SI24485A

SI24485A - Process of identifying the person who enters into the space

Info

Publication number: SI24485A
Application number: SI201300281A
Authority: SI
Inventors: MatjaĹľ GAMS; Rok Piltaver; Hristijan Gjoreski; AleĹˇ Moljk; Igor Gornik; Janez Polje; Mitja Virant
Original assignee: Intech Les, Razvojni Center, D.O.O.
Priority date: 2013-09-19
Filing date: 2013-09-19
Publication date: 2015-03-31
Also published as: WO2015040503A1

Abstract

Predlagani izum se nanaša na postopek identifikacije osebe, ki vstopa v prostor z omejenim dostopom, kot na primer v stanovanje in podobno. Po predlaganem izumu je predvideno, da postopek obsega zbiranje, s pomočjo senzorjev (2, 3), podatkov o odpiranju in zapiranju vstopne točke (1), ki se jih shranjuje v podatkovno zbirko (4). Temu sledi obdelava shranjenih podatkov in luščenje značilk o odpiranju in zapiranju vstopne točke (1) s pomočjo centralne procesne enote (5). Zatem se zgradi model vsakokratne osebe, ki vstopa in/ali izstopa skozi vstopno točko (1), ugotovi identiteto osebe, ki je odprla in/ali zaprla vstopno točko (1) in izbiroma posodobi model z na novo pridobljenimi podatki o osebi, ki je odprla in/ali zaprla vstopno točko (1).The proposed invention relates to the process of identifying a person entering a restricted access space, such as an apartment, and the like. According to the present invention, it is intended that the method comprises collecting, by means of sensors (2, 3), data on opening and closing the entry point (1) stored in the database (4). This is followed by the processing of stored data and exfoliation of the characteristics of opening and closing the entry point (1) using the central processing unit (5). Then, the model of each person entering and / or exiting through the access point (1) is built, the identity of the person who opened and / or closed the access point (1) and optionally updates the model with newly acquired data on a person who is opened and / or closed the access point (1).

Description

INTECH-LES, razvojni center d.o.o.INTECH-LES, Development Center d.o.o.

Postopek identifikacije osebe, ki vstopa v prostorThe process of identifying the person entering the space

Predlagani izum se nanaša na postopek identifikacije osebe, ki vstopa v prostor z omejenim dostopom, kot na primer v stanovanje in podobno.The present invention relates to a process for identifying a person entering a restricted space, such as an apartment and the like.

Postopek identifikacije osebe, ki vstopa v prostor z omejenim dostopom, je znan iz več dokumentov kot na primer iz mednarodne patentne prijave štev. PCT/AU2005/001285, patenta štev. US 8,232,860 B2, mednarodne patentne prijave štev. WO 2011/011282, patentne prijave štev. EP 1 753 206 A. Rešitve, razkrite v omenjenih dokumentih imajo kar nekaj slabosti, saj si mora uporabnik zapomniti dostopno geslo, poleg tega pa mora uporabnik pred vstopom v prostor z omejenim dostopom vsakokrat izvesti določeno identifikacijsko aktivnost, kot na primer identificirati se s pomočjo prstnega odtisa, posebne kartice in podobno. Nekateri dokumenti pa razkrivajo zgolj postopek ugotavljanja delovanja vstopnih vrat v okvari in podobno.The process of identifying a person entering a restricted space is known from several documents, such as from international patent application no. PCT / AU2005 / 001285, patent no. US 8,232,860 B2, International Patent Application Nos. WO 2011/011282, Patent Application Nos. EP 1 753 206 A. The solutions disclosed in the aforementioned documents have several disadvantages, since the user has to remember the access password and the user must perform certain identification activity before entering a restricted space, such as identifying himself with the help of fingerprints, special cards and the like. Some documents, however, only disclose the procedure for determining the functioning of the front door in a malfunction and the like.

Naloga predlaganega izuma je ustvariti nov postopek identifikacije osebe, ki vstopa v prostor z omejenim vstopom, ki je prijazen do uporabnika in s katerim so odpravljene pomanjkljivosti znanih.rešitev.It is an object of the present invention to provide a new process for identifying a user that enters a user-friendly space with limited entry, which eliminates known deficiencies.

