GB2536507A

GB2536507A - Lighting means and motion detection

Info

Publication number: GB2536507A
Application number: GB1504824.2A
Authority: GB
Inventors: Somaraju Abhinav
Original assignee: Tridonic GmbH and Co KG
Current assignee: Tridonic GmbH and Co KG
Priority date: 2015-03-16
Filing date: 2015-03-16
Publication date: 2016-09-21
Anticipated expiration: 2035-03-16
Also published as: GB201504824D0; GB2536507B

Abstract

Means for identifying the presence or absence of a person in an area monitored by a motion sensor 1 that operates lighting controller 2 to turn light source 4 on or off. The motion sensor autonomously detects if large, small or no movements have occurred and alters detection thresholds accordingly increasing or decreasing sensitivity. The history of the detection data may be utilised possibly using a hidden Markov model and further applying a Viterbi algorithm to the model to detect motion in the monitored area.

Description

Lighting means and motion detection * * * * * * * * * * * * * * * * * * * * . * * * * * * * * * The present invention relates to the field of motion detection and of motion sensors or presence sensors. The invention particularly relates to a method for detecting motion and to a device and system for performing such io detection.

The use of a motion sensor in combination with lighting means is already known in the state of the art. So-called motion or presence sensors actually react on motion and are connected to lighting means in such a way that light is turned on when motion is detected by the motion sensor. On the other hand, light is turned off after a fixed, user configured, delay time during which no motion is detected. *

The task of the motion sensor is to detect and signal a motion in order to control the lighting means according to this signalled motion. Thereby the sensor detects a motion in a known manner as soon as a motion signal is above a threshold, said motion signal reflecting the amount or intensity of motion within the area covered by the sensor. If the motion signal or its amplitude is below the threshold, no motion is detected. On the other hand, a motion signal above said threshold will trigger a motion detection and a corresponding control of the lighting 3o means. It is known that this threshold can be configured so as to achieve a desired sensitivity of the motion sensor.

The behaviour of lighting means controlled on the basis of a motion sensor thus depends on the sensitivity of the motion sensor and on the time delay after which the lighting means can be turned off. It is known to configure the sensitivity and the time delay with potentiometers, dip-switches or with proprietary configuration tools like e.g. IR remote controls.

This known configuration is problematic in that io potentiometers require the sensor to be accessible, and IR remote controls require the sensor to be visible. Further on, both solutions require additional hardware components. * * * A further drawback is that the choice of sensitivity and time delay is left to the installer, which means additional work for the installer. Such a choice is rather * * * difficult and cannot be accurate. It can be repeatable * * * * * * * * * * * * * * only with coarse settings. * * * * * * * * * The settings also depend on the environment in which the motion sensor is installed. Long time delays and high sensitivity are e.g. necessary in applications where occupants will spend long time in a room without generating large motion, like in an office. On the other hand, short delay time is necessary in corridors where lights are needed only when someone is passing by. That is, it is disadvantageous that each application requires proper, different configuration.

3o In any case the current solution will be triggered by a person leaving the monitored area and will thus waste energy. * * * *

The present invention proposes an improved method for avoiding the disadvantages mentioned above.

According to an aspect of the invention, it is proposed a method for autonomously detecting a presence state or a non-presence state of people in an area covered by a motion sensor. The motion sensor changes autonomously its behaviour depending on ongoing motion sensing.

lo This implies for example that it is not necessary for an installer to configure the detection and the motion sensor. This is an improvement vis-a-vis sensors of the state of the art that need to be configured in the field since the user that configures a sensor has no way to determine what the optimal settings are and for this operation the sensor has to be accessible or even visible. The autonomous behaviour change also eliminates the disadvantage that with state of the art sensors the lights always remain on for a time as long as the timer delay after the occupants have gone away.

According to a further aspect of the invention, it is proposed a motion sensor for detecting a presence state or a non-presence state of people in an area covered by a motion sensor. The motion sensor is adapted to change autonomously its behaviour depending on ongoing motion sensing.

According to a further aspect of the invention, it is 3o proposed a system comprising such motion sensor, and a controlling means connected to the motion sensor for controlling a lighting means. The controlling means is adapted to control the lighting means depending on the state detected by the motion sensor. Preferably, the * * * * * * * e * * * 0 * * * * * * * * * * * * * * * * * * * * controlling means can be adapted to switch on the lighting means if the presence state is detected and to switch it off if the non-presence state is detected.

