CN105718971A - RFID-based multi-target passive-type indoor activity identification method - Google Patents

Abstract

The invention discloses an RFID-based multi-target passive-type indoor activity identification method, which takes perturbation of RFID tag phases caused by human activities as fingerprint information and obtains activity situations of multiple targets by using perturbation of different tag phases caused by activities of many persons. However, current indoor activity identification methods are mainly for a single target, and cannot be applied or lead to big activity identification errors in multi-target indoor scenes. The method of the invention realizes multi-target activity identification without the requirement that targets carry devices, solves the defects of indoor activity identification in terms of target quantity, achieves good dynamic adaptability to the environment, and is low in economic cost.

Description

A kind of multiple target passive type indoor activity recognition methods based on RFID

Technical field

The invention belongs to wireless activity identification field, relate to a kind of multiple target activity recognition method, be specifically related to a kind of multiple target passive type indoor activity recognition methods based on RFID.

Background technology

Now in a lot of application scenarios, particularly indoor scene, activity recognition technology becomes particularly important.Activity recognition technology can provide most basic facility condition for Medicare, optimum management, Smart Home, healthy tracking etc..Traditional activity recognition method is generally through photographic head, radar, or gives the hardware devices such as detected person's wearable sensors.All there is either large or small defect in these methods: the method based on video camera is the most obvious by the impact of illumination, and the secret protection of people is also existed serious threat by video camera；With regard to radar, the radar low cost solution opereating specification of 60GHz only has limited tens centimetre；One maximum defect of method based on wearable sensors is exactly " must dress ", and this certainly will will bring much unnecessary trouble for attendee.

Current RFID passive label is because the characteristic of its cheap (5～10 cents) is subject to the favor of vast research team, and a lot of research teams RFID technique does location and the tracking of target.RFID (RadioFrequencyIdentification) technology, refers to unlimited REID.

Now in overwhelming majority activity recognition method, being typically all the activity recognition of passive type single goal, few people do the multiobject activity recognition of passive type.Reason mainly has 2 points: cannot realize multiobject activity recognition in detection region；Precision is too poor, it is impossible to accurately estimated multiobject activity recognition.

To sum up, the existing passive type following defect of activity recognition method ubiquity: 1) multiobject activity recognition can not be realized in monitored area；2) equipment is not cheap；3) precision can not meet people's demand.

Summary of the invention

For the defect existed in above-mentioned prior art or deficiency, it is an object of the invention to, a kind of multiple target passive type indoor activity recognition methods based on RFID is provided, the method can carry out multiobject movable estimation accurately and identify in surveillance area, the dynamic adaptable of environment is better, and Financial cost is low.

In order to realize above-mentioned task, the present invention by the following technical solutions:

A kind of multiple target passive type indoor activity recognition methods based on RFID, comprises the following steps:

Step one, disposes multiple RFID passive label in monitored area, constitutes passive label array, then disposes a pair rfid interrogator；

Step 2, monitored area is put into movable target, obtaining and preserve the movable target off-line finger print information when monitored area activity, off-line finger print information includes the region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and is affected degree；

Step 3, when target to be detected enters activity in monitoring, obtain the set being affected label that each rfid interrogator is corresponding, determine according to the set being affected label and multiple be affected label subregion, and then determine target data to be detected, thus obtaining the online finger print information that each target to be detected is corresponding, online finger print information includes the tag ID of object effects to be detected and the phase information of these labels being affected；

Step 4, is utilized off-line finger print information and online finger print information, is obtained the Activity Type of target to be detected by matching primitives.

Further, the detailed process of step 2 includes:

Step 2.1, two RFID of deployment send radiofrequency signal simultaneously, and phase information when not having target in monitored area that all passive labels return is measured and recorded to the aerial array of each rfid interrogator；

Step 2.2, is divided into the square area that multiple size is identical by monitored area, utilizes a movable target to carry out different activities in each square area, obtains the phase information of all passive labels that each rfid interrogator receives；

Step 2.3, the phase information that phase information step 2.1 obtained and step 2.2 obtain compares, and respectively obtains movable target and carries out the label of impact during different activity in each square area；Region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and the degree that is affected are stored in server database as off-line finger print information.

Further, the detailed process of step 3 includes:

Step 3.1, when more than one target to be detected enters monitored area activity, the aerial array of two rfid interrogators reads the phase information of all labels respectively, phase information is stored in database server；

Step 3.2, the radiofrequency signal that two rfid interrogators send arrives each label by different paths, two routing information set can be obtained: set one is first rfid interrogator phase information to all labels, set two is second rfid interrogator phase information to all labels, utilize dynamic time warping algorithm DTW and consider that the phase information of label is changed the impact produced by noise, label phase information during target is not had to compare with monitored area respectively the phase information of the label in set one and set two, obtain two set being affected label；

Step 3.3, for each rfid interrogator, corresponding is affected in the set of label and has multiple subregion being affected label, the overlapping cases being affected label subregion according to two rfid interrogator feedbacks, calculates the target numbers of target to be detected；For each target to be detected, the corresponding phase information being affected tag ID and these labels constitutes the online finger print information of this target to be detected.