Zastavljena naloga je po izumu rešena z značilnostmi, razkritimi v značilnostnem delu 1. patentnega zahtevka. Podrobnosti izuma so razvidne iz pripadajočih podzahtevkov. Opisani postopek rešuje problem uporabniku prijazne identifikacije oseb ob vstopu v prostor z omejenim dostopom kot na primer v sobo, hišo, zgradbo, ipd. Opisani postopek za identifikacijo uporablja vsaj en senzor, prednostno inercialni senzor, npr. pospeškomer, nameščen na vstopni točki, npr. na vratnem krilu, pri čemer se od uporabnika, razen običajnega vstopa, ne zahteva nobene dodatne akcije.The task according to the invention is solved by the features disclosed in the characteristic part of claim 1. Details of the invention can be seen in the corresponding sub-claims. The described procedure solves the problem of user-friendly identification of persons when entering a space with restricted access, such as a room, house, building, etc. The process described for identification uses at least one sensor, preferably an inertial sensor, e.g. an accelerometer mounted at an entry point, e.g. on the door leaf, with no additional action required from the user other than normal entry.

Za izvedbo postopka po izumu so potrebni vsaj naslednji elementi, ki so potrebni za izvedbo postopka po izumu.In order to carry out the process according to the invention, at least the following elements are required which are necessary to carry out the process according to the invention.

- Vstopna točka, npr. vrata, opremljena z vsaj enim senzorjem, ki je zmožen zaznavati lastnosti gibanja premičnega dela vstopne točke, npr. vratnega krila, med odpiranjem. Omenjeni senzor je lahko izbran kot na primer pospeškomer, žiroskop, magnetomer, merilnik hitrosti, merilnik kotne hitrosti, senzor gibanja, ipd. Omenjeni senzor je izbiroma lahko t.i. eno-, dvo- ali triosni MEMS pospeškomer.- Entry point, e.g. a door equipped with at least one sensor capable of detecting the movement characteristics of the moving part of the entry point, e.g. of the door leaf, during opening. Said sensor may be selected such as an accelerometer, gyroscope, magnetometer, speedometer, angular velocity meter, motion sensor, and the like. Said sensor may optionally be i.e. single, dual or triaxial MEMS accelerometer.

- Podatkovna zbirka za shranjevanje podatkov, ki jih posredujejo omenjeni senzorji, značilk vstopa in modelov vstopa.- Database for storing data transmitted by said sensors, entry characteristics and entry models.

- Centralna procesna enota, ki izvaja računski del predlaganega postopka z uporabo algoritmov umetne inteligence, to je segmentacijo zbranih podatkov, izračun značilk, strojno učenje oz. gradnjo modela vstopanja in identifikacijo na podlagi zgrajenega modela.- Central processing unit, which performs the computational part of the proposed process using artificial intelligence algorithms, ie segmentation of collected data, calculation of characteristics, machine learning, or. building an entry model and identification based on the model built.

- Vmesnik, na primer grafični uporabniški vmesnik, vmesnik do pametnega doma ali alarmnega sistema, ipd., ki posreduje rezultat identifikacije neprikazanim zunanjim sistemom.- An interface, such as a graphical user interface, an interface to a smart home or an alarm system, etc., that transmits the result of identification to non-displayed external systems.

