Device and computer-implemented method for processing sensor data detected by means of sensor element, in particular worn on body

Abstract

A device and a computer-implemented method for processing sensor data detected by means of sensor elements worn on a human or animal body, in which each sensor element is provided with a time sequence of sensor data of the sensor element, a feature associated with the time sequence is determined for each time sequence from these time sequences, determining, for each time sequence, a plurality of features associated with the time sequence from a respective portion of the time sequence, and determining, for each time sequence, a value associated with the time sequence from the features associated with the time sequence, a time series value associated with the time series is determined for each time series from the values associated with the time series, and a time series value associated with the time series is determined for each time series from a portion of the values associated with the time series. At least one predicted value associated therewith of the values of the at least one time series is determined for the at least one time series from the time series values associated therewith.

Description

Technical Field

[0001] The present invention relates to a device and a computer-implemented method for processing sensor data detected by means of sensor elements, particularly those worn on the body. Background Technology

[0002] The processing of sensor data is useful for a wide range of downstream tasks. Summary of the Invention

[0003] A computer-implemented method for processing sensor data detected by means of sensor elements, particularly those worn on the body of a person or animal, said sensor data being sensor data detected by means of accelerometers, rotational speed sensors, velocity sensors, magnetometers, pressure sensors, blood pressure sensors, V2 sensors, and pulsation sensors, the method proposes: providing time series of sensor data for each sensor element, wherein features associated with each time series are determined based on these time series, wherein multiple features associated with each time series are determined based on portions of each time series, wherein values ​​associated with each time series are determined based on the features associated with each time series, wherein time series values ​​associated with each time series are determined based on the values ​​associated with these time series, wherein time series values ​​associated with each time series are determined based on portions of the values ​​associated with these time series, and wherein for at least one time series, at least one predicted value associated with the value of the at least one time series is determined based on the time series values ​​associated with the at least one time series. Sensor data contained in multivariate time series is processed into predicted values. These predicted values ​​are values ​​generated from the context of sensor data contained in multivariate time series. The value is particularly suitable for further processing, such as further processing related to activity evaluation.

[0004] For example, multiple time series can be associated with a sensor that includes sensor elements for detecting the multiple time series, wherein for at least one time series, a time series value associated with the time series is determined based on a portion of the time series associated with the sensor. This enables a more refined evaluation of the temporal characteristics of the corresponding time series.

[0005] For example, when detecting a time series using a sensor element, metadata is provided to at least one sensor element, particularly the wearing location or specifications of the sensor element, or demographic information about the wearer of the sensor element. This metadata is mapped to a representative of the metadata, and a signal for the time series is determined based on the representative and at least one predicted value of the time series. This enables the fusion with additional information.

[0006] For example, downstream processing proposes classifying the signal, predicting future values ​​of the time series based on the signal, or reconstructing the values ​​of the time series based on the signal.

[0007] For example, the values ​​provided by sensor data for a time series when detecting sensor data are normalized through a transformation, wherein the time series includes normalized values, wherein future values ​​are predicted in a normalized manner, and wherein future values ​​are determined by an inverse transformation of the normalized predicted values, or wherein reconstructed values ​​are reconstructed in a normalized manner, and wherein reconstructed values ​​are determined by an inverse transformation of the normalized reconstructed values. This takes into account statistical differences in multivariate time series. This reduces the impact of sensor drift in the corresponding sensor elements, especially when the predicted values ​​of different multivariate time series are calculated sequentially multiple times using this method.

[0008] To generate training data, for example, it is proposed to provide each sensor element with multiple sets of time series and metadata, wherein each set comprises sensor data detected within different time periods, and wherein a signal is determined for each set. The signal can be used to train a classifier, particularly for classifying signals, and / or to train for predicting or reconstructing at least one value from one or more time series.

[0009] For example, for training, an artificial neural network is provided, wherein a first part of the artificial neural network is configured to: determine a signal based on a time series of sensor data from a sensor element and based on metadata, wherein a second part of the artificial neural network is configured to: process, in particular, classify a signal, or predict or reconstruct a value of one of the time series based on a signal, wherein the artificial neural network, in particular the first part of the artificial neural network, is trained with the first part of the group to process, in particular, classify a signal or predict or reconstruct a value of one of the time series based on a signal (312), and wherein the second part of the artificial neural network is subsequently trained with the second part of the group to process, in particular, classify a signal or predict or reconstruct a value of one of the time series based on a signal (312).

