WO2025008350A1

WO2025008350A1 - Luminaire with precipitation estimation

Info

Publication number: WO2025008350A1
Application number: PCT/EP2024/068605
Authority: WO
Inventors: Peter Deixler; Jin Yu
Original assignee: Signify Holding BV
Current assignee: Signify Holding BV
Priority date: 2023-07-06
Filing date: 2024-07-02
Publication date: 2025-01-09
Anticipated expiration: 2026-01-06

Abstract

A luminaire has a precipitation sensor for detecting precipitation at the location of the luminaire based on detected sounds. The precipitation sensor receives information from other luminaires in the vicinity of the luminaire, wherein the luminaire and the other luminaires in the vicinity form a set of luminaires. The detected sounds at the luminaire and the received information are combined to derive a precipitation estimate. In particular, a subset of the set of luminaires is selected and the information associated with that subset is used to derive the precipitation estimate.

Description

LUMINAIRE WITH PRECIPITATION ESTIMATION

FIELD OF THE INVENTION

This invention relates to luminaires, such as road lighting luminaires, and it relates in particular to the detection of rain or other precipitation using a luminaire.

BACKGROUND OF THE INVENTION

Heavy rain intensity is known to reduce visibility and cause many road accidents. It would be desirable to be able to detect rain with high granularity (i.e., hyperlocal detection) so that action can be taken in response to local rain conditions. It would be of particular interest to be able to use an existing infrastructure, such as the streetlighting infrastructure, for this purpose.

Various precipitation sensing techniques are known, including a tippingbucket rain gauge, an optical (infra-red) rain gauge (commonly used in automotive windshields), and a capacitive sensor.

It is also known to use a microphone that measures rainfall intensity based on the sounds made when the rain strikes a sensor. For example, WO 2016/156563 discloses the detection of precipitation in the outdoor environment of a luminaire. The luminaire has a precipitation detection module that is configured to convert an accelerometer output signal and/or microphone output signal into frequency domain waveforms by execution of a Fast Fourier Transform algorithm. Once the frequency domain representations of the accelerometer output signal and/or microphone output signal are obtained, the frequency domain waveforms are compared to reference precipitation information to detect the occurrence of precipitation in the outdoor environment of the luminaire. The reference precipitation information stored includes a priori information on one or more frequency ranges and associated amplitude levels which is indicative of precipitation.

In the case of streetlighting, the most obvious architecture is to use a microphone-equipped outdoor sensor bundle that is mounted below the luminaire and use the audio data to detect the sound made by the rain drops on the top of the luminaire. However, tree branches above a streetlight will strongly influence the amount of rain falling onto the top of the luminaire housing. Leaves will result in a delay between the onset of rain and the rain drops falling on the luminaire. Similarly, after the rain has stopped, leaves will continue to drip water droplets onto the luminaire. Compared to the unobstructed rain, the presence of leaves will also modify how many raindrops fall on the luminaire housing, as well as the size of individual raindrops.

This problem may be addressed by calibrating the audio-based rain detection algorithm of the individual luminaires to take into account different luminaire housings as well as the rain-shielding by tree branches and objects. However, the calibration of an audioresponse curve of rainfall is cumbersome, and the calibration parameters will be very sensitive to changes in the environment of the streetlight (e.g., as the tree canopy above the streetlights changes over the seasons; or when overhanging branches of trees are regularly pruned for example to protect electricity lines).

There is a need for an improved rain (or other precipitation) sensing approach.

SUMMARY OF THE INVENTION

The invention is defined by the claims.

According to examples in accordance with an aspect of the invention, there is provided a luminaire, comprising: a light source; and a precipitation sensor for detecting precipitation at the location of the luminaire, wherein the precipitation sensor comprises: a microphone for detecting sounds at the luminaire; an input for receiving information derived from detected sounds at other luminaires in the vicinity of the luminaire, wherein the luminaire and the other luminaires in the vicinity form a set of luminaires; an output for delivering information derived from detected sounds at the luminaire to said other luminaires in the vicinity of the luminaire; and a processor for processing the detected sounds at the luminaire and the received information and adapted to derive a precipitation estimate therefrom, wherein the processor is adapted to use the information associated with a subset of the luminaires to derive the precipitation estimate.