Predlagani postopek identifikacije po izumu se da uporabiti za identifikacijo uporabnikov za personalizirano upravljanje sistemov v pametni hiši, povečanje varnosti pri vstopni kontroli, avtomatizacijo registracije časa prisotnosti zaposlenih oz. preverjanja identitete ob registraciji in podobno, pri čemer našteta področja uporabnosti ne predstavljajo nikakršne omejitve. Predlagani postopek po izumu deluje v dveh fazah, pri čemer prvo fazo predstavlja faza učenja, medtem ko drugo fazo predstavlja faza identifikacije. V prvi fazi se za vsako osebo, ki naj bi se jo identificiralo, zbere senzorske podatke o zadostnem številu prehodov skozi vstopno točko, na primer skozi vrata prostora z omejenim dostopom. Podatke o posameznem prehodu se segmentira oz. razdelijo v dva dogodka in sicer odpiranje premičnega dela vstopne točke, na primer vratnega krila, in zapiranje le-tega. Zatem se iz zbranih in segmentiranih podatkov izlušči značilke, na podlagi katerih je mogoče ločiti med vstopi in izstopi posameznih oseb skozi vstopno točko. Značilke se izluščijo ločeno za vsakokraten dogodek odpiranja in zapiranja. Značilke se izračuna iz zbranih senzorskih podatkov v časovnem in frekvenčnem prostoru ter združi v vektor značilk, ki se shrani v podatkovno bazo kot učni podatek. Ko je za vsako osebo posebej zbranih dovolj vektorjev značilk o prehodih skozi vstopno točko, se z uporabo enega ali več algoritmov strojnega učenja zgradita modela za identifikacijo osebe, ki je povzročila vsakokratni dogodek odpiranja in zapiranja. Na ta način se ločeno modelira način odpiranja in zapiranja vstopne točke. S tem je prva faza, to je faza učenja, zaključena in prične se lahko druga faza, to je faza identifikacije, ki dejansko opravlja identifikacijo oseb ob prehodu skozi vstopno točko. Ob vsakem prehodu skozi vstopno točko se ponovi postopek iz prve faze vse do trenutka, ko se shrani vektorja značilk za dogodka odpiranja in zapiranja vstopne točke. Ko se vstopna točka zapre oz. ko po odpiranju preteče določen maksimalen čas, se uporabi modela za identifikacijo, ki sta zgrajena z uporabo algoritma za strojno učenje, na vektorjih značilk, izluščenih ob trenutnem prehodu skozi vstopno točko, na podlagi katerih se identificira osebo, ki je odprla, in osebo, ki je zaprla vstopno točko.The proposed identification process according to the invention can be used to identify users for personalized management of systems in a smart house, increase security during entry control, automate the registration of time attendance of employees or. identity checks at registration and the like, with the above mentioned areas of application no limit. The present method according to the invention operates in two phases, the first phase being the learning phase while the second phase is the identification phase. In the first phase, sensory information is collected for each person to be identified for a sufficient number of passages through the entry point, for example through the door of a restricted space. The data about each transition is segmented or. they divide into two events, namely, opening and closing a moving part of an entry point, such as a door leaf. Then, from the collected and segmented data, features are extracted, which can be used to distinguish between entrances and exits of individuals through the entry point. The badges are extracted separately for each opening and closing event. The characteristics are calculated from the collected sensor data in the time and frequency space and combined into a vector of features, which is stored in the database as a learning data. When sufficient vectors of pass-through feature entry points have been collected for each person, models using one or more machine learning algorithms are constructed to identify the person who caused each opening and closing event. In this way, the method of opening and closing the entry point is modeled separately. This is the first phase, this is the learning phase, completed and the second phase can begin, which is the identification phase that actually performs the identification of persons as they pass through the entry point. Each time you pass through the entry point, the procedure from the first stage is repeated until the moment vector of the entry and closing event of the entry point is stored. When the entry point closes or. when a certain maximum time has elapsed after opening, identification models built using the machine learning algorithm shall be used on feature vectors extracted at the moment of passage through the entry point to identify the opening person and person, which closed the entry point.

Izum je v nadaljevanju podrobneje opisan na osnovi neomejujočega izvedbenega primera in s sklicevanjem na priložene skice, kjer kaže sl. 1 izvedbeni primer sistema identifikacije osebe, ki vstopa v prostor z omejenim vstopom, sl. 2 shemo delovanja postopka identifikacije, sl. 3 primer podatkov o radialnem pospešku premičnega dela vstopne točke v časovni domeni, pri čemer so omenjeni podatki razdeljeni na dogodek odpiranja in dogodek zapiranja vrat.The invention is further described in the following with reference to a non-limiting embodiment and with reference to the accompanying drawings, in which FIG. 1 is an embodiment of a system for identifying a person entering a restricted space, FIG. 2 is a diagram of the operation of the identification process, FIG. 3 is an example of radial acceleration data of a moving part of an entry point in a time domain, said data being divided into an opening event and a closing event.