[0010] For regularization, it can be proposed that, in addition to the correlation of time series values ​​associated with at least one time series, the predicted value of at least one time series is determined based on the values ​​associated with the time series.

[0011] To make predictions over a time span, it can be proposed that, for each time series, multiple time series values ​​associated with the time series are determined, particularly multiple time series values ​​that follow each other, wherein at least one predicted value for the time series is determined based on the multiple time series values ​​determined for each time series among the time series values.

[0012] For regularization, it can be proposed that, in addition to the correlation with features, at least one time series-related value is determined based on the time series.

[0013] An apparatus for processing sensor data detected by means of sensor elements, particularly those worn on the body, such as accelerometers, rotational speed sensors, velocity sensors, magnetometers, pressure sensors, blood pressure sensors, VO2 sensors, and pulsation sensors, the apparatus comprising at least one processor and at least one memory, wherein the at least one memory includes instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the apparatus performs the method.

[0014] A computer program includes instructions that can be executed by a computer, and the method is performed by executing the instructions by the computer. Attached Figure Description

[0015] Other exemplary embodiments can be derived from the following description and accompanying drawings. In the drawings:

[0016] Figure 1 A schematic diagram is shown of a device for processing sensor data detected by means of sensor elements, particularly those worn on the body.

[0017] Figure 2 A schematic diagram of a block used for processing sensor data is shown.

[0018] Figure 3 An exemplary architecture comprising blocks is shown for processing sensor data.

[0019] Figure 4 This illustrates an exemplary processing of sensor data within a block.

[0020] Figure 5 A flowchart illustrating the steps of a method for processing sensor data is shown. Detailed Implementation

[0021] exist Figure 1 The diagram schematically illustrates a device 100 for processing sensor data. The sensor data is sensor data detected by means of sensor elements 102, which are typically worn on the body.

[0022] Examples of sensor element 102 are accelerometer, rotation speed sensor, speed sensor, magnetometer, pressure sensor, blood pressure sensor, VO2 sensor, and pulse sensor.

[0023] It can be proposed that: sensor element 102, i.e., acceleration sensor, or sensor element 102, i.e., rotation speed sensor, or sensor element 102, i.e., speed sensor, or sensor element 102, i.e., magnetometer, or sensor element 102, i.e., pressure sensor, is worn by the main body of a machine, robot, vehicle, household appliance, personal assistance system, or tool.

[0024] A sensor may be provided that includes multiple sensor elements 102. For example, a three-axis accelerometer may be provided, wherein each axis includes a sensor element 102 for detecting acceleration as a sensor signal. For example, a three-axis rotational speed sensor may be provided, wherein each axis includes a sensor element 102 for detecting yaw rate as sensor data.

[0025] The device 100 includes at least one processor 102 and at least one memory 104.

[0026] At least one memory 104 includes instructions executable by at least one processor 102, which, when executed by at least one processor 102, enable the device 100 to perform a method for processing sensor data.

[0027] exist Figure 2 The diagram shows a schematic of block 200 for processing sensor data. Block 200 constitutes a time series 202 for processing multiple variables. For a time series 202 of multiple variables including sensor data (detected by means of sensor element 102, particularly worn on a human body), processing the time series 202 of multiple variables with the help of block 200 can be used to: provide a true signal of human activity. Block 200 can also be correspondingly used for sensor element 102 worn on an animal body.

[0028] Block 200 is configured to process f time series 202. Each of the f time series 202 has a length w. For each of the f sensor elements 102, there are f time series 202, each time series including sensor data detected by one of the sensor elements 102.

[0029] Block 200 includes means 204 for determining global features 206, which are associated with one of the time series 202.

[0030] The apparatus 204 for determining global feature 206 is configured to: map the corresponding time series to the corresponding global feature associated with the corresponding time series.