This luminaire measures a precipitation level, such as rainfall. The precipitation level can be used to control the lighting as a function of the weather conditions, for example to provide different lighting for different road conditions when the luminaire is part of a streetlight.

The system of the invention estimates the precipitation level using a subset of luminaires rather than simply at an individual luminaire or simply by combining data from a set of luminaires. Instead, an intelligent selection is made of luminaires that should be used to contribute to an overall precipitation level estimate. The luminaires of the set from which the subset is chosen are in the same general vicinity, i.e., expected to experience the same precipitation level. The set is for example the luminaires of streetlights along a specific street. The subset is for example those luminaires which have the least probability to be obstructed by tree branches or leaves as well as the least exposure to audio noise (other than the rain), and hence will give a best estimation of the precipitation level.

The subset is for example determined by an installer during an installation.

However, the system (i.e., the processor) may itself learn which luminaires should be in the subset. It may also dynamically very the subset over time. For example, the precipitation level estimated by a luminaire which is covered by leaves will be different to the precipitation level estimated by other luminaires. Thus, the system itself can deduce which luminaires should best be selected to form the subset.

The sensor hardware is typically already provided in luminaires, for example luminaires are known with a radar sensor, tilt sensor, vibration sensor, microphone, and temperature sensor. These sensors are together known as a sensor bundle.

Unsupervised or lightly-supervised machine learning may be used to analyze the audio signals of the set of luminaires located in the vicinity of each other, and the luminaires of the set passes rainfall-audio-related messages to each other.

The precipitation level is for example determined based on audio-effects of the drops on the housing of the luminaire. The processor of an individual luminaire for example applies a self-learning algorithm to estimate the precipitation level, and this learning takes account of the particular design of the luminaire.

The precipitation sensor is for example a single design of sensor bundle which can be used on many different luminaire types (e.g., with different surface area exposed to the rain as well as different audio response when a raindrop does hit the top surface). Thus, the self-learning process will result in different processing algorithms being used for different luminaire designs.

If the subset is fixed for example at installation, each luminaire in the subset provides a precipitation estimate. If the processor is able to determine, and dynamically adapt, the subset then all luminaires in the vicinity (whether in the subset or not) may provide their local precipitation estimate.

The subset is for example selected as those luminaires which have no blocking from leaves bridges, tunnels etc. There is no need for a master luminaire, but a luminaire could start the message passing and it will also be the last luminaire to receive messages passed back from its neighboring luminaires. Each luminaire preferably has the capability to estimate the overall precipitation level by processing the inputs from its neighboring luminaires.

The precipitation estimate for example comprises an estimate of one or more of: the occurrence of precipitation; an intensity of precipitation; a precipitation droplet size.

Thus, the probability of precipitation, as well as the type and intensity of precipitation, may be estimated.

The processor is for example adapted to select a time-varying subset. The subset may vary for example between summer and winter, depending on the level of foliage. Branches also will grow over time and also occasionally are cut back, so the currently best suited subset of luminaires may be selected. If the tree branches above a certain streetlight are cut, this streetlight may then become suitable for detecting rain.

The time-varying subset may have a different number of members at different times. For instance, in winter when there are no leaves, fewer luminaires may be used in the sensing group to save connectivity cost (e.g., cellular cost for sending data from each of the luminaires to the cloud).

The precipitation sensor for example further comprises a vibration sensor.

This may be used to detect road usage and hence infer road noise and noise cancellation may be employed.

The precipitation sensor for example further comprises a wind speed sensor. This may be used to enable wind related noise to be cancelled.

The precipitation sensor for example further comprises a temperature sensor. This may be used to detect weather conditions, and hence enable precipitation to be detected by additional sensing, for example following a rapid decrease in temperature. The precipitation sensor for example further comprises a humidity sensor. Humidity changes accompany weather changes, and time-series humidity information may enable precipitation to be detected.