Further, the method for the matching primitives described in step 4 includes:

Note step 2 is off-line phase, and step 3 is on-line stage, then the computing formula of goal activities to be detected coupling is as follows:

Score(R_finger,F_finger)=N_sameλ-Dis(R_finger,F_finger)+NL_lack·NA_lack

$Dis\left(R_{finger},F_{finger}\right)=\frac{\sqrt{\underset{i=1}{\overset{N_{same}}{\mathrm{\Σ}}}{\left(\frac{\underset{j=1}{\overset{NA}{\mathrm{\Σ}}}{\mathrm{DTW}}_{ij}}{NA}\right)}^2}}{N_{same}}$

Score_target=argmax (Score (R_finger,F_finger(i))),i∈(1,…,N_rf)

Herein above in three formula:

R_fingerRepresent online finger print information, F_fingerRepresent off-line finger print information, N_sameRepresenting the on-line stage number of tags that be affected identical with in each group of off-line phase, NA is number of antennas, N_rfFor off-line finger print information number；

λ represents identical and is affected label weight in the matching process of position；

Dis(R_finger,F_finger) represent online finger print information and the one group of off-line phase finger print information comprehensive differences in all identical phase informations being affected label；

DTW_ijRepresent the identical difference in phase information that is affected label i on jth antenna；

NL_lackWhat in expression off-line phase and on-line stage, same antenna signal was incomplete is affected number of labels, i.e. two kinds of situations: situation one, off-line phase label Label is for being affected label, and on-line stage label Label cannot be collected the relevant information of this label by aerial array；Situation two, off-line phase label Label cannot be collected the relevant information of this label by aerial array, and on-line stage label Label is for being affected label；

NA_lackRepresent the weight that aerial signal is incomplete；

Score_targetRepresent for each target to be detected, calculate the Score (R of online fingerprint corresponding to this target to be detected and all off-line fingerprints respectively_finger,F_finger) value, obtain maximum Score (R_finger,F_finger) off-line fingerprint corresponding to value, then the goal activities type in this off-line fingerprint is required goal activities type.

Further, in step 2.3, it is determined whether the method for being affected label includes:

Note for store phase information file have two: D_α(i) and D_β(i), then the difference value of two phase informations is:

${\mathrm{DTW}}_{\mathrm{\α},\mathrm{\β}}=min\underset{i,j=1}{\overset{L}{\Σ}}|D_{\mathrm{\α}}\left(i\right)-D_{\mathrm{\β}}(j\left)\right|$

In above formula, L is the number of phase information file intermediate value；

Assume that ξ was that a upper moment is for judging that whether label phase information is by the threshold value of movable object effects, DTW_newRepresent that current time judges that whether label phase information is by the threshold value of movable object effects,Represent threshold deviation, then calculate the difference Δ of current time threshold value and previous moment threshold value:

Δ=DTW_new-ξ

Use current time threshold value update ξ andObtain:

ξ=(1-0.125) ξ+0.125 DTW_new

$\∂=(1-0.25)\·\∂+0.25\·\left|\mathrm{\Δ}\right|$

Two formula on abbreviation, draws:

ξ=ξ+0.125 Δ

$\∂=\∂+0.25\·\left(\right|\mathrm{\Δ}|-\∂)$

Thus up-to-date threshold value is:

$Th=\mathrm{\ξ}+4\∂$

If recording phase information difference value DTW when label has movable target in monitored area and do not have movable target_α,βMore than threshold value Th, then it is assumed that this label is for being affected label.

The present invention compared with prior art has techniques below feature:

1, the activity recognition method of the present invention only needs RFID passive label cheap on a small quantity just multiple unknown objects can be carried out activity recognition, and namely required Financial cost is low, breaches and uses expensive device could realize multiobject activity recognition in traditional sense.

2, the activity recognition method of the present invention using people to the disturbance of phase information as finger print information.

3, the present invention adopts selectivity matching strategy, if current online fingerprint and certain off-line fingerprint do not have identical to be affected label, then without coupling, largely reduces time complexity.

4, the present invention ingenious utilize rfid interrogator and label in communication process target block the situation that the signal caused cannot receive, rather than try every possible means and avoid this situation, there is universality and practicality.

5, the activity recognition method of the present invention is very good to the dynamic adaptable of environment, therefore, it is possible to well adapt to the dynamic change of application scenarios, this activity recognition method has certain universality.

Accompanying drawing explanation

Fig. 1 is that people is to phase information disturbance explanation figure；

Fig. 2 is the region recognition principle explanation figure of same movable diverse location, and wherein Fig. 2 (a) is the schematic diagram that a rfid interrogator and 12 passive type RFID tag are deployed in region to be detected；Fig. 2 (b) for label unmanned, people at phase information graphic differences schematic diagram when position G2 of position G1, people；Fig. 2 (c) is for people when same position, and different labels are by the schematic diagram of people's influence degree；

Fig. 3 is the activity recognition principle explanation figure of same position difference activity, and wherein Fig. 3 (a) is the schematic diagram that a rfid interrogator and 12 passive type RFID tag are deployed in region to be detected；Fig. 3 (b) is for people when sitting (S), pushing hands (P), lying down actions such as (F), and the phase information that label returns changes schematic diagram；

Fig. 4 is backscatter communication schematic diagram；

Four examples in Fig. 5 object count process, wherein Fig. 5 (a) for target range farther out time method of counting schematic diagram；Fig. 5 (b) is the schematic diagram occurring vanishing target situation under a rfid interrogator；Fig. 5 (c) is the schematic diagram eliminating vanishing target situation under two rfid interrogators；Fig. 5 (d) is target range schematic diagram time nearer；

Fig. 6 is the activity recognition flow chart of the present invention；

Fig. 7 is the counting precision of true experiment under 480cm*480cm, 600cm*600cm, 720cm*600cm monitored area；

Fig. 8 is 480cm*480cm 1.2m, 720cm*600cm 1.2m,

480cm*480cm 0.6m, 720cm*600cm 0.6m truly test under counting precision, wherein 0.6m and 1.2m is label spacing.

Fig. 9 is the precision in laboratory (Lab), spacious hall (Lobby) and outdoor two various activity recognition of target of (Outside) scene.