Preferably, the detection of the presence state or the non-presence state can depend on the history of the motion sensing. This means that the probability of being in the presence state or non-presence state depends on the history of the observed motion.

Preferably, the motion sensor can measure a signal reflecting the amount of motion in the covered area, and classify the measured signal into at least three different event categories comprising a no motion event, a small motion event and a large motion event. The detection of the presence state or the non-presence state can depend on * * . * .. * * * *** * * * . ** * the event category. *

Preferably, a no motion event can be identified if the. ..

amplitude of the signal is below a no motion threshold, a large motion event can be identified if the signal amplitude is above a large motion threshold, and a small motion event can be identified if the signal amplitude is between the no motion threshold and the large motion * * * ** threshold.

Preferably, the presence state and the non-presence state are detected by means of a hidden Markov model, the presence state and the non-presence state being states of 3o the hidden Markov model.

Preferably, the hidden Markov model comprises a third transition state between the presence state and the non-presence state.

Preferably, the transition probability from the transition state to the transition state can be zero.

Preferably, the transition probability between the presence state and the non-presence state can be zero.

Preferably, the estimated transitions from the transition state to the presence state or the non-presence state can lo be monitored. Depending on this monitoring it is deduced e.g. by the sensor whether the motion sensor is used in an office application or a corridor application. Preferably the transition probabilities can be adapted accordingly. * * * * . .

0 * * * Preferably, the no motion event, the small motion event and the large motion event are observations of the hidden Markov model. *

0 * * Preferably, the large motion event observation can have an * * * * * * * * * 20 emission probability of zero for the presence state and * the non-presence state.

Preferably, the highest emission probability of the transition state can correspond to the large motion event 25 observation.

Preferably, the set of observations can have different emission probabilities depending on the state of the hidden Markov model.

Preferably, the motion sensor can be adapted to detect a particular user movement, like the user waving his hand(s), interpret this movement as an indication that a * * *** * * * ** * * detected state is not correct, and update the emission probabilities of the observations accordingly.

Preferably, the presence state (P) and the non-presence state are detected by applying a Viterbi algorithm on the hidden Markov model.

The invention will be explained in the followings together with Figures.

Figure 1 shows a schematic representation of a system according to the present invention. * .

* * * * * * * * * * * * * * * * * * * * * * * * * * Figure 2 is a schematic representation of an is implementation of a statistical model according to the present invention.

Figure 3 shows an algorithm for determining the most likely state in the model of Figure 2. * 0.

Figure 1 shows a schematic representation of a system 10 according to the present invention comprising a motion sensor or motion detector 1, a control device 2 as well as a light source or lighting means 4. The light source 4 shown in Figure 1 is representative of any kind of light source and can e.g. correspond to an LED in form of an organic or inorganic light emitting diode, or to a gas discharge lamp. Instead of one light source 4, the system may comprise a plurality of light sources that can be 3o connected in series, in parallel or according to a more complex combined serial and parallel arrangement.

The light source 4 is connected to the control device 2 by means of a dedicated connection i.e. dedicated wires 3.

The control device 2 can also be referred to as an operating device or a ballast and has the function of supplying the light source 4 with electrical energy, i.e. with current. The control device 2 is connected to mains power supply 5 and is adapted to transfer energy to the light source 4 in a known manner. In order to operate the light source 4, the control device 2 can be in the form of a switched-mode power supply and can comprise a switched converter for generating a desired voltage for the light io source 4 and a desired current through the light source 4. The switched converter can be a buck converter, a boost converter, a flyback converter, or e.g. a resonance converter. * **

* ** *** * The motion sensor 1 is adapted to detect moving objects * * and particularly people. The detection is performed in a given area that is covered by the motion sensor 1, said * * covered area being the field of view of the sensor.

** * Advantageously, the covered area corresponds to the zone * * * * ..

illuminated by the light source 4. Alternatively, the covered area comprises said illuminated zone or is comprised in said illuminated zone. The motion sensor 1 is particularly adapted to sense a change in position of an object or a person within said covered area. The method used by the motion sensor 1 to detect motion is known in the art and can be based on a technology like e.g. passive infrared (PIR) motion detection, microwave, radar, ultrasonic or tomographic motion detection, or video camera motion detection.

