data streams are generated in an independent manner, one data point at a time, with an underlying probability distribution. Since it is impractical to give the data stream processing algorithm enough memory to store the entire stream history, this solution uses a change detection algorithm based on a two-window paradigm. This algorithm compares the data in the “feature window” with the data in the “slide window”. Each window contains a fixed number of consecutive data points. The “slide window” slides forward every time a data point is received. The “feature window” is a reference window and is updated each time a change is detected. As shown in FIG. 1, there are a total of K pairs of slide windows and feature windows. Using the K pairs of slide windows and feature window pairs, a system for executing a parallel change determination method for detecting a state change by comparing parallel windows is constructed.

As shown in FIG. 1, this state change detection system includes K difference test means 101-1, 101-2,... 101-K-1 and 101-K. Window length of the K difference test means, respectively _{L 1,} L _2, a _{... L K-1, L K} . The system further includes a position data stream update unit 102, a parallel state change determination unit 103, and a latest state change point notification unit 104. The position data stream update unit 102 updates the external position data stream by including the latest position data. This latest position data is supplied to K difference test means 101-1, 101-2,... 101-K-1 and 101-K, whereby the sliding window in the means is updated. Each of the difference test means 101-1, 101-2,... 101-K-1 and 101-K slides the sliding window forward to add the latest position data and delete the old data. An update unit is provided. The sliding window is updated each time new data arrives. Each of the difference test means 101-1, 101-2,... 101-K-1 and 101-K further comprises a feature window update unit that updates the feature window each time a new change is detected. The feature window is a static window that is always moved to the last detected state change point and remains static until a new change is detected. Each of the difference test means 101-1, 101-2,... 101-K-1 and 101-K further sets the distance between the slide window data and the feature window data as d _i = D _i (SW _i , FW _i ) ( A distance calculation unit for calculating i = 1, 2,... K) is provided. The parallel state change determination means 103 receives distance calculation results from the individual difference test means 101-1, 101-2,... 101-K-1 and 101-K having different window lengths, and K inequality. Check in parallel whether these distances exceed the corresponding threshold (ie, whether d _i > t _i (i = 1, 2,... K)). When a state change is detected at several branches (that is, when a branch of d _i > t _i is detected), the parallel state change determination unit 103 executes a trade-off of the result at the branch and changes the state. Determine if it is a point. Thereafter, the latest state change point notifying unit 104 notifies the application (not shown) and each feature window of the latest state change point determined by the parallel state change determining unit 103.

しきい値は、ｔ_１＝ｔ_２＝１に設定される。図２Ｂは、ＳＷ_１とＦＷ_１（ウィンドウ長は４）を使用して実行された差異テストの段階的テスト結果である。図２Ｂの結果によれば、ウィンドウ長が４のとき、状態変化はｔ_１０で検出されている。図２Ｃは、ＳＷ_２とＦＷ_２（ウィンドウ長は１）を使用して実行された差異テストの段階的テスト結果である。図２Ｃの結果によれば、ウィンドウ長が１のとき、状態変化はｔ_４で検出されている。 Specific examples are shown in FIGS. 2A, 2B, and 2C in order to facilitate understanding of the system according to the prior art shown in FIG. FIG. 2A is an example of a simple position data stream composed only of position data “1” and “2”. The parallel difference test is performed using two paths (ie, two feature windows with window lengths L ₁ = 4 and L ₂ = 1). The two-path feature window is initialized to FW ₁ = { _{2, 2, 2, 2} } and FW ₂ = {2} according to the most recent detection result of the data stream features. The initial sliding windows are SW ₁ = { _{2, 2, 2, 2} } and SW ₂ = {2}. The sliding window slides forward one point at a time. The distance function used in the parallel test is as follows.

The threshold is set to t ₁ = t ₂ = 1. FIG. 2B is a stepwise test result of a difference test performed using SW ₁ and FW ₁ (window length is 4). According to the results of Figure 2B, when the window length is 4, the state change is detected by the t _10. FIG. 2C is a stepwise test result of a difference test performed using SW ₂ and FW ₂ (window length is 1). According to the results of FIG. 2C, when the window length is 1, the state change is detected by t _4.

