Sleeping Posture Recognition Method for Smart Pillow Based on Probability Estimation

Abstract

The present invention discloses a method for recognizing sleeping postures of a smart pillow based on probability estimation. The smart pillow includes a bottom plate, a weighing sensor is arranged on the upper part of the bottom plate, an airbag is arranged on the upper part of the weighing sensor, a sponge is arranged on the upper part of the airbag, the airbag is connected to an air inlet valve and a pressure sensor through a Y-shaped air pipe, the air inlet valve is connected to an air pump through an air pipe, and the weighing sensor, the pressure sensor, the air inlet valve and the air pump are connected to a control circuit. The control circuit sets a method for recognizing sleeping postures, including the steps of: setting flat lying weight parameters and side lying weight parameters P1, W1, W2, P2, W3, W4; when the airbag pressure P = P1, based on the measured weight w i and Δw i , using the Bayesian formula to estimate the probability pi of lying on the back and the probability p i ～ of lying on the side. When p i < p i ～ and |p i ～ - p i |> K, it is determined that the sleeping posture is lying on the side; in the same way, when calculating the airbag measured pressure P = P2, the probability p j of lying on the back and the probability p j ～ of lying on the side are calculated. When p j > p j ～ and |p j - p j ～ |> K, it is determined that the sleeping posture is lying on the back.

Description

Technical Field

[0001] This patent relates to a method for recognizing sleeping postures of an intelligent pillow based on probability estimation, belonging to the technical fields of pattern recognition and artificial intelligence. Background Art

[0002] Pillows have been accompanying the development of human civilization. The materials have evolved from stones and woods to plant fibers and plant seeds. Nowadays, ordinary pillows are generally made of textiles, including cotton, linen, and chemical fibers. However, the function of pillows is still limited to providing a support for people in a sleeping state to keep the body in a relaxed state. Now people's requirements are not only that, but also hope that the height of the pillows they use can be adjusted to adapt to different sleeping postures. In order to adjust the height of the pillow according to the sleeping posture, it is necessary to accurately detect whether the user is lying flat or on the side. The currently used methods include detecting the pressure distribution of the user's head on the pillow or judging by the magnitude of the pressure of the head on the pillow. However, the measurement of the weight or pressure of the head is different from the measurement of body weight, and there are many influencing factors, resulting in various interferences in the measurement data. Judging based on such data will lead to frequent misjudgments.

[0003] In view of this, this patent is specifically proposed. Summary of the Invention

[0004] In view of the above problems, this patent provides a method for recognizing sleeping postures of an intelligent pillow based on probability estimation. Probability estimation is performed according to the currently measured and historical weights and the weight change values to judge the user's sleeping posture, and then the height of the pillow is adjusted, overcoming the drawback that judging based on the currently measured data is prone to frequent misjudgments.

[0005] The technical solution adopted by this patent to solve its technical problems is as follows:

[0006] A method for recognizing sleeping postures of an intelligent pillow based on probability estimation. The intelligent pillow includes a bottom plate for support. A weighing sensor is arranged on the upper part of the bottom plate. An airbag is arranged on the upper part of the weighing sensor. A sponge for supporting the head is arranged on the upper part of the airbag. The outside of the bottom plate, the weighing sensor, the airbag, and the sponge is wrapped with a pillowcase. The airbag is connected to an intake valve and a pressure sensor through a Y-shaped air pipe. The intake valve is connected to an air pump through an air pipe. The weighing sensor, the pressure sensor, the intake valve, and the air pump are connected to a control circuit. The control circuit measures the weight borne by the intelligent pillow through the weighing sensor, measures the air pressure value of the airbag through the pressure sensor, inflates and deflates the airbag through the intake valve and the air pump, and the control circuit is used to implement the sleeping posture recognition method. The sleeping posture recognition method includes the following steps:

[0007] (1) Set the flat-lying weight parameter and the side-lying weight parameter: Measure the weight W1 and air pressure P1 when the user feels comfortable in the flat-lying state, keep the air pressure at P1, and measure the weight W2 when the user is in the side-lying position; Measure the weight W4 and air pressure P2 when the user feels comfortable in the side-lying state, keep the air pressure at P2, and measure the weight W3 when the user is in the flat-lying position.