Sistem (sl. 1) za izvedbo postopka identifikacije osebe, ki vstopa v prostor z omejenim dostopom, kot na primer v stanovanje in podobno, obsega vstopno točko 1, vsaj en senzor 2, 3, ki zaznava lastnosti gibanja premičnega dela vstopne točke 1, podatkovno zbirko 4, ki hrani senzorske podatke, vektorje značilk in modela za identifikacijo, centralno procesno enoto 5, ki izvaja obdelavo zbranih senzorskih podatkov in identifikacijo na podlagi algoritmov strojnega učenja, in vmesnik 6, ki posreduje rezultata identifikacije zunanjim sistemom. Našteti sestavni elementi, ki sestavljajo zgoraj omenjeni sistem, predstavljajo zgolj minimalno konfiguracijo in se jih da dopolniti z dodatnimi podobnimi in/ali sorodnimi elementi, ne da bi se s tem oddaljili od smisla in obsega izuma.The system (Fig. 1) for carrying out the process of identifying a person entering a restricted space, such as an apartment and the like, comprises an entry point 1, at least one sensor 2, 3 that detects the motion characteristics of the movable part of the entry point 1, a database 4 that stores sensor data, feature vectors, and an identification model, a central processing unit 5 that processes the collected sensor data and identification based on machine learning algorithms, and an interface 6 that transmits identification results to external systems. The enumerated constituent elements that make up the aforementioned system represent only a minimal configuration and can be supplemented by additional similar and / or related elements without departing from the spirit and scope of the invention.

Omenjeni sistem je postavljen okrog vstopne točke 1, na primer vrat prostora z omejenim dostopom. Vstopna točka 1 je opremljena senzorji 2, 3, ki zaznavajo na primer pospešek, hitrost, kot odprtosti in/ali gibanje premičnega dela vstopne točke 1. Senzorji 2, 3 so izbiroma lahko MEMS triosni pospeškomeri. Na vstopno točko 1 mora biti nameščen vsaj en senzor, prednostno pa vsaj dva ali več senzorjev 2, 3. Opis postopka v nadaljevanju je osnovan na primeru dveh senzorjev 2, 3. Prvi senzor 2 je nameščen na odpiralno sredstvo vstopne točke 1, na primer kljuko vrat, in zaznava gibanje odpiralnega sredstva. Drugi senzor 3 je nameščen na premični del vstopne točke 1, na primer na zgornji kot vratnega krila, ki je kolikor mogoče odmaknjen od osi vrtenja vratnega krila, in zaznava gibanje v primeru vstopa skozi vstopno točko 1, na primer trkanje na vratno krilo ali odpiranje letega. Senzor 2 na odpiralnem sredstvu je lahko nameščen znotraj le-tega ali pa na zunanji strani odpiralnega sredstva. Senzor 3 na premičnem delu vstopne točke 1 je lahko nameščen znotraj le-te, na primer med plastmi vratnega krila, ali na zunanji strani premičnega dela vstopne točke 1.The said system is arranged around entry point 1, such as the door of a restricted space. Entry Point 1 is equipped with sensors 2, 3 that detect, for example, acceleration, speed, opening angle and / or movement of the moving part of entry point 1. Sensors 2, 3 can optionally be MEMS three-axis accelerometers. The entry point 1 must have at least one sensor, preferably at least two or more sensors 2, 3. The procedure below is based on the example of two sensors 2, 3. The first sensor 2 is mounted on the opening means of the entry point 1, for example door handle, and detects movement of opening means. The second sensor 3 is mounted on the movable part of the entry point 1, for example, on the upper corner of the door leaf as far as possible from the axis of rotation of the door leaf, and detects movement when entering through the entry point 1, such as knocking on the door leaf or opening flying. The sensor 2 on the opening means may be located inside it or on the outside of the opening means. The sensor 3 on the movable part of the entry point 1 may be located inside it, for example between the layers of the door leaf, or on the outside of the movable part of the entry point 1.

Omenjena senzorja 2, 3 sta povezana s podatkovno zbirko 4, ki shranjuje zajete senzorske podatke, izluščene vektorje značilk in modela za identifikacijo osebe, ki je odprla in zaprla vstopno točko. Centralna procesna enota 5 obdeluje zbrane podatke, zgradi ali posodobi model za identifikacijo in končno identificira osebo pri vstopu skozi vstopno točko 1. Po vstopu osebe se izvede procesiranje zbranih podatkov in identifikacijo osebe, ki se preko vmesnika 6 posreduje zunanjim sistemom, na primer uporabniškem vmesniku, pametnemu domu ali alarmnem sistemu. Podatkovna zbirka 4, procesna enota 5 in vmesnik 6 so bodisi kot enovita naprava bodisi kot ločene naprave razporejeni na ali ob vstopni točki 1.Said sensors 2, 3 are linked to a database 4 that stores sensory data captured, extracted feature vectors, and a model to identify the person who opened and closed the entry point. Central Processing Unit 5 processes the collected data, constructs or updates the model for identification, and finally identifies the person at entry through entry point 1. Upon entry, the person processes the collected data and identifies the person, which is transmitted to external systems via interface 6, such as the user interface. , a smart home, or an alarm system. Database 4, process unit 5 and interface 6 are either arranged as a single device or as separate devices at or at entry point 1.