The control device 2 comprises an input connection D1, D2 to connect the motion sensor 1. The motion sensor 1 is connected to the mains power supply 5 as well as to the control device 2, such that it can apply to the input connection DI, D2 of the control device 2 a coded voltage that transmits information regarding a detected motion or presence in the covered area. The voltage applied to the input connection DI, D2 is coded in that it can indicate to the control device 2 whether or not a motion has been detected, or whether or not a presence has been identified. E.g. if no motion or no presence is detected, the motion sensor 1 can apply a zero voltage to said input DI, D2, while a voltage of a defined amplitude can be applied when a motion or a presence is detected. E.g. the voltage applied in case of a detected motion or detected presence can be the mains supply voltage. The voltage applied to the input connection DI, D2 can be coded in a different way.

The coded voltage applied at the input connection D1, D2 is decoded by the control device 2. After decoding, the control device 2 controls the light source 4 depending on the decoded information. E.g. the light source 4 is turned on if the decoded information relates to a presence state P, i.e. when the presence of people is detected, and is turned off if the decoded information indicates a no presence state NP. In case of a no presence state, i.e. in case the motion sensor 1 has identified that nobody is present in the covered area, the light source can alternatively be dimmed.

According to the present invention, the presence state P and the non-presence state NP related to the presence of 3o people in the area covered by the motion sensor 1 are detected autonomously by the motion sensor.

The behaviour of the motion sensor 1, i.e. particularly the detection of the presence or no presence state, may * * * * . * * . change depending on ongoing motion sensing of the motion sensor 1. The detection of the presence state or the non-presence state may depend on the history of the motion sensing. * ** * * * *^** * **** * **** * * * ** * * * * * **

The motion sensor 1 is adapted to measure a signal reflecting the amount of motion in the covered area. Such signal is known in the art. Further on, the sensor can classify the measured signal into at least three different to event categories comprising following events: - a no motion event that reflects the fact that no motion occurs in the covered area, - a small motion event that corresponds to a detected 15 motion having a low intensity or amplitude, and - a large motion event corresponding to a detected motion with a higher intensity or amplitude than the small motion.

The no motion event is identified e.g. by the motion sensor 1 if the amplitude of the measured signal reflecting the amount of motion is below a first threshold referred to as no motion threshold. The large motion event can be identified if the signal amplitude is above a second threshold referred to as large motion threshold. The small motion event can be identified if the signal amplitude is between the no motion threshold and the large motion threshold.

3o The detection of the presence state P or the non-presence state NP depends on the event category, i.e. depends on whether a no motion event, a small motion event, or a large motion event has been identified.

Figure 2 is a schematic representation of an implementation of a model 20 or statistical model according to the present invention.

The model 20 shown is particularly a Markov model that comprises a chain of states with transitions between the different states. The model 20 comprises two states that are a presence state P corresponding to a presence of at least one person in the area covered by the motion sensor 1, as well as a non-presence state NP reflecting the fact that nobody is present in the area covered by the motion sensor 1. The model 20 of the present invention as shown in Figure 2 advantageously comprises at least three * * * different states that are the non-presence state NP, the. * * * * * * * presence state P and a third transition state TS. * * * * * * * The states NP, TS, P of the model 20 are hidden in that * * * * they cannot be directly detected by the motion sensor 1. * In other words, the motion sensor 1 cannot directly * * * * * * * * I measure the presence or not of individuals or persons in * * * the covered area. The invention thus advantageously makes use of a hidden Markov model. The model 20 shown in Figure 2 is in the form of such a hidden Markov model. In said Figure, the hidden states NP, TS, P are connected with dashed arrows representing a state transition probability. The dashed arrow of a transition probability starts from a first state and ends at a second state, wherein the second state can correspond to the first state or be different from the first state. The transition probability gives the 3o probability that, starting from the first state at time T1, the second state shall be chosen at time T. Starting from the non-presence state NP at time T-1, there is a state transition probability of 1-p that the state is kept unchanged at time T, as well as a transition probability of p that the next state is the transition state TS at time T. Similarly, starting from the presence state P, there is a transition probability of 1-p that the state is kept unchanged, and a transition probability of p that the next state is the transition state TS. Starting from the transition state TS, there is respectively a probability of 1/2 that the next state is the non-presence state NP or the presence state P. Advantageously, there is no transition between the non-presence state NP and the presence state P. This means that a direct transition from the non-presence state NP to * ** * * * the presence state P is not possible. Rather, in order to *** * * 15 carry out a transition between the two non-presence and * * U..