[0008] (2) During the steady-state operation process, that is, when the intake valve and the air pump are in the closed state and the measured air pressure P of the airbag is P1, the measured weight w of the weighing sensor i , i = 0, 1, 2, 3...., calculate p(w i |A), p(w i |~A):

[0009] is the probability that the weight is w i under the condition of the flat-lying state A when the air pressure P = P1;

[0010] is the probability that the weight is w i under the condition of the side-lying state ~A when the air pressure P = p1;

[0011] Calculate p(A|~A, Δw i ), p(~A|A, Δw i ), p(A|A, Δw i ), p(~A|~A, Δw i ):

[0012] Δw i = w i - w i-1 , w i-1 is the measured weight of the previous measurement period;

[0013] is the probability of converting from the side-lying state ~A to the flat-lying state A when the air pressure P = P1 and the weight change is Δw i ;

[0014] is the probability of converting from the supine state A to the side-lying state ~A when the air pressure P = P1 and the weight change is Δw i ;

[0015] is the probability that the sleeping position state does not change when the air pressure P = P1 and the weight change is Δw i ;

[0016] Then, use Bayes' formula to estimate the posterior probability:

[0017] Supine probability p i= p(A|w i , Δw i , w i-1 , Δw i-1 ...)

[0018] = p(w i |A)·[p(A|~A, Δw i )·p i-1 ～ + p(A|A, Δw i )·p i-1 / η i ;

[0019] Side - lying probability p i ～ = p(~A|w i , Δw i , w i-1 , Δw i-1 ...)

[0020] = p(w i |~A)·[p(~A|A, Δw i )·p i-1 + p(~A|~A, Δw i )·p i-1 ～ / η i , where η i is the normalization coefficient,

[0021] η i = p(w i |A)·[p(A|~A, Δw i )·p i-1 ～ + p(A|A, Δw i )·p i-1 + p(w i |~A)·[p(~A|A, Δw i )·p i-1 + p(~A|~A, Δw i )·p i-1 ～ ,

[0022] For the first - time estimation, i = 0,

[0023] p0 = p(A|w0)

[0024] = p(w0|A)·p(A) / η0;

[0025] p0 ～ = p(~A|w0)

[0026] = p(w0|~A)·p(~A) / η0,

[0027] where p(A) = p(~A) = 0.5,

[0028] η0 = p(w0|A)·0.5 + p(w0|~A)·0.5;

[0029] When p i < p i ～ and |p i ～ - p i | > K, where K is a judgment threshold, it is judged that the sleeping posture is lying on the side, the control circuit opens the intake valve, and starts the inflator pump. When P = P2 is satisfied, the intake valve and the inflator pump are closed;

[0030] (3) During the steady-state operation process, that is, when the intake valve and the inflator pump are in the closed state and the measured air pressure P of the airbag = P2, the measured weight w of the weighing sensor i , j = 0, 1, 2, 3...., calculate p(w j |A), p(w j |~A):

[0031] is the probability of the weight being w j under the condition of lying flat state A when the air pressure P = P2;

[0032] is the probability of the weight being w j under the condition of lying on the side state ~A when the air pressure P = P2;

[0033] Calculate p(A|~A, Δw j ), p(~A|A, Δw j ), p(A|A, Δw j ), p(~A|~A, Δw j ):

[0034] Δw j = w j - w j-1 , w j-1 is the measured weight of the previous measurement period;

[0035] is the probability of converting from the lying-on-the-side state ~A to the lying-flat state A when the air pressure P = P2 and the weight change is Δw j ;

[0036] is when the air pressure P = P2 and the weight change is Δw jThe probability of transitioning from the supine state A to the side-lying state ~A;

[0037] When the air pressure P = P2, the weight change is Δw j The probability that the sleeping posture state does not change;

[0038] Then, use Bayes' formula to estimate the posterior probability:

[0039] The supine probability p j = p(A|w j , Δw j , w j-1 , Δw j-1 ...)