Detailed description of the invention

Referring to Fig. 1, the mode that RFID device gathers environmental information is as follows: rfid interrogator sends radiofrequency signal by aerial array, and RFID tag, after radiofrequency signal, incidentally goes up the label information of self, return signal, to aerial array, obtains the relevant information of label after treatment.When people stands between antenna and label, by the change of the phase information that people causes, it is possible to determine that people is substantially in which region, and it is movable which type of is carrying out.

For more visible the cardinal principle providing the present invention, principles of the invention is described by we by two simply examples.

As in figure 2 it is shown, mainly for same activity is described, when people is in zones of different, signal fluctuation has obvious difference, so the situation according to signal, it is possible to infer the general areas residing for people.Disposing shown in schematic diagram 2 (a), a rfid interrogator and 12 passive type RFID tag are deployed in region to be detected.Phase information Fig. 2 (b) display label is very big at position G1, people's phase information graphic differences when position G2 nobody, people, so people is in different regions, the information being affected label return has obvious change.Therefore the influence degree of the phase information of different labels can substantially be inferred the region at target place by us according to target.It addition, from Fig. 2 (c) it can be seen that people is when same position, different labels are also had obvious difference by people's effect.Therefore the influence degree of the phase information of different labels can substantially be inferred the region at target place by us according to target.

As it is shown on figure 3, mainly for the same area is described, people carries out different activities, the phase information that label returns also has obvious change, so the situation according to signal intensity, it is possible to infer which kind of activity people is doing.Disposing shown in schematic diagram 3 (a), a rfid interrogator and 12 passive type RFID tag are deployed in region to be detected, and a target carries out different activities respectively in region 1.Phase information Fig. 3 (b) shows when people is when sitting (S), pushing hands (P), lying down actions such as (F), the phase information that label returns there will be obvious change, so the phase information returned according to label, it is concluded which kind of activity people is carrying out.The present invention is the activity recognition that the phase information utilizing label to return and fingerprint matching realize target based on the multiple target passive type indoor activity recognition methods of RFID, and merely with cheap RFID passive label, be a kind of New activity recognition methods do not inquired in activity recognition field.

Briefly, a kind of multiple target passive type indoor activity recognition methods based on RFID that the present invention proposes mainly comprises two stages: destination number calculates and activity recognition, as shown in Figure 6.

In the object count stage, the present invention adopts DTW algorithm to detect whether region to be detected interior label phase information changes, and continuous print is affected label is divided into subregion, and different targets has the different of correspondence to affect subregion.Then, use two rfid interrogators to obtain the region at target substantially place, and utilize detection edge length to eliminate vanishing target situation, thus realizing multiple target counting.

In the activity recognition stage, first, finger print information according to each target that on-line stage collects, namely it is affected label and is affected the phase information of label, all of for off-line phase finger print information is mated with current on-line stage finger print information, identical at least one of which off-line phase finger print information being affected label is extracted and puts a set under.In set, calculate they comprehensive differences of being affected label identical with on-line stage finger print information respectively.Finally, calculating the mark that mates of off-line phase finger print information and on-line stage finger print information according to DTW algorithm, mark is more high, and matching degree is more strong.Choosing the off-line finger print information that wherein coupling mark is the highest, the position in this off-line finger print information is the position of current goal.

The present invention specifically comprises the following steps that

A kind of multiple target passive type indoor activity recognition methods based on RFID, comprises the following steps:

Step one, scene setting: dispose multiple RFID passive label in monitored area, constitute passive label array, then dispose a pair rfid interrogator；

In this example, adjacent label forms matrix label with 1.2m spacing, then two rfid interrogators is positioned over label matrix arbitrary neighborhood both sides, each rfid interrogator connects one group of aerial array, and aerial array is made up of to antenna the whole day that 4 spacing are 0.15m.

Step 2, monitored area is put into movable target, obtain and preserve the movable target off-line finger print information when monitored area activity, off-line finger print information includes the region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and is affected degree, and detailed process includes:

Step 2.1, two RFID of deployment send radiofrequency signal simultaneously, and phase information when not having target in monitored area that all passive labels return is measured and recorded to each aerial array of each rfid interrogator；By principal component analytical method PCA (PrincipalComponentAnalysis), the phase value obtained is utilized to obtain each rfid interrogator phase information to all labels；

The computational methods of above-mentioned acquisition label phase value are as follows:

Rfid interrogator adopts backscatter communication mode, as shown in Figure 4, rfid interrogator uses the aerial array of four omnidirectional antenna compositions to continue to send radiofrequency signal, RFID tag is additional from a label information after receiving radiofrequency signal, then radiofrequency signal is reflected back, and the antenna array receiver of rfid interrogator is after generation signal, the tag ID number that each antenna receives is calculated by rfid interrogator, corresponding phase value θ, signal strength values RSSI etc., phase value θ computing formula is:

$\mathrm{\θ}=(\frac{2\mathrm{\π}}{\mathrm{\λ}}*2d+n)\mathrm{mod}2\mathrm{\π}$

N=θ_T+θ_R+θ_ref

In upper two formulas, λ is the wavelength of radiofrequency signal, and n is system noise, θ_T, θ_R, θ_refThe respectively reflection coefficient receiving circuit and label of the radiating circuit of rfid interrogator, rfid interrogator.Owing to there is system noise, so calculated phase value also can exist deviation, normally behave as the phase information of different time points in Same Scene and have random fluctuation.

