CN104125337A - Smart phone falling detection and alarming method - Google Patents

Abstract

The invention discloses a smart phone falling detection and alarming method. Positive data samples and false positive data samples are acquired through practical falling tests and approximate falling tests, and further, the samples are trained and tested by using a support vector machine algorithm to obtain a support vector machine model; during use, an acceleration value of an acceleration sensor of a smart phone and an angular acceleration value of a gyroscope are real-timely monitored; when a weighting resultant acceleration variation value and a resultant angular acceleration value of the mobile phone are larger than a preset threshold value, the acceleration value and the angular acceleration value are further calculated according to the support vector machine algorithm, whether calculated result data are positive data or false positive data is judged according to the support vector model obtained through training, and further, whether falling occurs and whether early warning and alarming are started are judged, so that users can be timely helped when falling. According to the smart phone falling detection and alarming method, falling detection and alarming are achieved on the smart phone, a third-party device is not required, and the mobile phone is more convenient to use.

Description

A kind of fall detection of smart mobile phone and alarm method

Technical field

The present invention relates to a kind of detection method, particularly a kind of acceleration transducer and gyrostatic fall detection and alarm method based on smart mobile phone.

Background technology

China just progressively enters aging society, and the elderly's action safety becomes social focal issue day by day.For the problem of Falls in Old People, except taking precautions against, if in the situation that falling, it is also the important measures that reduce Falls in Old People injury that the very first time obtains medical treatment in advance.Along with the transducer of the universal of smart mobile phone and the high progresses that get more and more is integrated in smart mobile phone, carry out fall detection with the smart mobile phone of carrying and there is stronger practicality and social benefit.

Summary of the invention

The present invention, for addressing the above problem, provides a kind of fall detection and alarm method of smart mobile phone, and its fall detection and warning are all to complete on smart mobile phone, and without third party device, and detection is more accurate, more approaches actual conditions.

For achieving the above object, the technical solution used in the present invention is:

A fall detection method for smart mobile phone, comprises the following steps:

10. obtain positive data sample by the test of falling of reality, and obtain false positive data sample by the approximate test of falling;

20. adopt support vector machine algorithm the positive data sample getting and false positive data sample to be carried out to training and testing, supported vector machine model;

The accekeration of the acceleration transducer of 30. Real-Time Monitoring smart mobile phones and gyrostatic angular acceleration values, and calculate current weighting resultant acceleration changing value and close angular acceleration values;

40. judge weighting resultant acceleration changing value and close angular acceleration values whether be greater than predetermined threshold, if be greater than threshold value, continue to record accekeration and the gyrostatic angular acceleration values of acceleration transducer, and according to support vector machine algorithm, described accekeration and angular acceleration values are calculated; If be less than or equal to reservation threshold, return to step 30.

The 50. support vector machine models that obtain according to training in step 20 carry out the whether positive data of calculation result data or the false positive data in determining step 40, and then judge whether to fall.

Preferably, in described step 40, according to support vector machine algorithm, described accekeration and angular acceleration values are calculated, mainly comprise the following steps:

41. when weighting resultant acceleration changing value with close angular acceleration values while whether being greater than predetermined threshold, puts t writing time ₁;

42. continue to record the accekeration of acceleration transducer and calculate its weighting resultant acceleration changing value; When weighting resultant acceleration changing value reaches substantially constant or overturns, put t writing time ₂;

43. continue to record gyrostatic angular acceleration values and calculate it and close angular acceleration values; Level off to 0 when closing angular acceleration values, put t writing time ₃;

44. as time point t ₂with time point t ₃all arrive, carry out following calculating:

441. are greater than the initial time difference Δ t to resultant acceleration substantially constant or reversion of changes of threshold from resultant acceleration ₁, wherein Δ t ₁=(t ₂-t ₁);

442. at Δ t ₁in time period, the mean value of weighting resultant acceleration changing value;

443. at Δ t ₁in time period, the variance of weighting resultant acceleration changing value;

444. is 0 time difference Δ t from closing angular acceleration change initial to closing angular acceleration substantially ₂, wherein Δ t ₂=(t ₃-t ₁);

445. at Δ t ₂in time period, close the mean value of angular acceleration change;

446. at Δ t ₂in time period, close the variance of angular acceleration change.

