JP2012017988A - Event detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a highly accurate event detection device which can distinguish a difference between statuses of a person or an object that exists there, by monitoring the status of a radio wave.SOLUTION: The event detection device includes: a plurality of antennas 21 for receiving a radio wave transmitted from a transmitter; correlation matrix arithmetic operation means 22 for arithmetically operating a correlation matrix from a received vector, with a signal received by the plurality of the antennas 21 as the received vector; characteristic vector arithmetic operation means 23 for arithmetically operating a characteristic vector of stretching a signal part space, by developing the correlation matrix arithmetically operated by the correlation matrix arithmetic operation means 22 as a characteristic value; an SVM (support vector machine) 24 for discriminating the status, by inputting the characteristic vector operated arithmetically by the characteristic vector arithmetic operation means 23; and event-detecting means 25 for detecting event in consideration of the continuity of the output of the SVM 24.

Description

  According to the present invention, in a predetermined area such as a bathroom, a transmitter that transmits radio waves and a receiver that receives radio waves transmitted from the transmitter are arranged, and a person loses consciousness based on the reception characteristics of the radio waves. The present invention relates to an event detection device that detects an event such as a fall or a fall.

The present inventors have proposed the following event detection device (see Patent Document 1).
A plurality of antennas for receiving radio waves transmitted by the transmitter, a correlation matrix calculating means for calculating a correlation matrix from the received vector using signals received by the plurality of antennas as reception vectors, and a calculation performed by the correlation matrix calculating means An event detection apparatus comprising: eigenvector computing means for computing eigenvectors that expand a signal subspace by expanding eigenvalues of a correlation matrix; and event detection means for detecting an event by detecting a temporal change of the eigenvector computed by the eigenvector computing means.

JP 2008-216152 A

  However, in the event detection device described above, even though a physically distinct phenomenon such as the presence or absence of an object can be identified, it exists there, such as whether the person is simply stationary or falls. It was not possible to have enough precision to identify the state of the person who is. Such a thing can be easily identified visually, but a person has to be monitored with care, and in situations where there is psychological resistance to being monitored such as in a bathroom, There was a problem that monitoring would not stop.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide an event detection device with high accuracy capable of discriminating a difference in the state of a person or an object present by monitoring the state of radio waves. And

  The event detection apparatus of the present invention includes a plurality of antennas that receive radio waves transmitted by a transmitter, a correlation matrix calculation unit that calculates a correlation matrix from the reception vectors using signals received by the plurality of antennas as reception vectors, Eigenvector calculation means for calculating eigenvectors that expand the eigenvalues of the correlation matrix calculated by the correlation matrix calculation means and extending the signal subspace, and a support vector machine function for determining an event by inputting the eigenvector calculated by the eigenvector calculation means And event detection means for detecting an event based on the event determined by the support vector machine function.

  In addition, the support vector machine function distinguishes between a normal bathing state and a loss of consciousness in a bathtub in the bathroom. Thus, it is possible to detect abnormality abnormally and reliably.

  In addition, the event detection unit can detect an abnormal state with higher accuracy by detecting an event based on the continuity of the event determined by the support vector machine function.

  ADVANTAGE OF THE INVENTION According to this invention, the event detection apparatus with the high precision which can also identify the difference in the state of the person and thing which exists there by monitoring the state of an electromagnetic wave can be provided.

It is a figure which shows the structure of the event detection apparatus by one Example of this invention. It is a figure which shows the environment where the experiment of the present Example was performed in the conference room. It is a figure which shows the example (the 1) of the evaluation result in a meeting room. It is a figure which shows the example (the 2) of the evaluation result in a meeting room. It is a figure which shows the example (the 3) of the evaluation result in a meeting room. It is a figure which shows the environment which performed the experiment of the present Example in the bathroom. It is a figure which shows the example of the evaluation result which identified each state of "static state" "moving" "during bathing" "losing consciousness" in a bathroom. It is a figure which shows the example of the evaluation result which identified each state of "static state" "moving" "washing a shower and head" "falling" in the bathroom. It is a figure which shows the example of the evaluation result which identified each state of "an empty state" and "suspicious behavior" around a passenger car.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a diagram showing a configuration of an event detection apparatus according to an embodiment of the present invention. The event detection apparatus of this embodiment includes a transmitter 10 and a receiver 20. The transmitter 10 and the receiver 20 are installed in a predetermined area in order to detect an event such as a change in the state of a person. A closed space such as a bathroom is desirable, but it may be an open area. The transmitter 10 transmits radio waves. The receiver 20 includes an array antenna 21, a correlation matrix calculation unit 22, an eigenvector calculation unit 23, an SVM 24, and an event detection unit 25. The array antenna 21 includes a plurality of antenna elements, and each antenna element receives a radio wave transmitted by the transmitter 10. Here, an example will be described in which the antenna elements are arranged on a straight line. The reception signal of the array antenna 21 is represented by a reception vector x → (t) having the reception signal of each array antenna 21 as an element. Here, “→” indicates that the left character in the sentence is a vector.
x → (t) = a → (θ) s (t) + n → (t) (1)
However, a → (θ): L-dimensional vector when the number of antenna elements is L s (t): received signal at the reference point n → (t): noise

