TECHNICAL FIELD

The present disclosure relates to the control of a utility, such as lighting, based on sensing the presence of a being or object in a sensing region associated with the utility. BACKGROUND

"Connected lighting" refers to a system of luminaires which are controlled not by (or not only by) a traditional stand-alone electrical circuit, but rather via a wired or wireless network using a digital communication protocol. For instance, each of a plurality of luminaires may be equipped with a respective control unit, a respective presence sensor, and a respective wireless transceiver or interface to a wired network. Using this transceiver or interface, the control unit on each luminaire can then share its sensor readings with its neighbors, or with a central controller. E.g. if the sensor of one of the luminaries detects presence then all the luminaries in a same group or room as that luminaire, or within a certain radius of it, may be turned on together in response.

For instance, outdoor lighting plays a key role in providing safety and comfort to citizens, and allows cities to express their unique identities. With financial pressure on municipalities mounting, however, the costs associated with public outdoor lighting are weighing increasingly on tight municipal budgets. By combining LED light sources with intelligent networked controls capable of adapting the light to dynamically varying conditions, the energy efficiency of outdoor systems can be substantially increased, allowing financial benefits to go hand in hand with environmental advantages such as lower greenhouse emissions and reduced light pollution.

Different control strategies can be employed to improve the energy efficiency of street lighting. One strategy is to control light levels on the basis of fixed or programmable time schedules (schedule-based control) is well-established. As another strategy, light levels can be adapted on the basis of traffic volumes (traffic-adaptive control), either provided by locally installed sensors or external information sources, following recommended practices for dynamic lighting class assignment. A third strategy is to use presence-adaptive controls light levels are set using locally installed presence sensors that immediately activate preconfigured groups of luminaires in their vicinity to an increased (e.g. 100% ) light level when presence (e.g. motion) is detected, in order to maintain safety and comfort to citizens. When no-one is present, the lights are turned off or dimmed to low levels to preserve energy.

Various types of suitable presence sensor for these or similar applications will be familiar to a person skilled in the art. For example, types of presence sensor include various types of motion sensor, such as those based on the Doppler effect. These work by emitting a wave such as an RF or ultrasound wave into a region of space in which presence is to be detected, then receiving back a reflection of that wave. A Doppler shift in the signal can be used to detect that the wave was reflected from a moving being or object (the target), and optionally to give an estimate of its speed. Motion detection techniques also extend to more advanced radar detectors, such as FSK (frequency shift keying) and FMCW (frequency modulated continuous wave) radar detectors, which modulate the frequency of the emitted signal. These are more expensive, but also more robust and accurate. Another type of presence sensor is a passive infrared (PIR) sensor which detects the heat of a living being. The PIR sensor can detect motion based on differences in the passively sensed infrared radiation. Other types of presence sensors are suitable for sensing presence regardless of whether the target is moving or static. E.g. the above-mentioned advanced FSK and FMCW radar sensors can also sense range which can be used to detect static targets. Or as another example, a 2D or depth-aware camera operating in the infrared or visible spectrum can be used to detect either moving or static targets, by applying an image recognition such as a facial recognition, body recognition or object recognition algorithm to the captured images.

Connected lighting systems equipped with motion sensors allow the right amount of light to be created when and where it is needed (light-on-demand). This enables substantial energy savings without compromising the safety and comfort of the end-user. An example of such a system is an outdoor lighting system such as street lighting.

US 2009/0262189 Al discloses a lighting system which both illuminates and monitors an area, wherein the monitoring is performed using an image sensor. The light output of the lighting system is adapted based on events detected based on said monitoring.

DE 102012100080 Al discloses a motion detection method using presence sensors. The method for controlling a lighting system based on a probablity of further movement (e.g. based on a determined velocity).

WO 2013/179175 Al discloses a method in a presence sensing system, whereby a physical characteristic of a sensed object is determined and a three-dimensional physical characteristic is derived from this physical characteristic. It is determined whether the object is a true object, such that the detection of its presence is to be accepted, or a false object or person, such that the detection of its presence is to be rejected.

US 2015/0254570 Al discloses a method for probabilistic semantic sensing in a sensory network. A service is performed based on a confidence level of a derived event; i.e. an event that is likely to occur based on analyses of aggregated data.

However, widely employed presence sensors (e.g. PIR, or Doppler radar sensors) are prone to high false positive rates (e.g. due to adverse weather conditions, moving vegetation, animals, and/or ambient noise). These false positives inadvertently activate sets of luminaires resulting in unnecessary energy dissipation and dynamic behavior that can be annoying to users (e.g. local residents).

As mentioned, presence detection can be executed using diverse sensor modalities including (thermal) cameras, radar, time-of-flight, pressure and active or passive infra-red sensors. Out of these, passive infra-red sensors are most widely deployed due to their attractive cost price. However, in the challenging conditions of an outdoor environment, presence sensors are also prone to errors. In particular, passive infrared sensors are known to produce large amounts of false triggers, for example due to moving vegetation, small animals or precipitation. These false triggers activate sets of luminaires unnecessarily and in extreme cases, may prevent parts of the lighting installation from dimming altogether even though no- one is present. Undesired light activations do not only result in wasteful energy dissipation, unnecessary switching behavior can be highly annoying for local residents and negatively affects the innovative image of the system.