The precipitation sensor may further comprise a pressure sensor. Air pressure changes accompany weather changes, and time series pressure information may enable precipitation to be detected as well as precipitation type.

The precipitation sensor for example further comprises a daylight sensor.

A daylight sensor can be used to recognize whether the streetlight is overgrown by a tree (e.g. if the streetlight is in the sun in winter but in shade in the rest of the year).

The precipitation sensor for example further comprises a radar sensor. Radar sensing can be used to detect the presence and distance to trees or other obstacles.

The precipitation sensor for example further comprises a RF sensor. RF sensing can be used to detect atmospheric conditions and it can also be used to detect tree branches.

The luminaire may further comprise a radio system for delivering and receiving information to and from the other luminaires. A Bluetooth low energy (BLE) mesh network is already available in known city lighting systems. Other radio communications systems may however be used.

The information received and delivered for example comprises a precipitation level probability distribution. This is a probability of precipitation based on the sensed conditions.

The subset of luminaires is for example selected based on, for each luminaire, one or more of: obstruction above by vegetation; obstruction above by man-made structures; ambient non-precipitation noise levels.

In one example, the precipitation sensor comprises a radar sensor, and the processor is adapted to select the subset of luminaires based on the proximity of trees or branches as determined by the radar sensor.

The processor is for example adapted to derive a shared/fused precipitation level for the area of the set of luminaires. The multiple probabilities are for example combined using any suitable statistical approach, such as a Bayesian fusion to combine multiple probabilities into a better estimation. Multiple luminaires in one area are thus used to estimate a shared precipitation level for that area. A message-passing approach reduces the impact of a single luminaire being (partially) blocked by tree branches or another luminaire being located close to a (building) structure which partially shields the rain drops. Luminaires in the set may then determine a maximum likelihood rain level for the area.

The processor is for example adapted to cluster the set of luminaires into luminaires of different types. In this way, the precipitation level may be estimated using different algorithms, each tailored to different luminaire types and/or mounting orientations.

The precipitation estimate is for example an estimated rainfall and/or rain drop size.

The invention also provides a streetlight comprising: a pole; and the luminaire as defined above, wherein the luminaire is mounted to said pole. These and other aspects of the invention will be apparent from and elucidated with reference to the embodiment s) described hereinafter.

BRIEF DESCRIPTION OF THE DRAWINGS

For a better understanding of the invention, and to show more clearly how it may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings, in which:

Fig. 1 shows a luminaire with precipitation sensing functionality;

Fig. 2 show a method for a luminaire to sense its local precipitation level; and Fig. 3 shows a method for a luminaire to derive an overall precipitation level based on monitoring by multiple luminaires in the vicinity.

DETAILED DESCRIPTION OF THE EMBODIMENTS

The invention will be described with reference to the Figures.

It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the apparatus, systems and methods, are intended for purposes of illustration only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, systems and methods of the present invention will become better understood from the following description, appended claims, and accompanying drawings. It should be understood that the Figures are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the Figures to indicate the same or similar parts.

The invention provides a luminaire with a precipitation sensor for detecting precipitation at the location of the luminaire based on detected sounds. The precipitation sensor receives information from other luminaires in the vicinity of the luminaire, wherein the luminaire and the other luminaires in the vicinity form a set of luminaires. The detected sounds at the luminaire and the received information are combined to derive a precipitation estimate. In particular, a subset of the set of luminaires is selected and the information associated with that subset is used to derive the precipitation estimate.

Thus, instead of looking at an individual luminaire, the audio signals of a population of luminaires located in direct vicinity of each other are analyzed, and the population of luminaires passes rainfall-audio-related messages to each other. The radius of the vicinity is such that the multitude of luminaires experience similar amount of rainfall. An automatic selection is made (based on selection criteria) which subset of luminaires has the least probability to be obstructed by tree branches/leaves and hence will give the best-faith estimation of the rainfall in this neighborhood. As branches will grow over time and also occasionally are cut back, the algorithm is preferably able dynamically to select the currently best suited subset of luminaires equipped with suitable outdoor sensor bundles. For instance, a different subset of luminaires may be used in winter vs. summer vs. spring.