Step 2.2, is divided into the square area that multiple size is identical by monitored area, utilizes a movable target to carry out different activities in each square area, obtains the phase information of all passive labels that each rfid interrogator receives；

In the present embodiment, dividing monitored area for the identical grid of 1.2m*1.2m size, the movable target in this step can adopt the people that a physique is medium, stands in each grid, all doing an activity to be identified in each grid, these activities include:

Seat, pushing hands, lie down, bend over, semi-girder etc., make the action of correspondence according to actual needs；These movable behaviors for follow-up judgement target to be measured；In each grid, each group of activity is done 20～30 seconds, obtains each rfid interrogator phase information to all labels；Someone stands when standing in grid with nobody in grid, and the phase information acquisition methods of all labels is identical.

Step 2.3, the phase information that phase information step 2.1 obtained and step 2.2 obtain compares, and respectively obtains movable target and carries out the label of impact during different activity in each square area；Region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and the degree that is affected are stored in server database as off-line finger print information.

Assume that the file being used for storing phase information has two, be D_α(i) and D_β(i):

D_α=D_α(0),…D_α(i),…D_α(180)

D_β=D_β(0),…D_β(i),…D_β(180)

After PCA method processes, have L=180 value in each phase information file, then the difference value of the two phase information is:

${\mathrm{DTW}}_{\mathrm{\α},\mathrm{\β}}=min\underset{i,j=1}{\overset{L}{\Σ}}|D_{\mathrm{\α}}\left(i\right)-D_{\mathrm{\β}}(j\left)\right|$

In above formula, L is the number of phase information file intermediate value；

In calculating process, i needs not be equal to j, and the difference of two phase informations is smallest match value；

Because system noise also can make the phase information of label change, in order to judge that the change of phase information is due to the impact of movable target or due to the impact of system noise: assume that ξ was that a upper moment is for judging that whether label phase information is by the threshold value of movable object effects, DTW_newRepresent that current time judges that whether label phase information is by the threshold value of movable object effects,Represent threshold deviation, then first calculate the difference Δ of current time threshold value and previous moment threshold value:

Δ=DTW_new-ξ

Use current time threshold value update ξ andObtain:

ξ=(1-0.125) ξ+0.125 DTW_new

$\∂=(1-0.25)\·\∂+0.25\·\left|\mathrm{\Δ}\right|$

Wherein, 0.125 and 0.25 is the proportion shared by up-to-date threshold value and deviation；Two formula on abbreviation, draws:

ξ=ξ+0.125 Δ

$\∂=\∂+0.25\·\left(\right|\mathrm{\Δ}|-\∂)$

Thus up-to-date threshold value is:

$Th=\mathrm{\ξ}+4\∂$

If recording phase information difference value DTW when label has movable target in monitored area and do not have movable target_α,βMore than threshold value Th, then it is assumed that this label is for being affected label.

Now, the region residing for target, the Activity Type of target, the RFID tag ID being affected, the phase information being affected label and influence degree are stored in server database as off-line finger print information.

Step 3, when target to be detected enters activity in monitoring, obtain the set being affected label that each rfid interrogator is corresponding, determine according to the set being affected label and multiple be affected label subregion, and then determine target data to be detected, thus obtaining the online finger print information that each target to be detected is corresponding, online finger print information includes the tag ID of object effects to be detected and the phase information of these labels being affected；Detailed process includes:

Step 3.1, when more than one target to be detected (people) enters monitored area activity, the aerial array of two rfid interrogators reads the phase information of all labels respectively, phase value after adopting the PCA method mentioned in step 2.1 to obtain denoising after phase value is carried out denoising, is stored in phase information in database server；

Step 3.2, the radiofrequency signal that two rfid interrogators send arrives each label by different paths, two routing information set can be obtained: set one is first rfid interrogator phase information to all labels, set two is second rfid interrogator phase information to all labels, utilize dynamic time warping algorithm DTW and consider that the phase information of label is changed the impact produced by noise, label phase information during target is not had to compare with monitored area respectively the phase information of the label in set one and set two, obtain two set being affected label；

Step 3.3, for each rfid interrogator, corresponding is affected in the set of label and has multiple subregion being affected label, the overlapping cases being affected label subregion according to two rfid interrogator feedbacks, calculates the target numbers of target to be detected；For each target to be detected, the corresponding phase information being affected tag ID and these labels constitutes the online finger print information of this target to be detected.Wherein, subregion number is number of targets to be detected.

As shown in Fig. 5 (a), for rfid interrogator X, what there are two separation is affected label subregion, then it represents that there are two targets to be detected.

But real process there will be certain and be affected the special circumstances that label subregion is excessive, namely the number of labels in this subregion is too much, experiments show that, if being affected number of labels more than upper limit R=6, then it is likely to this region and there is the situation of vanishing target, subregion I has two targets in Fig. 5 (b), now, need to use another one rfid interrogator Y, as shown in Fig. 5 (c), original subregion I can be divided into subregion I and subregion IV by rfid interrogator Y, and such vanishing target can be detected.

If rfid interrogator Y can not detect, as shown in Fig. 5 (d), rfid interrogator can not detect two targets in subregion I, now needs to calculate the length of side L that this subregion is the longest, then destination number calculating process is as follows:

L=l_tag·(n_tag-1)

$N=\frac{L}{1.2}$

Wherein, l_tagRepresent the spacing of adjacent two labels；N_tagRepresent the number of labels in subregion longest edge length；N represents destination number.

Said process obtain target numbers to be detected corresponding with each target to be detected be affected label subregion, for each target, the corresponding phase information being affected tag ID and these labels forms the online finger print information of this target.

Using rfid interrogator more many in principle, the number of vanishing target is more few.But use two or more rfid interrogators can produce read write line conflict simultaneously, cause the acquisition phase value time longer, it is impossible to meet real-time requirement.Reason is that passive label disposal ability is more weak, when label receives the signal of two or more rfid interrogators simultaneously, it is impossible to processes, can only be used as abnormal signal and discard, and so may require that the long period obtains same amount of phase value.