Kenji Yamanishi and Junichi Takeuchi “A unified framework for detecting outliers and change points from non-stationary time series data” (Japanese paper title: Outlier detection and change point detection using forgetting learning algorithm) Shai Ben-David et al. “Paper: Detecting Changes in Data Streams”

  From the above, in the conventional technique proposed by Shai Ben-David et al., When the difference test is executed using different window lengths, the detection points of the state change are different, so the determination becomes difficult. Furthermore, it is difficult to balance between “high speed” and “accuracy” with this method. That is, if the window size is increased, detection is delayed, and if the window size is decreased, a false alarm is caused. Therefore, there are many tradeoffs required between different change detection results.

  Furthermore, in the difference test with K parallel branches, not only the calculation amount is enormous, but also the calculation complexity is extremely high. In addition, since it is difficult to select a window size, a distance function, and a threshold value in the conventional method, it is difficult for a user to use it.

  The present invention is proposed in view of the above problems. In the present invention, not only the occurrence of a state change in the position stream can be detected, but also the location of the state change point and the state after the change of the object can be detected.

According to one aspect of the present invention, there is provided a method for detecting a change in the state of a target from a position data stream, the step of recording the state immediately before the target as _BL , and position data from the position data stream one by one receiving, each time it receives the position data, each state from the previous state _{B L B i (i = 1,2} , ... I) determining an accumulated state changing trend of the target to have been determined Detecting a presence or absence of a state change point in the position data stream based on the cumulative state change trend.

According to another aspect of the present invention, there is provided a system for detecting a change in a state of a target from a position data stream, the immediately preceding state recording means for recording a state immediately before the target as _BL , and a position from the position data stream. Position data receiving means for receiving data one by one, and each time the position data is received, the cumulative state change of the object from the previous state _BL to each state B _i (i = 1, 2,... I) A cumulative state change trend determining means for determining a trend; and a state change point detecting means for detecting the presence or absence of a state change point in the position data stream based on the determined cumulative state change trend. A featured system is provided. The system according to an embodiment of the present invention further includes description means for describing the position of the state change point and the state after the change of the object when the state change point is detected.

  In order to solve the problems of the prior art, the present invention defines two concepts, “state probability model” and “state change hypothesis”. Among them, the “state probability model” represents the probability that each position data corresponds to each target state, the “state change hypothesis” is used for modeling the state change, and each hypothesis corresponds to one type of state change.

  The method and system for detecting an object state change according to the present invention has a number of advantages over the prior art. According to the present invention, the detection of the state change is sequentially performed each time new position data arrives, and thus is faster than the prior art. Furthermore, in the present invention, since changes with different degrees of change can be adaptively detected without using parallel windows, the detection accuracy is improved. Further, the method of the present invention is highly descriptive because the specific position of the state change point and the state of the target after the change are also detected. The method of the present invention is lightweight. Furthermore, the user does not have to worry about window size and distance function.

  Other advantages and features of the present invention will become apparent upon reference to the following detailed description and drawings. However, it should be noted that the present invention is not limited to the examples shown in the drawings or specific embodiments.

The invention will be more clearly understood from the following detailed description of embodiments of the invention and the accompanying drawings. In the accompanying drawings, similar parts are denoted by the same reference numerals.
It is a structural block diagram which shows the state change detection system in a prior art. It is the schematic for demonstrating the principle of operation of the prior art system shown in FIG. It is the schematic for demonstrating the principle of operation of the prior art system shown in FIG. It is the schematic for demonstrating the principle of operation of the prior art system shown in FIG. 1 is a structural block diagram illustrating a state change detection system 300 according to an embodiment of the present invention. 4 is a flowchart illustrating an example of an operation process of the system 300 illustrated in FIG. 3. It is the schematic for demonstrating the principle of operation of the system 300 shown in FIG. FIG. 6 is a structural block diagram illustrating a state change detection system 600 according to another embodiment of the present invention. It is the schematic for demonstrating the operation principle of the system 600 shown in FIG.

  As described above, in the present invention, the detection of the state change of the target is facilitated by using the two concepts of the “state probability model” and the “state change hypothesis”. According to the present invention, the state change of the object is detected by sequentially evaluating the latest position data in the position data stream and finding a significant state change point that has changed significantly from the immediately preceding state.