[0031] For example, the apparatus 204 for determining global features 206 is configured to: determine w global features associated with each time series. The apparatus 204 for determining global features 206 may be configured to: determine fewer than w or more than w global features associated with each time series.

[0032] Block 200 includes means 208 for determining local features 210, which are associated with one of the time series 202.

[0033] The apparatus 208 for determining local features 210 is configured to: map a portion of a corresponding time series to a corresponding local feature associated with the corresponding time series.

[0034] For example, the apparatus 208 for determining local features 210 is configured to: determine w local features associated with each time series. The apparatus 204 for determining local features 210 may be configured to: determine fewer or more than w local features associated with each time series.

[0035] For example, the corresponding time series can be divided into multiple parts. These parts may be of equal length.

[0036] Block 200 is configured to: determine future or reconstructed values ​​212, in particular, based on global features 206 and local features 210.

[0037] Value 212 for each time series contains at least one value associated with that time series. It can be proposed that value 212 for each time series includes multiple values, particularly those that follow each other, associated with the corresponding time series.

[0038] For example, for w' = 1, block 200 is configured to: determine a single value for each time series based on the features associated with the corresponding time series in global feature 206 and the features associated with the corresponding time series in local feature 210. For example, block 200 is configured to: determine a single value for each time series based on the sum of the features associated with the corresponding time series in global feature 206 and the features associated with the corresponding time series in local feature 210.

[0039] For example, block 200 is configured to: for w'>1, determine individual values ​​of multiple values ​​for each time series as described for a single value.

[0040] Optionally, for the case where block 200 is configured to determine w' = w global features 206 and w' = w local features 210 for each time series respectively, block 200 is configured to: in addition to the correlation with the global features 206 and the local features 210, also determine the values ​​associated with the corresponding time series based on the corresponding time series, especially the sum of the corresponding time series and the global features 206 and the local features 210 determined for the corresponding time series.

[0041] Block 200 includes means 214 for determining a first time series value 216.

[0042] The apparatus 214 for determining the first time series value 216 is configured to: determine w' first time series values ​​216 respectively based on the values ​​212 of f time series.

[0043] The apparatus 214 for determining the first time series value 216 is configured to: for w' = 1, determine a single time series value 216 of the first time series value 216 for each time series based on value 212. For example, the apparatus 214 for determining the first time series value 216 is configured to: for w' > 1, determine multiple first time series values ​​216 of the first time series value 216 for each time series, wherein each first time series value is determined as described for a single first time series value.

[0044] Block 200 includes means 218 for determining a second time series value 220.

[0045] The apparatus 218 for determining the second time series value 220 is configured to: determine w' second time series values ​​220 respectively based on a portion of the values ​​212 of the f time series.

[0046] The apparatus 218 for determining the second time series value 220 is configured, for example, to: for w' = 1, determine a single second time series value 220 for each time series based on a portion of value 212. For example, the apparatus 218 for determining the second time series value 220 is configured to: for w' > 1, determine multiple second time series values ​​220 for each time series, wherein each second time series value is determined as described for a single second time series value.

[0047] Block 200 is configured to: determine at least one predicted value 222 based on time series 202, a first time series value 216, and a second time series value 220. The at least one predicted value 222 may include, for example, future values ​​or reconstructed values ​​of the corresponding time series.

[0048] For example, for w' = 1, block 200 is configured to: determine a single predicted value for each time series based on the corresponding time series, a single first time series value associated with the corresponding time series in the first time series value 216, and a single second time series value associated with the corresponding time series in the second time series value 220. For example, block 200 is configured to: determine multiple predicted values ​​in predicted values ​​222 for each time series for w' > 1, wherein each predicted value is determined as described for each predicted value.

[0049] For example, the apparatus described in block 200 is configured as a layer of an artificial neural network.

[0050] The layers of the neural network are configured to: map the multivariate time series 202 at the input of block 200 to at least one predicted value 222 at the output of block 200.

[0051] The devices are configured, for example, as single layers of an artificial neural network, and particularly as dense feedforward layers.

[0052] A dense feedforward layer is one example of a usable layer. Other types of layers can also be used. The device can also be implemented using multiple layers.

[0053] exist Figure 3 The diagram shows an exemplary architecture 300 for processing sensor data, including block 200.

