Search

EP-4609149-B1 - METHOD AND SYSTEM OF CALIBRATION OF A SENSOR OR A NETWORK OF SENSORS

EP4609149B1EP 4609149 B1EP4609149 B1EP 4609149B1EP-4609149-B1

Inventors

  • MISHRA, AMIT KUMAR

Dates

Publication Date
20260513
Application Date
20231024

Claims (15)

  1. A method of calibration of a sensor, wherein the sensor is in a given environment, the method comprising: in a reliable calibration phase after calibration of the sensor, determining (101) an environment response function (h) in a first time period based on a known sensor response function (f), the determining carried out when obtaining measured values (y) by the sensor of a measurand field (x) of an event (e) in the first time period, wherein the environment response function is a spatio-temporal response of factors of the environment representing the environmental response operating on the event to result in the measurand field; in an unreliable calibration phase after the first time period, estimating (102) a current sensor response function (g) based on the environment response function (h) determined in the reliable calibration phase, the estimating carried out when obtaining measured values (y) by the sensor of the measurand field (x) of the event (e) in a second time period; and outputting (103) a calibration function based on an inverse of the estimated current sensor response function (g) for updating calibration of the sensor.
  2. The method of claim 1, wherein determining the environment response function (h) determines one of the group of: a value of the function, a shape of the function determined from results of one sensor; a shape of the function determined from environmental response function results of multiple sensors.
  3. The method of claim 1 or claim 2, including: measuring selected factors of the environment over the first time period at the sensor to determine the environmental response function (h).
  4. The method of any one of claims 1 to 3, wherein the method is a semi-blind calibration using a controlled perturbation in the measurand field (x) of the event to determine the environment response function (h) based on the known sensor response function (f).
  5. The method of any one of claims 1 to 3, wherein the method is a blind calibration including modelling (111) the method as a two stage autoencoder network with each stage implemented by a block of convolutional neural networks, wherein the modelling is trained in the first time period for an event (e) and a second time period is selected with closely matching environment factors to the first period to replicate the event (e).
  6. The method of claim 5, wherein: a first stage is an observation process block (242) modelling a data dependency from the measured data (y) to the measurand field (x), wherein the first stage models a calibration function; and a second stage is an environment response block (241) modelling a data dependency from the measurand field (x) to the event field (e), wherein the second stage models the environment response function.
  7. The method of claim 5 or claim 6, including training a model using measured data (y) from the sensor including: in the reliable calibration phase, training (112) both the first stage and the second stage of the model using the measured data (y) and the measurand field (x), wherein the measurand field is determined using the known sensor response function (f); and in the unreliable calibration phase, the second stage is not trained and the first stage is trained (113) using only the measured data (y).
  8. The method of any one of claims 5 to 7, wherein, after the unreliable calibration phase, an updated observation process block of the first stage is used as an updated calibration process.
  9. The method of any one of the preceding claims, including adding (213, 223) a representation (204) of measurement noise to the known sensor response function and the current sensor response function.
  10. A method of calibration of a network of sensors, wherein the each of the sensors of the network has a known relative position in the given environment, the method comprising the method of any one of claims 1 to 9 applied to each of the sensors.
  11. A system for calibration of a sensor, wherein the sensor is in a given environment, the system comprising: an environment response determining component (402) for determining an environment response function (h) in a first time period in a reliable calibration phase after calibration of the sensor, with the environment response function (h) based on a known sensor response function (f), the determining carried out when obtaining measured values (y) by the sensor of a measurand (x) of an event (e) in the first time period, wherein the environment response function is a spatio-temporal response of factors of the environment representing the environmental response operating on the event to result in the measurand field; a current sensor response estimating component (403) for estimating in an unreliable calibration phase after the first time period, a current sensor response function based on the environment response function (h) determined in the reliable calibration phase, the estimating carried out when obtaining measured values (y) by the sensor of the measurand field (x) of the event (e) in a second time period after the first time period; and an output component (404) for outputting a calibration function based on an inverse of the estimated current sensor response function for updating calibration of the sensor.
  12. The system of claim 11, wherein the environmental response determining component (402) includes: an environment factor measuring component (407) for measuring selected factors of the environment over the first time period at the sensor to determine the environmental response function (h).
  13. The system of any one of claims 11 to 12, wherein the system is a blind calibration system including a two stage autoencoder network (430) with each stage implemented by a block of convolutional neural networks, wherein the autoencoder network is trained in the first time period for an event (e) and a second time period is selected with closely matching environment factors to the first period to replicate the event (e).
  14. The system of claim 13, including a training component (430) for training a model using measured data (y) from the sensor including: a reliable calibration phase training component (431) for training both the first stage and the second stage of the model using the measured data (y) and the measurand field (x), wherein the measurand field is determined using the known sensor response function (f); and an unreliable calibration phase training component (432) for training the first stage using only the measured data (y) and refraining from training the second stage.
  15. A computer program stored on a computer readable medium and loadable into an internal memory of a digital computer, comprising software code portions, when said program is run on a computer, for performing the method steps of any of the claims 1 to 10.