  FIG. 3 is a structural block diagram illustrating a state change detection system 300 according to one embodiment of the present invention. As shown in FIG. 3, the system 300 includes a position data stream updating unit 301, a position data receiving unit 302, a previous state recording unit 303, a cumulative state change tendency determining unit 304, a state change point detecting unit 305, A description unit 306 (arbitrary), an importance check unit 307 (arbitrary), a latest state change point notification unit 308, and a storage device 309 for storing a state probability model and a state change hypothesis are provided. In FIG. 3, the description means 306 and the importance check means 307 included in the preferred embodiment of the present invention are shown as wavy lines as optional modules. The storage device 309 stores a probability relationship of changes from position data to a target state, which is designated in advance for each application. In this system, the operations of the position data stream updating unit 301 and the latest state change point notifying unit 308 are respectively the means for updating the position data stream in the prior art system shown in FIG. This is the same as the operation of the means for notifying the state of the later target. Here, detailed description of these components is omitted.

  The position data receiving means 302 receives new position data xn (n = w, w + 1,... N), which are subject to state detection, from the position data stream update means 301 one by one. The immediately preceding state recording unit 303 records the immediately preceding state BL of the target at the latest state change detection point xw and inputs it to the accumulated state change tendency determining unit 304. Accumulated state change tendency determination means 304 accumulates from state BL immediately before the target to each state Bi (i = 1, 2,... I) each time position data xn (n = w, w + 1,... N) arrives. Determine state change trends. For example, the cumulative state change tendency determining means 304 calculates the cumulative state change probability ratio (detailed below) of the object from the immediately previous state BL to each state Bi (i = 1, 2,... I). Run specific sequential hypothesis tests for. Thereafter, the state change point detection unit 305 checks the cumulative state change tendency determined for each state by the cumulative state change tendency determination unit 304 to determine whether or not there is a target state change. For example, when there is no significant cumulative state change tendency in any state (that is, when the cumulative state change probability ratio calculated in any state is zero), the state change point detection unit 305 determines the target state. It is determined that there is no change. Conversely, when the cumulative state change probability ratio calculated for a certain state is not zero, the state change point detection unit 305 determines that a state change point exists in the position data stream. If there is a state change point, the description unit 306 can optionally describe the position of the state change point and the state after the change of the object. In the present invention, by adding the description means 306, the descriptiveness of the state change detection becomes higher than that of the prior art. For example, in one embodiment of the present invention, the description unit 306 determines a state corresponding to the most prominent cumulative state change tendency of the target as a new state of the target from various states, and corresponds to this state. The initial change point of the target to be used is used as the state change point.

  Furthermore, when the state change point detection unit 305 detects a state change point, the importance check unit 307 may be arbitrarily used to check the importance of this type of state change. For example, the importance checking unit 307 determines whether the state change is significant by comparing the most prominent one of the calculated cumulative state change trends with a predetermined threshold value. When the comparison result indicates that the change tendency exceeds the threshold value, the importance check unit 307 determines that the state change is significant. If the change tendency does not exceed the threshold value, the importance checking unit 307 ignores the change in state. By this importance check, only the significant state change point in the position data stream and the corresponding target state after the change are notified by the latest state change point notifying unit 308 to the immediately preceding state recording unit 303, thereby The previous state is updated and used to detect future state changes.

  FIG. 4 is a flowchart illustrating an example of an operation process of the system 300 illustrated in FIG. As shown in FIG. 4, the process 400 is step 401, that is, the position data receiving means 302 receives the latest position data xn from the position data stream, and the immediately preceding state recording means 303 records the immediately preceding state BL of the object. start from. Subsequently, based on the state probability model and the state change hypothesis stored in advance in the storage device 309, the cumulative state change tendency determining unit 304 determines each state Bi (i = 1, 1) from the previous state BL up to now. 2,... I) The cumulative state change tendency of the object is determined.

As described above, the state probability model defines the probability of occurrence of a change from position data to each target state Bi (i = 1, 2,... I). This is defined by the following three.
1. Target state set: {Bi, i = 1, 2,... I}
2. Position data set: {xj, j = 1, 2,... J}
3. State probability model: {P (xj | Bi), i = 1, 2,... I; j = 1, 2,. This represents the probability that the object is in state Bi when the position data is xj. This definition can be made, for example, if the user is in the office (for example, if the location data is 1), the probability that the user is working (likelihood) is 90%. .

Characterization of the subject's state change is performed using a state change hypothesis. For example, it is possible to define as follows.
H0: The target state does not change.
Hi: The target state changes from BL to Bi (i = 1, 2,... I).

