CN106503643A - Tumble detection method for human body - Google Patents

Abstract

The present invention relates to a kind of tumble detection method for human body, gathers view data with video camera as medium, the image for gathering is processed, background area is distinguished, is obtained human body target；Human body target is defined by minimum area boundary rectangle on this basis, then rectangle the ratio of width to height, mass center of human body highly ratio and trunk inclination angle three-type-person's body characteristicses are calculated with the external square of minimum area；By the fuzzy construction of function characteristics of human body BPA of Generalized Triangular, using improved D S evidence theory fusions characteristics of human body BPA, final testing result is obtained.Compared with relying on detection scheme of the sensor as information media, the fall detection scheme is not only solved carries out the low problem of fall detection accuracy rate by image procossing, also reduces testing cost with the market.The fall detection scheme can be generalized to the places such as intensive old man hospital of old man colony, home for destitute, with higher actual application value.

Description

Tumble detection method for human body

Technical field

The present invention relates to a kind of image application technology, more particularly to a kind of tumble detection method for human body.

Background technology

Chinese total population quantity surpasses 13.6 hundred million people within 2015, and 60 one full year of life ageds 2.1 hundred million account for the 15.5% of total population, China has stepped into aging society.Analyze according to related data, tumble is the key factor for causing the elderly's disability even dead One of.Falls Among Old People is detected timely, medical expenses can not only be reduced, old man can more succoured in time.

At present, two class methods are broadly divided into human body fall detection both at home and abroad：(1) human body wearable device is based on；(2) it is based on Video image analysis.Based on the fall detection of human body wearable device, gathered data is come as medium with sensor mainly.Common biography Sensor has acceleration transducer, gyroscope, pressure sensor etc..The seat rising sun just etc. have devised a kind of based on surface myoelectric (sEMG) With the fall detection system of plantar pressure signal fusion, system needs are in human body wearing electromyographic signal collection instrument and pressure sensing Device, system average recognition rate have reached 91.7%.Jin Wang etc. gather trunk when human body is fallen by wireless senser and incline Whether angle, the change of cardiac rate, analysis human body fall, rate of accuracy reached to 97.5%.Song Fei etc. is by by the fortune that extracts Moving-target image is divided into three regions, analyzes the change of zones of different center of gravity line, and whether detection human body falls, experiment card Bright fall detection rate reaches 90.5%.M.Et al. track target using oval marks, analyze elliptic motion target The direction of major and minor axis and the change of eccentricity, experiment accuracy rate can reach 90%.

Fall detection accuracy rate based on human body wearable device is higher, but its cost is also of a relatively high, wears related product Product are also cumbersome.Based on the fall detection low cost of video image analysis, but accuracy rate is relatively low, easily receives external condition Impact.

Content of the invention

The present invention be directed to based on the low problem of the fall detection accuracy rate of video image analysis, it is proposed that a kind of human body falls Detection method, gathers view data with video camera as medium, judges whether human body falls by image processing techniques analysis. There is good environmental suitability and higher fall detection accuracy rate.

The technical scheme is that：A kind of tumble detection method for human body, specifically includes following steps：

1) view data is gathered as medium with video camera, the image for gathering is processed, first, by the face of pixel Color information is transformed into hsv color space from RGB color, then by setup parameter value, background area is distinguished, is obtained Human body target；

2) in step 1) on the basis of human body target is defined by minimum area boundary rectangle, then use minimum area External square calculates rectangle the ratio of width to height, mass center of human body highly ratio and trunk inclination angle, the characteristic quantity for judging of falling as human body；

3) by the basic probability assignment BPA of the fuzzy construction of function characteristics of human body of Generalized Triangular, build human body respectively and falling Its human body the ratio of width to height feature corresponding membership function when, squatting down, standing；

4) by step 2) three-type-person's body object feature value for detecting is updated to step 3) in the generalized fuzzy function that constructs, Generate the strategy of BPA：

A:When sample is in the range of certain single fuzzy number abscissa, the ordinate of the sample point is the BPA of the proposition Value m_i(U), i is evidence number, takes 1,2,3；U is identification framework, takes S, T, and F stands, squats down for human body, 3 propositions of falling；

B:When sample is in the range of multiple fuzzy number abscissas, multiple ordinate values that the sample point meets are multiple lives BPA values m of topic_i(U)；

C:Single BPA values m for generating_i(U), when being less than 1, generating other BPA isk Take 1,2；l_kFor m_i(U) characteristic value abscissa and other two evidence Triangular Fuzzy Number central points, that is, be 1 with ordinate Abscissa distance；

D:The BPA sums of generationDuring less than 1, generate

E:Generate all of BPA sumsDuring more than 1, then normalize each BPA values；

5) according to the calculating of step 4, each proposition has 3 evidences to be respectively m₁, m₂, m₃, demonstrate,proved by dual weighted average According to source, information fusion is carried out to human body evidence, the fusion probable value of three-type-person's body state, judges most probable value in analytical evidence State be body state.