[0054] Architecture 300 proposes: using block 200 to determine predicted values ​​222 based on time series 202. In this example, w predicted values ​​are determined for each of the f time series.

[0055] Architecture 300 proposes providing a wearing position 302 for sensor element 102 when detecting f time series 202. Wearing position 302 is the name of the location where the corresponding sensor element 102 is positioned on the body, such as the name of a part of the body. This means that, in this example, sensor element 102 is positioned at the same location on the body.

[0056] In this example, the wearing location involves the body of a living organism, such as a human or animal. In the case of a human, the wearing location could be, for example, the wrist, chest, hip, ankle, waist, sac, or head.

[0057] Architecture 300 proposes that the wearing position 302 be mapped to the representative 306 using encoder 304. In this example, w identical representatives 306 are determined for a time series 202 of length w.

[0058] Architecture 300 proposes that, based on representative 306, an extended representative 310 is determined by multiplier 308, wherein the extended representative includes the representative 306 associated with the corresponding prediction value for each prediction value 222 at the wearing position 302.

[0059] Architecture 300 proposes: determining signal 312 based on predicted value 222 and extended representation 310. Signal 312 is, for example, a real signal of human activity.

[0060] For example, a signal 312 is determined based on the sum of each predicted value 222 and each value of the extended representative 310, wherein the predicted values ​​and the values ​​are associated with each other.

[0061] For example, architecture 300 is configured as layers of an artificial neural network. The layers of the neural network are configured to: map the multivariate time series 202 and meta-information (wearing position 302 in this example) at the input of architecture 300 onto the signal 312 at the output of architecture 300.

[0062] Architecture 300 proposes that the output variable 318 is determined based on the signal 312 and the header 316. This means that no further data collection is required; the output variable 318 is provided based on the multivariate time series 202.

[0063] For example, the head 316 constitutes one or more layers of an artificial neural network.

[0064] The head 316, for example, constitutes a classifier for classifying the signal 312.

[0065] For example, without the need for further data collection, the classifier can be trained based on signal 312 to identify or track various activities of a person wearing sensor element 102 for detecting multivariate time series 202.

[0066] For example, architecture 300 proposes to classify signal 312 using a classifier.

[0067] For example, as in block 200, header 316 is configured to: predict future values ​​of at least one time series in time series 202 based on signal 312 and / or reconstruct values ​​of at least one time series in time series 202 based on signal 312.

[0068] The portion in front of the head 316 of the architecture 300 forms a base model. This base model is configured to: in particular, identify characteristic patterns in the multivariate time series 202 based on metadata, such as wearing position 302. Wearing position 302 is an example of metadata regarding the placement of sensor element 102. The metadata may also include the specifications of the corresponding sensor element 102 or demographic information about the wearer of the sensor element 102. Specifications include, for example, the sampling rate, sensitivity, or measurement range of the corresponding sensor element 102. Demographic information includes, for example, the age, gender, height, and weight of the wearer of the sensor element 102. The head 316 can be replaced according to the task. Multiple different heads 316 can be provided simultaneously for different tasks. Multiple heads 316 can be trained on each other simultaneously based on signal 318.

[0069] In this example, architecture 300 has block 200 before fusion with metadata. Architecture 300 may also have multiple blocks 200 before fusion. For example, architecture 300 has block 200 as header 316 after fusion with metadata. Architecture 300 may also have multiple blocks 200 as header 316 after fusion.

[0070] exist Figure 4 The example shown in block 200 illustrates the processing of sensor data for m time series:

[0071] x 1,1 ,…,x 1,n

[0072] x 2,1 ,…,x 2,n

[0073] x m-1,1 ,…,x m-1,n

[0074] x m,1 ,…,x m,n

[0075] The values ​​are of length w = n, and a single predicted value x for each time series is schematically shown for predicted value 222. 1,n+1 ,…,x m,n+1 .

[0076] Global feature 206 in Figure 4 China with the help of Indicated. Local feature 210 in Figure 4 China with the help of This indicates that the value 212, determined for the corresponding time series based on global feature 206 and local feature 210, is obtained using... express.

[0077] exist Figure 4 With the help of This represents the first time series value, 216. Based on the value... For each time series, a first time series value of 216 is determined.