（２）
ただし、θ：アンテナアレイ２１の並びの方向に対する電波到来方向
ｄ：アレイアンテナ２１の各素子の間隔
λ：電波の波長
ここで、Ｍ個の到来波が平面波として到来するとき、
ｘ→(ｔ)＝Ａ→ｓ→(ｔ)＋ｎ→(ｔ) （３）
ただし、Ａ→：Ｍ個のベクトル（ステアリングベクトルという）を列としたＬ×Ｍ行列
ｓ→(ｔ)：各到来波の複素振幅を要素としたＭ次元ベクトル

(2)
Where θ is the direction of arrival of radio waves relative to the direction in which the antenna array 21 is arranged d: the distance between the elements of the array antenna 21 λ: wavelength of the radio waves where M incoming waves arrive as plane waves,
x → (t) = A → s → (t) + n → (t) (3)
However, A →: L × M matrix having M vectors (referred to as steering vectors) as columns s → (t): M-dimensional vector having complex amplitude of each incoming wave as an element

（４）

(4)

（５）
ただし、Ｔは転置を表す。
と表せる。

(5)
However, T represents transposition.
It can be expressed.

  The correlation matrix calculation means 22 calculates a correlation matrix R → xx from the received vector x → (t).

（６）
ただし、Ｅ→［・］：集合平均
Ｈ：複素共役転置
ここで、雑音は到来波と無関係であり、素子に独立であるので、

(6)
However, E → [·]: Collective average H: Complex conjugate transposition Here, noise is independent of the incoming wave, and is independent of the element.

（７）
ただし、σ：雑音の分散
Ｓ→：波源相関行列＝Ｅ→［ｓ→(ｔ)ｓ→(ｔ)^Ｈ］
また、

(7)
However, σ: Noise variance S →: Wave source correlation matrix = E → [s → (t) s → (t) ^H ]
Also,

（８）
から得られる固有値λi、それに対応する固有ベクトルｖ→iを用いて、

(8)
Using the eigenvalue λi obtained from Eq. And the corresponding eigenvector v → i,

（９）
と固有値展開できる。ここで、

(9)
And eigenvalue expansion. here,

（１０）
である。ここでdiagは行列の対角要素を並べたものである。

(10)
It is. Here, diag is an array of diagonal elements of a matrix.

  Here, the eigenvalues of the received data correlation matrix R → xx are K signal eigenvalues corresponding to the sum of the number of coherent wave groups and incoherent waves, and LK noise eigenvalues whose magnitude is equal to the noise power. Can be divided into That is,

（１１）

(11)

  From the above, it was shown that the correlation matrix R → xx generated from the received vector can be divided into a signal subspace and a noise subspace by expanding eigenvalues. The eigenvector v → and the steering vector a → (θ) that span the signal subspace span the same space and can be expressed as the other linear combination. That is, the eigenvector spanning the signal subspace can be expressed by a linear combination of steering vectors including the direction of arrival information, and can be said to represent a radio wave propagation structure.

  Here, the first eigenvector v → 1 is an eigenvector corresponding to λ1 indicating the highest eigenvalue, and is always the basis of the signal subspace as long as the signal reaches the receiver.