Figure 1 shows a luminaire 10, comprising a light source 12 such as a LED arrangement and a precipitation sensor for detecting precipitation at the location of the luminaire. The precipitation sensor may have one or more sensing modalities as discussed below.

As a minimum, the precipitation sensor comprises a microphone 14 for detecting sounds at the luminaire. The microphone is for example an existing sensor of a sensor bundle of an existing luminaire, so that the precipitation sensing may be added to an exiting infrastructure. Existing streetlights equipped with sensor bundles are well equipped for hyperlocal detection of rain fall. The invention makes use of selection criteria to purposefully pick a subset of luminaires to use at a given time for making a rain-intensity inference.

The local sounds can be interpreted to determine a local precipitation probability as sensed by the luminaire. However, in addition, the precipitation sensor comprises an input for receiving information derived from detected sounds at other luminaires in the vicinity of the luminaire. This input is represented by the receiver 16. There is also an output for delivering information derived from detected sounds at the luminaire to said other luminaires in the vicinity of the luminaire. This output is represented by the transmitter 18.

In this example, the luminaire is part of a streetlight. The luminaire and the other luminaires in the vicinity form a set of luminaires. This set is a geographically associated set of luminaires, such as all luminaires along a particular street or section of a street.

A processor 20 is provided for processing the locally detected sounds at the luminaire and the received information and adapted to derive a precipitation estimate therefrom. The processor in particular uses the information associated with a subset of the luminaires to derive the eventual precipitation estimate.

The invention makes use of a four-stage approach to estimate the rain precipitation level based on the audio sensors, and optionally also vibration sensors, embedded in the luminaire.

Figure 2 shows the method performed by an individual luminaire to derive a probability of precipitation locally at the luminaire.

In step 30, the sensor signals are collected which enable rain detection.

In step 32, a data pre-processing algorithm is applied to the signals from each sensor bundle. A classification algorithm detects whether the detected sound is due to rain drops or not. This pre-processing algorithm is for example similar to the Voice Activity Detection (VAD) in speech recognition. A filter is for example applied to mitigate for wind gusts as well as the traffic noise due to passing cars or trucks. Each sensor bundle may include vibration sensors, and this enables correlations to be analyzed between the timeseries audio events and the time-series vibrations associated with wind gusts and passing trucks. Raindrops will not cause any significant vibration of the streetlight head or pole. Hence, the use of a vibration sensor of the sensor bundle allows non-rain-related audio noise to be filtered out. This data pre-processing is possible when training data for the specific streetlighting installation has been collected.

The luminaires present in the set may then be clustered per luminaire type. The precise luminaire type may be available as data forming part of the lighting system network, or it could be obtained from satellite images or Google Streetview images.

A lighting installation in a city may have many different luminaire types (with different surface area exposed to the rain as well as vastly different audio response when a raindrop does hit the luminaire' s top surface). When it is not possible to select luminaires of exactly the same make, different luminaire makes may be selected which have similar audio-response characteristics to rain (e.g. similar housing material, similar size of the luminaire top surface, similar structure of the luminaire top surface).

Conversely, in a more advanced embodiment, two (or more) luminaire makes may be selected which have very different audio response to rain. This may enable the rain classification algorithm to look at different rain-related audio features. For instance, a first luminaire housing may generate detectable audio sounds already with very light rain (however the first luminaire housing may create too much sound for detecting larger raindrops, especially when many rain drops per second fall onto the luminaire). The first luminaire housing however may be well suited to count a small number of individual heavy rain drops falling per second on the luminaire).

A second luminaire housing may generate audio features which are well suited for classifying heavy rain.

In such a case, a machine learning algorithm may be used to ingest rain related audio features from both the first and second luminaire housings and thereby is able to detect both the occurrence of rain as well as the intensity of rain as well as the features of the rain.