[0040] = p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ +p(A|A, Δw j )·p j-1 / η j ;

[0041] The side-lying probability p j ～ = p(~A|w j , Δw j , w j-1 , Δw j-1 ...)

[0042] = p(w j |~A)·[p(~A|A, Δw j )·p j-1 +p(~A|~A, Δw j )·p j-1 ～ / η j , where η j is the normalization coefficient,

[0043] η j = p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ +p(A|A, Δw j )·p j-1 +p(w j |~A)·[p(~A|A, Δw j )·p j-1 +p(~A|~A, Δw j )·p j-1 ～ ,

[0044] For the first estimation, j = 0,

[0045] p0 = p(A|w0)

[0046] = p(w0|A)·p(A) / η0;

[0047] p0 ～ = p(~A|w0)

[0048] = p(w0|~A)·p(~A) / η0

[0049] where p(A) = p(~A) = 0.5

[0050] η0 = p(w j |A)·0.5 + p(w j |~A)·0.5;

[0051] When p j > p j ～ and |p j - p j ～ |> K, it is determined that the sleeping position is lying on the back, and the control circuit opens the intake valve. When P = P1 is satisfied, the intake valve is closed.

[0052] The beneficial effects of this patent are mainly manifested in: probability estimation is carried out based on the current and historical measured weights and the weight change values, which enhances the accuracy of sleeping position judgment, filters out measurement errors, and avoids the oscillation of the output height of the intelligent pillow caused by single measurement errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic structural diagram of the present invention;

[0054] Figure 2 is a schematic diagram of probability distribution modeling based on weight parameters of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0055] The present patent will be further described below with reference to the accompanying drawings:

[0056] As Figure 1-2 shown, for the intelligent pillow sleeping position recognition method based on probability estimation, the intelligent pillow includes a bottom plate 1 for support, a weighing sensor 3 is arranged on the upper part of the bottom plate 1, an airbag 4 is arranged on the upper part of the weighing sensor 3, a sponge 2 for supporting the head is arranged on the upper part of the airbag 4, and the outside of the bottom plate 1, weighing sensor 3, airbag 4 and sponge 2 is wrapped with a pillowcase 5. The weighing sensor 3 is used to measure the weight applied by the head on the intelligent pillow.

[0057] The described airbag 4 changes the height of the intelligent pillow by inflating or deflating, achieving the best support for the head and neck. The airbag 4 is connected to an intake valve and a pressure sensor through a Y-shaped air tube, and the intake valve is connected to an air pump through an air tube. When the intake valve is opened and the air pump works, the airbag 4 is inflated; when the intake valve is opened and the air pump does not work, the airbag 4 is deflated. The pressure sensor can measure the air pressure value in the airbag 4, which is in a proportional relationship with the height of the intelligent pillow.

[0058] For centralized control, the weighing sensor 3, the pressure sensor, the intake valve and the air pump are connected to a control circuit. The control circuit measures the weight borne by the intelligent pillow through the weighing sensor 3, measures the air pressure value of the airbag 4 through the pressure sensor, and inflates and deflates the airbag 4 through the intake valve and the air pump. Therefore, on the premise of correctly judging the user's sleeping posture, the control circuit can adjust the height of the intelligent pillow to improve the quality of the user's sleep.

[0059] The control circuit is used to implement a sleeping posture recognition method, and the sleeping posture recognition method includes the following steps:

[0060] (1) Set the flat-lying weight parameter and the side-lying weight parameter: Measure the weight W1 and air pressure P1 when the user feels comfortable in the flat-lying state, keep the air pressure at P1, and measure the weight W2 when the user is lying on the side; Measure the weight W4 and air pressure P2 when the user feels comfortable in the side-lying state, keep the air pressure at P2, and measure the weight W3 when the user is lying flat.