[0078] The second time series value 216 is in Figure 4 China with the help of express.

[0079] exist Figure 4 In the example shown, the first two time series are associated with the same sensor. This means: the first time series... The second time series value is based on the value associated with the first time series. and values ​​associated with the second time series This means: the second time series. The second time series value is based on the value associated with the first time series. and values ​​associated with the second time series To determine. In Figure 4 In the example shown, the latter two time series are associated with the same sensor. This means: the penultimate time series The second time series value is based on the value associated with the penultimate time series. and values ​​associated with the last time series This means: the final time series. The second time series value is based on the value associated with the penultimate time series. and values ​​associated with the last time series To determine.

[0080] For example, in an application with a triaxial accelerometer, it is proposed to associate three time series of sensor data from the three sensor elements 102 of the three axes, i.e., the three sensor elements 102 of the triaxial accelerometer, with the same sensor, and to determine the corresponding second time series value based on the three time series values ​​associated with the three time series.

[0081] For example, in an application with a triaxial speed sensor, it is proposed to associate three time series of sensor data from the three-axis sensor elements 102, i.e., the three sensor elements 102 of the triaxial speed sensor, with the same sensor, and to determine the corresponding second time series value based on the three time series values ​​associated with the three time series.

[0082] Each predicted value x i,n+1 Based on the value determined for the corresponding time series i Based on the first time series value determined for the corresponding time series i And the second time series value determined for the corresponding time series i To determine. This is in Figure 4 For example, the predicted value x for the first time series i=1 1,n+1 And the predicted value x of the last time series i=m m,n+1 As shown.

[0083] exist Figure 5 The flowchart shows the steps of a method for processing sensor data.

[0084] The method includes step 502.

[0085] In step 502, a time series of sensor data for each sensor element 102 is provided. For example, m time series with n values ​​each are provided:

[0086] x 1,1 …x 1,n ;…;x m,1 …,x m,n

[0087] The method includes step 504.

[0088] In step 504, features associated with each time series are determined based on the time series.

[0089] The method includes step 506.

[0090] In step 506, for each time series, multiple features associated with the time series are determined based on each part of the time series.

[0091] The method includes step 508.

[0092] In step 508, for each time series, based on the features associated with the time series... Determine the values ​​associated with the time series

[0093] For example, besides features In addition to the correlation, it is also based on the time series x i,1 …x i,n Determine the value associated with at least one time series i.

[0094] The method includes step 510.

[0095] In step 510, based on the values ​​associated with the time series For each time series, determine the time series values ​​associated with that time series.

[0096] The method includes step 512.

[0097] In step 512, based on the values ​​associated with the time series As part of this process, for each time series, the associated time series values ​​are determined.

[0098] For example, multiple time series are associated with a sensor including sensor element 102 for detecting the multiple time series. For example, for at least one time series i associated with the sensor, a time series value is determined based on a portion of the time series associated with the sensor.

[0099] The method includes step 514.

[0100] In step 514, for time series i, based on the time series values ​​associated with the at least one time series i... Determine the predicted value x associated with the at least one time series. i,n+1 .

[0101] One approach is to determine the predicted values ​​of multiple time series.

[0102] One approach is to determine multiple predicted values ​​for each time series.

[0103] For example, for a time series, determining the corresponding time series x 1,n+1 ,…,x m,n+1 The predicted value associated with the corresponding time series.

[0104] For example, for at least one time series i, in addition to the time series values ​​associated with at least one time series. In addition to correlation, it is also based on the values ​​associated with the time series. Determine the predicted value x i,n+1 .

[0105] It can be proposed that the method includes step 516 in order to take into account the wearing position.

[0106] In step 516, for at least one sensor element 102, the wearing position 302 of the sensor element 102 is provided when detecting a time series through the sensor element 102.

[0107] It can be proposed that the method includes step 518 to take into account the wearing position.

[0108] In step 518, wearing position 302 is mapped to representative 306 of wearing position 302.

[0109] It can be proposed that the method includes step 520 to take into account the wearing position.

[0110] In step 520, the signal 312 of the time series is determined based on the representative 306 of the wearing position 302 and at least one predicted value 222 of the time series.