Description

FIELD OF THE INVENTION This invention relates to sensor calibration. In particular, the invention relates to calibration of a single sensor or a network of sensors that are in a given environment. BACKGROUND TO THE INVENTION Calibration is a major component of any metrological system especially for sensors which are installed in remote places. Without a thorough investigation and methodology around sensor calibration, the data collected from the sensors are often not reliable. Sensors may be calibrated to an initial high standard; however the quality of the data often quickly deteriorates as the calibration loses accuracy. This is a major pain point in the current day of ubiquitous sensing. Calibration efforts for individual sensor types are extensive. Most of these processes need a reference sensor or some ground-truths. Running a reference-based calibration for remote sensors is a costly task. Also, it does not scale up; i.e., when the number of sensors is in the hundreds, the task becomes impossible to be carried out on a regular basis. One of the solutions to this challenge has been to treat the battery of sensors as a single system. This system can, then, be calibrated as a whole rather than focusing on individual sensors in this network. A physics-based model may be used to do this and data from the sensor network is expected to fit the model as closely as possible. Hence, the individual sensor-calibration parameters are fine-tuned to force this fit. Many following works have used this approach and modified it as well. Though efficient, this approach is not fully blind. Calibration of sensors when a ground-truth or references are not used is referred to as "blind calibration". Some blind sensor methods require a large number of sensors to work. For example, one method uses an assumption of the existence of spatial oversampling to give an elegant solution which has been leveraged upon by many other works. Unfortunately, in most real-life cases, the number of sensors available is usually limited. US10008770B2 discloses calibration of sensors or sensor arrays with beam forming based calibration of antennas. The method leverages on beam forming algorithms and then this is used through a Gaussian source to calibrate the antennas. This is a well-known method of calibration in radar domain. WO2020191980A1 discloses a blind calibration method for wireless sensor network data drift. This uses a Kalman filter to track the drift in sensor drift. This is not a long-term strategy, because, after a while Kalman-filter will drift itself and will need reference data. Hence, this method cannot correct indefinitely. The paper of BALZANO L. et Al : "Blind calibration of sensor networks", Information Processing in Sensor Networks, 2007. IPSN 2007. 6th Int. Symp, 25 April 2007, relates to blind calibration of a wireless sensors network. The preceding discussion of the background to the invention is intended only to facilitate an understanding of the present invention. It should be appreciated that the discussion is not an acknowledgment or admission that any of the material referred to was part of the common general knowledge in the art as at the priority date of the application. SUMMARY OF THE INVENTION According to an aspect of the present invention there is provided a method of calibration of a sensor, wherein the sensor is in a given environment, the method comprising: in a reliable calibration phase after calibration of the sensor, determining an environment response function (h) in a first time period based on a known sensor response function (f), the determining carried out when obtaining measured values (y) by the sensor of a measurand (x) of an event (e) in the first time period, wherein the environment response function is a spatio-temporal response of factors of the environment representing the environmental response operating on the event to result in the measurand field; in an unreliable calibration phase after the first time period, estimating a current sensor response function (g) based on the environment response function (h) determined in the reliable calibration phase, the estimating carried out when obtaining measured values (y) by the sensor of the measurand field (x) of the event (e) in a second time period; and outputting a calibration function based on an inverse of the estimated current sensor response function (g) for updating calibration of the sensor. Determining the environment response function (h) may determine a value of the function or a shape of the function. Determining the environment response function (h) as a shape of the function may be determined from environmental response function results of multiple sensors. The method may include measuring selected factors of the environment over the first time period at the sensor to determine the environmental response function (h). In one embodiment, the method may be a semi-blind calibration using a controlled perturbation in the measurand fi