The present invention weighs time cost and rfid interrogator quantity, adopts two rfid interrogators to carry out object count.Meanwhile, for reducing the time cost that conflict brings, the present invention controls the radiofrequency signal transmission frequency of two RFID reader so that obtaining phase value as much as possible within the identical time, concrete grammar is as follows:

The present invention uses for reference ALOHA agreement, and this agreement is a kind of effective anticollision protocol.It is Poisson process that ALOHA agreement sends signal depending on transmitting terminal within the unit interval, and tag reflection signal is as Poisson process in the present invention.After label filters out collision signal, the signaling reflex not conflicted is returned rfid interrogator, then rfid interrogator success sample rate F_successFor:

F_success=P. σ

Wherein, P is by the probability of label success reflected signal, and σ is defined as:

σ=e^-2P

Then rfid interrogator success sample rate F_successCan be expressed as:

F_success=P e^-2P

It addition, number of labels is n, in the unit interval, average each label success reflected signal is η, then be represented by by the probability P of label success reflected signal:

P=n η

Thus, rfid interrogator success sample rate F_successFor:

F_success=n η e^-2nη

Theoretical according to data sampling, and if only if when sampling interval S meets certain condition, and the data of collection could reconstruct target distribution, then satisfy condition into:

$S=\frac{1}{F_{success}}$

It practice, S is defined as:

S=n e lnn+r

Owing to r is the random number between 1～n, then:

$n\&CenterDot;e\&CenterDot;\ln n<\frac{e^{2n\mathrm{\&eta;}}}{n\mathrm{\&eta;}}$

Therefore, the present invention, by controlling the tranmitting frequency of RFID reader, adjusts suitable η, it becomes possible to obtain maximum phase values within the unit interval, reduces the time of real-time counting as far as possible.

Step 4, is utilized off-line finger print information and online finger print information, is obtained the Activity Type of target to be detected by matching primitives.

The label of each object effects is different, then not having any relation between each target and each target, so the calculation that each goal activities coupling is estimated is identical with process, note step 2 is off-line phase, step 3 is on-line stage, then the computing formula of goal activities to be detected coupling is as follows:

Score(R_finger,F_finger)=N_sameλ-Dis(R_finger,F_finger)+NL_lack·NA_lack

$Dis\left(R_{finger},F_{finger}\right)=\frac{\sqrt{\underset{i=1}{\overset{N_{same}}{\mathrm{\&Sigma;}}}{\left(\frac{\underset{j=1}{\overset{NA}{\mathrm{\&Sigma;}}}{\mathrm{DTW}}_{ij}}{NA}\right)}^2}}{N_{same}}$

Score_target=argmax (Score (R_finger,F_finger(i))),i∈(1,…,N_rf)

Herein above in three formula:

R_fingerRepresent online finger print information, the tag ID affected including each target and the phase information of these labels, a corresponding one group of online finger print information in target location；

F_fingerRepresent off-line finger print information, including goal activities region, goal activities type, be affected tag ID and be affected the phase information of label, N_rfFor off-line finger print information number；

N_sameRepresent the on-line stage number of tags that be affected identical with in each group of off-line phase；

λ represents identical and is affected label weight in the matching process of position；

Dis(R_finger,F_finger) represent finger print information and one group of off-line phase finger print information comprehensive differences in all identical phase informations being affected label of on-line stage；

DTW_ijRepresent the identical difference in phase information that is affected label i on jth antenna；

NL_lackWhat in expression off-line phase and on-line stage, same antenna signal was incomplete is affected number of labels, i.e. two kinds of situations: situation one, off-line phase label Label is for being affected label, and on-line stage label Label cannot be collected the relevant information of this label by aerial array；Situation two, off-line phase label Label cannot be collected the relevant information of this label by aerial array, and on-line stage label Label is for being affected label；

NA_lackRepresent the weight that aerial signal is incomplete；

Score_targetRepresent for each target to be detected, calculate the Score (R of online fingerprint corresponding to this target to be detected and all off-line fingerprints respectively_finger,F_finger) value, obtain maximum Score (R_finger,F_finger) off-line fingerprint corresponding to value, then the goal activities type in this off-line fingerprint is required goal activities type.

First off-line finger print information is classified by the present invention, extracts containing only there being the off-line finger print information that be affected label identical with online finger print information, mates for next stage.Owing to the off-line finger print information after extracting is far fewer than all off-line finger print informations, thus greatly reducing match time.Secondly, DTW algorithm is used to calculate the online finger print information phase information similarity DTW that be affected label identical with in off-line finger print information_ij.Finally, use matching formula, obtain online finger print information R respectively_fingerWith off-line finger print information F_fingerCoupling mark.It practice, people's blocks the phase value that can cause that rfid interrogator can not read some label, namely can not construct the phase information of this label, it is impossible to carry out goal activities identification.Matching formula considers this phenomenon, but not avoids this phenomenon.

Embodiment:

The RFID system using two sets general completes test in Northwest University's information science communication experiment room with technical college, is placed with multiple tables and chairs, desktop computer, communication equipment etc. in laboratory, and indoor environment is more complicated.In this experiment, always having experiment three kinds different, for verifying effectiveness of the invention, robustness and the advantage compared with additive method, every kind of experiment has different experiment scenes, 7 volunteers complete, and continues 14 hours altogether, collects 1054400 data.

The present embodiment adopt experiment I first scene in, in scene one, be made up of the label matrix of 5*5 25 RFID tag, every pair of label be spaced apart 1.2m, thus forming the monitored area of 480cm*480cm.Two rfid interrogators and corresponding aerial array thereof place the adjacent both sides of monitored area.In this scene, always have 4 volunteers.