Similarly, when selecting the luminaires, it is advantageous to make sure that each luminaire type is represented with multiple luminaires in the same hyperlocal area. This allows the rain inference to be performed across the combined raw audio data from multiple luminaires of identical luminaire hardware design of the luminaire housing. k-means clustering may for example be used to first identify from the rain audio data (and optionally as well as from radar/RF-sensing sensor data) collected by the various sensor bundles how many luminaire classes are present in this specific streetlighting installation. K-means clustering may then also be used to segment the streetlights of a single luminaire type into different levels of shielding by vegetation (e.g., a first luminaire class is not covered by tree branches, a second luminaire class is partially covered by branches, and a third luminaire class is fully covered by branches). The classes may include different current densities of the leaf canopy.

In step 34, a self-learning algorithm is applied for estimating the rain precipitation level for each of the luminaire types present in this set. Similar to acoustic event detection, an Audio Spectrum Transformer (AST) with masks or Autoregressive Predictive Coding (APC) may be applied so that the machine learning system can learn which timeseries audio data is associated with rainfall. An AST or APC may further be used to classify the audio recordings (of rain on the luminaire housing) into different categories corresponding to the level of rain fall. Only very limited ground truth data is needed to fine tune and map these audio classes into the real rain fall levels. The ground truth data may come from a local weather station or may utilize the precipitation estimation from a passing-by satellite. Since a high-resolution satellite image could have a resolution from 30cm to 5m, this granularity can help not only to obtain the information for rain precipitation level for each luminaire but also check if a luminaire is blocked by any tree branches or others. However, satellite data is very expensive to acquire for a city, so it is preferably used only during the training phase of the low-cost rainfall inference system. The typical distance between luminaires equipped with sensor bundles is 30m to 50m.

The self-learning enables the sensor sensor bundle, including the microphone and any other sensors that may be used, to be applied to many different luminaire types (with different surface area exposed to the rain as well as different audio response when a raindrop does hit the luminaire' s top surface).

In step 36 mean pooling (or average pooling) is optionally carried out to provide fine tuning with a small batch of labelled data. This divides the input region (e.g., a pooling window or filter) into non-overlapping regions and calculates the average value within each region. The resulting output feature map has reduced spatial dimensions compared to the input.

In step 36, the algorithm generates a probability of precipitation, which takes into account the particular type of luminaire to which the sensing system has been applied.

The algorithm for example uses a simple linear layer or a linear transformation (called a linear head) applied to the learned representations from a pre-trained model. The linear head is often added on top of the pre-trained model to perform specific downstream tasks or to extract useful features. By adding a linear head, the pre-trained model's representations are mapped to the task-specific output space, enabling the model to perform well on the desired task. The linear head essentially adapts the pre-trained representations to the specific downstream task by learning task-specific weights.

This method is performed individually in each luminaire, or in each luminaire that has been selected as one that will report its local estimated precipitation level.

Figure 3 shows a method for combining these estimated precipitation levels from multiple luminaires. Thus, it shows the process undertaken by multiple luminaires and is used to explain the selection process for the subset of the luminaires. Figure 3 shows a set of luminaires LI to LN. Luminaire L2 is covered by a tree canopy and luminaire L3 is under a bridge. Thus, a suitable subset is LI, L4, LN (and others between numbers 4 and N). The full set may for example be all luminaires along a street or multiple streets, or a fraction of the luminaires along a (long) street.

The luminaires L2 and L3 are marked as being unsuitable for contributing to a combined precipitation estimate. Thus, they do not form part of a subset of luminaires that will be used to derive a more accurate precipitation estimate. These luminaires are selected based on predefined selection criteria. The luminaires are selected for the subset which have the lowest probability to be obstructed by tree branches/leaves as well as the least exposure to audio noise and hence will give the best-faith estimation of the rainfall.