ここで、Ｒ^ｎ（Ｈ_ｉ｜Ｈ_０）は、位置データｘ_ｎが着信したときに対象の状態が直前の状態Ｂ_Ｌから状態Ｂ_ｉに変化する確率比を表す。Ｒ^ｎ（Ｈ_ｉ｜Ｈ_０）の結果のうち、正の値は状態がＢ_ＬからＢ_ｉに変化する確率が高いことを意味し、負の値は状態がＢ_ＬからＢ_ｉに変化する確率が低いことを意味する。 Returning to FIG. 4, in step 402, the cumulative state change tendency determination means 304 calculates the log hypothesis check probability ratio Rn (Hi | H0) for each position data xn and the cumulative state change probability ratio Tin, for example. It is possible to represent various state change trends. When the new position data xn arrives, the cumulative state change tendency determination means 304 calculates a log hypothesis check probability ratio Rn (Hi | H0) of the position data xn using the following equation (1).

Here, R ⁿ (H _i | H ₀ ) represents a probability ratio that the target state changes from the previous state B _L to the state B _i when the position data x _n arrives. Among the results of R ⁿ (H _i | H ₀ ), a positive value means that there is a high probability that the state changes from B _L to B _i , and a negative value changes the state from B _L to B _i . This means that the probability is low.

Thereafter, in step 403, the cumulative change in state probability ratio T _i ⁿ is calculated by the following equation.

As described above, T _i ⁿ characterizes the cumulative state change tendency in which the target changes from the previous state B _L to the state B _i (i = 1, 2,... I) before the position data x _n arrives. Used for.

Cumulative state change tendency determining unit 304, when the position data _{x n} arrives, various states _{B i (i = 1,2, ...} I) and all of the accumulated state change probability from _T ^{1 n} relates to _T ^{I n} After calculating the ratio, in step 404, the state change point detecting means 305 detects the maximum value from the calculated cumulative state change probability ratio (that is, identifying the state change having the most prominent state change tendency). The maximum cumulative state change probability ratio T _{i *} ⁿ is obtained by the following equation.

Thereafter, the state change point detecting means 305 determines in step 405 whether or not the maximum cumulative state change probability ratio T _{i *} ⁿ is zero. If the maximum cumulative state change probability ratio T _{i *} ⁿ is zero, it means that there is no state change in the position data stream. In this case (when the result of step 405 is “No”), process 400 returns to step 401 to receive new position data and continue to detect state changes. If the maximum cumulative state change probability ratio T _{i *} ⁿ is not zero (if the result of step 405 is “Yes”), the process 400 proceeds to step 406. In step 406, based on the calculated maximum cumulative state change probability ratio T _{i *} ⁿ , using any description means 306 (shown in FIG. 3), the specific position of the state change point and the target of the change destination The state of is described.

Specifically, in step 406, the description means 306 first determines a state change hypothesis H _{i *} corresponding to the maximum cumulative state change probability ratio T _{i *} ⁿ , and from this state change hypothesis H _{i *} , The state is determined as B _{i *} . At the same time, the description means 306 uses the following equation (4) to specify the exact position λ ^* of the state change point.

In step 407, any importance checker 307 may determine the importance of the state change having the most prominent cumulative state change tendency detected by the state change point detector 305. For example, the importance checking unit 307 compares the maximum cumulative state change probability ratio T _{i *} ⁿ with a predetermined threshold Th. If T _{i *} ⁿ exceeds Th, the corresponding state change point is accepted and recorded in step 408. Conversely, if T _{i *} ⁿ is less than Th, this state change point is discarded (step 409). Thereafter, in step 410, and the position lambda ^* state change point accepted after the change target state, the last known state recording unit 303 by the last state change point notifying means 308, the target state immediately before B _L Is updated. The state immediately before the update is used for detection of the next state change.

FIG. 5 is a schematic diagram for explaining the operation principle of the system 300 shown in FIG. In this example, the state model has a state set {B ₁ , B ₂ } and a position data set {1, 2}, the position data x _w arrives, and the state immediately before the target is B _L = B _2. Suppose that The state probability model can be defined as follows, for example.
_{P (x i = 1 | B} 1) = 0.8; P (x i = 1 | B 2) = 0.2; P (x i = 2 | B 1) = 0.1; P (x i = 2 | B ₂ ) = 0.9

In this case, the state change hypothesis is defined as follows.
H ₀ : The state of the target does not change.
The state of the H1 object changes from B _L = B ₂ to B ₁ .