The multiple target passive type indoor activity recognition methods based on RFID of the present embodiment, specifically carries out according to following steps:

Step one, scene setting

Monitored area is disposed RFID passive label, adjacent label forms label matrix with 1.2m spacing, and 2 rfid interrogators are positioned over label matrix arbitrary neighborhood both sides, and each rfid interrogator connects one group of aerial array, aerial array is made up of the omnidirectional antenna that 4 spacing are 0.15m.

Label matrix is that 5 row 5 arrange, and label E PC is numbered 0001～0025, arranged in sequence.The RFID system communication distance of passive type is limited, is generally 10～20 meters, though here label spacing is 1.2m, but the communication range ability for RFID system, it has been reduce lower deployment cost as much as possible.In aerial array, antenna distance is 0.15m, arranges corresponding with the program generating phase information.

Step 2, obtains off-line phase finger print information and stores

Step 2.1:RFID read write line is provided with two, respectively XReader and YReader, two rfid interrogators send radiofrequency signal simultaneously, phase value when not having target in experiment scene that all labels return is measured and recorded to the aerial array of each rfid interrogator, and adopts PCA method to carry out the phase value after denoising obtains denoising；

Step 2.2: monitoring region is divided into multiple square area, the people selecting physique medium carries out different activities respectively according to movement reference Table A TT in each region, obtain each rfid interrogator phase information to all labels, and adopt the PCA method mentioned in step 2.1 to carry out denoising；

Step 2.3: consider that the phase information of label is changed the impact produced by system noise, phase information when having a target after denoising that when not having a target, after denoising, phase information and step 2.2 obtain step 2.1 obtained adopts DTW method to compare, respectively obtain target carry out in each region different movable time impact label；Using the region residing for target, target Activity Type, be stored in server database as off-line phase finger print information by the tag ID of object effects, the phase information being affected label and the degree that is affected.Specific implementation is as follows:

Also the change of label phase information can be caused, so needing to set the label multi-path information change that a suitable threshold value causes for judging people or environment noise due to environment noise.Concrete threshold calculations process is as follows:

Assume that ξ was that a upper moment is for judging the threshold value whether label phase information is affected by people, DTW_newRepresent that current time judges the threshold value whether label phase information is affected by people,Represent threshold deviation, then first calculate the difference Δ of current time threshold value and previous moment threshold value:

Δ=DTW_new-ξ

Use current time threshold value update ξ andObtain:

ξ=(1-0.125) ξ+0.125 DTW_new

$\&part;=(1-0.25)\&CenterDot;\&part;+0.25\&CenterDot;\left|\mathrm{\&Delta;}\right|$

Wherein, 0.125 and 0.25 is the proportion shared by up-to-date threshold value and deviation, and two formula on abbreviation draws:

ξ=ξ+0.125 Δ

$\&part;=\&part;+0.25\&CenterDot;\left(\right|\mathrm{\&Delta;}|-\&part;)$

Thus up-to-date threshold value is:

$Th=\mathrm{\&xi;}+4\&part;$

According to above-mentioned calculating, the scene lower threshold value Th obtaining the present embodiment is 0.4161.

If recording the label multi-path information difference value SLT when having people and no one_α,βMore than threshold value Th, then it is assumed that this label is for being affected label.

In order to reduce error as far as possible, the people selecting physique medium obtains the finger print information of off-line phase.Because during on-line stage activity recognition, the build of multiple targets is different, if people's difference on build that target now and off-line phase use is too big, can cause bigger activity recognition error.Off-line phase finger print information form is:

Position (x y), is affected label (tag1, tag2 ...), label multi-path information (tag1.dat, tag2.dat ...) }.

It addition, monitored area is 0.48m*0.48m, sizing grid is 0.4m, then always have 12*12=144 bar off-line finger print information.

Step 3, what acquisition on-line stage was corresponding with each rfid interrogator is affected tag set, determines that several are affected label subregion according to being affected tag set, and then determines target numbers to be detected；Obtain the online finger print information of each target to be detected simultaneously.Concrete methods of realizing includes:

Step 3.1:4 people to be detected enters detection region and carries out different activities, the aerial array of two rfid interrogators reads the phase value of all labels, and adopt the PCA method mentioned in step 2.1 that phase value carries out the phase value after denoising obtains denoising, finally the phase value information after these denoisings is stored in server database；

Step 3.2: the radiofrequency signal of two rfid interrogator transmissions is by different path to each label, namely obtaining two routing information set: set one is first rfid interrogator to the phase information after the denoising of all labels, gathering two is second rfid interrogator to the phase information after the denoising of all labels；Utilize the DTW algorithm mentioned in step 2.3 and consider that the phase information of label is changed the impact produced by system noise, label phase information in set one and set two is compared with label phase information in no one's situation respectively, obtains two set being affected label；

According to above-mentioned analysis and statistics, obtaining first tag set that is affected corresponding to rfid interrogator is:

0001,0002,0006,0007,00011,0012,0016,0019,0021,0022,0004,0005,0009,0010,0014,0015,0019,0020,0024,0024},

The tag set that is affected that second rfid interrogator is corresponding is:

{0004,0005,0003,0002,0001,0006,0007,0008,0009,00010,00016,0017,0018,0019,0020,0021,0022,0023,0024,0025}。

Step 3.3: for each rfid interrogator, corresponding is affected in tag set and has several to be affected label subregion, and subregion number is target numbers to be detected.