Preferably, described step 20 is mainly that the false positive data sample that positive data sample by choosing the event of typically falling and daily exercise close on the event of falling is trained, and the ratio of positive data sample and false positive data sample is 3: 1.

Preferably, in described step 20, adopting support vector machine algorithm to carry out training and testing to the positive data sample getting and false positive data sample, is mainly to utilize kernel function that all positive data samples and false positive data sample are mapped to the realization classification of high dimensional feature hypersphere space.

Preferably, the support vector machine algorithm in described step 20 is mainly adopt weighting resultant acceleration changing value and close angular acceleration values as basic data, and chooses lower train value as vector:

20a. changes the initial time difference to resultant acceleration substantially constant or reversion from resultant acceleration, is designated as the very first time poor;

20b. within poor time period very first time, the mean value of weighting resultant acceleration changing value;

20c. within poor time period very first time, the variance of weighting resultant acceleration changing value;

20d. is time difference of 0 from closing angular acceleration change initial to closing angular acceleration substantially, is designated as for the second time difference;

20e., within the time period of the second time difference, closes the mean value of angular acceleration change value;

20f., within the time period of the second time difference, closes the variance of angular acceleration change value.

Preferably, described weighting resultant acceleration changing value with the computational methods of closing angular acceleration values is:

The computing formula of weighting resultant acceleration changing value A is as follows:

$A=\sqrt[2]{w_x\mathrm{\Δ}{a_x}^2+w_y\mathrm{\Δ}{a_y}^2+w_2\mathrm{\Δ}{a_z}^2};$

The computing formula of closing angular acceleration values W is as follows:

$W=\sqrt[2]{{w_x}^2+{w_y}^2+{w_z}^2};$

Wherein, w _x, w _y, w _zto be distributed in x, y, the weight of z axle; Δ a _x, Δ a _y, Δ a _zrefer to the changing value of acceleration on x axle, y axle, z direction of principal axis.

Preferably, described high dimensional feature hypersphere space is to distinguish positive data sample and false positive data sample by defining a hyperspherical surface, and all positive data samples all concentrate in this hypersphere surface.

Preferably, the concrete steps of the positive data sample of described differentiation and false positive data sample further comprise:

21. hypothesis sample set X, X={ χ _i, i=1....l}, high-dimensional feature space by Nonlinear Mapping find centered by vector α, R is the up-to-date hypersphere of radius, the expression formula of its optimization problem is:

‖Φ(χ _i)-α‖ ²≤R ²+ξ _i，ξ _i≥0，i∈[1，l]；

Wherein, F is feature space, and ξ is slack variable, the volume and the sample set that determine the suprasphere of hypersphere definition can be classified number outward at suprasphere, v ∈ (0,1), and l represents sample size;

22. based on KKT condition, introduces kernel function

K(x，y)＝＜φ(x)·Φ(y)＞；

The expression formula of optimization problem is:

$\underset{\mathrm{\λ}}{\min }\underset{i,j}{\mathrm{\Σ}}{\mathrm{\λ}}_i{\mathrm{\λ}}_{j}K({\mathrm{\χ}}_i,{\mathrm{\χ}}_j)-\underset{i}{\mathrm{\Σ}}{\mathrm{\λ}}_{i}K({\mathrm{\χ}}_i,{\mathrm{\χ}}_j);$

Wherein, $0\≤{\mathrm{\λ}}_i\≤\frac{1}{\mathrm{vl}},{\mathrm{\Σ}}_i{\mathrm{\λ}}_i=1;$

Wherein hypersphere center is:

$a=\underset{i}{\mathrm{\Σ}}{\mathrm{\λ}}_i\mathrm{\Φ}\left({\mathrm{\χ}}_i\right);$

23. obtain hyperspherical radius R by training:

