TW201711002A - Detection method for determining fall by applying triaxial accelerometer and time sequence having advantages of sending a warning signal before the fall, performing real-time detection to improve the security, reducing the false determination rate and being easy to implement due to its simple circuit - Google Patents

Abstract

The present invention is characterized in that an acceleration information capturing unit collects three-axis acceleration vectors of a plurality of real-time triaxial accelerations of a user and respectively calculates a plurality of low-pass acceleration intensities (t) to be supplied into a shift register of a fall detection determination unit, when the low-pass acceleration intensities (t) fall within a threshold range thereof, the output of a comparator is +1 and the output of a processing unit is +<alpha>, and when the low-pass acceleration intensities (t) do not fall within its threshold range, the output of the comparator is -1, and the output of the processing unit is -<alpha>; and finally, the outputs of the N processing units are output to an adder for summation, when a result is greater than zero, the adder generates a fall determination signal, otherwise, no fall determination signal is generated. This invention has advantages of sending a warning signal before the fall, performing real-time detection to improve the security, reducing the false determination rate and being easy to implement due to its simple circuit.

Description

Fall detection method using three-axis accelerometer and time series judgment

The invention relates to a fall detection method using a three-axis accelerometer and a time series judgment, in particular, a warning signal can be issued before a fall, an instant detection can improve safety, a false positive rate can be reduced, and a circuit is simple and simple to implement. A three-axis accelerometer and a time series judgment fall detection method.

There are many fall detection methods. These methods can be divided into two types: image recognition and sensor detection. Most of the sensor detection methods use the three-axis acceleration information worn by the guard to determine whether there is any The fall condition occurs. When the fall occurs, the three-axis accelerometer worn by the guardian causes the triaxial acceleration information to change drastically due to the instantaneous collision, and then the fall threshold detection is completed after the single threshold value comparison judgment. This method is susceptible to noise and The influence of posture leads to misjudgment.

Since the process of falling is a continuous time action, the fall detection method of the Republic of China invention patent No. I419085 captures continuous multiple acceleration data, and substitutes the continuous multiple acceleration data into a fall comparison unit to determine whether the fall is satisfied. Condition to generate a fall detection signal. However, the method must be judged once at intervals, and it cannot be judged continuously and continuously. The empty window period of the interval is the personal danger of falling.

The point is that these methods are detected after the fall, and it is impossible to avoid the damage. It can only reduce the damage. If it can detect the fall early in the fall process, then Taking protective measures will directly prevent the occurrence of injuries and achieve the purpose of fall prediction.

In view of this, it is necessary to develop a technique that can solve the above disadvantages.

The object of the present invention is to provide a fall detection method using a three-axis accelerometer and a time series judgment, which can simultaneously issue a warning signal before the fall, improve the security by instant detection, reduce the false positive rate, and simplify the circuit. Implementation and other advantages. In particular, the problem to be solved by the present invention is that the conventional fall detection method belongs to the detection after the fall occurs, and the damage cannot be avoided, and the damage degree can only be reduced, and the fall cannot be detected early in the fall process, and then the protection measures are taken. It will be able to directly avoid the occurrence of injuries and achieve the purpose of the fall prediction.

The technical means for solving the above problem is to provide a fall detection method using a three-axis accelerometer and time series judgment, which includes: Preparation steps; two. Detection step; and three. Fall judgment step.

The above objects and advantages of the present invention will be readily understood from the following detailed description of the preferred embodiments illustrated herein.

The invention will be described in detail in the following examples in conjunction with the drawings:

11‧‧‧Preparation steps

12‧‧‧Detection steps

13‧‧‧ falls judgment step

20‧‧‧Acceleration information acquisition unit

21‧‧‧Three-axis accelerometer

30‧‧‧ Fall Detection Unit

31‧‧‧Processing unit

32‧‧‧Adder

311‧‧‧Shift register

312‧‧‧ Comparator

313‧‧‧ threshold value register

314‧‧‧weight value register

40‧‧‧Acceleration threshold threshold decision unit

90‧‧‧Users

K1, K2, K3, K4, K5, K6, K7, K8, K9, K10, K11‧‧‧ detection points

A‧‧‧First detection section

B‧‧‧Second detection section

C‧‧‧ third detection section

D‧‧‧Four detection section

E‧‧‧ fifth detection section

F‧‧‧Sixth detection section

G‧‧‧Seventh detection section

The first figure is a flow chart of the present invention

The second drawing is a schematic diagram of an application example of the present invention.