Luminaires may be marked as unsuitable for use in the subset during installation of the system. However, the system may also learn or determine by itself which are the luminaires that are unsuitable for inferring precipitation levels. For example, luminaires may be selected that are far from trees as determined by a radar sensor of the sensor bundle, as well as luminaires with the least exposure to audio noise (other than rain sounds) as determined based on audio analysis using the microphone.

The subset of luminaires may all be of the same type or there may be multiple luminaire types within the subset.

In this example luminaire LI is being used to derive an overall precipitation probability. Any luminaire could be used to derive the overall probability.

Luminaires LI, L4 and LN form the subset that will be used to derive the overall precipitation estimate. For each of those luminaires, a local detection method is performed.

Rain drop detection is performed in step 40 based on audio-effects of the drops on the selected luminaire housing and audio noise cancellation from traffic is performed in step 42.

The self-learning algorithm for each individual luminaire is applied in step 44 to estimate rain precipitation level.

The estimated rain precipitation level is in the form of a probability distribution 46, for example a respective probability for different types and/or intensities of precipitation.

During inference of the overall precipitation level, a message passing algorithm is applied within the selected subset of luminaires to obtain a robust rain level estimation. As shown in Figure 3, forward messages F are sent from luminaire LI to the other luminaires and backward reply messages B are returned. The messaging involves local out-of-band message passing which can easily be achieved using existing Bluetooth Mesh radios which are already installed in networked streetlights.

By way of example, a “max-product” based message passing inference system may be used, which utilizes the multiple luminaires in one area to estimate the rain precipitation for that area. This message-passing approach reduces the impact of a single luminaire being (partially) blocked by tree branches or another luminaire being located close to a (building) structure which partially shield the rain drops. A message passing loop is generated whereby the first luminaire passes its rain precipitation level prediction to its neighbors in the same sensing subset. This is repeated by all luminaires in the subset. The luminaires then determine as a group the maximum likelihood rain level for the area.

The multiple probabilities received at luminaire LI are for example combined using any suitable statistical approach, such as a Bayesian fusion to combine multiple probabilities into a better estimation.

In one option, all luminaires report their precipitation estimate, but only the estimates from the luminaires in the subset will be used for the overall precipitation level inference. It is for example possible for the system to learn additionally from the precipitation estimates (as well as from other sensing information as mentioned above) if one or more luminaires are not being exposed to the same rain conditions as their neighboring luminaires. Thus, the system can learn which luminaires of the set are suitable for inclusion in the subset which will be used to derive an overall precipitation estimate.

In another option, shown in Figure 3, only the luminaires in the subset perform their precipitation estimate. Luminaires L2 and L3 do not need to run the estimate algorithm.

The adaptation to the members of the subset may then be implemented manually by the system administrator.

The overall precipitation probability 50 is generated by luminaire LI. It is a real-time inference for the hyperlocal rain precipitation level. This information is then provided to a third-party road safety system.

As mentioned above, the subset may be defined dynamically. For example, if the system detects that one of the luminaires in the subset (i.e. the rain sensing group) is suddenly showing a much smaller or much larger rain precipitation level compared to its neighbors (e.g. a 3c deviation), the sensor bundle radar may be used to check if the luminaire has become shielded by tree branches or by a structure such as a large truck parked in front of this streetlight. In this way, it is possible to detect the accurate level of rain precipitation even if some luminaires in the street are (temporarily) shielded by tree branches or trucks from the raindrops.

Some refinements and alternative options for the system will now be discussed.

As mentioned above, the subset of luminaires may contain different type of luminaires e.g., at an intersection, a first street-lighting luminaire type is located on a residential side-street and a second road-lighting luminaire type located on a main-road. The algorithm can automatically cluster the two different luminaire types into a first classifier and second classifier and then employ first and second sets of neural net weights for performing the rain-fall inferences of the first and second luminaire types.

In addition, the algorithm may also identify which of the luminaires in a city are located in proximity of a public weather station. This can provide an accurate rain fall reading to be used for fine-tuning the rain inference system for the environmental conditions in the specific city as well as finetuning for the specific luminaire type used in the street lighting installation.