Referring to the operation process of FIG. 4, a state change in the position data stream is detected, and the detection result is shown in FIG. Since the non-zero cumulative state change probability ratio T _{i *} ⁿ is calculated at the positions of the position data x _{w + 5} , x _{w + 8} , x _{w + 9} , it may be determined that there is a high possibility that a state change has occurred at these positions. it can. If the importance check threshold is set to 1, a significant state change is detected at position x _{w + 9} , and this state change has occurred at x _{w + 7} .

  FIG. 6 is a structural block diagram illustrating a state change detection system 600 according to another embodiment of the present invention. Compared to the system 300 of FIG. 3, the system 600 differs in that it further includes a state change weight storage device 601.

  In the embodiment shown in FIG. 3, all state changes are treated equally, but in practical applications, different importance is often added to different state changes. In view of this fact, the present invention provides the embodiment shown in FIG. 6 to solve this problem. The state change weight storage device 601 is used to store weights corresponding to the importance of different state changes. The higher the importance of the state change, the greater weight is added. By this change, the present invention can realize another effect in addition to the above. The effect is that since different weights are assigned to different state changes, the probability that an important state change is detected increases, and the detection speed also increases.

As shown in FIG. 6, the state change weight storage device 601 stores weights corresponding to the importance of different state changes. For example, if hypothesis H ₀ (the target state does not change) and hypothesis H _i (the target state changes from B _L to B _i (i = 1, 2,... I)), hypothesis H _A weight w (B _i | B _L ) that represents a weight corresponding to _i is defined.

これにより、重要な状態変化が検出される確率が高まる。重要な状態変化については、重要性チェックの実行時に確認することができる。 The corresponding weight w (B _i | B _L ) is supplied to the cumulative state change tendency determining means 304, where the weight is added to the cumulative state change tendency calculated each time. Specifically, the accumulated state change tendency determining means 304, to calculate the cumulative change in state probability ratio T _i ^n, to change the above equation (2) as follows.

This increases the probability that an important state change is detected. Significant state changes can be confirmed when the importance check is performed.

Therefore, the example of FIG. 5 can be modified like the process shown in FIG. In this example, it is assumed that the weights are set as w (B ₁ | B ₂ ) = 2.0 and w (B ₂ | B ₁ ) = 1.0. From this setting, it can be seen that the state change from B ₂ to B ₁ is more important. In the embodiment of FIG. 5, since the state change detected by x w _{+ 5} does not pass the significance check not considered state change point. On the other hand, in the example of FIG. 7 in which weights are assigned, the state change from B ₂ to B ₁ detected at _{xw + 5} is an important state change and is accepted as a confirmed state change point. Thus, since the probability that an important state change is detected increases by setting the weight, the accuracy and speed of detection of the important state change can be increased.

  In the above, with reference to the attached drawings, a method and system for detecting a change in the state of an object according to the present invention has been described. The detection method of the present invention is faster, more accurate, and more descriptive than the prior art.

  Although specific embodiments of the present invention have been described with reference to the accompanying drawings, the present invention is not limited to the specific configurations and processes illustrated in the accompanying drawings. In the above description, details of known methods and techniques are omitted for the sake of brevity. Also, in the above examples, some specific steps have been illustrated, but the method and process of the present invention are not limited to the specific steps used in the description and illustration, so those skilled in the art are familiar. For example, once the spirit of the present invention is understood, various modifications, changes and additions can be made, and the order of the steps can be changed.

  Each element of the invention may be implemented as hardware, software, firmware, or a combination thereof and utilized within the system, subsystem, component, or subcomponent. When implemented as software, each element of the present invention is a program or code section for performing necessary tasks. These programs or code sections can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried on a carrier wave. "Machine readable medium" includes any medium that can store or transmit information. Examples of the machine-readable medium include an electronic circuit, a semiconductor storage device, a ROM, a flash memory, an EROM, a floppy disk, a CD-ROM, an optical disk, a hard disk, an optical fiber medium, and an RF link. The code section can be downloaded via a computer network such as the Internet or an intranet.

  The present invention can be implemented in various other forms without departing from the spirit and essential characteristics thereof. For example, the algorithm described in the embodiments can be modified as long as the system architecture does not depart from the basic spirit of the present invention. Accordingly, the above embodiments are considered in all respects to be illustrative and not restrictive. Since the scope of the present invention is defined by the appended claims rather than the foregoing description, any variation or equivalent that falls within the scope of the claims is included in the scope of the invention.