Similar issues may also occur in relation to other utilities such as heating or air conditioning which may also be triggered based on presence detection. SUMMARY

According to one aspect disclosed herein, there is provided a system comprising: a plurality of sensing nodes, each comprising a presence sensor; and control logic comprising one or more control units. The control logic is configured, for each respective one of the sensor nodes, to perform operations of: (a) identifying occurrence of a local sensor event in response to the presence sensor of the respective sensing node generating a sensor reading indicative of a person or object apparently being present in a sensing region of the respective sensing node; (b) identifying occurrence of at least a first remote sensor event in response to the presence sensor at a first remote one of said sensing nodes generating a sensor reading indicative of a person or object apparently being present in a sensing region of the first remote sensing node; (c) determining a time difference between the local and first remote sensor events; (d) determining a first hypothetical speed by dividing a distance between the local and first remote sensing nodes by the determined time difference between the local and first remote sensor events; (e) declaring a positive detection on condition that the local and first remote sensor events are determined to occur with said first hypothetical speed being within a predetermined propagation speed range (below an upper limit, or above lower limit, or both); and (f) in response to said positive detection, triggering an appliance (e.g. luminaire) to turn on or increase a level of a utility (e.g. illumination) provided into a region served by the respective sensing node.

The control logic is configured to maintain a respective confidence level associated with each of the sensing nodes, by increasing the respective confidence level each time the positive detection is declared by the respective sensing node, but decreasing the confidence level each time the local sensor event is identified without the positive detection being declared.

For instance, the control logic may be configured, for each of the sensing nodes, to declare said positive detection on further condition that the confidence level of one the local sensing node and the first remote sensing node is greater than a confidence threshold. In embodiments, the control logic may be configured, for each of the sensing nodes, to declare said positive detection on further condition that the confidence level of both the local sensing node and the first remote sensing node is greater than a confidence threshold. Optionally, the control logic may also be configured, for each of the sensing nodes, to declare said positive detection on further condition that the confidence level of the second remote sensing node is beyond a lower confidence threshold.

I.e. the process comprising operations (a)-(f) above, and optionally (g)-(i), is performed repeatedly with the confidence level being updated each time as described, then upon a or each subsequent repetition of this process the confidence level is used to determine whether the remote sensing event or events can reliably be used in the determination as to whether or not a positive declaration should now be declared in the present repetition of the process. E.g. this can be used to instantly decide whether or not to activate sets of lights having only the local detection.

Thus instead of considering sensor events in isolation, the analysis of consecutive detections by adjacent nodes in a sensor network allows the reliability of individual sensors to be estimated over time. By including sensor confidence levels, indicative of historic performance, in light activation messages, this allows adjacent nodes to adapt their behavior based on the reputation of the sender to improve the system performance in terms of energy savings and user experience. E.g. as a pedestrian walking along a street should generate a consistent spatio-temporal sequence of triggers across the deployed sensor network, comparing observations of individual sensors with those of its neighbors can provide insights into their performance. By accumulating such information over time and including information about a sensor's reliability in the light activation messages transmitted to its neighbors, this allows the adjacent nodes to adapt their behavior based on the reliability of the sender. Triggers of notorious sensor nodes can thus be ignored unless confirmed by more trustworthy neighbors, thus preventing unnecessarily switching of the luminaires and optimizing the achieved energy savings.

Thus advantageously, the disclosed system is able to distinguish between positive and false-positive triggers of presence based on consecutive detections by adjacent nodes. If a sensing event at a given node is indeed a true positive, it will be experienced shortly after or shortly before another similar sensing event at a nearby node, within a time window that is consistent with the likely speed of the target being or object being detected in the application in question (e.g. a person or a car). According to the present disclosure, this observation is advantageously exploited to detect and discard false positives.

The first remote sensor event which the control logic is configured to identify may occur before or after the local sensor event in time.

Preferably the first remote sensor event which the control logic is configured to identify occurs before the local sensor event in time; and the control logic is further configured to perform operations of: (g) identifying occurrence of a second remote sensor event in response to the presence sensor of a second remote one of said sensing nodes generating a sensor reading indicative of a person or object potentially being present in a sensing region of the second other sensing node, wherein the second remote sensor event occurs after the local sensor event in time, (h) determining a time difference between the local and second remote sensor events, and (i) determining a second hypothetical speed by dividing a distance between the local and second remote sensing nodes by the determined time difference between the local and second remote sensor events; wherein the control apparatus is configured to declare the positive detection on further condition that the local and second remote sensor events are determined to occur with said second hypothetical speed being within the predetermined propagation speed range.

Alternatively or additionally, the control logic may be configured, for each of the sensing nodes, to override said condition or conditions if the confidence level of the respective sensing node is beyond an upper confidence threshold, such that the positive detection is declared in response to identifying the local sensor event alone without verification by any other of the sensing nodes.

