WO2015040503A1

WO2015040503A1 - Identification method of a person entering a room

Info

Publication number: WO2015040503A1
Application number: PCT/IB2014/059576
Authority: WO
Inventors: Matjaž GAMS; Rok PILTAVER; Hristijan GJORESKI; Aleš MOLJK; Igor GORNIK; Janez POLJE; Mitja VIRANT
Original assignee: Intech-Les, Razvojni Center, D.O.O.
Priority date: 2013-09-19
Filing date: 2014-03-10
Publication date: 2015-03-26
Also published as: SI24485A

Abstract

The present invention relates to a method of identifying a person entering a room with limited access, i.e. and apartment and the like. According to the invention it is designed that the method includes collecting information about opening and closing access point (1) by using sensors (2, 3), which are stored in a database (4). This is followed by processing the stored information and gathering indicators about opening and closing access point (1) by using a central processing unit (5). Afterwards a model of every person entering and/or exiting through access point (1) is designed, the identity of a person, opening and/or closing access point (1) is defined, and optionally, a model with newly acquired information about a person opening and/or closing access point (1), is upgraded.

Description

Identification method of a person entering a room

The present invention refers to an identification method of a person entering a room with limited access, i.e. into an apartment and the like.

The identification method of a person entering a room with limited access has been disclosed in several documents, such as the international patent application No. PCT/AU2005/001285, patent No. US 8,232,860 B2, international patent application No. WO2011/011282 and patent application No. EP 1753 206 A. However, the solutions disclosed in the said documents have a number of disadvantages, as the user is supposed to remember an access code, and additionally, prior to entering a room with limited access, the user has to perform a particular identification activity every time, i.e. fingerprint identification, use a special card, etc. Certain documents disclose only a method, defining the function of the access door in malfunction and the like.

The object of the present invention is to create a new identification method of a person entering a room with limited access which is user-friendly and eliminates the deficiencies of the already known solutions.

According to the invention the object as set above is solved by means of the characteristics, disclosed in the characterizing portion of the claim 1. The invention details are further disclosed in sub-claims. Said method solves the problem of a user-friendly identification of a person upon entering a room with limited access, i.e. into a room, house, building, etc. Said method uses at least one sensor for identification, preferably an inertial sensor, i.e. an accelerometer, located on the access point, e.g. on a door leaf, whereby a user is not required to perform any additional task other than entering a room. According to the invention the following set of elements is required to carry out said method:

Access point, i.e. the door, equipped with at least one sensor, capable of detecting movement characteristics of a moving part of the access point, .i.e. a door leaf during opening. The said sensor may be selected as for example, an accelerometer, a gyroscope, a magnetometer, a speedometer, an angular velocity gauge, movement sensor and the like. The said sensor my optionally be a so called one-, two- or three-axial MEMS accelerometer.

Database for storing data, transmitted by the said sensors, entry indicators and entry models.

Central processing unit, performing the calculations of the present method by using algorithms of artificial intelligence, i.e. segmenting the collected data, calculating indicators, mechanical learning or designing the entry model and identification on the basis of the model designed.

- Interface, i.e. a graphic user interface, a smart home or alarm system interface and the like, transmitting the identification result to the external systems not shown.

According to the invention the present identification method may be used to identify users for personalized system management in a smart house, to increase entry control safety, to automate time registration for the employees or to check the identity upon registration and the like, whereby the said applicability areas do not present any limitation whatsoever. According to the invention the present method works in two stages, whereby the first stage is a learning stage and the second an identification stage. In the first stage sensory information on sufficient number of passages made through the access point (i.e. through the door, entering a room with limited access) is collected for every person to be identified. Information on individual passage is segmented or divided into two events: opening a moving part of the access point, i.e. a door leaf, and closing it. Afterwards, indicators are gathered from the collected and segmented information providing the possibility to distinguish between entries and exits made by individual persons through the access point. The indicators are gathered separately for each opening and closing event. They are calculated from the collected sensory information in a time and frequency domain and combined into a vector or indicators, which is stored in a database as a learning piece of information. When enough vectors of indicators on passages made through the access point are gathered for every person respectively, two models are designed by using one or more algorithms of mechanical learning for the identification of a person, respectively for the events of opening and closing. Therefore the mode of opening and closing the access point is separately designed. This ends the first, learning stage and the second, identification stage can begin, which actually performs the identification of the persons passing through the access point. At each passing through the access point the first stage process is repeated until the vectors of indicators for opening and closing events of access points are stored. When the access point closes, i.e. when a certain amount of maximum time passes after opening, the identification models are used, designed by algorithms for mechanical learning on vectors of indicators, gathered at current passages made through the access point, on the basis of which a person, opening and closing the access point, is identified.