The third A diagram is a system block diagram of the present invention

The third B diagram is a schematic diagram of the fall detection judging unit of the present invention

The fourth figure is a waveform diagram of a plurality of three-axis acceleration vectors of the instant three-axis acceleration of the present invention.

The fifth figure is a waveform diagram of the intensity of the plurality of triaxial acceleration values of the present invention.

The sixth figure is a waveform diagram of a plurality of low-pass acceleration intensities of the present invention.

The seventh figure is a waveform diagram of the plurality of detection processes of the present invention.

The eighth picture is the data chart of the seventh series according to the time series sampling by batch.

The ninth A is a schematic diagram of the training set of the present invention for collecting 100 samples of falling samples and 100 pens without falling samples for threshold values.

Figure 9 is a schematic diagram of the error rate of Figure 9A.

The tenth A is a schematic diagram of a test set for collecting a fall sample of 20 pens and a non-falling sample 20 pens to perform a threshold range value.

Figure 10B is a schematic diagram of the error rate of Figure 10A.

Referring to the first, second, third, and third B diagrams, the present invention is a fall detection method using a three-axis accelerometer and time series judgment, and includes the following steps after starting: The preparation step 11: preparing an acceleration information capturing unit 20, a fall detection determining unit 30, and an acceleration threshold value determining unit 40, the acceleration information capturing unit 20 is internally built with a mathematical formula: And being used for loading on a user 90; the fall detection determining unit 30 includes N serially connected processing units 31 and an adder 32, and N is a positive integer of ≧5; each processing unit 31 The system includes a shift register 311, a comparator 312, a threshold value register 313, and a weight value register 314; the threshold value register 313 has a threshold value θ _N built therein (see the table). First, the threshold range value θ _N is between 63.5 and 74.5), and the weighting register 314 has a weight value α _N built therein (refer to Table 1, the weight value α _{N is} set to 1); Detecting step 12: The acceleration information capturing unit 20 is provided with a three-axis accelerometer 21 for collecting a plurality of three-axis acceleration vectors of the instantaneous triaxial acceleration of the user 90 [ x ( t ), y ( t ) , z ( t )] ^T (as shown in the fourth figure), and then the intensity m ( t ) of the plurality of triaxial acceleration values (as shown in the fifth figure) is processed by a low-pass filter to become a plurality of low Passing acceleration ( t ) (as shown in Figure 6), used as a basis for judgment of fall detection; Fall judgment step 13: the low-pass acceleration intensity that is supplied to the shift register 312 when the fall detection determination unit 30 supplies ( t ) falls within the threshold range value θ _N , the comparator 312 outputs +1, the processing unit 31 outputs + α ; and the low-pass acceleration is immediately supplied to the shift register 312 strength ( t ) does not fall within the threshold range value θ _N , then the comparator 312 outputs -1, the processing unit 31 outputs -α ; finally the output of the N processing units 31 is re-inputted to the adder 32 When the result is greater than zero, the adder 32 generates a judgment of the fall signal. When the result is less than zero, the adder 32 does not generate a judgment of the fall signal; wherein the shift register 312 of the N processing units 30, Sustained N consecutive continuous low-pass accelerations ( t ) the comparator 312 sequentially feeding the N series of processing units 30 is compared with the threshold range value θ _N , so that the N -pass continuous low-pass acceleration intensity is simultaneously applied at any time. ( t ) Perform a comparison to determine if it has fallen.

In practice, in the detecting step 12, the low-pass filtering process is selected from one of two ways:

The first type: the low-pass filtering process obtains the average number of low-pass accelerations by taking the intensity m ( t ) of the instantaneous plurality of triaxial acceleration values as an average value thereof ( t ).

The second type: the low-pass filtering process obtains the plurality of low-pass acceleration intensities by filtering the high-frequency portion of the intensity m ( t ) of the plurality of three-axis acceleration values in the instant ( t ).

The processing unit 31 is provided with at least five.