Such adaptive control is a particularly advantageous application of the sensor confidence levels. Confidence levels are preferably maintained in order to deal with first detections (those that do not have preceding remote events), e.g. as a person is entering the monitored area and is being picked up by the system. In that case, the local detection generates an activation of nearby light sources (or other appliances) because of its historic reputation. If it has been performing reliably (consistent later confirmations of neighbors), its detections are trusted and executed without waiting for further confirmation. For instance the above behavior is advantageous in dealing with the detectability of, and response, to first- time detections. Note that in such embodiments, the detection is not strictly classified as positive (in the sense that it will increase the sensor confidence level by itself). Instead the receiving nodes are configured to trust and hence execute activations from trustworthy nodes.

Note that in embodiments, these events are not strictly declared positive detections in the sense that they will increase the sensor confidence level. Rather, they are positive in the sense that the system is configured to act upon them if the confidence level is above the upper threshold (e.g. execute activation messages with sensor confidence levels above the threshold). It is possible that these events are not confirmed later on, resulting in a decrease of the sensor confidence level in spite of the fact that lights were switched on.

In a further alternative or additional application of the confidence level, sensor confidence levels can be remotely monitored to identify defects and provide adequate maintenance services and can be of support in the context of performance-contracting.

In further embodiments, the control logic may comprise a plurality of distributed control units, a respective one of the control units being implemented at each of the sensing nodes; the respective control unit at each of the sensing nodes may be configured to perform, for the respective sensing node, said identification of the first remote sensing event, said determinations of time difference and hypothetical speed, and said declaration; and the respective control unit at each of the sensor nodes may be configured to perform the identification of the first remote sensor event based on a message from the first remote node reporting the first remote sensor event. The confidence level of the first remote node may also be communicated to the control unit of the local node in this message.

Further, where a second remote node is involved, the respective control unit at each of the sensor nodes may be configured to perform the identification of the second remote sensor event based on a message from the second remote node. The confidence level of the second remote node may also be communicated to the control unit of the local node in this message.

In yet further embodiments, the system may further comprises a respective instance of said appliance (e.g. a respective luminaire) at each of the sensing nodes, each arranged to provide said utility into a different respective service region; and the respective control unit at each of the sensor nodes may be configured to perform said triggering by triggering the respective instance of the appliance to turn on or increase the level of said utility (e.g. the emitted illumination) in the respective service region.

In embodiments, each of the luminaires may be mounted on a respective outdoor light pole. In embodiments, the light poles may be street lights.

According to another aspect disclosed herein, there is provided a method comprising, for each respective one of a plurality of sensing nodes, each comprising a presence sensor, performing steps of: identifying occurrence of a local sensor event in response to the presence sensor of the respective sensing node generating a sensor reading indicative of a person or object apparently being present in a sensing region of the respective sensing node; identifying occurrence of at least a first remote sensor event in response to the presence sensor at a first remote one of said sensing nodes generating a sensor reading indicative of a person or object apparently being present in a sensing region of the first remote sensing node; determining a time difference between the local and first remote sensor events; determining a first hypothetical speed by dividing a distance between the local and first remote sensing nodes by the determined time difference between the local and first remote sensor events; declaring a positive detection on condition that the local and first remote sensor events are determined to occur with said first hypothetical speed being within a predetermined propagation speed range; and in response to said positive detection, triggering an appliance to turn on or increase a level of a utility provided into a region served by the respective sensing node.

The method further comprises maintaining a respective confidence level associated with each of the sensing nodes, by increasing the respective confidence level each time the positive detection is declared by the respective sensing node, but decreasing the confidence level each time the local sensor event is identified without the positive detection being declared; and initiating servicing (e.g. maintenance or realignment) of one or more of the sensing nodes based on the respective confidence levels. In embodiments, the method may further comprise steps in accordance with any of the system features disclosed herein.

According to another aspect disclosed herein, there is provided a computer program product for operating one of a plurality of sensing nodes, each comprising a presence sensor, the computer program product comprising code embodied on a computer- readable storage medium and configured so as when run on one or more processing units of the sensor node to perform operations of: identifying occurrence of a local sensor event in response to the presence sensor of the respective sensing node generating a sensor reading indicative of a person or object apparently being present in a sensing region of the respective sensing node; identifying occurrence of at least a first remote sensor event in response to the presence sensor at a first remote one of said sensing nodes generating a sensor reading indicative of a person or object apparently being present in a sensing region of the first remote sensing node; determining a time difference between the local and first remote sensor events; determining a first hypothetical speed by dividing a distance between the local and first remote sensing nodes by the determined time difference between the local and first remote sensor events; declaring a positive detection on condition that the local and first remote sensor events are determined to occur with said first hypothetical speed being within a predetermined propagation speed range; and in response to said positive detection, triggering an appliance to turn on or increase a level of a utility provided into a region served by the respective sensing node.