Postopek identifikacije po izumu je shematsko predstavljen na sl. 2. Postopek lahko deluje bodisi v fazi učenja ali fazi identifikacije. Zajem in segmentacija podatkov, luščenje značilk v časovnem in frekvenčnem prostoru, sestavljanje končnega vektorja značilk in shranjevanje le-tega ter gradnja modelov za identifikacijo z algoritmi za strojno učenje se izvajajo enako v obeh fazah delovanja postopka. Razlika med fazami je le v enem koraku, namreč modela za identifikacijo se prvič zgradita v fazi učenja, potem ko je zbranih dovolj učnih podatkov o prehodih vsake osebe, ki jo želimo identificirati. Potem ko je model zgrajen, postopek preide v drugo fazo, to je fazo identifikacije, v kateri se obstoječa modela samo še posodablja z novimi podatki o prehodih skozi vstopno točko 1 ter ju uporablja za identifikacijo osebe, ki je odprla, in osebe, ki je zaprla vstopno točko 1, na podlagi vektorjev značilk, ki sta bil pridobljena ob trenutnem prehodu skozi vstopno točko 1. Postopek identifikacije po izumu je razdeljen v fazo učenja in fazo identifikacije. V obeh fazah se postopek prične z zaznavanjem in zbiranjem podatkov, ki jih posredujejo senzorji 2, 3, o odpiranju in zapiranju vstopne točke 1. V prvem koraku 7 se zazna odpiranje vrat, ki mu sledi korak 8 zbiranja podatkov, to pomeni, da se prične zbiranje in shranjevanje podatkov, ki jih posredujejo senzorji 2, 3. Omenjeni korak 8 traja še določen čas, na primer ls, po zaznanem koncu zapiranja vstopne točke 1, ko so vrata popolnoma zaprta, ali po pretečenem predpisanem maksimalnem času začetka odpiranja. Pretek omenjenega določenega časa se zazna v koraku 9 zaznavanja zaprtja vrat. V naslednjem koraku 10 se zbrane podatke razdeli v podatke o dogodku odpiranja in podatke o dogodku zapiranja. Dogodek odpiranja traja od začetka odpiranja do trenutka, po katerem se vstopna točka 1 več ne odpira, na primer kot odprtosti vratnega krila se do konca zbranih podatkov več ne poveča. Dogodek zapiranja traja od trenutka, po katerem se vstopna točka 1 začne zapirati in je kasnejši od trenutka, ko se konča dogodek odpiranja. Dogodek zapiranja traja še kratek čas, na primer ls, po tem, ko se vstopna točka 1 popolnoma zapre ali mine predpisan maksimalen čas. Nato se s pomočjo centralne procesne enote 5 za vsak dogodek odpiranja in/ali zapiranja ločeno izvede korak 11 luščenja značilk v časovnem prostoru. Sočasno se izvaja tudi korak 12, ki podatke, pridobljene iz senzorjev 2, 3, iz časovnega prostora transformira 13 v frekvenčni prostor in izlušči še značilke za oba dogodka v frekvenčnem prostoru. V naslednjem koraku 14 se oba tipa značilk, to je značilk v časovnem prostoru in značilk v frekvenčnem prostoru, združi v končna vektorja značilk za vsak posamezen dogodek, ki se shranita v podatkovno zbirko 4. Če je zbranih dovolj vektorjev značilk, kar je v fazi identifikacije 15 vedno res, se zgradi oz., če sta že zgrajena, samo posodobi 16 modela za identifikacijo osebe, ki je povzročila odpiranje, in osebe, ki je povzročila zapiranje vstopne točke 1. Faza identifikacije se od faze učenja razlikuje le v tem, da se pred posodabljanjem modelov le-ta uporabi za identifikacijo 15 na podlagi vektorjev značilk trenutnega prehoda skozi vstopno točko 1. Podrobnosti o posamezni fazi postopka za identifikacijo so pojasnjene v nadaljevanju.The identification process of the invention is schematically illustrated in FIG. 2. The process can work either in the learning phase or the identification phase. Data capture and segmentation, peeling of features in time and frequency space, assembly and storage of the final vector of tags, and construction of models for identification by machine learning algorithms are performed equally in both phases of the process operation. The phase difference is only one step, namely, the identification models are first built in the learning phase after sufficient learning data about the transitions of each person we want to identify has been collected. Once the model is built, the process goes into the second phase, which is the identification phase, in which the existing model is only updated with new pass-through data from entry point 1 and used to identify the person who opened it and the person who opened it. closed entry point 1, based on feature vectors obtained at the current passage through entry point 1. The identification process of the invention is divided into a learning phase and an identification phase. In both phases, the process begins by sensing and collecting the data provided by sensors 2, 3, about opening and closing the entry point 1. In the first step 7, the opening of the door is detected, followed by the data collection step 8, that is, Collecting and storing of data transmitted by sensors 2, 3 begins. The aforementioned step 8 takes a certain amount of time, such as LS, after the detected end of closing the entry point 1 when the door is fully closed, or after the prescribed maximum opening time has elapsed. The expiration of said specified time is detected in step 9 of the door closing detection. In the next step 10, the collected data is divided into opening event data and closing event data. The opening event lasts from the beginning of the opening to the moment after which entry point 1 no longer opens, for example, as the door leaf is opened, it no longer increases until the end of the collected data. The closing event lasts from the moment after which entry point 1 begins to close and is later than the moment when the opening event ends. The closing event takes a short time, such as ls, after the entry point 1 has completely closed or the prescribed maximum time has passed. Thereafter, step 11 of peeling off the features in the time space is separately performed with the help of the central processing unit 5 for each opening and / or closing event. At the same time, step 12 is performed which transforms the data obtained from sensors 2, 3 from a time space 13 into a frequency space and extracts characteristics for both events in the frequency space. In the next step 14, both feature types, ie time space and frequency space tags, are combined into finite feature vectors for each individual event, which are stored in database 4. If enough feature vectors are collected, which is in phase Identification 15 is always true, build or, if already built, only update 16 models to identify the person who caused the opening and the person who caused the entry point to close 1. The identification phase differs from the learning phase only in that that, prior to updating the models, the model is used to identify 15 based on the vectors of the current transition through entry point 1. The details of each stage of the identification procedure are explained below.