Referring to the seventh figure, for example, assuming that the device is actually measured, the number N of the processing units 31 is reduced to five for convenience of illustration. Sampling 10 times per second, that is, sampling every 0.1 seconds (of course, this sampling frequency can also be modified to 100 times per second or other values), between the 5.5th and 7.0 seconds, there are 16 detection points (K1, K2) K3, K4, K5, K6, K7, K8, K9, K10, K11, ...), can be organized into a data map as shown in the eighth figure (sample by batch in time series, including a first detection section A a second detecting section B, a third detecting section C, a fourth detecting section D, a fifth detecting section E, a sixth detecting section F and a seventh detecting The detection process of the five processing units 31 is as follows:

[a] First detection section A (see table 1 below):

Capture the low-pass acceleration of the first to fifth detection points (K1~K5) ( t ), which are: 74, 76, 74, 73 and 74, the discriminating process is as follows: Low-pass acceleration intensity of the first detection point (K1) ( t ) is 74, within the threshold θ _N (73.5~74.5), the comparator 312 outputs +1; the low-pass acceleration of the second detection point (K2) ( t ) is 76, within the range of entry threshold θ _N (72.5~73.5), the comparator 312 outputs -1; the low-pass acceleration intensity of the third detection point (K3) ( t ) is 74, within the threshold of entry threshold θ _N (68.5~69.5), the comparator 312 outputs -1; the low-pass acceleration of the fourth detection point (K4) ( t ) is 73, within the range of entry threshold θ _N (66.5~67.5), the comparator 312 outputs -1; the low-pass acceleration intensity of the fifth detection point (K5) ( t ) is 74, within the range of the threshold value θ _N (63.5~64.5), the comparator 312 outputs -1; then the outputs of the five processing units 31 are re-inputted to the adder 32, and the result is obtained. The adder 32 does not generate a judgment fall signal for +1-1-1-1-1=-3 (<0).

[b]Second detection section B (see Table 2 below):

Capture the low-pass acceleration of the second to sixth detection points (K2~K6) ( t ), which are: 76, 74, 73, 74, and 73, respectively. The discriminating process is as follows: the low-pass acceleration intensity of the second detecting point (K2) ( t ) is 76, within the range of entry threshold θ _N (73.5~74.5), the comparator 312 outputs -1; the low-pass acceleration of the third detection point (K3) ( t ) is 74, within the range of the entry threshold θ _N (72.5~73.5), the comparator 312 outputs -1; the low-pass acceleration of the fourth detection point (K4) ( t ) is 73, within the range of the entry threshold θ _N (68.5~69.5), the comparator 312 outputs -1; the low-pass acceleration of the fifth detection point (K5) ( t ) is 74, within the range of the entry threshold θ _N (66.5~67.5), the comparator 312 outputs -1; the low-pass acceleration of the sixth detection point (K6) ( t ) is 73, the falling threshold value θ _N (63.5~64.5), the comparator 312 outputs -1; then the outputs of the five processing units 31 are re-inputted to the adder 32, and the result is For -1-1-1-1-1=-5 (<0), the adder 32 does not generate a judgment of the fall signal.

The principle of the third detecting section E to the seventh detecting section G is as described above, and is not described here. Only the fifth detecting section E indicates that the adder 32 generates a judgment of the falling signal:

[c] Fifth detection section E (see Table 3 below):

Capture the low-pass acceleration of the pens from the fifth to the ninth detection points (K5~K9) ( t ), which are: 74, 73, 69, 67, and 64, respectively, and the discrimination process is as follows: the low-pass acceleration intensity of the fifth detection point (K5) ( t ) is 74, within the threshold θ _N (73.5~74.5), the comparator 312 outputs +1; the low-pass acceleration of the sixth detection point (K6) ( t ) is 73, within the threshold θ _N (72.5~73.5), the comparator 312 outputs +1; the low-pass acceleration of the seventh detection point (K7) ( t ) is 69, within the threshold θ _N (68.5~69.5), the comparator 312 outputs +1; the low-pass acceleration of the eighth detection point (K8) ( t ) is 67, within the threshold θ _N (66.5~67.5), the comparator 312 outputs +1; the low-pass acceleration of the ninth detection point (K9) ( t ) is 64, within the falling threshold range θ _N (63.5~64.5), the comparator 312 outputs +1; then the outputs of the five processing units 31 are re-inputted to the adder 32, and the result is +1+1+1+1+1=5 (>0), the adder 32 generates a judgment fall signal.