In embodiments, the computer program product may be further configured so as when run to perform operations in accordance with any of the system or method features disclosed herein. BRIEF DESCRIPTION OF THE DRAWINGS

To assist understanding of the present disclosure and to show how embodiments may be put into effect, reference is made by way of example to the

accompanying drawings in which:

Figs. 1 a- lc provide schematic illustrations of a lighting system, Figs. 2a-2b provide further schematic illustrations of a lighting system,

Fig. 3 is a schematic block diagram of a luminaire,

Fig. 4 is a flow chart of a data handling function, and

Fig. 5 is a flow chat of a timer tick function. DETAILED DESCRIPTION OF EMBODIMENTS

The present disclosure provides a system and method for distinguishing between positive and false-positive triggers of motion sensors or other presence sensors in a lighting network, based on consecutive detections by adjacent nodes.

Each node in the lighting network temporarily stores sensor activation attributes of both its local sensor as well as of remote sensor(s) (nearby nodes that share sensor data). The attributes of remote activations that are stored comprise the activation timestamp and its distance to the current node. For local observations, only activation timestamps need to be stored. New events are classified as true or false positive by comparing their time stamps against remote events stored locally. The ratio of the respective sensor distances and time differences results in a hypothetical propagation speed associated with this particular pair of observations. If one of the hypothetical propagation speeds falls within predefined acceptance thresholds, the activation is classified as a true positive.

If none of the local and remote event pairs results in an acceptable speed estimate, the event may not yet be classified as a false positive. It could be a new 'target' where remote observations will succeed the local activation. The method can therefore use two separate loops: comparing against preceding remote events and comparing against succeeding remote events. As adaptive lighting control is a highly responsive application, light activation messages are preferably transmitted immediately upon sensor activation. For lighting control, it is therefore beneficial to split the event classification in two separate analysis loops (one capable of immediately comparing the local activation against preceding events and a second after expiry of a time window to compare against succeeding events).

In more advanced embodiments, smoothing is applied based on historically determined confidence levels: the sensor confidence level is raised every time when a local activation is classified as a true positive and decreased when the event is classified as a false positive.

In order to benefit from the locally computed confidence levels, this information is shared with the neighboring nodes by including it in the light activation messages. Alternatively, sensor nodes could inform each-other about their performance on a regular basis using dedicated messages at the expense of using additional bandwidth.

Receiving lighting nodes can then adapt their behavior based on the confidence level of the activated sensor node, for example ignoring messages of nodes with a confidence level below a certain threshold. Figures 1 a to 1 c schematically illustrate a system in accordance with embodiments disclosed herein. The system comprises a plurality of luminaires 4. The luminaires 4 are arranged to illuminate an environment 2, e.g. an outdoor space such as a street, road or park; or an indoor space such as one to more rooms of a building. Each of the luminaires 4 comprises a respective one or more lamps (illumination element), plus any associated socket, support and/or housing. In one embodiment each of the luminaires 4 is mounted on an outdoor light pole, e.g. so as to form a street light.

Each of the luminaires 4 is associated with a different respective sensing node SI ... Sn (only n=5 nodes are shown here for illustrative purposes, but it will be appreciated there may be many more). For instance, each of the luminaires 4 may be co-located with its respective sensing node SI ... Sn. The plurality of sensing nodes SI ... Sn are arranged to sense the presence of a being or object (the "target") 6 such as a car or pedestrian, which may move throughout the environment 2. The sensing nodes SI ... Sn are deployed throughout at least a substantive part of the environment 2, being substantially separated from one another in space (where "substantive" or the like here means on a scale that is significant - i.e. non negligible - compared to distances over which the target is expected to potentially move). E.g. the sensing nodes SI ... Sn may be spatially separated from one another by a radius of at least 50cm, or at least lm, or at least 2m, or at least 5m, or at least 10m, depending on application.

Each of the sensing nodes is arranged to sense presence within a respective sensing region (i.e. a respective spatial are or volume within the environment 2). Further, each of the luminaires 4 is arranged to illuminate a respective service region LI ... Ln (i.e. again a respective spatial are or volume within the environment 2). Preferably each of the service region LI ... Ln is approximately coincident with the sensing region of its respective sensing node, and may be referred to interchangeably in the following. However, it will be appreciated that this is not limiting to all possible embodiments. E.g. as an alternative, a sensing region could consist of one or more smaller zones located at expected entry and/or exit points of its respective service region.

Each of the sensing nodes LI ... Ln is configured so as when it detects presence of a target 6, to be able to trigger its respective luminaire 4 to turn on or dim up its illumination, as well as to transmit a lighting activation message to one or more neighboring nodes to trigger its/their respective luminaries 4 to turn on or dim up as well. The luminaire 4 at each node LI ... Ln then turns off or dims down after a predetermined timeout period has elapsed, running from the time when presence is no longer sensed by the respective node, or the time when no activation message is received from any of its neighbors. Thus as the target 6 moves about within the environment, a plurality of luminaries 4 turn on or dim up around the target, but turn off elsewhere, thus forming a "lighting bubble" which follows the target around.