[0061] The weights W1, W2, W3, and W4 are characteristic parameters of the user lying flat and on the side when the intelligent pillow is in the high and low states, and are used to establish a probability distribution model of the sleeping posture.

[0062] (2) During the steady-state working process, that is, when the intake valve and the air pump are in the closed state and the measured air pressure P of the airbag 4 = P1, the measured weight w of the weighing sensor 3 i , i = 0, 1, 2, 3...., using the standard normal probability distribution, calculate p(w i |A), p(w i |~A):

[0063] is the probability of the weight being w i under the condition of the flat-lying state A when the air pressure p = p1;

[0064] is the probability of the weight being w iProbability;

[0065] Similarly, calculate p(A|~A, Δw i ), p(~A|A, Δw i ), p(A|A, Δw i ), p(~A|~A, Δw i ) using the standard normal probability distribution:

[0066] Δw i = w i - w i-1 , where w i-1 is the measured weight in the previous measurement period. When the user's sleeping position changes, Δw i will change: when changing from lying flat to lying on the side, Δw i is negative; when changing from lying on the side to lying flat, Δw i is positive;

[0067] is the probability of changing from the side-lying state ~A to the lying-flat state A when the weight change is Δw i at the air pressure p = p1;

[0068] is the probability of changing from the lying-flat state A to the side-lying state ~A when the weight change is Δw i at the air pressure p = p1;

[0069] is the probability that the sleeping position state does not change when the weight change is Δw i at the air pressure P = P1;

[0070] Then, use Bayes' formula to estimate the posterior probability:

[0071] The probability of lying flat p i = p(A|w i , Δw i , w i-1 , Δw i-1 ...)

[0072] = p(w i |A, Δw i , w i-1 , Δw i-1 ...)·p(A, Δw i , w i-1 , Δw i-1 ...) / p(w i , Δw i , w i-1 , Δw i-1 ...)

[0073] Since the weight data measured before and after are uncorrelated, then

[0074] p(w i |A, Δw i , w i-1 , Δw i-1 ...) = p(w i |A), and according to the total probability formula, we get

[0075] p(A, Δw i , w i-1 , Δw i-1 ...)

[0076] = p(A|Δw i , w i-1 , Δw i-1 ...)·p(Δw i , w i-1 , Δw i-1 ...)

[0077] = [p(A|~A, Δw i , w i-1 , Δw i-1 ...)·p(~A|w i-1 , Δw i-1 ...)+p(A|A, Δw i , w i-1 , Δw i-1 ...)·p(A|w i-1 , Δw i-1 ...)]·p(Δw i , w i-1 , Δw i-1 ...)

[0078] = [p(A|~A, Δw i )·p(~A|w i-1 , Δw i-1 ...)+p(A|A, Δw i )·p(A|w i-1 , Δw i-1 ...)]·p(Δw i , w i-1 , Δw i-1 ...)

[0079] = [p(A|~A, Δw i )·p i-1 ～ +p(A|A, Δw i )·p i-1 ·p(Δw i , w i-1 , Δwi-1 ...)

[0080] For supine and lateral lying, p(w i , Δw i , w i-1 , Δw i-1 ...) and p(Δw i , w i-1 , Δw i-1 ...) are the same. Therefore, p i = p(w i |A)·[p(A|~A, Δw i )·p i-1 ～ + p(A|A, Δw i )·p i-1 / η i , where η i is the normalization coefficient;

[0081] Similarly, the lateral lying probability p

[0082] can be obtained as i ～ = p(~A|w i , Δw i , w i-1 , Δw i-1 ...)