The step 5) by dual weighted average evidence source, information fusion is carried out to human body evidence, specifically include as follows Step：

m₁(S)、m₂(S)、m₃(S) 3 evidences that stands for human body, m₁(T)、m₂(T)、m₃(T) demonstrate,prove for 3 that human body is squatted down According to m₁(F)、m₂(F)、m₃(F) 3 evidences that falls for human body,

(1) average evidence is calculated：

m_ave(S)=(m₁(S)+m₂(S)+m₃(S))/3；

(2) distance of each evidence and average evidence under single proposition is calculated：

d_s=[(m₁(S)-m_ave(S))²+(m₂(S)-m_ave(S))²+…(m₃(S)-m_ave(S))²]^1/2；

d_T=[(m₁(T)-m_ave(T))²+(m₂(T)-m_ave(T))²+…(m₃(T)-m_ave(T))²]^1/2；

d_F=[(m₁(F)-m_ave(F))²+(m₂(F)-m_ave(F))²+…(m₃(F)-m_ave(F))²]^1/2

(3) nearer with average evidence distance, weight is bigger, so defining the weights omega of each evidence under single proposition_S、ω_T、 ω_F：

ω_S=1/d_S/((1/d_S)+(1/d_T)+(1/d_F))；

ω_T=1/d_T/((1/d_S)+(1/d_T)+(1/d_F))；

ω_F=1/d_F/((1/d_S)+(1/d_T)+(1/d_F))；

(4) weight calculation using each evidence goes out new average evidence m_ave(S)、m_ave(T)、m_ave(F)：

m_ave(S)=ω_S*m₁(S)+ω_T*m₂(S)+ω_F*m₃(S)；

m_ave(T)=ω_S*m₁(T)+ω_T*m₂(T)+ω_F*m₃(T)；

m_ave(F)=ω_S*m₁(F)+ω_T*m₂(F)+ω_F*m₃(F)；

(5) each evidence is calculated with new average evidence apart from D₁、D₂、D₃：

D₁=[(m₁(S)-m_ave(S)²+(m₁(T)-m_ave(T))²+(m₁(F)-m_ave(F))²]^1/2；

D₂=[(m₂(S)-m_ave(S)²+(m₂(T)-m_ave(T))²+(m₂(F)-m_ave(F))²]^1/2

D₃=[(m₃(S)-m_ave(S)²+(m₃(T)-m_ave(T))²+(m₃(F)-m_ave(F))²]^1/2

(6) nearer with average evidence distance, weight is bigger, so calculating the weight of each evidence：

(7) weight calculation using each evidence goes out final weighted average evidence：

(8) finally to m₄(S)、m₄(T)、m₄(F) calculating 2 times is iteratively repeated with D-S rules of combination, draw fusion results, D- S rule of combination iteration fusion calculation evidence m₄As follows：

K=m in formula₄(S)*m₄(T)+m₄(S)m₄(F)+m₄(T)m₄(S)+m₄(T)m₄(F)+m₄(F)m₄(S)+m₄(F)m₄ (T) the evidence m for, obtaining₅Again with m₄Carry out D-S fusions and obtain final evidence m₆.

The beneficial effects of the present invention is：Tumble detection method for human body of the present invention, has higher accuracy rate.With on the market with Rely on the detection scheme that sensor is information media to compare, the fall detection scheme is not only solved and carried out by image procossing The low problem of fall detection accuracy rate, also reduces testing cost.It is intensive that the fall detection scheme can be generalized to old man colony Old man hospital, the place such as home for destitute, with higher actual application value.