For instance, Figures la-lc show an example of basic presence-adaptive lighting controls using intelligent sensor-equipped luminaires. A pedestrian 6 moving along the street will be detected by a first sensor S2, upon which a command is broadcast to activate a preconfigured group of luminaires 4 serving regions L1-L3 for a predefined time in order to illuminate the space around the pedestrian 6. As the pedestrian 6 continues his or her movement along the street (Figures lb to lc), consecutive detections activate new sets of luminaires thus ensuring that the pedestrian 6 is enclosed in a bubble of light of configurable extent and duration.

However, Figures 2a to 2b illustrates an example of the problem of false triggers. These are undesired activations not associated with actual traffic passing by, resulting in undesired switching behavior of sets of luminaires 4, and in turn unnecessary energy use and irregular behavior that can be highly annoying to local residents. For instance, a false trigger could be caused by a non-target object 8 such as a tree, carrier bag or tumbleweed blowing in the wind, or a non-target being such as an urban fox or pet cat.

Hence as will be discussed in more detail in the following, in accordance with the present disclosure the sensing nodes SI ... Sn are arranged so that mere detection of local presence alone is not enough to trigger the luminaire nor the transmission of a lighting activation message to its neighbors.

Figure 3 gives a schematic block diagram of sensing node according to embodiments of the present disclosure. Each of the multiple sensing nodes SI ... Sn may be configured in accordance with Figure 3.

As mentioned, each of the sensing nodes SI ... Sn comprises (or is at least associated with) a respective luminaire 4. The luminaire 4 comprises one or more lamps 14, and a driver 16 connected to the one or more lamps 14. The luminaire 4 further comprises a wired or wireless transceiver 18, and a local control unit 20 coupled to both the wireless receiver 18 and the driver 16. In embodiments, the luminaire 4 may be mounted on a light pole 1 1, e.g. the pole of a streetlight. The transceiver 18 is operable to communicatively connect the local control unit 20 of the respective node S to one or more others of the nodes SI ... Sn, either via a wired network connecting the nodes or via a wireless (e.g. radio) access technology. The local control unit 20 may be implemented in the form of code stored on a memory of the luminaire 4 and arranged to run on a processing apparatus of the luminaire 4 (the memory comprising one or more storage media implemented in one or more memory units, and the processing apparatus comprising one or more processing units). Alternatively the local control unit 20 may be implemented in dedicated hardware circuitry, or configurable or reconfigurable hardware circuitry such as a PGA or FPGA, or any combination of hardware and software.

Further, each sensing node SI ... Sn comprises a respective presence sensor 24, operatively coupled to the local control unit 20 of its respective luminaire. The presence sensor may for example be mounted externally on the housing of the luminaire 4, or internally within the housing of the luminaire 4, or separately on the light pole 1 1, or indeed elsewhere in association with the luminaire 4. The presence sensor 24 may take any suitable form such as a radar or ultrasound based Doppler sensor, or a passive infrared sensor, or a 2D or 3D (depth aware) camera for capturing images in the infrared and/or visible spectrum. Suitable presence sensors will in themselves be familiar to a person skilled in the art.

Whatever modality it employs, the presence sensor 24 is arranged to sense the presence of a target being or object 6 in the respective sensing region LI ... Ln of the respective sensing node SI ... Sn. To do this, the presence sensor 24 continually generates a sensor reading which it supplies to the local control unit 20. The control unit 20 monitors this sensor reading to determine whether it is characteristic of presence of the target being or object 6 according to one or more local predetermined criteria. For instance if the presence sensor 24 is a motion sensor which outputs signal representing a current value of a single metric measuring a degree of motion sensed, then the predetermined criteria may be that the signal exceeds a predetermined threshold. Or if the presence sensor 24 is a camera, the control unit 20 may comprise an image recognition algorithm (e.g. facial recognition algorithm, body recognition algorithm or object recognition algorithm) configured to detect whether the captured image matches a certain signature characteristic of the target being or object 6. Suitable techniques for sensing presence will in themselves be familiar to a person skilled in the art.

Conventionally, when the control unit 20 detects a sensor reading from its respective presence sensor 24 that is indicative of presence according to the predetermined local criterion or criteria, it controls the lamp(s) 14 of its respective luminaire 4, via the driver 16, to switch on or dim up (i.e. begin emitting illumination or dim up the emitted

illumination). It also operates the transceiver 18 to transmit a lighting activation message to the control unit 20 of one or more neighboring nodes SI ... Sn, triggering it/them to switch on or dim up as well. Also, in a complementary fashion, if the local lighting control unit 20 receives such a message from one of its neighbors, it will switch on or dim up. Preferably the lighting control messages are transmitted wirelessly, and will be described as such by way of example in the following.

However, as mentioned, this conventional arrangement on its own can lead to an undesirable number of false positives. Hence according to the presence disclosure, the local control unit 20 at each node SI ... Sn is configured to apply at least one additional conditions which must be met in order for the respective luminaire 4 to be triggered to switch on or dim up, and in order for a lighting activation message is transmitted to the neighboring node(s). Examples of this will be set out in the following.