[0083] = p(w i |~A)·[p(~A|A, Δw i )·p i-1 + p(~A|~A, Δw i )·p i-1 ～ / η i ,

[0084] Therefore,

[0085] η i = p(w i |A)·[p(A|~A, Δw i )·p i-1 ～ + p(A|A, Δw i )·p i-1 + p(w i |~A)·[p(~A|A, Δw i )·p i-1 + p(~A|~A, Δw i )·p i-1 ～ ;

[0086] For the first estimation, i = 0, and there is no historical measurement data. Therefore,

[0087] p0 = p(A|w0)

[0088] = p(w0|A)·p(A) / p(w0)

[0089] = p(w0|A)·p(A) / η0;

[0090] Similarly,

[0091] p0 ～ = p(~A|w0)

[0092] = p(w0|~A)·p(~A) / η0

[0093] where p(A) = p(~A) = 0.5, indicating that the probabilities of the user lying flat and lying on the side are the same during the first estimation.

[0094] η0 = p(w0|A)·0.5 + p(w0|~A)·0.5;

[0095] Finally, based on the values of p i and p i ～ make a sleeping posture judgment: when p i < p i ～ and |p i ～ - p i | > K, judge that the sleeping posture is lying on the side. The control circuit opens the intake valve and starts the inflator pump. When P = P2 is satisfied, close the intake valve and the inflator pump, where K is the judgment threshold. Increasing the value of K can increase the judgment stability, but the response speed decreases;

[0096] (3) During the steady-state operation process, that is, when the intake valve and the inflator pump are in the closed state and the measured air pressure P of the airbag 4 = P2, the measured weight w of the load cell 3 j , j = 0, 1, 2, 3...., adopt the standard normal probability distribution to calculate p(w j |A), p(w j |~A):

[0097] is the probability of the weight being w j under the condition of lying flat state A when the air pressure P = P2;

[0098] is the probability of the weight being w j under the condition of lying on the side state ~A when the air pressure P = P2;

[0099] Similarly, calculate p(A|~A, Δw j ), p(~A|A, Δw j ), p(A|A, Δw j ), p(~A|~A, Δw j ) using the standard normal probability distribution:

[0100] Δw j = w j - w j-1 , where w j-1 is the measured weight in the previous measurement cycle. When the user's sleeping position changes, Δw j will change: when changing from lying flat to lying on the side, Δw j is negative; when changing from lying on the side to lying flat, Δw j is positive;

[0101] is the probability of changing from the side-lying state ~A to the lying-flat state A when the weight change is Δw j at air pressure P = P2;

[0102] is the probability of changing from the lying-flat state A to the side-lying state ~A when the weight change is Δw j at air pressure P = P2;

[0103] is the probability that the sleeping position state does not change when the weight change is Δw j at air pressure P = P2;

[0104] Then, use Bayes' formula to estimate the posterior probability:

[0105] The probability of lying flat p j = p(A|w j , Δw j , w j-1 , Δw j-1 ...)

[0106] = p(w j |A, Δw j , w j-1 , Δw j-1 ...)·p(A, Δw j , w j-1 , Δw j-1 ...) / p(w j , Δw j , w j-1 , Δw j-1 ...)

[0107] Since the weight data measured before and after are uncorrelated, then

[0108] p(w j |A, Δw j , w j-1 , Δw j-1 ...) = p(w j |A), and according to the law of total probability, we get p(A, Δw j , w j-1 , Δw j-1 ...)

[0109] = p(A|Δw j , w j-1 , Δw j-1 ...)·p(Δw j , w j-1 , Δw j-1 ...)

[0110] = [p(A|~A, Δw j , w j-1 , Δw j-1 ...)·p(~A|w j-1 , Δw j-1 ...)+ p(A|A, Δw j , w j-1 , Δw j-1 ...)·p(A|w j-1 , Δw j-1 ...)]·p(Δw j , w j-1 , Δw j-1 ...)

[0111] = [p(A|~A, Δw j )·p(~A|w j-1 , Δw j-1 ...)+ p(A|A, Δw j )·p(A|w j-1 , Δw j-1 ...)]·p(Δw j , w j-1 , Δw j-1 ...)