transition state TS. The use of the transition state TS * * allows for a more fine-grained representation of the * model. This means that the transition probability from NP * * * * to P or from P to NP is zero. *

Advantageously, there is also no direct transition from the transition state TS to the transition state TS. In other words, the transition probability from TS to TS is zero. This means that the transition state TS can only be a transition that occurs between the non-presence and presence states NP, P. Beside the hidden states that cannot be directly measured, 3o the model 20 also comprises observations that can be measured and that are thus known. The motion sensor 1 is adapted to measure a motion within the covered area and the observation depends on the measured motion. A motion signal generated by the motion sensor 1 is related to the presence states NP, P, it is necessary to pass through the presence state. The model 20 comprises a set of observations that have different probabilities depending on the state of the model.

In the embodiment of Figure 2, the model comprises three observations linked to the level of motion in the area covered by the motion sensor 1. A first observation BNO reflects the fact that a no motion event is detected by the motion sensor. A second observation SMO corresponds to lo a small motion event, and a third observation LMO to a large motion event.

The non-dashed arrows of Figure 2 represent emission or * * * output probabilities, i.e. they represent how likely an * * * * * * observation is detected by the motion sensor 1 depending on the state of the model. * * * * * * * * * * * * In the non-presence state NP i.e. if nobody is present in the area covered by the motion sensor 1, there is a * probability of no motion BNO corresponding to 1-s, and a probability s for a small motion SMO. However, the probability of a large motion LMO is O. . * In the presence state P i.e. if someone is present in the * * * area covered by the motion sensor 1, there is a probability of 1-t for a small motion SMO and a probability of t for a no motion BNO observation. Further on, the probability of a large motion is 0. * * * 3o In the transition state TS, the three observations are possible: there is a probability of 1-epsilon-r that a large motion LMO is detected, a probability of epsilon that no motion BNO is detected and a probability of r that a small motion SMO is detected. The large motion * * * observation LMO is thus preferably only possible in the transition state TS. In the transition state TS there is a non-zero emission probability for the three observations, while in the non-presence NP and the presence P states the large motion observation LMO is not possible.

On the basis of the model 20 shown in Figure 2, the presence or not of individuals in the area covered by the motion sensor 1 can be estimated with an algorithm that lo provides the most likely sequence of states given a sequence of observations.

A particular embodiment of such estimation makes use of * ** * * . the Viterby algorithm. In this respect, Figure 3 shows an *** * ^ 15 implementation of the Viterby algorithm for the model 20 * * of Figure 2. Figure 3 particularly shows a trellis diagram O00 * * O011141 for a given sequence of observed events. The observations * * made at time T+1, T+2 and T+3 are respectively a no motion 00 * observation BNO, a large motion LMO observation and a * * . **..

small motion SMO observation.

At time T, it is assumed that nobody is present in the area covered by the motion sensor 1, which corresponds to the non-presence state NP.

At time T+1, a no motion observation BNO is measured, such that the transition probability is (1-p).(1-s) for reaching the non-presence state NP and is p.epsilon for reaching the transition state TS. The presence state P is 3o not possible at time T+1 since no direct transition from NP to P is provided.

At time T+2, a large motion observation LMO is detected by the motion sensor 1. A transition probability from NP at time T+1 to TS at time T+2 is [(1-p).(1-s)].[p.(1-epsilonr)]. This is in fact the only possible transition at this time. Indeed, a transition from NP at time T+1 to NP at time T+2 is not possible since the emission probability of 5 the large motion observation LMO in the non-presence state NP is zero. Further on, a transition from TS at T+1 to NP or P at T+2 is not possible according to the model 20 of Figure 2. Further on, a transition from TS at T+1 to TS at T+2 is not possible since the transition probability from 10 TS to TS is zero in the model 20 of Figure 2.