V fazi učenja se izvaja zbiranje učnih podatkov za nadzorovano strojno učenje modelov za identifikacijo oseb. Več uporabnikov, to je vsi uporabniki, ki jih želimo identificirati z opisanim postopkom, mora večkrat iti skozi vstopno točko 1, pri čemer sta model in posledično identifikacija tem bolj natančna, čim več prehodov je zbranih. Za vsak primer prehoda skozi vstopno točko 1 se najprej zazna odpiranje vstopne točke 1, ki sproži shranjevanje senzorskih 2, 3 podatkov v podatkovno bazo 4, ki traja še kratek čas po tem, ko je zaznano popolno zaprtje vstopne točke 1, ali poteče vnaprej predpisani maksimalni čas po odprtju vstopne točke 1. Po tem se opravi segmentacija 10 zbranih senzorskih podatkov na dogodek 17 odpiranja in dogodek 18 zapiranja vstopne točke 1. Dogodek 17 odpiranja traja od začetka zbranih podatkov do časa ob katerem je zaznano maksimalno odprtje vstopne točke 1 v oknu zbranih podatkov. Dogodek 18 zapiranja traja od časa, ko se po maksimalni odprtosti vstopne točke 1 zazna začetek zapiranja le-te, in do konca zbranih podatkov za trenutni prehod skozi vstopno točko 1. Implementacija detektorjev začetka in konca odpiranja in zapiranja je odvisna od konkretnih senzorjev 2, 3. Začetek in konec odpiranja je npr. mogoče zaznati z magnetnim kontaktnim senzorjem odprtosti vrat, konec odpiranja in začetek zapiranja pa s senzorjem kota odprtosti vstopne točke 1, na primer primerno nameščenim potenciometrom. Kot je prikazano na sl. 3 se iz senzorjev pridobi podatke o dogodku 17 vstopa in dogodku 18 izstopa v časovnem prostoru. Te podatke se z uporabo hitre Fourierjeve transformacije 13 preračuna še v podatke predstavljene v frekvenčnem prostoru. V naslednjem koraku se izračuna relevantne značilke o načinu odpiranja in zapiranja vstopne točke 1 v časovnem 11 in frekvenčnem 12 prostoru, ki so primerne za gradnjo modelov za identifikacijo in omogočajo ločevanje med prehodi posameznih oseb. V časovnem prostoru se izlušči statistične značilke, kot so na primer minimalna, maksimalna in povprečna vrednost, integral, smerni koeficient linearne interpolacije, standardna deviacija ter povprečna razdalja od linearne interpolacije, za vsako podfazo vsakega dogodka na vsakem izmed direktno izmerjenih ali posredno izračunanih parametrov pridobljenih iz senzorskih podatkov v časovnem prostoru, na primer kotni pospešek, kotna hitrost, pospešek, hitrost, kot odprtosti. Kot značilke se upošteva tudi trajanje posameznih podfaz za posamezen parameter in dogodek. Vsak parameter za vsak dogodek je razdeljen na podfaze po naslednjem pravilu: nova podfaza se začne, ko vrednost parameter doseže lokalni maksimum, minimum ali spremeni predznak (črtkane črte na sl. 3). Vrednosti magnitude pri posameznih frekvencah so uporabljene kot značilke v frekvenčnem prostoru. Značilke v časovnem in frekvenčnem prostoru se za vsak dogodek združi 14 v po en vektor značilk, ločeno za odpiranje in zapiranje vstopne točke, ki se ju shrani v podatkovno zbirko 4 kot podatka za nadzorovano strojno učenje modelov za identifikacijo. Postopek se ponovi za vsak prehod vsakokratne osebe, ki naj se jo identificira. Ko je zbranih dovolj vektorjev značilk prehoda skozi vstopno točko 1 za vsakokratno osebo, se faza učenja zaključi z gradnjo modelov za identifikacijo z algoritmom za strojno učenje, na primer C4.5, SVM, naivni Bayesov klasifikator, metoda najbližjih sosedov z uporabo t.i. Dynamic Time Warping-a kot mere za razdaljo, ipd.. V koraku 15 se gradi ločena modela za identifikacijo osebe, ki je odprla vstopno točko 1, in identifikacijo osebe, ki jo je zaprla, saj je mogoče, da sočasno vstopi več kot ena oseba, pri čemer oseba, ki je odprla vstopno točko 1, ni ista kot oseba, ki jo je zaprla. Oba modela se • · · · · shrani v podatkovno zbirko 4, zatem pa postopek po izumu preide v fazo identifikacije.During the learning phase, learning data is collected for supervised