Description of the drawings

Fig. 1 is tumble detection method for human body flow chart of the present invention；

Fig. 2 is video extraction frame figure of the present invention；

Fig. 3 is hsv color space diagram of the present invention；

Fig. 4 is removed for the present invention and scheme before shade；

Fig. 5 is removed for the present invention and scheme after shade；

Fig. 6 is human body the ratio of width to height ambiguity function of the present invention definition figure；

Fig. 7 is that mass center of human body of the present invention height is schemed than ambiguity function definition；

Fig. 8 is trunk inclination angle of the present invention ambiguity function definition figure；

Fig. 9 is that human body tumble state of the present invention is schemed in fact；

Figure 10 is human body tumble state diagram after the inventive method process.

Specific embodiment

The present invention is mainly analyzed with characteristics of human body and D-S information fusions, using human body minimum area boundary rectangle and vertically Boundary rectangle is marked to the human body target for detecting, and analyzes rectangle the ratio of width to height, the mass center of human body height in human body target region Than the changing features with trunk inclination angle.According to judgement sensitivity of each feature to human body different conditions, it is proposed that one Plant the method based on the human body fall detection for improving D-S evidence theory.The program has good environmental suitability and higher Fall detection accuracy rate.

As shown in figure 1, the tumble detection method for human body is comprised the following steps that：

Step 1：View data is gathered as medium with video camera, video extraction frame figure as shown in Figure 2, to the image for gathering Processed, first, the colouring information of pixel is transformed into hsv color space, HSV face as shown in Figure 3 from RGB color Colour space figure.Shadow region is detected according to formula (1), the forward and backward figure of shade is removed as shown in Figure 4,5.In formula (1), I (x, y) Represent bianry image in coordinate x, the value at y, H_I、S_I、V_IRepresent value of the present image in each component in hsv color space, H_B、S_B、 V_BRepresent value of the background two field picture in each component in hsv color space, T_V、T_S、T_HThe brightness of setting, saturation degree, tone are represented respectively Parameters threshold value, by the excursion of parameters before and after statistics shadow region, arranges T_V=0.3, T_S=0.15, T_H= 0.25.The pixel for meeting condition is set to 1, will the shade that measures of flase drop be converted to black background region in binary map.

The shadow interference that illumination variation is brought under background subtraction is eliminated using formula (1), and then obtains complete human body Profile, as shown in Figure 5.

Step 2：By being calculated the characteristic quantity for carrying out that human body tumble judges.In the present invention, using human body the ratio of width to height, people The characteristic quantity that body height of center of mass ratio and trunk inclination angle are judged as human body tumble.

When object boundary is unknown, it is simplest method to describe target with boundary rectangle.Thing for any direction Body, it is necessary to determine the main shaft of object, now describes target with the rectangle of horizontal or vertical direction, will no longer have that reference Property.Therefore, a kind of human body minimum area boundary rectangle that human body target can be described more fully is defined.By the human body target for detecting Profile border is rotated in the range of 90 ° with 3 ° or so every time of increment.Often rotate and once record once on its coordinate system direction Minimum and maximum x, y value of boundary rectangle boundary point.After rotating to some angle, the area of boundary rectangle reaches minimum.Take The parameter of the minimum boundary rectangle of area is the length and width under main shaft meaning.Four summits of rectangle are labeled as in order successively R1(X_R1, Y_R1), R2 (X_R2, Y_R2), R3 (X_R3, Y_R3), R4 (X_R4, Y_R4).Then the wide W and high H of boundary rectangle can pass through following equation Calculate：

Human body the ratio of width to height can be calculated by formula (4)：

In formula, W, H are respectively the wide and high of human body minimum area boundary rectangle.

Then, center (rectangle diagonal) Z for defining minimum enclosed rectangle is mass center of human body position.Through Shadows Processing In target detection binary map P (x, y) afterwards, it is target area (pixel is 0) pixel to define two-dimensional array set [row, col] Set, wherein row, col are respectively transverse and longitudinal coordinate value set of the target area pixel in binary map.

It is level ground to arrange straight line L, and wherein L is that pixel (0, max (row)) arrives pixel (640, max (row) straight line).640 are the maximums that can get of row under experiment resolution ratio of camera head herein herein.

Next, defining barycenter Z to straight line L apart from Z_LFor height of center of mass, Z is defined_LWith Human Height H_PRatio R_ZFor people Body height of center of mass ratio, as shown in formula (5)：

H in formula (5)_PFor human body normal stand when calculated height average in binary map, be set to 370.

Trunk inclination angle phi is finally defined, and (6) calculate φ values according to the following formula.