To customize the rain detection system for a specific luminaire, few-shot learning may be applied, utilizing the data from both the public weather station and its closest-by unobstructed luminaire. This enables the weights of the rain-inference neural network to be tuned for a specific luminaire. In other words, the audio response curve for each luminaire type and each sensor bundle local environment can be fine tuned (wind speed, traffic noise, current tree canopy etc.).

The same streetlight luminaire type in practice will be mounted in a first street at a first height and angles on the pole and in a second street on a second height and angles on the pole, hence resulting in different readings of the audio sensor bundle despite the same rain full.

The rain algorithm may also make use of wind speed data. The wind speed data may for example be derived from the pole swing (utilizing a tilt sensor in the sensor bundle). The time series wind data is for example used to selectively mask out from the rain detection audio data set those moments from the audio data when a strong wind gust has hit the sensor bundle.

The amplitude of the time domain waveform output by the microphone is generally proportional to the amount of precipitation (e.g., increases during periods of rain). This enables precipitation to be detected and the type/intensity of the precipitation. However, sometimes the precipitation inferences made based on microphone data alone may yield inconclusive results between two or more precipitation types. Temperature data may for example be used to help classify detected precipitation as hail instead of heavy rain if the temperature is less than 0°C. Similarly, a rapid decrease in temperature often occurs during periods of precipitation. Thus, the temperature sensor can be used to detect the onset of rain as well as other precipitation types from the temperature drop in the time-series temperature data.

If temperature sensing is being used, luminaires may be selected to form part of the subset whose direct environment is unaffected by nearby air drafts from a building (e.g. HVAC exhaust, sliding doors of a retail shop, air exhausts from a subway).

Another sensor modality which may be used is RF sensing. Indeed, it has been proposed to detect atmospheric conditions using RF sensing performed by streetlights.

Another sensor modality which may be used is motion sensing. WO 2016/156563, mentioned above, makes use of an accelerometer to detect precipitation from the vibration experienced by a luminaire-embedded accelerometer or even an accelerometer inside the luminaire itself. Hence, an accelerometer sensor may additionally be used to detect the onset of rain as well as other precipitation types from the time-series accelerometer data.

If accelerometer data is being used, luminaires may be selected for the subset whose direct environment is unaffected by vibrations from traffic, machinery, subways etc. which may add noise to the accelerometer data.

An outdoor humidity sensor may additionally be used, to detect rain as well as other precipitation types from the change in the time-series humidity data. Instead of a dedicated humidity sensor, it has been proposed to sense air humidity levels using RF sensing.

If humidity sensing sensing is being used, luminaires may be selected to form part of the subset whose direct environment is unaffected by nearby air exhausts from heating and ventilation systems, sliding doors of a retail shop or air exhausts from a subway, which all may alter the humidity of the air at a nearby luminaire.

Another sensor modality which may be used is pressure sensing to detect rain as well as other precipitation types based on the time series pressure change. Instead of a classical pressure sensor, changes in air pressure may be indirectly inferred with a piezoelectric sensor means positioned at the housing of the luminaire. Openings of the housing of an outdoor luminaire are sealed such that water ingress into the housing is prevented. This leads to a pressure differential between the inside and the outside of the housing of the luminaire whenever the air pressure is changing fast, as is common with the onset of rain.

The luminaire housing may for example comprise at least a flexible part and a rigid part, wherein the flexible part is more flexible than the rigid part, and wherein the piezoelectric sensor is adapted to sense a deformation of the housing by sensing the deformation of the flexible part of the housing.

Another sensor modality which may be used is a daylight sensor. A daylight sensor can be used to recognize whether the streetlight is overgrown by a tree (e.g. if the streetlight is in the sun in winter but in shade in the rest of the year).

As mentioned above, it is possible to infer from precipitation estimates from a group of luminaires how much a specific streetlight is overgrown by tree branches. This data can be used by the system to map which streetlights are located underneath a (large) tree. This information may be used to monitor tree overgrowth as part of a highway monitoring function.