machine learning of models for identifying persons. The more users, that is, all the users we want to identify by the procedure described, must repeatedly go through entry point 1, the more accurate the model and, consequently, the identification, the more transitions are collected. In each case of entry point 1 opening, entry point 1 is detected first, triggering the storage of sensor 2, 3 data in the database 4, which persists for a short time after the entry point 1 is completely closed or expires in advance. the maximum time after entry point 1 is opened. Segmentation of 10 collected sensor data is then performed on the opening event 17 and the entry point closing event 18. The opening event 17 lasts from the beginning of the collected data until the time at which the maximum opening point 1 is detected in the window is detected of the data collected. The closing event 18 runs from the time when the start point of the entry point 1 is detected after the maximum openness of the entry point and until the end of the collected data for the current passage through the entry point 1. The implementation of the start and end detectors of the opening and closing depends on the specific sensors 2, 3. The beginning and the end of the opening is e.g. can be detected by the magnetic contact sensor of the door opening and the end of the opening and the beginning of the closing by the sensor of the opening angle of the entry point 1, for example a suitably fitted potentiometer. As shown in FIG. 3, information is obtained from the sensors about the entry event 17 and the exit event 18 in the time space. Using this fast Fourier transform 13, this data is converted to the data presented in the frequency space. In the next step, the relevant characteristics of the method of opening and closing entry point 1 in time 11 and frequency 12 are calculated, which are suitable for the construction of models for identification and allow for separation between the transitions of individual persons. In time space, statistical characteristics such as minimum, maximum and average values, integral, directional linear interpolation coefficient, standard deviation and average distance from linear interpolation are extracted for each subphase of each event on each of the directly measured or indirectly calculated parameters obtained. from sensory data in temporal space, for example, angular acceleration, angular velocity, acceleration, velocity, angle of openness. The duration of each subphase for each parameter and event is also taken into account as characteristics. Each parameter for each event is divided into sub-phases according to the following rule: a new sub-phase begins when the parameter value reaches the local maximum, minimum or changes sign (dashed lines in Fig. 3). Magnitude values at individual frequencies are used as characteristics in the frequency space. Time and frequency space tags are combined for each event 14 into one feature vector, separately for opening and closing the entry point, which is stored in database 4 as data for supervised machine learning of identification models. The process is repeated for each passage of each person to be identified. When enough vectors of pass-through feature entry 1 have been collected for each person, the learning phase ends with the construction of models for identification with a machine learning algorithm, for example C4.5, SVM, naive Bayes classifier, nearest-neighbor method using i.i. Dynamic Time Warping as distance measures, etc. In Step 15, a separate model is built to identify the person who opened Entry Point 1 and identify the person who closed, as more than one can enter at the same time. a person whose entry point 1 is not the same as the person who closed it. Both models are stored in database 4, after which the process of the invention proceeds to the identification phase.