The invention is further described in detail on the basis of a non-limiting embodiment and by referring to the accompanied drawings, where: Fig.l shows an embodiment of the identification system of a person entering a room with limited access,

Fig.2 shows a schematic operation of the identification method,

Fig.3 shows an example of information on radial acceleration of the moving part of the access point in time domain, whereby the said information is divided to the opening and closing events.

The system (Fig.l) for carrying out an identification method of a person entering a room with limited access, i.e. an apartment and the like, comprises an access point 1, at least one sensor 2, 3, detecting movement characteristics of a moving part of access point 1, database 4, saving sensor information, vectors of indicators and identification models, central processing unit 5, processing the collected sensor information and identification on the basis of algorithms of mechanical learning and an interface 6, transmitting the results of identification to the external systems. The above mentioned integral part forming the said system represent only a minimal configuration and can be supplemented with additional similar and/or related elements without departing from the spirit and scope of the invention.

The said system is placed around access point 1, i.e. the door of a room with limited access. Access point 1 is equipped with sensors 2, 3, detecting i.e. acceleration, speed, such as opening and/or movement of the moving part of access point 1. Sensors 2, 3 can optionally be MEMS three-axial accelerometers. At least one sensor must be placed on access point 1, preferably at least two or more sensors 2, 3. The first sensor 2 is placed on the opening element of access point 1, i.e. the door handle and detects movement of the opening element. The second sensor 3 is placed on the moving part of access point 1, i.e. on the upper angle of the door leaf, which is located as far away from the rotation axis of the door leaf as possible and detects movement in case of entering through access point 1, i.e. knocking on the door or opening it. Sensor 2 on the opening element may be placed on the internal or external side of the opening element. Sensor 3 on the moving part of access point 1 may be placed inside it, i.e. between layers of door leaf or on the external side of the moving part of access point 1.

Said sensors 2, 3 are connected to the database 4, which stores the collected sensor information, gathered vectors of indicators and identification models of a person opening and closing the access point. Central processing unit 5 processes the collected information, designs or upgrades the identification model and finally identifies a person entering access point 1. Upon entry of the person the processing of collected information and person identification runs through and is transmitted to external systems via interface 6, i.e. to a user interface, a smart home or an alarm system. Database 4, processing unit 5 and interface 6 are either a uniform device or separate devices, allocated on or at the access point 1.