$R^2=\underset{i,j}{\mathrm{\Σ}}{\mathrm{\λ}}_i{\mathrm{\λ}}_{j}K({\mathrm{\χ}}_i,{\mathrm{\χ}}_j)+K({\mathrm{\χ}}_s,{\mathrm{\χ}}_s)-2\underset{i}{\mathrm{\Σ}}{\mathrm{\λ}}_{i}K({\mathrm{\χ}}_i,{\mathrm{\χ}}_s);$

24. for any one support vector, and decision function is as follows:

f(χ)＝sgn(R ²-∑ _i，jλ _iλ _jK(χ _i，χ _j)-K(χ，χ)+2∑ _iλ _iK(χ _i，χ _s))；

Kernel function is:

$K(x,z)=\exp (-\frac{{\left|\right|x-z\left|\right|}^2}{{\mathrm{\σ}}^2});$

25. for all samples, if f (χ) > 0, sample belongs to positive data sample, if f (χ) < 0, sample belongs to false positive data sample.

In addition, the present invention also provides a kind of alarm method of falling that adopts above-mentioned smart mobile phone fall detection method, it is characterized in that, comprises the following steps:

Whether 70. detections fall, and if it is start early warning;

80. judge whether pre-warning time exceedes Preset Time, if overtime and user does not cancel early warning, start and report to the police.

Preferably, described step 70 is mainly to fall while occurring when mobile phone detects, will eject early warning interface at mobile phone, plays warning simultaneously and points out and vibrate mobile phone.

Preferably, if in described step 80 in Preset Time user select to cancel early warning; return normal mode; If user does not cancel early warning in Preset Time, start following alarm behavior:

81. obtain Current GPS position, and current GPS positional information is sent to predetermined group of contacts by network or short message;

Default group of contacts is dialed in 82. circulations, until the people that is related connects, after the people that is related connects, plays predefined early warning voice messaging.

The invention has the beneficial effects as follows:

The fall detection of a kind of smart mobile phone of the present invention and alarm method, it is tested by falling of reality and positive data sample and false positive data sample are obtained in the approximate test of falling, and then adopt support vector machine algorithm sample to be carried out to training and testing, supported vector machine model, the accekeration of the acceleration transducer of Real-Time Monitoring smart mobile phone and gyrostatic angular acceleration values when use, when its weighting resultant acceleration changing value with close angular acceleration values while being greater than predetermined threshold, further according to support vector machine algorithm, described accekeration and angular acceleration values are calculated, and the support vector machine model obtaining according to training judges the whether positive data of above-mentioned calculation result data or false positive data, and then judge whether to fall and whether start early warning and warning, thereby user can be succoured in time in the time falling, and fall detection of the present invention and warning are all to complete on smart mobile phone, without third party device, use more convenient.

Brief description of the drawings

Accompanying drawing described herein is used to provide a further understanding of the present invention, forms a part of the present invention, and schematic description and description of the present invention is used for explaining the present invention, does not form inappropriate limitation of the present invention.In the accompanying drawings:

Fig. 1 is the general flow chart of the fall detection method of a kind of smart mobile phone of the present invention;

Fig. 2 is the general flow chart of the alarm method of falling of a kind of smart mobile phone of the present invention.

Embodiment

In order to make technical problem to be solved by this invention, technical scheme and beneficial effect clearer, clear, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.

What fall is to fall and produce to topple over, the duration of falling is comparatively of short duration, than general squatting down, lie down etc. and to want short a lot, the moment of toppling at health, can produce accekeration variation toward the direction of body inclination, health, in the moment of falling, also has the variation of angular acceleration simultaneously.According to this principle, in order to distinguish normal motor behavior and the abnormal behavior of falling, the invention provides a kind of fall detection method of smart mobile phone, it mainly adopts support vector machine algorithm (One-class SVM) to detect the behavior of falling.As shown in Figure 1, comprise the following steps:

10. obtain positive data sample by the test of falling of reality, and obtain false positive data sample by the approximate test of falling;