Assume a plurality of low-pass acceleration intensities ( t ) The numerical system is in the range of 10.5 to 6.5, and assumes the low-pass acceleration intensity ( t ) A value of 7 series of fall collisions has occurred, collecting 100 pens of falling samples, 10 sampling points per pen, and 100 pens without falling samples, each with 10 sampling points, see Figure 9A, The error rate of the training process for the threshold of the present invention is that when there is only one comparator at the beginning, the error rate is 0.1250, and then when two are used, the error rate is also 0.1250, and then the comparison is performed to three, four, and five. The error rate is 0.0250, 0.0750, and 0 (finally compiled into ninth B). Please refer to the 10th and 10th B pictures, which are 40 tests (including 20 falls and 20 non-falls). When using five comparators, the error comes to 0.1, which is 90% correct. This means that as long as half of the sampling points are used, for example, at time t-5, it is predicted that a fall will occur at time t, and a fall prediction will be completed in advance.

The advantages and functions of the present invention are as follows:

[1] A warning signal can be issued before a fall. The invention can issue a warning signal before the user falls to reduce the risk of personal fall. Therefore, a warning signal can be issued before the fall.

[2] Instant detection improves security. The invention continuously detects in real time, instead of detecting every predetermined time, which can reduce the risk of falling due to detecting the empty window period. Therefore, instant detection improves security.

[3] can reduce the false positive rate. The invention draws the continuous multi-stroke low-pass acceleration intensity and judges through the continuous multi-door threshold comparison to complete the fall detection judgment. Therefore, the false positive rate can be reduced.

[4] The circuit is simple and easy to implement. The fall detection judging unit of the present invention is provided with a plurality of serially connected processing units and an adder. Each processing unit only designs a shift register, a comparator, a threshold register, and a weight register. Overall, there is not a large number of complicated circuits. Therefore, the circuit is simple and easy to implement.

The present invention has been described in detail with reference to the preferred embodiments of the present invention, without departing from the spirit and scope of the invention.

11‧‧‧Preparation steps

12‧‧‧Detection steps

13‧‧‧ falls judgment step

Claims (3)

A fall detection method using a three-axis accelerometer and time series judgment includes: Preparing steps: preparing an acceleration information capturing unit, a fall detection determining unit, and an acceleration threshold determining unit, wherein the acceleration information capturing unit has a built-in mathematical formula: And the device is mounted on a user; the fall detection judging unit includes N serial processing units and an adder, and N is a positive integer of ≧5; each processing unit includes a shift a bit register, a comparator, a threshold register, and a weight value register; the threshold value register has a threshold value θ _N built in the weight register, and the weighting register has a built-in weight value α _N ; two. Detection step: the acceleration information acquisition unit is provided with a three-axis accelerometer for collecting the three-axis acceleration vectors of the user's multiple instantaneous triaxial accelerations [ x ( t ), y ( t ), z ( t ) ] ^T, and equation are transmitted through the operation into a plurality of three-axis acceleration intensity value m (t), then the intensity of the plurality of three-axis acceleration values of m (t) through a low-pass filter processing, a complex low- Passing acceleration ( t ), used as a basis for judgment of fall detection; a fall determination step: the low-pass acceleration intensity supplied to the shift register by the fall detection judging unit ( t ) falls within the threshold range value θ _N , then the comparator output is +1, the processing unit outputs + α ; and the low-pass acceleration intensity immediately supplied to the shift register ( t ) does not fall within the threshold range value θ _N , then the comparator output is -1, the processing unit output is -α ; finally the output of the N processing units is re-inputted to the adder, when the result is greater than Zero, the adder generates a judgment of the fall signal, and when the result is less than zero, the adder does not generate a judgment of the fall signal; wherein the shift register in the N processing units is capable of continuously continuing N consecutive complex numbers Pen low-pass acceleration intensity ( t ) sequentially comparing the comparators in the N series of processing units with the threshold range value θ _N , so that the N -continuous low-pass acceleration intensity is simultaneously for N consecutive times ( t ) Perform a comparison to determine if it has fallen. The fall detection method using the three-axis accelerometer and the time series judgment as described in claim 1, wherein in the detecting step, the low-pass filter processing is the strength of the plurality of triaxial acceleration values m ( t ) takes the average value to obtain the plurality of low-pass acceleration intensities ( t ). The fall detection method using the three-axis accelerometer and the time series judgment as described in claim 1, wherein in the detecting step, the low-pass filter processing is the strength of the plurality of triaxial acceleration values after m (t) of which filter out the high frequency component obtained by the plurality of low-pass acceleration intensity ( t ).