[0111] The method includes step 522.

[0112] In step 522, for example, the predicted value 222 is classified, or the signal 312 is classified to take into account the wearing position 302.

[0113] For example, head 316 is trained based on signal 312 to identify or track various activities of a person wearing sensor element 102 for detecting multivariate time series 202. For example, head 316 is trained based on time series data 202 to classify time series data 202 according to activities performed by the wearer of sensor element 102 when detecting sensor data. For example, head 316 is trained based on time series data 202 to classify time series data 202 according to the wearer's physical state when detecting sensor data.

[0114] In step 522, for example, based on the predicted value 222, the future value of the time series is predicted or the value of the time series is reconstructed.

[0115] To take into account the wearing position 302, for example, in step 522, the future value of the time series is predicted or the value of the time series is reconstructed based on the signal 312.

[0116] For example, the head 316 is trained based on the signal 312 to predict future values ​​of the time series or reconstruct the values ​​of the time series. For example, the head 316 is trained to reconstruct or predict the shape of at least one multivariate time series in the multivariate time series 202.

[0117] For prediction, the head 316 is trained, for example, on the multivariate time series 202, to predict the next n values ​​of at least one multivariate time series in the multivariate time series 202.

[0118] For reconstruction, the head 316 is trained, for example, on the multivariate time series 202, to predict n mask values ​​for at least one multivariate time series in the multivariate time series 202.

[0119] It can be proposed that the head 316 of a three-axis accelerometer or a three-axis speed sensor is trained based on a multivariate time series 202 for the following: determining the time series of sensor data for the third axis based on the time series of sensor signals from two of the three axes.

[0120] It can be proposed that signal 312 be further processed using multiple heads 316. For example, based on signal 312, multiple heads can be used to classify, reconstruct, and / or predict values ​​of signal 312. It can also be proposed that multiple heads 316 be trained on each other simultaneously based on signal 318.

[0121] The training, for example, involves performing basic training on layers of an artificial neural network based on a multivariate time series 202 and on metadata, where the metadata includes, for example, the positions of the corresponding sensor elements 102.

[0122] For example, the training could be configured to specifically train the head 316, i.e., one or more output layers of the neural network, for a specific task. It could be proposed that this specific training can be performed without meta-information.

[0123] This means that the predicted value 222 and the signal 312 can be determined without meta-information.

[0124] It can be suggested that step 502 be performed subsequently.

[0125] Step 502 may propose: normalizing the sensor data values ​​provided for the time series when detecting sensor data by means of a transformation.

[0126] For example, sensor data can be normalized using a reverse instance normalization layer derived from RevIN. Normalization removes the mean value of the sensor data from the corresponding sensor element and unifies the variance in the sensor data values ​​of the corresponding sensor element.

[0127] This means that time series data includes normalized values.

[0128] For example, predict future values ​​in a normalized manner, and determine future values ​​through the inverse transformation of the normalized predicted values.

[0129] Sensor data can be normalized, for example, using a reverse instance denormalization layer derived from RevIN. For instance, the previously removed average value can be added to the sensor data.

[0130] For example, the reconstructed values ​​are reconstructed in a normalized manner, and the reconstructed values ​​are determined by the inverse transformation of the normalized reconstructed values.

[0131] Step 512 may involve determining multiple time series values ​​associated with each time series, particularly multiple time series values ​​that follow each other. For example, in step 514, based on the multiple time series values ​​determined for each time series, at least one predicted value for the time series is determined.

Claims

1. A computer-implemented method for processing sensor data detected, particularly by means of sensor elements (102) worn on the body of a person or animal, said sensor data being particularly detected by means of accelerometers, rotational speed sensors, velocity sensors, magnetometers, pressure sensors, blood pressure sensors, VO2 sensors, and pulsation sensors, characterized in that, For each sensor element (102), a time series (x) of sensor data of the sensor element (102) is provided (502). 1,1 …x 1,n ,…,x m,1 …,x m,n (504) features associated with each time series are determined based on these time series. For each time series, multiple features associated with the time series are determined (506) based on a portion of the time series. For each time series, based on the features associated with the time series Determine (508) the values ​​associated with the time series. Based on the values ​​associated with these time series For each time series, determine (510) the time series value associated with the time series. For each time series, the values ​​associated with these time series are used as a basis. A portion of the time series values ​​associated with the time series are determined (512). And wherein for at least one time series, the time series value associated with the at least one time series is used. Determine (514) at least one predicted value (x) associated with the at least one time series. 1,n+1 ,…,x m,n+1 ).