At time T+3, the motion sensor 1 detects a small motion * * * observation SMO. The path to the non-presence state NP has a probability of A.1/2.s, while the path to the presence * * * state P has a probability of A.1/2.(1-t), with A=[(1-p).(1-s)].[p.(1-epsilon-r)]. * * * * * * A,** O * * * ** When a new observation is carried out by the motion sensor 1, the probability of the new state at time T is given by 20 the following equation: p(new state) -p(old state).t(old state, new state).e(new observation, new state) with p(old state) the probability of the old state at T-1, t(old state, new state) the transition probability from the old state to the new state, and e(new observation, new state) the emission probability of the new observation for the new state Then, for each state, the path with the highest probability is selected. The use of a hidden Markov model thus allows for estimating presence state from motion data. * * * * * * * * * *

II * * According to the invention, the probability of being in the presence state P or non-presence state NP depends on * * * the observed motion. So, the modified behavior of the ** * * 5 motion sensor 1 is a behavior that adapts to the actual conditions simply measuring the actual emission and transition probabilities while operating. The invention preferably proposes an estimation of the presence state with a Markov model. The invention can be based on a lo mathematical model -like the hidden Markov model -that considers the fact that in order to switch from the presence state P to the no presence state NP, the system has to go through a transition state TS where large motion most likely occurs. The hidden Markov model can quite accurately determine the correct state even with loosely * * *** * * correct parameters, making it robust in different * ** ** operating environments. * *

* The basic idea of the invention is that people do not * * * * * * appear from thin air. Correspondingly, transitions between the presence and non-presence states require motion. Preferably, it can be assumed that the motion during this transition is often large or huge. This is why a switch between the presence state P and the no presence state NP requires a passage through the transition state TS. This models the fact that motion is needed for the state to change between presence and no presence.

According to the invention, the observations BNO, LMO, SMO 3o are identified by some thresholds like the first no motion threshold and the second large motion threshold mentioned above. This is different from the state of the art approach in which a presence is detected or not based on one threshold, i.e. motion/presence is detected when the signal from the sensing element exceeds said one threshold.

Further on, in the state of the art the threshold is defined at the time of installation following some general guidelines, and a careful choice of threshold and timer delay is important to minimize discomfort -light turning off when there is presence but no motion -and power consumption -light remaining on after the last person io left the room. In the present invention, no configuration is necessary because of the determination of the most likely sequence of states depending on the observations, i.e. based on the transition probabilities and the emission probabilities. Advantageously, these probabilities are inferred according to the readings of the motion sensor, such that no configuration should be necessary. * * * ** * *

* * ** * * * ** * **** The system can learn because the Viterbi algorithm * * * provides the most likely sequence of states given the observations: «future» observations do affect the most likely states of the gpast». Thus, extracting the estimated states a bit later improves the accuracy of the hidden state estimation. The statistic distributions can * * * be correctly captured. * * * The thresholds can be loosely determined while still obtaining the correct estimation of the presence state. Although dynamic behaviour suffers, that's why it is 3o possible to improve the behaviour directly from the sensor readings.

The used algorithm, e.g. the Viterby algorithm shown in Figure 3, updates the probability of being in the presence state P or non-presence state NP according to the observed motion. It does so using instantaneous values of the motion signal -very similar to the normal threshold and timer delay approach of the state of the art -but also a moving average -e.g. the average over 5 seconds slots. The reaction to instantaneous values provides the responsiveness to large motion and it is possible to determine the threshold that defines such an event, like e.g. "max value/2". On the other hand "continuous" motion io above another threshold (this time considering the moving average) will lead to a change of the estimated state (from non-presence to presence) after a certain number of time slots. * * *

* * . *** . * The implementation of the algorithm of the invention can * * try to keep the statistical distribution of the * * **** observations as close as possible to the emission * * * probabilities that are hardcoded in the software. *** * * * * * **

Also, the algorithm can take into consideration additional information -like user complaining -in order to change the emission probabilities, and the thresholds accordingly. For example if the sensor 1 estimates that the room is vacant and the light is turned off, the user can complain by waving his hands and the sensor understands it made a mistake and updates the parameters of the Markov model accordingly. The user thus can complain: if the light is turning off, the user can wave his hand and thus generate an event whose probability is 3o extremely low (motion while fading) this is an additional type of observation that resets the probability to 1 (certain) and allows fine tuning of the emission matrix. The system in this way can adapt to the habits of the user, and particularly to an active user or a static user. * * * * *

This rudimental interaction with the user provides additional information to fine tune the hidden Markov model 20 parameters.