[0112] = [p(A|~A, Δw j )·p j-1 ～ + p(A|A, Δw j )·p j-1 ·p(Δw j , w j-1 , Δw j-1 ...),

[0113] For supine and lateral lying, p(w j , Δw j , w j-1 , Δw j-1 ...) and p(Δw j , w j-1 , Δw j-1 ...) are the same. Therefore, p j = p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ + p(A|A, Δw j )·p j-1 / η j , where η j is the normalization coefficient;

[0114] Similarly, the lateral lying probability p

[0115] can be obtained as j ～ = p(~A|w j , Δw j , w j-1 , Δw j-1 ...)

[0116] = p(w j |~A)·[p(~A|A, Δw j )·p j-1 + p(~A|~A, Δw j )·p j-1 ～ / η j ,

[0117] Therefore, η

[0118] is obtained as j = 1 / {p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ + p(A|A, Δw j )·p j-1 + p(w j |~A)·[p(~A|A, Δw j )·p j-1 + p(~A|~A, Δw j )·p j-1 ～},

[0119] For the first estimate, j = 0 and there is no historical measurement data. Therefore,

[0120] p0 = p(A|w0)

[0121] = p(w0|A)·p(A) / p(w0)

[0122] = p(w0|A)·p(A) / η0

[0123] Similarly,

[0124] p0 ～ = p(~A|w0)

[0125] = p(w0|~A)·p(~A) / η0

[0126] where p(A) = p(~A) = 0.5, indicating that at the first estimation, the probabilities of the user lying on the back and lying on the side are the same;

[0127] η0 = 1 / [p(w j |A)·0.5 + p(w j |~A)·0.5].

[0128] Finally, based on the values of p j and p j ～ perform sleeping posture judgment: when p j > p j ～ and |p j - p j ～ | > K, judge that the sleeping posture is lying on the back, and the control circuit opens the intake valve. When P = P1 is satisfied, close the intake valve.

[0129] In summary, the solution disclosed in the present invention performs probability estimation based on the currently measured and historical weights and the weight change value, judges the user's sleeping posture, and then adjusts the pillow height to improve the supporting effect of the pillow on the head and neck. This solution filters out measurement errors, enhances the accuracy of sleeping posture judgment, and solves the problem of frequent misjudgment caused by only relying on currently measured data for judgment.