If all streetlights in a neighborhood are at least partially covered by trees, a subset of streetlights may be selected which exhibit similar levels of blocking of the rain by the overhead tree canopy. To increase the robustness of our rainfall inference, the detected rainfall over the subset of streetlights is averaged, and then the rainfall estimate is corrected to take account of the tree canopy, based on ground truth data. The ground truth data may for instance be obtained from one local weather station somewhere in the city.

The algorithm required by the invention is not computationally intensive. It may be unsupervised or lightly-supervised.

Besides detecting rain, the detected precipitation can take the form of one of a plurality of precipitation types e.g., rain, snow, hail, sleet etc.

The rainfall monitoring for many applications, such as road safety applications, does not require the same quality of rainfall data as a professional weather station, as long as the rain sensing data is sufficiently accurate to infer rough classes of rainfall intensity, especially for heavy rain.

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.

Functions implemented by a processor may be implemented by a single processor or by multiple separate processing units which may together be considered to constitute a "processor". Such processing units may in some cases be remote from each other and communicate with each other in a wired or wireless manner (optional)

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, (optional)

If the term "adapted to" is used in the claims or description, it is noted the term "adapted to" is intended to be equivalent to the term "configured to". If the term "arrangement" is used in the claims or description, it is noted the term "arrangement" is intended to be equivalent to the term "system", and vice versa.

Any reference signs in the claims should not be construed as limiting the scope.

Claims

CLAIMS:

1. A luminaire (10), comprising: a light source (12); and a precipitation sensor (14,16,18,20) for detecting precipitation at the location of the luminaire, wherein the precipitation sensor comprises: a microphone (14) for detecting sounds at the luminaire; an input (16) for receiving information derived from detected sounds at other luminaires in the vicinity of the luminaire, wherein the luminaire and the other luminaires in the vicinity form a set of luminaires; an output (18) for delivering information derived from detected sounds at the luminaire to said other luminaires in the vicinity of the luminaire; and a processor (20) for processing the detected sounds at the luminaire and the received information and adapted to derive a precipitation estimate therefrom, wherein the processor is adapted to: select a subset of the luminaires based on at least one predefined selection criterion; use the information associated with the subset of the luminaires to derive the precipitation estimate.

2. The luminaire of claim 1, wherein the at least one predefined selection criterion comprises one or more of: obstruction above by vegetation; obstruction above by man-made structures; ambient non-precipitation noise levels.

3. The luminaire of any one of claims 1 to 2, wherein the precipitation estimate comprises an estimate of one or more of: the occurrence of precipitation; an intensity of precipitation; a precipitation droplet size.

4. The luminaire of claim 3, wherein the processor (20) is adapted to select a time-varying subset.

5. The luminaire of claim 4, wherein the time-varying subset has a different number of members at different times.

6. The luminaire of any one of claims 1 to 5, wherein the precipitation sensor further comprises one or more of: a vibration sensor; a wind speed sensor; a temperature sensor; a humidity sensor; a pressure sensor; a daylight sensor; a RF sensor.

7. The luminaire of any one of claims 1 to 6, wherein the precipitation sensor further comprises a radar sensor.

8. The luminaire of claim 7, wherein at least one predefined selection criterion comprises a proximity of trees or branches as determined by the radar sensor.

9. The luminaire of any one of claims 1 to 8, comprising a radio system (16,18) for delivering and receiving information to and from the other luminaires.

10. The luminaire of any one of claims 1 to 9, wherein the information received and delivered comprises a precipitation level probability distribution.

11. The luminaire of any one of claims 1 to 10, wherein the processor is adapted to derive a shared precipitation level for the area of the set of luminaires.

12. The luminaire of any one of claims 1 to 11, wherein the at least one predefined selection criterion comprises a type and/or mounting orientation of the other luminaires.

13. The luminaire of any one of claims 1 to 12 wherein the precipitation estimate is an estimated rainfall and/or rain drop size.

14. A streetlight comprising: a pole; and the luminaire according to any of claims 1 to 13, wherein the luminaire is mounted to said pole.