102: Position data stream update means 101-1 to K: Difference test means 103: Parallel state change determination means 104: Latest state change point notification means 301: Position data stream update means 302: Position data reception means 303: Immediate state recording means 304: Cumulative state change tendency determination means 305: State change point detection means 306: Description means 307: Importance check means 308: Latest state change point notification means 309: Storage device 601: State change weight storage device

Claims (13)

A method for detecting a state change of an object from a position data stream,
Recording the previous state of the subject as _BL ;
Receiving position data one by one from the position data stream;
Determining the cumulative state change tendency of the object from the previous state _BL to each state B _i (i = 1, 2,... I) each time position data is received;
Detecting the presence or absence of a state change point in the position data stream based on the determined cumulative state change tendency.   The method according to claim 1, further comprising a step of describing a position of the state change point and a state after the change of the object when the state change point is detected.   The method of claim 1, further comprising performing an importance check on the state change point to check for an important state change point when the state change point is detected. Determining the cumulative state change trend comprises:
Based on a pre-stored state probability model and state change assumption, a log hypothetical check probability ratio R ⁿ (H _i | H ₀ ) is calculated for each received position data x _n by the following equation:

(Here, the state probability model is used to characterize the probability that each location data is mapped to each target state, and the state change assumption H _i (i = 0, 1,… I) characterizes the change in the target state. H ₀ means that the state of the object does not change, and H _i (i = 1,… I) is the state B _i from the previous state B _L to the state of the object.
(i = 1, 2,… I) means)
When receiving the position data x _n, to calculate the cumulative change in state probability ratio T _i ⁿ by the following equation

(Cumulative change in state probability ratio T _i ^n, upon receiving the position data x _n, characterize the accumulated state change tendency of the object to be changed from the previous state B _L to the state B _i)
The method according to claim 1. The detecting step comprises:
From the calculated cumulative state change probability ratio T _i ⁿ (i = 1, 2, ... I) when the position data x _n is received

Find the maximum value T _{i *} ⁿ obtained with
T _{i *} ⁿ
≠ 0
If so, it indicates that there is a state change point in the position data stream. When detecting the state change point, determine the position of the state change point by the following formula,

The method of claim 5, wherein the determining the target state after state change on the basis of the maximum cumulative state change probability ratio T _{i *} state change assumptions corresponding to the ⁿ H _{i *.} Comparing the maximum cumulative state change probability ratio T _{i *} ⁿ with a predetermined threshold;
The method further comprises the step of accepting the corresponding state change point λ ^* if the maximum cumulative state change probability ratio T _{i *} ⁿ exceeds a predetermined threshold, and discarding the state change point otherwise. The method of claim 6. In the step of determining the cumulative state change tendency,
Each state B _i from the previous state B _L
A weight is added to the determined cumulative state change trend of the target to (i = 1, 2,… I) (where the weight reflects the importance of the target state change, the higher the importance of the state change, , Higher weight is added to the state change)
The method according to claim 1. Calculating the cumulative change in state probability ratio T _i ⁿ by the following equation

(W (B _i | B _L ) is the state B _i from the previous state B _L.
(It is the weight of the target state change to i = 1, 2,… I)
The method according to claim 4. A system for detecting a state change of an object from a position data stream,
Immediately preceding state recording means for recording the immediately preceding state of the object as _BL ;
Position data receiving means for receiving position data from the position data stream one by one;
A cumulative state change tendency determination means for determining a target cumulative state change tendency from the previous state _BL to each state B _i (i = 1, 2,... I) each time position data is received;
And a state change point detecting means for detecting the presence / absence of a state change point in the position data stream based on the determined cumulative state change tendency.   11. The apparatus according to claim 10, further comprising description means for describing a position of the state change point and a state after the change of the object when the state change point detection means detects the state change point. System.   In the case where a state change point is detected, in order to check an important state change point, an importance check unit is further provided for executing an importance check on the state change point detected by the state change point detection unit. The system according to claim 10. A state change weight storage means for storing weights corresponding to various state changes of the object;
The more important the state change, the more weight is added to the state change,
The cumulative state change tendency determining means determines a target from the immediately preceding state B _L to each state B _i (i = 1, 2, ... I) using the corresponding weight stored in the state change weight storage means. The system according to claim 10, wherein a weight is added to the cumulative state change tendency.

Images