2. The method according to claim 1, characterized in that, Multiple time series are associated with a sensor including a sensor element (102) for detecting the multiple time series, wherein for at least one time series, a time series value associated with the time series is determined (512) based on a portion of the time series associated with the sensor.

3. The method according to any one of the preceding claims, characterized in that, When the time series is detected by the sensor element (102), (516) meta-information is provided to at least one sensor element (102), particularly the wearing position (302) or the specifications of the sensor element (102) or demographic information about the wearer of the sensor element (102), wherein the meta-information is mapped (518) onto a representative (306) of the meta-information, wherein a signal (312) for the time series is determined (520) based on at least one predicted value (222) of the representative (306) and the value of the time series.

4. The method according to claim 3, characterized in that, The signal (312) is classified (318)(522), or the future value (318) of the time series is predicted (522) based on the signal (312), or the value (318) of the time series is reconstructed (522) based on the signal.

5. The method according to claim 4, characterized in that, The sensor data is normalized (502) when the sensor data is detected, and the time series includes normalized values, wherein the future value is predicted in a normalized manner and the future value is determined (522) by an inverse transformation of the normalized predicted value, or wherein the reconstructed value is reconstructed in a normalized manner and the reconstructed value is determined (522) by an inverse transformation of the normalized reconstructed value.

6. The method according to claim 4 or 5, characterized in that, For each sensor element (102), multiple sets of time series and metadata are provided (502), wherein each set includes sensor data detected in different time periods, and wherein a signal (514) is determined for each set (312).

7. The method according to claim 6, characterized in that, An artificial neural network is provided, wherein a first portion (200, 304, 308) of the artificial neural network is configured to: determine the signal (312) based on a time series (202) of sensor data from the sensor element (102) and based on the meta-information (302), wherein a second portion (316) of the artificial neural network is configured to: process (318), particularly classify the signal (312), or predict or reconstruct the value of one of the time series based on the signal (312), wherein the artificial neural network, particularly the first portion (200, 304, 308), is trained with the first portion of the group to process, particularly classify the signal (312) or predict or reconstruct the value of one of the time series based on the signal (312), and wherein the second portion (316) of the artificial neural network is subsequently trained with the second portion of the group to process, particularly classify the signal (312) or predict or reconstruct the value of one of the time series based on the signal (312).

8. The method according to any one of the preceding claims, characterized in that, In addition to time series values ​​associated with at least one time series In addition to the correlation, it is also based on the values ​​associated with the time series. Determine (514) the predicted value (x) of the at least one time series. i,n+1 ).

9. The method according to any one of the preceding claims, characterized in that, For each time series, (512) a plurality of time series values ​​associated with the time series are determined, in particular a plurality of time series values ​​that follow each other, wherein (514) at least one predicted value of the time series is determined based on the plurality of time series values ​​determined for each time series among the time series values.

10. The method according to any one of the preceding claims, characterized in that, In addition to the features mentioned above In addition to the correlation, it is also based on the time series (x) i,1 …x i,n (508) Determine (508) the value associated with at least one time series.

11. An apparatus (100) for processing sensor data detected by means of sensor elements (102) particularly worn on the body, said sensor data being sensor data detected particularly by means of accelerometers, rotational speed sensors, velocity sensors, magnetometers, pressure sensors, blood pressure sensors, VO2 sensors, and pulsation sensors, characterized in that, The device (100) includes at least one processor (102) and at least one memory (104), wherein the at least one memory (104) includes instructions executable by the at least one processor (102), and when the instructions are executed by the at least one processor (102), the device (100) performs the method according to any one of claims 1 to 10.

12. A computer program, characterized in that, The computer program includes instructions executable by a computer, which, when executed by the computer, run the method according to any one of claims 1 to 10.

Images