Additionally the transition probabilities can be measured during operation and adapted to the actual operating condition.

In the followings, some examples are shown as how, depending on the motion detection history, the sensor according to the invention changes its behaviour. This behaviour change is related to how the thresholds, emission and transition probabilities are adapted. 00 * * WOO *

0.15000 * * * 004.8 * * * 01, * * ** ** * ** * * 1. Office application; a user that spends most of his time sitting with small motions and seldom goes away from the desk; in this condition the system will observe a few TS-NP transitions and roughly the same number of TS-P transitions. The system can count these transitions and update the transition probabilities accordingly. In the same situation, it can happen that the user doesn't move much (because he types a lot or reads long documents remaining always in the same position), then the algorithm might make a wrong estimation and think that nobody is there and turns off the light; the user can thus complain, waving one hand generating a very unlikely situation which the system uses to adjust the emission probabilities, correctly identifying that those long periods of inactivity are indeed a presence state.

2. Office application with additional people moving; in this scenario, a person stays at his desk for long periods, but there are other people moving around; in this condition there will be few TS-NP and much more TS-P transitions. The system can count these transitions and update the transition probabilities, so that now after a large motion the system is more likely to keep the light on for a longer time. This because it is more probable that someone is still there.

3. Corridor application; in this case the vast majority of transitions from TS will end up in NP. Indeed, corridors are usually empty except when someone is going lo through them, very seldom people stop in corridors. Also in this case the transition probabilities can be measured and used to determine what the most likely state is. * ..

* * * Changing the transition probabilities affects how quickly *** * or slowly the light will turn off; in the office applications it will naturally take more time for the light to turn off after a large motion; in the corridor * * application, the light will turn off much more quickly. * * * * * . * * *

The Markov model used here gives an estimation of the most * likely state given the observations, the additional intelligence in choosing the parameters of the Markov model slightly changes its behaviour in order to make it more comfortable (more stable, slow changing) or more 25 efficient (more reactive, faster turn-off). It has to be noted that turning off the light source 4 can consist in dimming the light source 4 progressively over a turn-off duration.

3o An important aspect of the invention is that without the third transition state TS, the estimation will not have the locking characteristic that it is necessary to capture the fact that people do not appear from thin air, but have to move in/out of an environment for the presence state P to change.

Monitoring the estimated transitions from TS to P or NP can tell whether it is an office application or a corridor application. The transition matrix can then be adapted accordingly.

Monitoring the transitions P->TS->NP and the other way io round and correlating them with the observations gives additional information about the installation e.g. whether it is likely that the change of state can happen without a large motion. * * . * * * * * * * A preferred application is an LED module with a microcontroller (not shown), so that the presence * * * * * * * * detection requires only the addition of the sensing * * * * * element, possibly mounted on the LED module itself. The * * algorithm used to determine the most likely state can be * * * * * * * * * implemented in said microcontroller A standalone sensor is * * * * * also feasible.

The presence of a daylight sensor is useful for a better learning of the hidden Markov model parameters. This can be done through "interaction" with the user. In this respect, an "Interaction" means that the sensor estimates that the room is vacant and turns off the light. The user can then complain as shown above by e.g. waving his hands and the sensor understands it made a mistake and updates 3o the parameters of the Markov model accordingly. If the light is not necessary (e.g. because there is already enough natural light) the new information on the model's parameters is not so reliable, hence the need of a daylight sensor.

The present invention is advantageous in that it is not necessary to define at the time of installation the timer delay together with the threshold. * ** * * .

* * * * * * * *** * * * **^** * * * ** . * * * ** Further on, if a person does not generate a signal large enough within the timer delay, the light will go off according to the state of the art although the person is still in the area covered by the sensor. The present lo invention is advantageous in that in such a case the light stays on.

Withrthe present invention, it is possible to turn off the light when a user has left the covered area, i.e. when no presence is detected. On the other hand, according to the state of the art the light stays on for the whole timer delay, which implies a waste of energy.

Also, the parameters of the models that are the transition and the emission probabilities are inferred according to the readings of the motion sensor and some empirical observations related to the practical meaning of the different parameters. Therefore no configuration should be necessary.