What first rfid interrogator X was corresponding be affected tag set can be divided into two separated subregions, is namely affected label subregion:

I{0001,0002,0006,0007,00011,0012,0016,0019,0021,0022},

It is affected label subregion:

II{0004,0005,0009,0010,0014,0015,0019,0020,0024,0024}。

For rfid interrogator X, two separated subregions indicate two targets, but owing to the label number that is affected of every sub regions is all higher than 6, then every sub regions all exists and is hidden target.Now, can be calculated second rfid interrogator corresponding be affected label subregion:

III{0004,0005,0003,0002,0001,0006,0007,0008,0009,00010},

Subregion:

IV{00016,0017,0018,0019,0020,0021,0022,0023,0024,0025}。

Now subregion I and II can be divided into four intersecting areas separated by subregion III and IV, and namely A{0001,0002,0006,0007}, B{0021,0022,0016,0017}, C{0004,0005,0009,0010}, D{0019,0020,0025,0024}.Label number in each intersecting area is respectively less than 6, then be absent from being hidden label, and now target number is the number of intersecting area, is 4；And the label in tetra-intersecting areas of A～D is the label that is affected of each target, corresponding label phase information can be extracted from any one rfid interrogator.

Step 4, utilizes the online finger print information that target each to be detected that the off-line finger print information that step 2 obtains obtains is corresponding with step 3, and matching primitives obtains the Activity Type of target to be detected.

The computing formula of one goal activities coupling estimation is as follows:

Score(R_finger,F_finger)=N_sameλ-Dis(R_finger,F_finger)+NL_lack·NA_lack

$Dis\left(R_{finger},F_{finger}\right)=\frac{\sqrt{\underset{i=1}{\overset{N_{same}}{\mathrm{\&Sigma;}}}{\left(\frac{\underset{j=1}{\overset{NA}{\mathrm{\&Sigma;}}}{\mathrm{DTW}}_{ij}}{NA}\right)}^2}}{N_{same}}$

Score_target=argmax (Score (R_finger,F_finger(i))),i∈(1,…,N_rf)

According to above-mentioned calculating, obtaining 4 target locations in this example is F7 (1,0.2), F32 (4.2,0.6), F114 (0.6,4.6), and F143 (4.6,4.4.2), wherein zero is 0001 label position.Fn represents the station location marker in off-line fingerprint, and n represents sequence number, always has 144 off-line finger print informations in this scene, and F7 represents that position (1,0.2) is the position in the 7th article of off-line finger print information.

Activity recognition compliance test result:

Experiment I:

Experiment I aims at checking effectiveness of the invention.First, setting up 3 monitored areas varied in size, i.e. 480cm*480cm, 600cm*600cm, 720cm*600cm, target numbers, number of tags and matrix information are as shown in table 1.In experimentation, gathering phase information corresponding to each label 200, arrange the interval between target is 0.3m～1.5m simultaneously, and label spacing is 1.2m.

Table 1 experiment parameter is arranged

	Target numbers	Number of tags	Area size (cm)	Label matrix size
					Monitored area 1	4	25	480*480	5*5
Monitored area 2	5	36	600*600	6*6
					Monitored area 3	6	42	720*600	7*6

The test effect of experiment I:

Object count precision: by calculating under different target spacer conditions, obtain different object count precision.Experimental result is as it is shown in fig. 7, along with the increase of target distance, the counting precision of three monitored areas all constantly promotes, and when target distance is more than 1m, object count precision is 100%.The present invention regards the target of multiple hypotelorisms as a target, but, this shortcoming of the present invention can be tolerated, because in this problem ubiquity behavior identity system (such as infrared ray).

Experiment II:

The robustness aiming at the checking present invention of experiment II.This experiment carries out in the monitored area be sized to 480cm*480cm, 720cm*720cm.Arranging the interval between target is 0.3m～1.5m, and label spacing is 0.6m, 1.2m.Target numbers, number of tags and matrix information are as shown in table 1.

The test effect of experiment II:

Object count precision: as shown in Figure 8,480*480-1.2m with 720*600-1.2m is illustrated respectively in label and is spaced apart the counting precision of two monitored areas during 1.2m, similarly, 480*480-0.6m and 720*600-0.6m is illustrated respectively in label and is spaced apart the counting precision of two monitored areas during 0.6m.Can drawing from figure, reducing label spacing and improve counting precision and inconspicuous, therefore, object count does not Qiang Zhiyaoqiu in label spacing, so the present invention has robustness in object count.

Experiment III:

In experiment III, the present invention is respectively in laboratory, spacious hall, this invention of the outdoor test robustness situation to environment.Monitoring of environmental size is 720cm*720cm, and destination number is 7, interval 1.2m between label.

The test effect of experiment III:

As can be seen from Figure 9, this system environmental testing in spacious hall effect out is best, except (B) this action of brushing teeth, other action seems sit down (S), lie down (F), pushing hands (P), take out mobile phone (T), jump (J) and do not have activity (E), and it can well be identified by this RFID system.Recognition success rate will be higher than 80% substantially.

Claims (5)

1. the multiple target passive type indoor activity recognition methods based on RFID, it is characterised in that comprise the following steps:

Step one, disposes multiple RFID passive label in monitored area, constitutes passive label array, then disposes a pair rfid interrogator；

Step 2, monitored area is put into movable target, obtaining and preserve the movable target off-line finger print information when monitored area activity, off-line finger print information includes the region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and is affected degree；

Step 3, when target to be detected enters activity in monitoring, obtain the set being affected label that each rfid interrogator is corresponding, determine according to the set being affected label and multiple be affected label subregion, and then determine target data to be detected, thus obtaining the online finger print information that each target to be detected is corresponding, online finger print information includes the tag ID of object effects to be detected and the phase information of these labels being affected；

Step 4, is utilized off-line finger print information and online finger print information, is obtained the Activity Type of target to be detected by matching primitives.