According to the invention the identification method is schematically presented by Fig. 2. The method may function either in a learning stage or in the identification stage. The collection process and segmentation of information, gathering indicators in time and frequency domain, assembling the final vector of indicators, storing it and designing identification models with algorithms for mechanical learning are performed in the same way in both stages of the method functioning. The difference between the stages is only in one step, being the point when identification models are first designed in the learning stage after sufficient learning information is collected about the passages made by every person we wish to identify. After the model is designed the method proceeds to a second stage, the identification stage, in which the existing models are only upgraded with new information about passages made through access point 1 and are used for the identification of a person opening and the person closing access point 1, on the basis of the vectors of indicators, which were acquired at the moment of passing through access point 1. According to the invention the identification method is divided into learning stage and identification stage. In both stages the method begins by detecting and collecting information, transmitted by the sensors 2, 3, about opening and closing access point 1. In the first step 7 the opening of the door is detected, followed by step 8 of collecting the information, i.e. the collecting and storing of information, transmitted by sensors 2, 3 then begin. The mentioned step 8 lasts for a particular time, e.g. 1 s after detecting the closing of access point 1 when the door is completely closed or after the set maximum time already passed, when initial opening of the door is detected. The passing of the set maximum time is detected in step 9, which detects closing of the door. In step 10 the collected information is divided into the information about an opening event and information about a closing event. The opening event lasts from the beginning of opening until the moment after access point 1 no longer opens, i.e. when the door leaf is open, the size of the collected information no longer increases. The moment of closing lasts from the moment, when access point 1 starts to close and appears later than the moment of ending the opening event. The closing event lasts for a short while, e.g. 1 s, after access point 1 closes completely or a set maximum time passes. Afterwards step 11 of gathering indicators in time domain is performed by using a central processing unit 5 for every opening and/or closing separately. Simultaneously step 12 is also in progress, transforming 12 the information, collected from sensors 2, 3 from time domain into the frequency domain and furthermore gathers the indicators for both events in the frequency domain. In the next step, step 14, both types of indicators, i.e. indicators in time domain and indicators in frequency domain are combined into final vectors of indicators for every individual event and they are stored into database 4. If sufficient vectors of indicators are collected, which is always the case in the identification stage 15, identification models are designed; if already designed, models of identification of a person opening and a person closing access point 1, are only upgraded. The identification stage differs from the learning stage in the fact that prior to upgrading the models the identification stage is used for identification 15 on the basis of vectors of indicators of current passage made through access point 1. The details about individual stages of identification method are explained in the continuation. In the learning stage collecting of learning information is performed for controlled mechanical learning of identification models of persons. More users, i.e. all the users that are to be identified by using the said method, should pass through access point 1 several times, whereby the model and consequently the identification become more accurate, if as many passages as possible are collected. For every case of passing through access point 1 first opening of access point 1 is detected, triggering storing of sensor 2, 3 information into the database 4, which lasts for a while after a complete closure of access point 1 is detected or preliminary set maximum time sequence passes after opening access point 1. Afterwards, the segmentation 10 of the selected sensor information related to event 17 of opening and event 18 of closing access point 1 is performed. Opening event 17 lasts from the beginning of the collected information until the moment, when maximum opening of access point 1 is detected in the window of collected information. Closing event 18 lasts from the moment, when after maximum opening of access point 1 the beginning of its closing is detected and until the end of the collected information for the current passage through access point 1. Implementation of detectors, measuring beginning and ending of opening and closing, depends of definite sensors 2, 3. The beginning and ending of opening can be detected with magnetic contact door opening sensor, whereas the ending of opening and the beginning of closing can be detected with the sensor, measuring the angle of openness of access point 1, i.e. with appropriately placed potentiometer. As shows Fig.3 information is acquired through sensors about the event 17 of entering and event 18 of exiting within time domain. By using a fast Fourier transform 13 such information is converted into information, presented within frequency domain. Relevant indicators about the mode of opening and closing of access point 1 in time domain 11 and frequency domain 12 are calculated; these indicators are appropriate for designing identification models and provide the possibility to distinguish between the passages made by individual persons. Statistical indicators are gathered in time domain, as for example a minimal, maximum and mean value, integral, direction coefficient of linear interpolation, standard deviation and average distance from linear interpolation for every sub-stage of every event on each of the directly measured or indirectly calculated parameters, acquired from sensor information in time domain, i.e. angular acceleration, angular velocity, acceleration, speed, angle of openness. Indicators also include the period of how long individual sub-stages last for individual parameter and event. Every parameter for every event is divided to sub- stages according to the following rule: a new sub-stage begins, when the parameter value reaches a local maximum, minimum or changes its advance sign (dashed lines of Fig.3). Magnitude values in individual frequencies are used as indicators in frequency domain. Indicators in time and frequency domain are combined 14 for every event into one vector of indicators, separately made for opening and closing of access point, which are stored into data base 4 as two pieces of information for controlled mechanical learning of identification models. The method is repeated for every passage made by every person that should be identified. When enough indicator vectors of passages through access point 1 for every person are collected, the learning stage ends by designing identification models with algorithms for mechanical learning, i.e. C4.5, SVM, naive Bayes classifier, the nearest neighbour method by using a so called Dynamic Time Warping as a distance measure and the like. In step 15 separate identification models for a person opening access point 1 are designed and the identification of a person closing it, as it is possible that more than just one person makes an entrance, whereby the person opening access point 1 is not the same as the person closing it. Both models are stored into data base 4 and afterwards the method, according to the invention, proceeds into identification stage.

In the identification stage an unknown person goes through access point 1, the information from sensors 2, 3 are collected 8 and processed in the same way as in the learning stage. First the opening 7 is detected and sensor information is collected 8 until closing 9 is detected or until time limit passes. Afterwards the information is divided 10 into event 17 of opening and event 18 of closing; for each of them, the described vectors of indicators are gathered, containing the indicators, calculated in time and frequency domain. On the basis of a gathered vector and designed identification model the identification 15 of a person, causing the event of opening, runs through together with the identification of a person causing the event of closing access point 1. The identities are transmitted to external systems via interface 6 and the vectors of indicators are used together with previously collected and stored vectors of indicators in order to upgrade the identification models; this provides automatic adaptation to changes in the mode of using access point 1. This gradually increases the accuracy of the operation of the said identification method.

Claims

Method for identifying a person entering a room with limited access, i.e. into an apartment and the like, characterized in that it comprises

a) collecting information about opening and closing access point (1) by using sensors (2, 3);

b) storing information about opening and closing access point (1), collected by using sensors (2, 3), into a database (4);

c) processing collected information and gathering (11, 12) indicators about opening and closing access point (1) by using a central processing unit (5); d) designing (16) a model for every person entering and/or exiting through access point (1)

e) defining (15) the identity of a person, opening and/or closing access point (1); and

f) upgrading (16) the model with newly acquired information about a person opening and/or closing access point (1).

Method according to claim 1, characterized in that the processing of stored information regarding opening and closing access point (1) by using a central processing unit (5) comprises the following:

a) segmenting (10) information about opening and closing access point (1), collected by using sensors (2, 3) for the event (17) of opening and event (18) of closing;

b) acquiring (11) vector of indicators about opening and/or closing access point (1) in time dependence;

c) transformation (13) of information, acquired by following a previous step, into frequency domain;

d) acquiring (12) vector of indicators in frequency domain; and

e) combining (14) vector of indicators in time domain and vector of indicators in frequency domain into a uniform vector of indicators.