The system could be helped though, putting it on the right track by e.g. informing the system what kind of application is currently used. Applications can be e.g. corridor applications or office applications. In the case 3o of a corridor application, the system would not need to infer the transition matrix but could still make small changes to better adapt to the actual usage of the corridor.

Claims

Claims 1. Method for detecting a presence state (P) or a non-presence state (NP) of people in an area covered by a motion sensor (1), wherein the motion sensor (1) changes autonomously its lo behaviour depending on ongoing motion sensing.
2. Method according to claim 1, wherein the detection of the presence state (P) or the non-presence state (NP) depends on the history of the 15 motion sensing.
3. Method according to any of the preceding claims, wherein the motion sensor (1) measures a signal reflecting the amount of motion in the covered area, and classifies the measured signal into at least three different event categories comprising a no motion event (BNO), a small motion event (SMO) and a large motion event (LMO), wherein the detection of the presence state (P) or the non-presence state (NP) depends on the event category.
4. Method according to claim 3, wherein a no motion event (BNO) is identified if the amplitude of the signal is below a no motion threshold, a large motion event (LMO) is identified if the signal 3o amplitude is above a large motion threshold, and a small motion event (SMO) is identified if the signal amplitude is between the no motion threshold and the large motion threshold. * . * *

* * * * *** ** * * * * * * * * *** * * * * * * co * ** * * ** *
5. Method according to any of the preceding claims, wherein the presence state (P) and the non-presence state (NP) are detected by means of a hidden Markov model (20), the presence state (P) and the non-presence state (NP) being states of the hidden Markov model (20).
6. Method according to claim 5, wherein the hidden Markov model (20) comprises a third transition state (TS).
7. Method according to claim 6, wherein the transition probability from the transition state (TS) to the transition state (TS) is zero. * *** * * *
8. Method according to any of claims 5 to 7, wherein the transition probability between the presence * ...* state (P) and the non-presence state (NP) is zero. * *4,0 *
9. Method according to any of claims 5 to 8, * *** 20 wherein the estimated transitions from the transition state (TS) to the presence state (P) or the non-presence state (NP) are monitored, and depending on this monitoring it is deduced whether the motion sensor (1) is used in an office application or a corridor application, wherein preferably the transition probabilities are adapted accordingly.
10. Method according to any of the claims 5 to 9 when depending on claim 3 or 4, 3o wherein the no motion event (BNO), the small motion event (SMO) and the large motion event (LMO) are observations of the hidden Markov model (20).
11. Method according to claim 10, * * wherein the large motion event (LMO) observation has an emission probability of zero for the presence state (P) and the non-presence state (NP).
12. Method according to claim 10 or 111, wherein the highest emission probability of the transition state (TS) corresponds to the large motion event (LMO) observation.
13. Method according to any of the claims 10 to 12, wherein the set of observations (BNO, LMO, SMO) has different emission probabilities depending on the state of the hidden Markov model (20).Is
14. Method according to any of the claims 10 to 13, wherein the motion sensor (1) is adapted to detect a particular user movement, like the user waving his hand(s), interpret this movement as an indication that a detected state is not correct, and update the emission probabilities of the observations (BNO, LMO, SMO) accordingly.
15. Method according to any of the claims 5 to 14, wherein the presence state (P) and the non-presence state 25 (NP) are detected by applying a Viterbi algorithm on the hidden Markov model (20).
16. Motion sensor (1) for detecting a presence state (P) or a non-presence state (NP) of people in an area covered 3o by a motion sensor (1), wherein the motion sensor (1) is adapted to change autonomously its behaviour depending on ongoing motion sensing. * ** * * 0 * * *** * * **** * **** * * *
17. System (10) comprising a motion sensor (1) according to claim 16, and a controlling means (2) connected to the motion sensor (1) for controlling a lighting means (4), wherein the controlling means (2) is adapted to control the lighting means (4) depending on the state detected by the motion sensor (1).
18. System (10) according to claim 17, wherein the controlling means (2) is adapted to switch on lo the lighting means if the presence state (P) is detected and to switch it off if the non-presence state (NP) is detected. ** * * * * * * * * * * ** * * * * *** * * an 0 b 7' * * 5*