（１２）
と表せる。お互いがコヒーレントである波が到来した場合はそのステアリングベクトルの線形結合が新しい１つのステアリングベクトルとなるので、上式の本質には影響しない。したがって、第一固有ベクトルはマルチパス環境の信号空間を表し、伝搬環境によって一意に決まる。そこで、固有ベクトル演算手段２３は、相関行列Ｒ→xxから第１固有ベクトルｖ→1を算出する。

(12)
It can be expressed. When waves that are coherent with each other arrive, the linear combination of the steering vectors becomes a new steering vector, so that the essence of the above equation is not affected. Therefore, the first eigenvector represents the signal space of the multipath environment and is uniquely determined by the propagation environment. Therefore, the eigenvector computing means 23 calculates the first eigenvector v → 1 from the correlation matrix R → xx.

  Here, the inner product of the first eigenvector v → none acquired in advance when the event has not occurred and the first eigenvector v → ob acquired during the event detection observation is used as the evaluation function P (t).

（１３）
とする。ただし固有ベクトルの大きさはどちらも１に正規化しておく。

(13)
And However, both eigenvector sizes are normalized to 1.

  Since the propagation environment does not change at the observation time when no event occurs, v → ob (tnone) shows a value very close to v → none, and is close to 1. On the other hand, at the observation time t = tevent when the event occurs, the propagation environment changes, and v → ob (tevent) shows a value different from v → none, and thus becomes a value smaller than 1.

  The SVM 24 is a known support vector machine (Support Vector Machine) and is one of identification methods using learning. In the present embodiment, an evaluation function P (t) for simulating each state to be identified (a plurality) is input to the SVM 24, learned in advance, and an evaluation function P (t for actual observation. ) Is input to the SVM 24 to determine an assumed state, that is, an event.

  FIG. 2 is a diagram illustrating an environment in which the experiment of the present example was performed in a conference room. A transmitter Tx and a receiver Rx were installed in the conference room, and the SVM 24 learned by simulating normal human behavior and falls.

  The event detection unit 25 in FIG. 1 detects a final event from the events determined by the SVM 24. The data of the eigenvector that spans the signal subspace used as the feature quantity for state identification is time-series data, and event detection is performed in consideration of the correlation between the state and time. In other words, the event detection result is not taken every momentary time, but is taken into consideration for the correlation between the state and the time every predetermined time width. Specifically, the time when “every hour” is measured is about 0.1 seconds, and the “time width” is 30 times the measured time (0.1 × 30 = 3 seconds). Thirty evaluation functions are input to the SVM 24 as feature vectors.

  Next, in response to the determination result of the SVM 24, the event detection means 25 performs processing with the following algorithm in order to eliminate a false alarm (determining that it is not an abnormal state but an abnormal state).

(1). First, values before and after the address to be checked (i.e., i) for checking a false alarm are used as reference values.
Example: {i-1, i, i + 1, i + 2}
(2). The previous value (i-1) and the subsequent value (i + 1, i + 2) of the i-th are the same, but if the previous value (i-1) and the i-th value are not the same, it is determined as a false value. The value of (i-1) is substituted for the i-th value.
(3). Unlike the previous value (i-1), even if the value of i is "abnormal", if the value of (i + 1) or (i + 2) is "normal", it is judged as a false alarm. The value of (i−1) is substituted for the i-th value. As a result, “normal” is assumed as long as the “abnormal” state does not continue for nine seconds (three times). For example, when a person falls down, it can be judged that a state in which the person has not fallen continuously (such as immediately after falling down, judging that the person has fallen due to noise or a device error) is normal. .
(4). While i is incremented by one, check the above (2) and (3).

  By the processing of the above algorithm, the event detection means 25 identifies “normal” (normal behavior) and “abnormal” (falling), and outputs a signal corresponding to each.

  3, 4, and 5 are diagrams illustrating examples of evaluation results in the conference room. Each figure (a) shows the horizontal axis: time (seconds), the vertical axis: the evaluation function P (t), and each figure (b) shows the horizontal axis: time (seconds), the vertical axis: “true "Label" (solid line) and "judgment label" (circled line) are shown. Label “−1” was normal behavior, and label “1” was fallen. As shown in the figure, it is considered difficult to accurately identify normal behavior and falls by the evaluation function P (t). However, according to the present example, the determination label completely matched the true label.

  FIG. 6 is a diagram showing an environment in which the experiment of this example was performed in a bathroom. A transmitter Tx and a receiver Rx were installed in the bathroom.