Consider a networked lighting system such as that described in relation to Figures 1 to 3, that comprises a set of luminaires 4, each equipped with an intelligent lighting controller 20 that comprises processing capabilities, memory and communication means 18 that allows information to be exchanged between nodes or to a central software backend. Furthermore, presence sensors 24 are installed on all or some of the luminaires 4. These may be integrated into the luminaire 4 and connected directly to the luminaire controller 20 or installed externally on the pole 1 1 and connected to a dedicated network node. In the following, it is assumed that the presence sensors 24 are binary, meaning that they are only capable of communicating whether or not presence is detected (yes/no), but are incapable of providing information about the fidelity or reliability of their observations. This is the case for most commonly applied commercially available presence sensors. Each lighting controller 20 has a local clock that may be synchronized at regular intervals by a central management system and it's the coordinates of its physical location (e.g. GPS coordinates) have been defined and stored in memory during configuration.

In absence of activity, the luminaires 4 dim to a low state to save energy and reduce light pollution. When a moving target 6 (e.g. a pedestrian) appears on the street (or the like), it will be detected by a first presence sensor (S2) that broadcasts a command to activate a set of associated nodes SI -S3 in its vicinity for a specified hold time (Figure la). As the target continues, it will be detected by consecutive presence detectors (S3, S4), each of which activates another set of luminaires 4 centered around its position (Figures lb and lc), thus creating a bubble of light accompanying the road user 6 sufficiently wide to warrant a sense of safety. When an actual target 6 (pedestrian, bicycle, car) traverses the street, this will result in a predictable spatio-temporal sequence of light activation messages, all of which are received by nearby luminaires 4. Motion sensors 24 activated by passing traffic can expect to receive light activation messages from nearby nodes both preceding and following the transmission of its own message by time differences governed by the sensor spacing and the target speed.

However, (particularly low-cost) motion sensors are also notoriously prone to false triggers, i.e. undesired activations not associated with the passing of relevant traffic. Common sources of false triggers include adverse weather conditions, pole swing, small animals and swaying vegetation 8. False triggers activate sets of luminaires 4 and can result in annoying dynamic behavior of the lighting system (Figure 2b). However, false triggers generally lack the spatio-temporal correlation that true positives exhibit and often occur in isolation. On the basis of these assumptions, it is possible to devise a distributed mechanism for each node S 1 ... Sn to classify its own activations as either true positive or false positive by comparing it against preceding or succeeding messages from a set of neighbors.

In order to adopt this approach, the local controller 20 at each sensor node SI ...Sn is configured to implement the following functionality.

EVENT STORAGE

In order to classify each new observation as a true or false positive, the observation is compared against observations from nearby sensor nodes. To do this, the nodes stores attributes of both local (from its locally attached sensor 24) and remote (received from nearby sensor nodes) sensor activations into dedicated queues. The attributes of remote activations that are to be stored include the activation timestamp and its distance to the current node. The latter can be achieved by including the activated sensor coordinates in the activation message payload or transmitting the ID and storing coordinates of nearby sensor IDs locally. For local observations, only activation timestamps need to be stored. To circumvent system synchronization issues, it may be beneficial to store the time at which a node received a message from a nearby node (instead of the timestamp of the remote node itself), such that all processing is executed in local time. This does assume that message delays are small enough to be ignored in relation to the speed of the road user 6.

Events do not need to be stored in memory indefinitely, but only for an analysis window T an, equal to the longest sensor distance considered divided by the slowest valid speed anticipated. In one example, one could limit the analysis to events from the direct neighbors of a node and the maximum storage time equals the sensor distance divided by the minimum pedestrian threshold taken into consideration (e.g. 0.5 m/s). EVENT CLASSIFICATION

New events are classified as true or false positive by comparing their time stamps against remote events stored locally. For each event stored in the remote queue (R- queue), the ratio of the respective sensor distances and time differences results in a hypothetical propagation speed associated with this particular pair of observations. If one of the hypothetical propagation speeds falls within predefined acceptance thresholds, the activation is classified as a true positive. This range of acceptable speeds may be defined by only a lower bound, or only an upper bound, or both a lower and upper bound. E.g. in a residential street a car is expected to drive between 15 and 50 km/h. Or a pedestrian may be expected to walk faster than 2km/h, or slower than 6km/h, or both.

If none of the local and remote event pairs results in an acceptable speed estimate, the event may not yet be classified as a false positive, because the sensor activation may represent a first observation of a newly appearing target (e.g. someone leaving his house or appearing from an uninstrumented side street) and remote observations will succeed the local activation. For that reason, local activations are stored for the same time interval T an as their remote counterparts. Upon expiry of this time interval, the local event is again compared against the events stored in the remote queue in the same manner as described above. Only if neither of the two analysis procedures fail to qualify the local event as a true positive, it is labelled as a false positive. Dividing the analysis procedure in two separate loops (comparing against preceding and succeeding remote events, respectively) is beneficial for certain lighting applications where immediate classification is desirable (see below). However, if the sensor confidence levels are not used in applications that require direct interaction, such as remote monitoring and maintenance, it may be more convenient to conduct the analysis in a single loop at the expiry of the local event only. In that case, remote events need to be stored twice as long, however, and more memory needs to be allocated for the respective queues.