20. adopt support vector machine algorithm the positive data sample getting and false positive data sample to be carried out to training and testing, supported vector machine model;

The accekeration of the acceleration transducer of 30. Real-Time Monitoring smart mobile phones and gyrostatic angular acceleration values, and calculate current weighting resultant acceleration changing value and close angular acceleration values;

40. judge weighting resultant acceleration changing value and close angular acceleration values whether be greater than predetermined threshold, if be greater than threshold value, continue to record accekeration and the gyrostatic angular acceleration values of acceleration transducer, and according to support vector machine algorithm, described accekeration and angular acceleration values are calculated; If be less than or equal to reservation threshold, return to step 30.

The 50. support vector machine models that obtain according to training in step 20 carry out the whether positive data of calculation result data or the false positive data in determining step 40, and then judge whether to fall.

In described step 40, according to support vector machine algorithm, described accekeration and angular acceleration values are calculated, mainly comprise the following steps:

41. when weighting resultant acceleration changing value with close angular acceleration values while whether being greater than predetermined threshold, puts t writing time ₁;

42. continue to record the accekeration of acceleration transducer and calculate its weighting resultant acceleration changing value; When weighting resultant acceleration changing value reaches substantially constant or overturns, put t writing time ₂;

43. continue to record gyrostatic angular acceleration values and calculate it and close angular acceleration values; Level off to 0 when closing angular acceleration values, put t writing time ₃;

44. as time point t ₂with time point t ₃all arrive, carry out following calculating:

441. are greater than the initial time difference Δ t to resultant acceleration substantially constant or reversion of changes of threshold from resultant acceleration ₁, wherein Δ t ₁=(t ₂-t ₁);

442. at Δ t ₁in time period, the mean value of weighting resultant acceleration changing value;

443. at Δ t ₁in time period, the variance of weighting resultant acceleration changing value;

444. is 0 time difference Δ t from closing angular acceleration change initial to closing angular acceleration substantially ₂, wherein Δ t ₂=(t ₃-t ₁);

445. at Δ t ₂in time period, close the mean value of angular acceleration change;

446. at Δ t ₂in time period, close the variance of angular acceleration change.

In the present embodiment, described step 20 is mainly that the false positive data sample that positive data sample by choosing the event of typically falling and daily exercise close on the event of falling is trained, and the ratio of positive data sample and false positive data sample is 3: 1.

In described step 20, adopting support vector machine algorithm to carry out training and testing to the positive data sample getting and false positive data sample, is mainly to utilize kernel function that all positive data samples and false positive data sample are mapped to the realization classification of high dimensional feature hypersphere space; Support vector machine algorithm in described step 20 is mainly adopt weighting resultant acceleration changing value and close angular acceleration values as basic data, and chooses lower train value as vector:

20a. changes the initial time difference to resultant acceleration substantially constant or reversion from resultant acceleration, is designated as the very first time poor;

20b. within poor time period very first time, the mean value of weighting resultant acceleration changing value;

20c. within poor time period very first time, the variance of weighting resultant acceleration changing value;

20d. is time difference of 0 from closing angular acceleration change initial to closing angular acceleration substantially, is designated as for the second time difference;

20e., within the time period of the second time difference, closes the mean value of angular acceleration change value;

20f., within the time period of the second time difference, closes the variance of angular acceleration change value.

Described weighting resultant acceleration changing value with the computational methods of closing angular acceleration values is:

The computing formula of weighting resultant acceleration changing value A is as follows:

$A=\sqrt[2]{w_x\mathrm{\&Delta;}{a_x}^2+w_y\mathrm{\&Delta;}{a_y}^2+w_2\mathrm{\&Delta;}{a_z}^2};$

The computing formula of closing angular acceleration values W is as follows:

$W=\sqrt[2]{{w_x}^2+{w_y}^2+{w_z}^2};$

Wherein, w _x, w _y, w _zto be distributed in x, y, the weight of z axle; Δ a _x, Δ a _y, Δ a _zrefer to the changing value of acceleration on x axle, y axle, z direction of principal axis.