In formula (6), max (row), min (row) represent maximum of the target in binary map vertical direction, minimum image respectively Plain coordinate value, max (col), min (col) represent maximum of the target in binary map horizontal direction, minimum pixel coordinate respectively Value.The rectangle surrounded by four coordinate values is target vertical boundary rectangle.

Judge that human body attitude can accurately not analyze the motion conditions of human body by single characteristics of human body, in particular cases Certain characteristics of human body even there is also the situation of mutation, by certain characteristics of human body, now single judges that human body attitude can be made Into detection error.How to merge three-type-person's body characteristicses information, be the key for accurately judging human motion attitude.

Step 3：According to the three-type-person's body characteristicses amount for the obtaining in step 2 detection sensitivity different to human body attitude, carry Go out the detection algorithm that a kind of D-S evidence theory merges characteristics of human body's information.

The three-type-person's body target signature for collecting does not meet the requirement of D-S evidence theory BPA, it is impossible to be directly substituted into D-S cards Information fusion is carried out according to theory, is needed by constructing BPA, using the physical characteristic data of statistics, the person in servitude for setting up Triangular Fuzzy Number Membership fuction, building method are simple and practical, and amount of calculation is less, generate BPA, the basic probability assignment BPA for representing human body target signature, Also referred to as m functions, represent the trusting degree for distributing to each proposition, and m (A) is substantially credible number, reflects the reliability size to A.

First, a kind of fuzzy construction of function characteristics of human body BPA of Generalized Triangular is defined.The generalized fuzzy of human body the ratio of width to height feature Function mu_QA () represents human body its human body the ratio of width to height feature when falling, squatting down, stand as shown in formula (7), (8), (9), respectively Corresponding membership function, the fuzzy functional arrangement of its Generalized Triangular is as shown in Figure 6.

Mass center of human body highly ratio, the generalized fuzzy function at trunk inclination angle are constructed in the same mannerμ_φ(c) such as formula (10), shown in (11), (12), (13), (14), (15), the fuzzy functional arrangement of its Generalized Triangular is as shown in accompanying drawing 7,8.

Step 4：The three-type-person's body object feature value for detecting is updated in the generalized fuzzy function that step 3 is constructed, raw Strategy into BPA is as follows：

(1) when sample is in the range of certain single fuzzy number abscissa, the ordinate of the sample point is the BPA of the proposition Value m_i(U) (i is evidence number, takes 1,2,3；U is identification framework, takes S, T, F stands, squats down for human body, 3 propositions of falling).

(2) when sample is in the range of multiple fuzzy number abscissas, multiple ordinate values that the sample point meets are multiple BPA values m of proposition_i(U).

(3) single BPA values m for generating_i(U), when being less than 1, generating other BPA is (k Take 1,2；l_kFor m_i(U) characteristic value abscissa and other two evidence Triangular Fuzzy Number central points (ordinate is 1 abscissa) Distance).

(4) the BPA sums for generatingDuring less than 1, generate

(5) all of BPA sums are generatedDuring more than 1, then normalize each BPA values.

Step 5：The BPA generated in step 4 is updated in improved D-S evidence theory rule, human body evidence is carried out Information fusion and judgement.D-S evidence theory is improved by the innovatory algorithm of dual weighted average evidence source.

S, T, F stand, squat down for human body, 3 propositions of falling, and according to the calculating of step 4, each proposition has 3 evidences Respectively m₁, m₂, m₃.Improved blending algorithm step is as follows：

(1) average evidence is calculated.

m_ave(S)=(m₁(S)+m₂(S)+m₃(S))/3；

(2) distance of each evidence and average evidence under single proposition S is calculated.

d_s=[(m₁(S)-m_ave(S))²+(m₂(S)-m_ave(S))²+…(m₃(S)-m_ave(S))²]^1/2；

d_T=[(m₁(T)-m_ave(T))²+(m₂(T)-m_ave(T))²+…(m₃(T)-m_ave(T))²]^1/2；

d_F=[(m₁(F)-m_ave(F))²+(m₂(F)-m_ave(F))²+…(m₃(F)-m_ave(F))²]^1/2

(3) nearer with average evidence distance, weight is bigger, so defining the weights omega of each evidence under single proposition_S、ω_T、 ω_F.