  FIG. 7 is a diagram illustrating an example of an evaluation result that identifies each state of “static state”, “moving”, “during bathing”, and “losing consciousness” in the bathroom. FIG. 7 (a) shows the horizontal axis: time (seconds), the vertical axis: evaluation function P (t), and FIG. 7 (b) shows the horizontal axis: time (seconds), the vertical axis: “true "Label" (solid line) and "judgment label" (circled line) are shown. Label “1” is static (washing area), label “2” is moving (washing area), label “3” is bathing (in the bathtub), and label “4” is unconscious (in the bathtub) . As shown in the figure, it is considered difficult to accurately identify these states by the evaluation function P (t). However, according to the present example, the determination label completely matched the true label. Thereby, it becomes possible to detect abnormalities mechanically and reliably in an environment where psychological resistance exists in visual observation of a bathroom or monitoring by a camera.

  FIG. 8 is a diagram illustrating an example of an evaluation result identifying each state of “static state”, “moving”, “washing the shower and head”, and “falling” in the bathroom. FIG. 8A shows the horizontal axis: time (seconds), the vertical axis: evaluation function P (t), and FIG. 8B shows the horizontal axis: time (seconds), the vertical axis: “true "Label" (solid line) and "judgment label" (circled line) are shown. Label “1” was in a static state (washing place), label “2” was moved (washing place), label “3” was washed with a shower and head (washing place), and label “4” was overturned (washing place). As shown in the figure, it is considered difficult to accurately identify these states by the evaluation function P (t). However, according to the present example, the determination label completely matched the true label.

  FIG. 9 is a diagram illustrating an example of an evaluation result identifying each state of “nothing state” and “suspicious behavior” in the vicinity of the passenger car. FIG. 9A shows the horizontal axis: time (seconds), the vertical axis: evaluation function P (t), and FIG. 9B shows the horizontal axis: time (seconds), the vertical axis: “true "Label" (solid line) and "judgment label" (circled line) are shown. Suspicious behavior A (dotted line) is “person's suspicious behavior when there are no cars in the vicinity (look into, hang around, etc.)”, and suspicious behavior B (solid line) is “ Peeking, wandering, etc.) ". The label “−1” is regarded as nothing, and the label “1” is regarded as suspicious behavior. As shown in the figure, it is considered difficult to accurately identify these states by the evaluation function P (t). However, according to the present example, the determination label completely matched the true label.

In addition, this invention is not limited to the said Example.
As long as the transmitter can generate radio waves and can be received by the array antenna, the transmitter used in other systems can be used in combination. For example, it corresponds to a wireless LAN base station. The signal may be a wide band or a narrow band.

  The antenna may be an antenna composed of a plurality of elements, and is not necessarily an array antenna.

  The eigenvector computing means may compute a plurality of eigenvectors, and is not necessarily limited to computing only the eigenvector corresponding to the maximum eigenvalue of the correlation matrix. Further, the present invention is not limited to the one that takes the inner product of eigenvectors, and for example, a difference or a ratio may be taken.

  In the above-described embodiment, the state is identified in consideration of the correlation between the state and the time, but this is not necessary unless high accuracy is required. Further, it may be considered in the input of SVM.

  In the above embodiment, the inner product of the first eigenvector is input to the SVM. However, the first eigenvector may be input as it is or an eigenvalue may be input depending on the application field.

DESCRIPTION OF SYMBOLS 10 Transmitter 20 Receiver 21 Array antenna 22 Correlation matrix calculation means 23 Eigenvector calculation means 24 SVM (support vector machine)
25 Event detection means 51 White board 52 Door 53 Window

Claims (3)

Multiple antennas to receive the radio waves transmitted by the transmitter,
Correlation matrix computing means for computing a correlation matrix from the received vector using signals received by the plurality of antennas as received vectors;
Eigenvector computing means for computing eigenvectors that expand the signal subspace by expanding the correlation matrix computed by the correlation matrix computing means;
A support vector machine function for inputting an eigenvector computed by the eigenvector computing means and discriminating an event;
An event detection apparatus comprising: event detection means for detecting an event based on an event determined by the support vector machine function.   The event detection apparatus according to claim 1, wherein the support vector machine function distinguishes between a normal bathing state and a loss of consciousness in a bathtub in a bathroom. The event detection device according to claim 1, wherein the event detection unit detects an event based on continuity of events determined by the support vector machine function.

Images