Described high dimensional feature hypersphere space is to distinguish positive data sample and false positive data sample by defining a hyperspherical surface, and all positive data samples all concentrate in this hypersphere surface, and concrete steps further comprise:

21. hypothesis sample set X, X={ χ _i, i=1....l}, high-dimensional feature space by Nonlinear Mapping find centered by vector α, R is the up-to-date hypersphere of radius, the expression formula of its optimization problem is:

‖Φ(χ _i)-α‖ ²≤R ²+ξ _i，ξ _i≥0，i∈[1，l]；

Wherein, F is feature space, and ξ is slack variable, the volume and the sample set that determine the suprasphere of hypersphere definition can be classified number outward at suprasphere, v ∈ (0,1), and l represents sample size;

22. based on KKT condition, introduces kernel function

K(x，y)＝＜φ(x)·Φ(y)＞；

The expression formula of optimization problem is:

$\underset{\mathrm{\&lambda;}}{\min }\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{\mathrm{\&lambda;}}_{j}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j)-\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_{i}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j);$

Wherein, $0\&le;{\mathrm{\&lambda;}}_i\&le;\frac{1}{\mathrm{vl}},{\mathrm{\&Sigma;}}_i{\mathrm{\&lambda;}}_i=1;$

Hyperspherical center is:

$a=\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i\mathrm{\&Phi;}\left({\mathrm{\&chi;}}_i\right);$

23. can obtain hypersphere radius R by training:

$R^2=\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{\mathrm{\&lambda;}}_{j}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j)+K({\mathrm{\&chi;}}_s,{\mathrm{\&chi;}}_s)-2\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_{i}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_s);$

24. for any one support vector, and decision function is as follows:

f(χ)＝sgn(R ²-∑ _i，jλ _iλ _jK(χ _i，χ _j)-K(χ，χ)+2∑ _iλ _iK(χ _i，χ _s))；

Kernel function is:

$K(x,z)=\exp (-\frac{{\left|\right|x-z\left|\right|}^2}{{\mathrm{\&sigma;}}^2});$

25. for all samples, if f (χ) > 0, sample belongs to positive data sample, if f (χ) < 0, sample belongs to false positive data sample.

As shown in Figure 2, the present invention also provides a kind of alarm method of falling that adopts above-mentioned smart mobile phone fall detection method, it is characterized in that, comprises the following steps:

Whether 70. detections fall, and if it is start early warning; Fall while occurring when mobile phone detects, will eject early warning interface at mobile phone, play warning simultaneously and point out and vibrate mobile phone.

80. judge whether pre-warning time exceedes Preset Time, if overtime and user does not cancel early warning, start and report to the police; If user selects to cancel early warning in Preset Time, return normal mode; If user does not cancel early warning in Preset Time, start following alarm behavior:

81. obtain Current GPS position, and current GPS positional information is sent to predetermined group of contacts by network or short message;

Default group of contacts is dialed in 82. circulations, until the people that is related connects, after the people that is related connects, plays predefined early warning voice messaging.

The present invention can effectively distinguish positive data and false positive data, thereby reduction False Rate, for user provides reliable service, it is accuracy of judgement not only, and response fast, and also increase early warning step detecting after falling, also can cancel early warning even if produce erroneous judgement, saved unnecessary trouble to user.

Above-mentioned explanation illustrates and has described the preferred embodiments of the present invention, as front, be to be understood that the present invention is not limited to disclosed form herein, should not regard the eliminating to other embodiment as, and can be used for various other combinations, amendment and environment, and can, in invention contemplated scope herein, change by technology or the knowledge of above-mentioned instruction or association area.And the change that those skilled in the art carry out and variation do not depart from the spirit and scope of the present invention, all should be in the protection range of claims of the present invention.