ω_S=1/d_S/((1/d_S)+(1/d_T)+(1/d_F))；

ω_T=1/d_T/((1/d_S)+(1/d_T)+(1/d_F))；

ω_F=1/d_F/((1/d_S)+(1/d_T)+(1/d_F))；

(4) weight calculation using each evidence goes out new average evidence m_ave(S)、m_ave(T)、m_ave(F).

m_ave(S)=ω_S*m₁(S)+ω_T*m₂(S)+ω_F*m₃(S)；

m_ave(T)=ω_S*m₁(T)+ω_T*m₂(T)+ω_F*m₃(T)；

m_ave(F)=ω_S*m₁(F)+ω_T*m₂(F)+ω_F*m₃(F)；

(5) each evidence is calculated with new average evidence apart from D₁、D₂、D₃.

D₁=[(m₁(S)-m_ave(S)²+(m₁(T)-m_ave(T))²+(m₁(F)-m_ave(F))²]^1/2；

D₂=[(m₂(S)-m_ave(S)²+(m₂(T)-m_ave(T))²+(m₂(F)-m_ave(F))²]^1/2

D₃=[(m₃(S)-m_ave(S)²+(m₃(T)-m_ave(T))²+(m₃(F)-m_ave(F))²]^1/2

(6) nearer with average evidence distance, weight is bigger, so calculating the weight of each evidence.

(7) weight calculation using each evidence goes out final weighted average evidence.

(8) finally to m₄(S)、m₄(T)、m₄(F) calculating 2 times is iteratively repeated with D-S rules of combination, draw fusion results.D- S rule of combination iteration fusion calculation evidence m₄As follows：

K=m in formula₄(S)*m₄(T)+m₄(S)m₄(F)+m₄(T)m₄(S)+m₄(T)m₄(F)+m₄(F)m₄(S)+m₄(F)m₄ (T) the evidence m for, obtaining₅Again with m₄Carry out D-S fusions and obtain final evidence m₆, calculating process such as formula (16), (17), (18). Analytical evidence m₆The fusion probable value of middle three-type-person's body state, judges the state of most probable value for body state.Fig. 9 is this Person of good sense's body tumble state is schemed in fact；Figure 10 is human body tumble state diagram after the inventive method process.

Claims (2)

1. a kind of tumble detection method for human body, it is characterised in that specifically include following steps：

1) view data is gathered as medium with video camera, the image for gathering is processed, first, the color of pixel is believed Breath is transformed into hsv color space from RGB color, then by setup parameter value, background area is distinguished, obtains human body Target；

2) in step 1) on the basis of human body target is defined by minimum area boundary rectangle, then with minimum area external Square calculates rectangle the ratio of width to height, mass center of human body highly ratio and trunk inclination angle, the characteristic quantity for judging of falling as human body；

3) by the basic probability assignment BPA of the fuzzy construction of function characteristics of human body of Generalized Triangular, build respectively human body falling, under Its human body the ratio of width to height feature corresponding membership function when squatting, standing；

4) by step 2) three-type-person's body object feature value for detecting is updated to step 3) in the generalized fuzzy function that constructs, generate The strategy of BPA：

A:When sample is in the range of certain single fuzzy number abscissa, the ordinate of the sample point is BPA values m of the proposition_i (U), i is evidence number, takes 1,2,3；U is identification framework, takes S, T, and F stands, squats down for human body, 3 propositions of falling；

B:When sample is in the range of multiple fuzzy number abscissas, multiple ordinate values that the sample point meets are multiple propositions BPA values m_i(U)；

C:Single BPA values m for generating_i(U), when being less than 1, generating other BPA isK takes 1, 2；l_kFor m_i(U) characteristic value abscissa and other two evidence Triangular Fuzzy Number central points, that is, with the horizontal stroke that ordinate is 1 The distance of coordinate；

D:The BPA sums of generationDuring less than 1, generate

E:Generate all of BPA sumsDuring more than 1, then normalize each BPA values；

5) according to the calculating of step 4, each proposition has 3 evidences to be respectively m₁, m₂, m₃, by dual weighted average evidence Source, carries out information fusion to human body evidence, and in analytical evidence, the fusion probable value of three-type-person's body state, judges most probable value State is body state.