Note, the acceptance criterion itself can be configured, for example by modifying the amount of local/remote pairs resulting in acceptable speeds (effectively requiring a longer target trajectory) or imposing further conditions within the trajectory in terms of speed and direction. In practice, however, the inventors have found that just requiring a local activation to be validated by a single remote event is suitable for basic lighting control applications. SENSOR CONFIDENCE LEVEL

Each node SI ...Sn maintains its own confidence level indicative of its historic reliability. It is convenient, but not imperative, to normalize the confidence level such that it ranges between 0 and 1, where 1 indicates a perfect sensor and 0 means a completely malfunctioning sensor. The sensor confidence level is raised every time when a local activation is classified as a true positive and decreased when the event is classified as a false positive. The adaptation factor can be defined in a number of ways, including fixed or relative factors.

Relative adaptation of the confidence level carries the benefit that the confidence level is directly proportional to the true positive rate. E.g. a sensor performing at a true positive rate of 80% results in a confidence level of 0.8. To this end, the confidence level CLi of sensor node i can to be modified in the following manner:

False positive: CLi = CLi - c*CLi

True positive: CLi = CLi + c*(l-CLi)

In the above, c is a configurable smoothing parameter. Decreasing c results in more stable behavior of the sensor confidence levels, whereas increasing c results in more rapid adaptation of confidence levels to recent sensor performance. In experiments, the inventors have found c=0.025 to provide decent results, but this is not limiting.

LIGHTING CONTROL ALGORITHM

In order to benefit from the locally computed confidence levels, this information is shared with the neighboring nodes by including it in the light activation messages. Alternatively, sensor nodes could inform each-other about their performance on a regular basis using dedicated messages at the expense of using additional bandwidth.

Receiving lighting nodes can then adapt their behavior based on the confidence level of the activated sensor node, for example ignoring messages of nodes with a confidence level below a certain threshold.

Adaptive lighting control is a highly responsive application, wherein light activation messages are preferably transmitted immediately upon sensor activation. For lighting control, it is therefore beneficial to split the event classification in two separate analysis loops (one capable of immediately comparing the local activation against preceding events and a second after expiry of Tan to compare against succeeding events). If a local activation is validated by a (set of) preceding event(s) from neighboring nodes, it is desirable that lights are activated even if the originating sensor often performs poorly. In other words, it is undesirable to ignore all messages generated by an unreliable sensor, as it may result in road users being poorly illuminated unnecessarily. Confidence levels of activation messages invoked by true positive events are thus set to 1 (to ensure their execution), whereas the confidence level of messages generated by (yet) unclassified events are set to the current sensor confidence levels. The benefit of this cautious approach is that it only ignores new events generated by sensor nodes with a poor historic reputation, in which case the lights will be activated once the road user 6 will be detected by a second sensor node. Once road users have been detected by some sensor nodes, passing of unreliable sensor nodes will have no effect on the propagation of the light bubble surrounding them. On the other hand, isolated light activation messages that are frequently generated by unreliable sensor nodes without validation by their surrounding will increasingly be ignored thus improving system behavior and increasing energy savings by preventing unnecessary activations.

By way of example, Figures 4 and 5 show a flow diagram of a two-step implementation of the procedures for the sensor node to enable confidence-based lighting control. The data handling function (Figure 4) is executed every time when the sensor is newly activated while the timer tick functionality (Figure 5) is executed for every heartbeat of the system. Lighting nodes receiving the transmitted activation messages can decide to execute or ignore the message based on the value of the received confidence level (e.g.

ignoring messages with confidence levels below a certain threshold).

Figure 4 shows the data handling function. The process begins by receiving a new local event. At step S 10 the process matches the local event to an event s in the remote queue (R-queue). At step S20 the process determines whether the local and remote events are within an acceptable velocity. If so the process proceeds to step S30 where the confidence level attached to a message is set to a maximum value CL max and transmitted, such that the lighting responds immediately. Even when the local sensor has an unreliable reputation, confirmed events will be executed (because the penalty of leaving people in the dark is much higher than that of an undesired activation). The confidence level of this particular sensor will be incremented when executing the process described by Figure 5.

The method then proceeds to step S40 where it is determined whether the remote queue is full (in practice, these algorithms may be implemented on resource constrained devices, e.g. memory constrained devices). If so the method branches to step S 100 where it is determined if the event currently being processed is the last event in the remote queue. If not, at step SI 10 the method proceeds to the next event in the remote queue, and loops back to step S 10 in relation to this next event. If this is not the last event in the remote queue on the other hand, the method proceeds to step S60 (described shortly).

If at step S40 the remote queue is determined not to be full however, the method proceeds to step S50. Here, a flag "flag ismatched" is set to true. This flag is used in the process of Figure 5 to prevent that matching is executed twice. The method then proceeds to step S60, where a message advertising a local detection is transmitted. Then the method proceeds to step S70 where it is determined whether the local queue (L-queue) is full (again because the method may be implemented on a resource constrained device). If so the method just ends at step S90, but if so the method proceeds step S80 where the event is added to the local queue before ending at step S90.