2. the described multiple target passive type indoor activity recognition methods based on RFID as claimed in claim 1, it is characterised in that the detailed process of step 2 includes:

Step 2.1, two RFID of deployment send radiofrequency signal simultaneously, and phase information when not having target in monitored area that all passive labels return is measured and recorded to the aerial array of each rfid interrogator；

Step 2.2, is divided into the square area that multiple size is identical by monitored area, utilizes a movable target to carry out different activities in each square area, obtains the phase information of all passive labels that each rfid interrogator receives；

Step 2.3, the phase information that phase information step 2.1 obtained and step 2.2 obtain compares, and respectively obtains movable target and carries out the label of impact during different activity in each square area；Region residing for target, the Activity Type of target, the tag ID being affected, the phase information of label being affected and the degree that is affected are stored in server database as off-line finger print information.

3. the described multiple target passive type indoor activity recognition methods based on RFID as claimed in claim 1, it is characterised in that the detailed process of step 3 includes:

Step 3.1, when more than one target to be detected enters monitored area activity, the aerial array of two rfid interrogators reads the phase information of all labels respectively, phase information is stored in database server；

Step 3.2, the radiofrequency signal that two rfid interrogators send arrives each label by different paths, two routing information set can be obtained: set one is first rfid interrogator phase information to all labels, set two is second rfid interrogator phase information to all labels, utilize dynamic time warping algorithm DTW and consider that the phase information of label is changed the impact produced by noise, label phase information during target is not had to compare with monitored area respectively the phase information of the label in set one and set two, obtain two set being affected label；

Step 3.3, for each rfid interrogator, corresponding is affected in the set of label and has multiple subregion being affected label, the overlapping cases being affected label subregion according to two rfid interrogator feedbacks, calculates the target numbers of target to be detected；For each target to be detected, the corresponding phase information being affected tag ID and these labels constitutes the online finger print information of this target to be detected.

4. the described multiple target passive type indoor activity recognition methods based on RFID as claimed in claim 1, it is characterised in that the method for the matching primitives described in step 4 includes:

Note step 2 is off-line phase, and step 3 is on-line stage, then the computing formula of goal activities to be detected coupling is as follows:

Score(R_finger,F_finger)=N_sameλ-Dis(R_finger,F_finger)+NL_lack·NA_lack

$Dis(R_{finger},F_{finger})=\frac{\sqrt{\underset{i=1}{\overset{N_{same}}{\&Sigma;}}{\left(\frac{\underset{j=1}{\overset{NA}{\&Sigma;}}{\mathrm{DTW}}_{ij}}{NA}\right)}^2}}{N_{same}}$

Score_target=argmax (Score (R_finger,F_finger(i))),i∈(1,…,N_rf)

Herein above in three formula:

R_fingerRepresent online finger print information, F_fingerRepresent off-line finger print information, N_sameRepresenting the on-line stage number of tags that be affected identical with in each group of off-line phase, NA is number of antennas, N_rfFor off-line finger print information number；

λ represents identical and is affected label weight in the matching process of position；

Dis(R_finger,F_finger) represent online finger print information and the one group of off-line phase finger print information comprehensive differences in all identical phase informations being affected label；

DTW_ijRepresent the identical difference in phase information that is affected label i on jth antenna；

NL_lackWhat in expression off-line phase and on-line stage, same antenna signal was incomplete is affected number of labels, i.e. two kinds of situations: situation one, off-line phase label Label is for being affected label, and on-line stage label Label cannot be collected the relevant information of this label by aerial array；Situation two, off-line phase label Label cannot be collected the relevant information of this label by aerial array, and on-line stage label Label is for being affected label；

NA_lackRepresent the weight that aerial signal is incomplete；

Score_targetRepresent for each target to be detected, calculate the Score (R of online fingerprint corresponding to this target to be detected and all off-line fingerprints respectively_finger,F_finger) value, obtain maximum Score (R_finger,F_finger) off-line fingerprint corresponding to value, then the goal activities type in this off-line fingerprint is required goal activities type.

5. the described multiple target passive type indoor activity recognition methods based on RFID as claimed in claim 2, it is characterised in that in step 2.3, it is determined whether the method for being affected label includes:

Note for store phase information file have two: D_α(i) and D_β(i), then the difference value of two phase informations is:

${\mathrm{DTW}}_{\mathrm{\&alpha;},\mathrm{\&beta;}}=min\underset{i,j=1}{\overset{L}{\&Sigma;}}|D_{\mathrm{\&alpha;}}\left(i\right)-D_{\mathrm{\&beta;}}\left(j\right)|$

In above formula, L is the number of phase information file intermediate value；

Assume that ξ was that a upper moment is for judging that whether label phase information is by the threshold value of movable object effects, DTW_newRepresent that current time judges that whether label phase information is by the threshold value of movable object effects,Represent threshold deviation, then calculate the difference Δ of current time threshold value and previous moment threshold value:

Δ=DTW_new-ξ

Use current time threshold value update ξ andObtain:

ξ=(1-0.125) ξ+0.125 DTW_new

$\&PartialD;=(1-0.25)\&CenterDot;\&PartialD;+0.25\&CenterDot;\left|\mathrm{\&Delta;}\right|$

Two formula on abbreviation, draws:

ξ=ξ+0.125 Δ

$\&PartialD;=\&PartialD;+0.25\&CenterDot;\left(\right|\mathrm{\&Delta;}|-\&PartialD;)$

Thus up-to-date threshold value is:

$\mathrm{Th}=\mathrm{\&xi;}+4\&PartialD;$

If recording phase information difference value DTW when label has movable target in monitored area and do not have movable target_α,βMore than threshold value Th, then it is assumed that this label is for being affected label.