2. tumble detection method for human body according to claim 1, it is characterised in that the step 5) by dual weighted average Evidence source, carries out information fusion to human body evidence, specifically includes following steps：

m₁(S)、m₂(S)、m₃(S) 3 evidences that stands for human body, m₁(T)、m₂(T)、m₃(T) 3 evidences that squats down for human body, m₁ (F)、m₂(F)、m₃(F) 3 evidences that falls for human body,

(1) average evidence is calculated：

m_ave(S)=(m₁(S)+m₂(S)+m₃(S))/3；

(2) distance of each evidence and average evidence under single proposition is calculated：

d_s=[(m₁(S)-m_ave(S))²+(m₂(S)-m_ave(S))²+…(m₃(S)-m_ave(S))²]^1/2；

d_T=[(m₁(T)-m_ave(T))²+(m₂(T)-m_ave(T))²+…(m₃(T)-m_ave(T))²]^1/2；

d_F=[(m₁(F)-m_ave(F))²+(m₂(F)-m_ave(F))²+…(m₃(F)-m_ave(F))²]^1/2

(3) nearer with average evidence distance, weight is bigger, so defining the weights omega of each evidence under single proposition_S、ω_T、ω_F：

ω_S=1/d_S/((1/d_S)+(1/d_T)+(1/d_F))；

ω_T=1/d_T/((1/d_S)+(1/d_T)+(1/d_F))；

ω_F=1/d_F/((1/d_S)+(1/d_T)+(1/d_F))；

(4) weight calculation using each evidence goes out new average evidence m_ave(S)、m_ave(T)、m_ave(F)：

m_ave(S)=ω_S*m₁(S)+ω_T*m₂(S)+ω_F*m₃(S)；

m_ave(T)=ω_S*m₁(T)+ω_T*m₂(T)+ω_F*m₃(T)；

m_ave(F)=ω_S*m₁(F)+ω_T*m₂(F)+ω_F*m₃(F)；

(5) each evidence is calculated with new average evidence apart from D₁、D₂、D₃：

D₁=[(m₁(S)-m_ave(S)²+(m₁(T)-m_ave(T))²+(m₁(F)-m_ave(F))²]^1/2；

D₂=[(m₂(S)-m_ave(S)²+(m₂(T)-m_ave(T))²+(m₂(F)-m_ave(F))²]^1/2

D₃=[(m₃(S)-m_ave(S)²+(m₃(T)-m_ave(T))²+(m₃(F)-m_ave(F))²]^1/2

(6) nearer with average evidence distance, weight is bigger, so calculating the weight of each evidence：

${\mathrm{\ω}}_{S_N}=1/D_1/\left(\right(1/D_1)+(1/D_2)+(1/D_3\left)\right);$

${\mathrm{\ω}}_{T_N}=1/D_2/\left(\right(1/D_1)+(1/D_2)+(1/D_3\left)\right);$

${\mathrm{\ω}}_{F_N}=1/D_3/\left(\right(1/D_1)+(1/D_2)+(1/D_3\left)\right);$

(7) weight calculation using each evidence goes out final weighted average evidence：

$m_4\left(S\right)={\mathrm{\ω}}_{S_N}*m_1\left(S\right)+{\mathrm{\ω}}_{T_N}*m_2\left(S\right)+{\mathrm{\ω}}_{F_N}*m_3\left(S\right);$

$m_4\left(T\right)={\mathrm{\ω}}_{S_N}*m_1\left(T\right)+{\mathrm{\ω}}_{T_N}*m_2\left(T\right)+{\mathrm{\ω}}_{F_N}*m_3\left(T\right);$

$m_4\left(F\right)={\mathrm{\ω}}_{S_N}*m_1\left(F\right)+{\mathrm{\ω}}_{T_N}*m_2\left(F\right)+{\mathrm{\ω}}_{F_N}*m_3\left(F\right);$

(8) finally to m₄(S)、m₄(T)、m₄(F) calculating 2 times is iteratively repeated with D-S rules of combination, draw fusion results, D-S groups Normally iteration fusion calculation evidence m₄As follows：

$m_5\left(S\right)=\frac{m_4\left(S\right)*m_4\left(S\right)}{1-K};$

$m_5\left(T\right)=\frac{m_4\left(T\right)*m_4\left(T\right)}{1-K};$

$m_5\left(F\right)=\frac{m_4\left(F\right)*m_4\left(F\right)}{1-K};$

K=m in formula₄(S)*m₄(T)+m₄(S)m₄(F)+m₄(T)m₄(S)+m₄(T)m₄(F)+m₄(F)m₄(S)+m₄(F)m₄(T), obtain The evidence m for arriving₅Again with m₄Carry out D-S fusions and obtain final evidence m₆.