Claims (11)

1. a fall detection method for smart mobile phone, is characterized in that, comprises the following steps:

10. obtain positive data sample by the test of falling of reality, and obtain false positive data sample by the approximate test of falling;

20. adopt support vector machine algorithm the positive data sample getting and false positive data sample to be carried out to training and testing, supported vector machine model;

The accekeration of the acceleration transducer of 30. Real-Time Monitoring smart mobile phones and gyrostatic angular acceleration values, and calculate current weighting resultant acceleration changing value and close angular acceleration values;

40. judge weighting resultant acceleration changing value and close angular acceleration values whether be greater than predetermined threshold, if be greater than threshold value, continue to record accekeration and the gyrostatic angular acceleration values of acceleration transducer, and according to support vector machine algorithm, described accekeration and angular acceleration values are calculated; If be less than or equal to reservation threshold, return to step 30.

The 50. support vector machine models that obtain according to training in step 20 carry out the whether positive data of calculation result data or the false positive data in determining step 40, and then judge whether to fall.

2. the fall detection method of a kind of smart mobile phone according to claim 1, is characterized in that: in described step 40, according to support vector machine algorithm, described accekeration and angular acceleration values are calculated, mainly comprise the following steps:

41. when weighting resultant acceleration changing value with close angular acceleration values while whether being greater than predetermined threshold, puts t writing time ₁;

42. continue to record the accekeration of acceleration transducer and calculate its weighting resultant acceleration changing value; When weighting resultant acceleration changing value reaches substantially constant or overturns, put t writing time ₂;

43. continue to record gyrostatic angular acceleration values and calculate it and close angular acceleration values; Level off to 0 when closing angular acceleration values, put t writing time ₃;

44. as time point t ₂with time point t ₃all arrive, carry out following calculating:

441. are greater than the initial time difference Δ t to resultant acceleration substantially constant or reversion of changes of threshold from resultant acceleration ₁, wherein Δ t ₁=(t ₂-t ₁);

442. at Δ t ₁in time period, the mean value of weighting resultant acceleration changing value;

443. at Δ t ₁in time period, the variance of weighting resultant acceleration changing value;

444. is 0 time difference Δ t from closing angular acceleration change initial to closing angular acceleration substantially ₂, wherein Δ t ₂=(t ₃-t ₁);

445. at Δ t ₂in time period, close the mean value of angular acceleration change;

446. at Δ t ₂in time period, close the variance of angular acceleration change.

3. the fall detection method of a kind of smart mobile phone according to claim 1, it is characterized in that: described step 20 is mainly that the false positive data sample that positive data sample by choosing the event of typically falling and daily exercise close on the event of falling is trained, and the ratio of positive data sample and false positive data sample is 3: 1.

4. the fall detection method of a kind of smart mobile phone according to claim 1, it is characterized in that: in described step 20, adopting support vector machine algorithm to carry out training and testing to the positive data sample getting and false positive data sample, is mainly to utilize kernel function that all positive data samples and false positive data sample are mapped to the realization classification of high dimensional feature hypersphere space.

5. the fall detection method of a kind of smart mobile phone according to claim 4, it is characterized in that: the support vector machine algorithm in described step 20 is mainly adopt weighting resultant acceleration changing value and close angular acceleration values as basic data, and choose lower train value as vector:

20a. changes the initial time difference to resultant acceleration substantially constant or reversion from resultant acceleration, is designated as the very first time poor;

20b. within poor time period very first time, the mean value of weighting resultant acceleration changing value;

20c. within poor time period very first time, the variance of weighting resultant acceleration changing value;

20d. is time difference of 0 from closing angular acceleration change initial to closing angular acceleration substantially, is designated as for the second time difference;

20e., within the time period of the second time difference, closes the mean value of angular acceleration change value;

20f., within the time period of the second time difference, closes the variance of angular acceleration change value.