[0130] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A smart pillow sleep posture recognition method based on probability estimation. The smart pillow includes a bottom plate for support. A weighing sensor is provided on the upper part of the bottom plate. An airbag is provided on the upper part of the weighing sensor. A sponge for supporting the head is provided on the upper part of the airbag. The outside of the bottom plate, weighing sensor, airbag and sponge is wrapped with a pillowcase. The airbag is connected to an air inlet valve and a pressure sensor through a Y-shaped air pipe. The air inlet valve is connected to an air pump through an air pipe. The weighing sensor, pressure sensor, air inlet valve and air pump are connected to a control circuit. The control circuit measures the weight borne by the smart pillow through the weighing sensor, measures the air pressure value of the airbag through the pressure sensor, and inflates and deflates the airbag through the air inlet valve and the air pump. It is characterized in that: The described control circuit is used to implement a sleeping posture recognition method, and the sleeping posture recognition method includes the following steps: (1) Set the flat-lying weight parameter and the side-lying weight parameter: Measure the weight W1 and air pressure P1 when the user feels comfortable in the flat-lying state, keep the air pressure as P1, and measure the weight W2 when the user is side-lying; Measure the weight W4 and air pressure P2 when the user feels comfortable in the side-lying state, keep the air pressure as P2, and measure the weight W3 when the user is flat-lying; (2) During the steady-state operation process, that is, when the intake valve and the charging pump are in the closed state and the measured air pressure P of the airbag is P1, the measured weight w of the weighing sensor i , i = 0, 1, 2, 3...., calculate p(w i |A), p(w i |~A): is the probability that, under condition A in the lying state when the atmospheric pressure P = P1, the weight is w i ; The probability when lying on the side in the state of air pressure P = P1 ~ under condition A with weight w i ; Calculate p(A|~A,Δw i ), p(~A|A,Δw i ), p(A|A,Δw i ), p(~A|~A,Δw i ): Δw i = w i - w i-1 ,where w i-1 is the measured weight in the previous measurement cycle; When the air pressure P = P1, the weight change is Δw i The probability of converting from the side-lying state to the supine state A in this case; When the air pressure P = P1, the weight change is Δw i The probability of transitioning from the supine state A to the side-lying state ~A in this case; When the air pressure P = P1, the weight change is Δw i The probability that the sleeping posture state does not change under the circumstances; Then, use Bayes' formula to estimate the posterior probability: Flat probability p i = p(A|w i , Δw i , w i-1 , Δw i-1 ...) = p(w i |A)·[p(A|~A, Δw i )·p i-1 ～ + p(A|A, Δw i )·p i-1 / η i ; Lying-on-side probability p i ～ = p(~A|w i ,Δw i ,w i-1 ,Δw i-1 ...) = p(w i |~A)·[p(~A|A, Δw i )·p i-1 + p(~A|~A, Δw i )·p i-1 ～ / η i , where η i is the normalization coefficient, η i = p(w i |A)·[p(A|~A,Δw i )·p i-1 ～ +p(A|A,Δw i )·p i-1 +p(w i |~A)·[p(~A|A,Δw i )·p i-1 +p(~A|~A,Δw i )·p i-1 ～ , For the first estimation, i = 0, p0 = p(A|w0) = p(w0|A)·p(A) / η0; p0 ～ = p(~A|w0) = p(w0|~A)·p(~A) / η0, where p(A) = p(~A) = 0.5, η0 = p(w0|A)·0.5 + p(w0|~A)·0.5; When p i <p i ～ And |p i ～ - p i | > K, where K is a judgment threshold value, it is judged that the sleeping posture is lying on the side, the control circuit opens the intake valve and starts the inflator pump, and when P = P2 is satisfied, the intake valve and the inflator pump are closed; (3) During the steady-state operation process, that is, when the intake valve and the charging pump are in the closed state and the measured air pressure P of the airbag is P2, the measured weight w of the weighing sensor j , j = 0, 1, 2, 3...., calculate p(w j |A), p(w j |~A): is the probability that, under condition A in the lying position when the air pressure P = P2, the weight is w j ; Probability when lying on side in state ~ A with air pressure P = P2 and weight w j ; Calculate p(A|~A,Δw j ), p(~A|A,Δw j ), p(A|A,Δw j ), p(~A|~A,Δw j ): Δw j = w j - w j-1 where w j-1 is the measured weight in the previous measurement cycle; When the air pressure P = P2, the weight change is Δw j The probability of transitioning from the side-lying state to the supine state A in this case; When the air pressure P = P2, the weight change is Δw j The probability of transitioning from the supine state A to the side-lying state ~A in the case; When the air pressure P = P2, the weight change is Δw j The probability that the sleeping posture state does not change under the circumstances; Then, use Bayes' formula to estimate the posterior probability: Flat probability p j = p(A|w j , Δw j , w j-1 , Δw j-1 ...) = p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ + p(A|A, Δw j )·p j-1 / η j ; Lying-on-side probability p j ～ = p(~A|w j ,Δw j ,w j-1 ,Δw j-1 ...) = p(w j | ~A)·[p(~A|A, Δw j )·p j-1 + p(~A|~A, Δw j )·p j-1 ～ / η j , where η j is a normalization coefficient η j = p(w j |A)·[p(A|~A, Δw j )·p j-1 ～ +p(A|A, Δw j )·p j-1 +p(w j |~A)·[p(~A|A, Δw j )·p j-1 +p(~A|~A, Δw j )·p j-1 ～ , For the first estimation, j = 0, p0 = p(A|w0) = p(w0|A)·p(A) / η0; p0 ～ = p(~A|w0) = p(w0|~A)·p(~A) / η0 where p(A) = p(~A) = 0.5 η0 = p(w j |A)·0.5 + p(w j |~A)·0.5; When p j > p j ～ And |p j - p j ～ | > K, it is determined that the sleeping position is lying flat, and the control circuit opens the intake valve. When P = P1 is satisfied, the intake valve is closed.

Images