V fazi identifikacije gre skozi vstopno točko 1 neznana oseba, podatki iz senzorjev 2, 3 pa se zajame 8 in obdela enako kot v fazi učenja. Najprej se zazna odpiranje 7 in zbira 8 senzorske podatke do zaznanega zaprtja 9 ali preteka časovne omejitve. Za tem se podatke ločijo 10 v dogodek 17 odpiranja in dogodek 18 zapiranja ter se za vsakega izmed njih izlušči opisana vektorja značilk, ki vsebujeta značilke izračunane v časovnem in frekvenčnem prostoru. Za tem se na podlagi izluščenega vektorja in zgrajenega modela za identifikacijo izvede sama identifikacija 15 osebe, ki je povzročila dogodek odpiranja, in osebe, ki je povzročila dogodek zapiranja vstopne točke 1. Identiteti se preko vmesnika 6 posreduje zunanjim sistemom, vektorja značilk pa se skupaj z prej zbranimi in shranjenimi vektorji značilk uporabi za posodabljanje modelov za identifikacijo, kar omogoča samodejno prilagajanje spremembam v načinu uporab vstopne točke 1. S tem se postopoma zvišuje točnost delovanja opisane metode za identifikacijo.In the identification phase, an unknown person goes through entry point 1 and the data from sensors 2, 3 is captured 8 and processed the same as in the learning phase. First, the opening 7 is detected and collects 8 sensor data until the detected closure 9 or the time limit expires. Thereafter, the data are separated 10 into the opening event 17 and the closing event 18, and for each of them the described feature vectors are extracted, containing the features calculated in the time and frequency space. After that, based on the hatching vector and the identification model built, the identification of the 15 person who caused the opening event and the person who caused the entry point 1 closure event is carried out. Identity is transmitted to external systems via interface 6 and the feature vector is combined uses the previously collected and stored feature vectors to update models for identification, allowing automatic adaptation to changes in the mode of use of entry point 1. This gradually increases the accuracy of the operation of the described identification method.

Claims

Patent claims

A process for identifying a person entering a restricted space, such as an apartment or the like, characterized in that it comprises (a) collecting, by means of sensors (2, 3), information about opening and closing the entry point (1) ;

(b) storing the opening and closing data of the entry point (1), collected by sensors (2, 3), in the database (4);

(c) processing the stored data and peeling (11, 12) of the opening and closing point (1) features by means of a central processing unit (5);

(d) constructing (16) a model of each person entering and / or exiting through the entry point (1);

(e) establishing (15) the identity of the person who opened and / or closed the entry point (1); and (f) updating (16) the model with newly acquired information about the person who opened and / or closed the entry point (1).

The method according to claim 1, characterized in that the processing of the stored data on opening and closing of the entry point (1) by means of the central processing unit (5) comprises (a) segmentation (10) of the data on opening and closing of the entry point (1) collected by means of sensors (2, 3), the opening event (17) and the closing event (18);

(b) obtaining (11) a vector of time-dependent opening and / or closing entry point (1);

(c) transforming (13) the data obtained in the previous step into a frequency space;

(d) obtaining (12) a feature vector in the frequency space; and (e) combining (14) the feature vector in the time space and the feature vector in the frequency space into a single feature vector.