6. according to the fall detection method of a kind of smart mobile phone described in claim 1 or 2 or 3 or 4 or 5, it is characterized in that: described weighting resultant acceleration changing value with the computational methods of closing angular acceleration values is:

The computing formula of weighting resultant acceleration changing value A is as follows:

$A=\sqrt[2]{w_x\mathrm{\&Delta;}{a_x}^2+w_y\mathrm{\&Delta;}{a_y}^2+w_2\mathrm{\&Delta;}{a_z}^2};$

The computing formula of closing angular acceleration values W is as follows:

$W=\sqrt[2]{{w_x}^2+{w_y}^2+{w_z}^2};$

Wherein, w _x, w _y, w _zto be distributed in x, y, the weight of z axle; Δ a _x, Δ a _y, Δ a _zrefer to the changing value of acceleration on x axle, y axle, z direction of principal axis.

7. the fall detection method of a kind of smart mobile phone according to claim 4, it is characterized in that: described high dimensional feature hypersphere space is to distinguish positive data sample and false positive data sample by defining a hyperspherical surface, and all positive data samples all concentrate in this hypersphere surface.

8. the fall detection method of a kind of smart mobile phone according to claim 7, is characterized in that: the concrete steps of the positive data sample of described differentiation and false positive data sample further comprise:

21. hypothesis sample set X, X={ χ _i, i=1....l}, high-dimensional feature space by Nonlinear Mapping find centered by vector α, R is the up-to-date hypersphere of radius, the expression formula of its optimization problem is:

‖Φ(χ _i)-α‖ ²≤R ²+ξ _i，ξ _i≥0，i∈[1，l]；

Wherein, F is feature space, and ξ is slack variable, the volume and the sample set that determine the suprasphere of hypersphere definition can be classified number outward at suprasphere, v ∈ (0,1), and l represents sample size;

22. based on KKT condition, introduces kernel function

K(x，y)＝＜φ(x)·Φ（y)＞；

The expression formula of optimization problem is:

$\underset{\mathrm{\&lambda;}}{\min }\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{\mathrm{\&lambda;}}_{j}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j)-\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_{i}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j);$

Wherein, $0\&le;{\mathrm{\&lambda;}}_i\&le;\frac{1}{\mathrm{vl}},{\mathrm{\&Sigma;}}_i{\mathrm{\&lambda;}}_i=1;$

Hyperspherical center α is:

$a=\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i\mathrm{\&Phi;}\left({\mathrm{\&chi;}}_i\right);$

23. by training, can obtain hyperspherical radius R:

$R^2=\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{\mathrm{\&lambda;}}_{j}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_j)+K({\mathrm{\&chi;}}_s,{\mathrm{\&chi;}}_s)-2\underset{i}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_{i}K({\mathrm{\&chi;}}_i,{\mathrm{\&chi;}}_s);$

24. for any one support vector, and decision function is as follows:

f(χ)＝sgn(R ²-∑ _i，jλ _iλ _jK(χ _i，χ _j)-K(χ，χ)+2∑ _iλ _iK(χ _i，χ _s))；

Kernel function is:

$K(x,z)=\exp (-\frac{{\left|\right|x-z\left|\right|}^2}{{\mathrm{\&sigma;}}^2});$

25. for all samples, if f (χ) > 0, sample belongs to positive data sample, if f (χ) < 0, sample belongs to false positive data sample.

9. the alarm method of falling that adopts the smart mobile phone fall detection method of the claims 1 to 8, is characterized in that, comprises the following steps:

Whether 70. detections fall, and if it is start early warning;

80. judge whether pre-warning time exceedes Preset Time, if overtime and user does not cancel early warning, start and report to the police.

10. the alarm method of falling of a kind of smart mobile phone according to claim 9, is characterized in that: described step 70 is mainly to fall while occurring when mobile phone detects, will eject early warning interface at mobile phone, plays warning simultaneously and points out and vibrate mobile phone.

The fall detection of 11. a kind of smart mobile phones according to claim 9 and alarm method, is characterized in that: if in described step 80 in Preset Time user select to cancel early warning; return normal mode; If user does not cancel early warning in Preset Time, start following alarm behavior:

81. obtain Current GPS position, and current GPS positional information is sent to predetermined group of contacts by network or short message;

Default group of contacts is dialed in 82. circulations, until the people that is related connects, after the people that is related connects, plays predefined early warning voice messaging.