The process of Figure 5 is executed at every timer tick. By storing the local and remote events long enough, the method can validate in a single loop against earlier and later occurring remote events. This is done when the local event has aged by T an.

At step T10 the method updates timers of the local queue, i.e. to maintain the relative age of the observed events. The method then proceeds to step T20 where it is determined whether any of the local queue times (for any of the events in the local queue) is equal to T an. If not the method just exists at step SI 20. If so however the method proceeds to step T30 where it is checked whether flag ismatched is true. If so the method branches to step T70 (described shortly). If not, the method proceeds to step T40 where it is determined whether the remote queue is full. If so, the method branches to step Tl 10 where the event is removed from the local queue, then the method ends at step T120.

Otherwise if the remote queue is not full, the method proceeds to step S50. Here the local evert is matched to an event in the remote queue. The method then proceeds to step S60 where it is determined whether the local and remote events are within an acceptable velocity. If so the process proceeds to step S70 where the confidence level is increased, and then the method proceeds to step Tl 10 where the remote event is removed from the local queue and then the method ends at step T120.

If at step T60 the local and remote events are determined not to be within an acceptable velocity on the other hand, the method proceeds to step T80 where it is determined whether the event is the last event in the remote queue. If so, the process proceeds to step S90 where the confidence level is decreased, and then the method proceeds to step Tl 10 where the remote event is removed from the local queue and then the method ends at step T120. Else if the event was not the last in the remote queue, the method loops back to step S50.

The inventors have tested the disclosed procedure based on a two-week database of events collected in a real-life residential street deployment with commercially available PIR sensors. The sensors were found generally to perform reasonably well with confidence levels generally above 0.9. However, some individual nodes exhibited lower confidence levels indicative of poor performance. Interestingly, these sensor nodes were positioned very closely to trees, a well-known source of false triggers in PIR detectors. The confidence levels provide insights into the system performance, and allows it to be serviced (for example by moving the poor sensors or trimming trees), as well as improving the system performance by incorporating the estimated confidence levels into lighting control.

ADVANTAGES AND APPLICATIONS OF CONFIDENCE LEVEL

The method disclosed above adopts the approach to event classification of imposing conditions on consecutive detections, but in addition, the individual sensor nodes aggregate their classification results over time to gather information about their historic performance. In the context of lighting control this can be used to achieve robust adaptive lighting control. In embodiments, it may also be used to achieve any one or more of the following.

Event validation: this can make the adaptive control even more robust and reliable. In prior systems, events are verified as true positives based on preceding detections only. In embodiments of the present disclosure however, events are verified based on both preceding and successive detections. This results in more accurate classification results (because more data is involved) and superior performance on the perimeter of the monitored area (where preceding detections are not always available).

First-time detections: the first detection of a target entering the monitored area cannot be verified. Without further measures, this results in a slow responsiveness of the lighting system to new entrants (for example, when somebody leaves his house, it may take about a minute to pass multiple sensors to validate and activate the street lights). In further embodiments of the present disclosure, the historic performance of the triggered sensor node is broadcast with the activation message, such that first-time triggers by sensors that typically perform well are executed, resulting in a high responsiveness necessary to maintain the perception of safety. Sensor performance for service application: aggregated sensor performance data cannot only be applied in the context of immediate lighting control, but also supports service concepts, such as maintenance and performance contracting not covered in the prior art.

CONCLUSION

It will be appreciated that the above embodiments have been described only by way of example.

For instance, the control logic for performing the above-described functionality does not necessarily have to be implemented (or only implemented) in a plurality of distributed local control units 20, with one at each luminaire 4 or sensing node SI ... Sn. Instead, the control functionality could be implemented by a centralized controller arranged to receive the sensor readings from each of the presence sensors 24 and to control each of the luminaires 4, or could be implemented in a combination of distributed and centralized control. Further, there need not necessarily be one sensing node S 1 ... Sn per luminaire 4. Instead, one, some or all of the sensing nodes could be arranged to serve a respective group of luminaires 4; or there could even be multiple sensing nodes per luminaire for one, some or all of the luminaires 4. Furthermore, the techniques disclosed herein are not limited to an application to controlling the illumination emitted by luminaires. More generally, the disclosed techniques could be used to control any kind of appliance providing any kind of private or public utility into the environment 2, be it an indoor utility such as a domestic or office utility, or an outdoor utility. E.g. the disclosed techniques may be applied to the control of appliances providing lighting, heating and/or air conditioning; or more generally, any transducing appliance which convert power from a power source (typically electrical power) into a utility in the form of light, sound, heat and/or motion of a fluid (such as the air in the environment). Thus although the above description has been primarily exemplified in relation to an intelligent outdoor lighting systems that comprises (mesh) network technology and motion-based occupancy sensors, more generally the disclosure can be applied in any networked lighting system or other system, including domains such as office or industry lighting. It is also relevant in the context of lighting services as it provides a means to monitor the quality of services and supports scheduling and optimizing of maintenance activities.

Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. A computer program may be stored/distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope.