Device and method for posture recognition, radar system and electronic equipment

Abstract

The invention relates to a device and method for posture recognition, a radar system and electronic equipment. An apparatus for gesture recognition is presented. The apparatus includes processing circuitry configured to obtain a data matrix indicative of a distance over a slow time based on radar data acquired by a radar sensor, determine a reduced-size compact data matrix by compressing the data matrix over the slow time, and transmit the data matrix to the radar sensor. And determining a distance interval of the target based on the compact data matrix. The processing circuitry is further configured to exclusively process the determined distance intervals in the data matrix for determining a movement of the target, extract a time series of at least one feature of the movement, and identify a pose based on the time series of at least one feature of the movement.

Description

Technical Field

[0001] The present disclosure relates to radar-based gesture recognition. Examples relate to an apparatus for gesture recognition, a method for gesture recognition, a radar system, and an electronic device. Background Art

[0002] Conventionally, gesture detection uses computationally intensive techniques to analyze range Doppler maps in great detail. These algorithms undergo complex processing steps, including multidimensional Fourier transforms, Doppler profile analysis, object segmentation, shape feature extraction, etc. These operations require a large amount of computing resources such as memory or processor resources and time-consuming calculations, resulting in high power consumption. Therefore, improved gesture detection may be needed. Summary of the invention

[0003] Some aspects of the present disclosure relate to an apparatus for gesture recognition, the apparatus including a processing circuit device configured to obtain a data matrix indicating a distance in slow time based on radar data acquired by a radar sensor, determine a compact data matrix of reduced size by compressing the data matrix in slow time, determine a distance interval of a target based on the compact data matrix, specifically process the determined distance interval in the data matrix for determining movement of the target, extract a time series of at least one feature of the movement, and recognize a gesture based on the time series of the at least one feature of the movement.

[0004] Some aspects of the present disclosure relate to a radar system, which includes the apparatus for gesture recognition as described herein, and a radar sensor, wherein the radar sensor is configured to acquire radar data.

[0005] Some aspects of the present disclosure relate to an electronic device including a radar system as described herein, and control circuitry configured to control operation of the electronic device based on a recognized gesture.

[0006] Some aspects of the present disclosure relate to a method for gesture recognition, the method comprising: obtaining a data matrix indicating distance in slow time based on radar data acquired by a radar sensor, determining a compact data matrix of reduced size by compressing the data matrix in slow time, determining a distance interval of a target based on the compact data matrix, specially processing the determined distance interval in the data matrix for determining movement of the target, extracting a time series of at least one feature of the movement, and recognizing a gesture based on the time series of the at least one feature of the movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Some examples of apparatus and / or methods will now be described, by way of example only, and with reference to the accompanying drawings, in which:

[0008] Figure 1 An example of an apparatus for gesture recognition is illustrated;

[0009] Figure 2 An example distance spectrum illustrating binning distances is shown;

[0010] Figure 3 A flow chart illustrating an example of a method for gesture recognition;

[0011] Figure 4 A flow chart illustrating another example of a method for gesture recognition;

[0012] Figure 5 a table illustrating an example of a time series of features associated with an example of a gesture;

[0013] Figure 6 An example of a radar system is illustrated;

[0014] Figure 7 An example of an electronic device is illustrated; and

[0015] Figure 8 A flow chart illustrating another example of a method for gesture recognition is shown. DETAILED DESCRIPTION

[0016] Some examples are now described in more detail with reference to the accompanying drawings. However, other possible examples are not limited to the features of these embodiments described in detail. Other examples may include modifications of features and equivalents and substitutes of features. In addition, the terms used herein to describe some examples should not limit other possible examples.

[0017] In the description of the entire drawings, the same or similar reference numerals refer to the same or similar elements and / or features, which may be identical or implemented in a modified form while providing the same or similar functions. For clarity, the thickness of the lines, layers and / or regions in the drawings may also be exaggerated.

[0018] When two elements A and B are combined using "or", unless explicitly defined otherwise in individual cases, this should be understood to disclose all possible combinations, i.e. only A, only B and A and B. As an alternative wording for the same combination, "at least one of A and B" or "A and / or B" can be used. The same applies to combinations of more than two elements.

[0019] If singular forms such as "a", "an", "said", and "the" are used, and the use of only a single element is not explicitly or implicitly defined as mandatory, other examples may also use several elements to achieve the same function. If the function is described below as being implemented using multiple elements, other examples may use a single element or a single processing entity to achieve the same function. It should also be understood that the terms "comprise", "comprising", "including", and / or "comprising" when used describe the presence of specified features, wholes, steps, operations, processes, elements, components, and / or groups thereof, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, processes, elements, components, and / or groups thereof.

[0020] Figure 1 A block diagram of an example of an apparatus 100 for gesture recognition is illustrated. Gesture recognition involves detecting and interpreting gestures or movements of a living being using radar-based sensing. The gestures recognized may vary depending on the specific application and system design. For example, gestures to be recognized may include waving, foot movement, finger movement, arm gestures, body movement, etc.

[0021] The apparatus 100 will be considered in the context of a radar sensor. For example, the apparatus 100 may be integrated into a radar system or may be external to a radar system including a radar system such as that described below with reference to Figure 6 In the former case, the apparatus 100 may be external to the radar sensor or integrated (eg, partially or fully) into the radar sensor.

[0022] The apparatus 100 includes a processing circuit device 110 and, optionally, an interface circuit device 120. Where the interface circuit device 120 is present, the interface circuit device 120 may be communicatively coupled (e.g., via a wired or wireless connection) to the processing circuit device 110, e.g., for data exchange between the interface circuit device 120 and the processing circuit device 110. The apparatus 100 may optionally include a memory (not shown) configured to store data, such as radar data, a data matrix (as explained below), or any results of processing by the processing circuit device 110.

[0023] The interface circuit device 120 can be any device or device for communicating or exchanging data. In the case where the device 100 includes the interface circuit device 120, the interface circuit device 120 can be configured to receive radar data acquired by a radar sensor (e.g., raw or preprocessed). For example, the interface circuit device 120 can be communicatively coupled to a radar sensor or a storage device storing radar data. For example, the interface circuit device 120 can receive radar data via a wired or wireless coupling to the radar sensor or storage device. Alternatively, the interface circuit device 120 can be configured to receive a data matrix indicating the distance within the slow time of the radar data. That is, the preprocessing of the radar data can be performed outside the device 100 to generate a data matrix.

[0024] According to the specific implementation, the device 100 can omit the interface circuit device 120: for example, the processing circuit device 110 can determine the radar data. For example, the processing circuit device 110 can be integrated into a radar sensor. The radar sensor can be configured to transmit a radio frequency signal into the field of view of the radar sensor and acquire radar data based on the reflection of the radio frequency signal. The radar data can indicate a signal received by a receiver of the radar sensor. For example, the radar data can indicate an IF (intermediate frequency) signal of the radar sensor. The radar sensor can acquire radar data by mixing the reflection with a local oscillator signal at a specific frequency. For example, the processing circuit device 110 can determine the radar data by sampling the IF signal and perform further processing of the radar data within the radar sensor. The processing circuit device 110 can optionally modify the sampled IF signal in a pre-processing step, for example, for noise reduction, DC removal (direct current), etc.

[0025] Alternatively, processing circuit device 110 may partially determine the radar data. For example, processing circuit device 110 may determine a first portion of the data, and at least one external processing circuit device may determine at least a second portion of the data. For example, processing circuit device 110 and the external processing circuit device may be connected within a distributed computing environment for jointly determining the data. In this case, the processing circuit device may be integrated into the radar sensor or may be external to the radar sensor. For example, processing circuit device 110 may receive the second portion of the radar data via an interface to the external processing circuit device (such as interface circuit device 120), and further process the first portion and the second portion of the radar data as described below.

[0026] In another alternative, processing circuit device 110 is partially integrated into the radar sensor and partially external to the radar sensor. In this case, interface circuit device 120 is optional. For example, processing circuit device 110 may include a first part (first processing circuit device) integrated into the radar sensor and a second part (second processing circuit device) external to the radar sensor. In this case, the determination and / or further processing of the radar data described below can be performed in a distributed manner by the first part and the second part of processing circuit device 110.

[0027] For example, processing circuit device 110 may be a single dedicated processor, a single shared processor, or multiple individual processors (some or all of which may be shared), digital signal processor (DSP) hardware, application specific integrated circuit (ASIC), microcontroller, or field programmable gate array (FPGA). For example, processing circuit device 110 may be selectively coupled to read-only memory (ROM), random access memory (RAM), and / or non-volatile memory for storing software.

[0028] The processing circuit device 110 is configured to obtain a data matrix indicating the distance in slow time based on radar data acquired by the radar sensor. The radar data can be preprocessed or raw radar data. For example, the processing circuit device 110 can determine the data matrix based on the radar data, acquire the data matrix according to external preprocessing, or partially determine the data matrix and receive the remaining data matrix. In the first case, the processing circuit device 110 can be configured to determine the data matrix by fast time processing, such as by performing Fourier transform along fast time or along different chirps of radar data. For example, the processing circuit device 110 can perform a range FFT (fast Fourier transform) using the radar data to generate a range (range profile, such as a complex range profile) in slow time. The range FFT can convert the IF signal value of the radar data into a range profile based on correlation and binning. Fast time processing can also include attenuation or removal of the complex mean and / or DC component of the distance to eliminate (quasi) static targets.

[0029] In some examples, the processing circuitry 110 may be configured to determine the data matrix using a moving target indicator (MTI). In the latter case, the distances of moving objects may remain substantially unchanged or may be amplified in the data matrix, while the distances of static objects may be attenuated or discarded.

[0030] Data matrix can refer to the abstract nature of data representation or formatting, that is, data matrix can refer to an array or matrix of data elements that keep distance in slow time. For example, data matrix can associate distance (for example, as a distance value of a fast time instance or bin) with a slow time instance of a corresponding slow time, so that a subset of distances associated with a certain slow time instance is arranged along a specific dimension, such as equivalent to a table format. Therefore, the data matrix can indicate the data points of the distance at a specific instance of slow time. This can allow distances to be processed more easily. Data matrix can optionally or alternatively refer to a storage format, such as when the distance in slow time is stored in a memory. For example, a data matrix can be stored in a continuous memory area, as a pointer array, a linked list, etc., to associate distance values ​​with their corresponding slow time instances.

[0031] Range in slow time may refer to a (radial) distance value measured by a radar sensor within its field of view. Thus, range may indicate the distance between a radar sensor and a surface or object around it. Slow time may refer to the temporal resolution of the duration over which a radar measurement is made, such as the duration of at least one radar pulse (e.g., chirp) transmitted by a radar sensor. Slow time differs from fast time, which may refer to the temporal resolution within a single pulse echo period or chirp.

[0032] The processing circuit device 110 is also configured to determine a compact data matrix of reduced size by compressing the data matrix in (along) slow time (or in chirp). The compact data matrix may have a reduced size in the sense that the compact data matrix has a smaller number of entries than the data matrix. For example, the length of the slow time dimension of the compact data matrix may be reduced. In some examples, the processing circuit device 110 may determine the compact data matrix by compressing the data matrix to a vector or a lower dimensional matrix (compared to the data matrix). In some examples, determining the compact data matrix may be performed without performing a Fourier transform on the data matrix.

[0033] (Slow time) compression may refer to any method for data compression, such as averaging at least two radar measurements, chirps or slow time instances in a data matrix. Thus, at least two radar measurements, chirps or slow time instances may be combined into a single measurement, chirp or slow time instance. This may reduce noise and improve the accuracy of the measurement. For example, such a method for slow time compression may be a simplified averaging method (e.g., integrating the distance over the slow time), a moving average method (e.g., averaging the distance over a sliding window along the slow time), a filtering method, a smoothing or weighting method, etc.

[0034] In addition, the processing circuit device 110 can be configured to determine a compact data matrix by compressing (e.g., averaging) the data matrix over multiple channels of radar data. For example, the data matrix can have an additional dimension (e.g., a third dimension) by which different channels are represented. The multiple channels of radar data can correspond to data generated based on corresponding channels of a radar sensor. For example, a radar sensor can include multiple receiver channels that capture corresponding parts of a radar signal from corresponding angles or viewpoints. The radar sensor can generate corresponding radar data for each receiver channel in a receiving channel. (Channel) compression can combine (align or synchronize) radar measurements of at least two channels (e.g., among all channels). Channel compression can reduce random noise or interference and improve the signal-to-noise ratio (SNR).

[0035] To combine slow time compression with channel compression, processing circuitry 110 may integrate the amplitude along the receive channel and the (complex) distance of the chirp. For example, the resulting compact data matrix may be a one-dimensional vector representing the reflected energy of a moving target along a resolvable distance.

[0036] The processing circuit device 110 is also configured to determine the distance interval of the target based on the compact data matrix. For the distance of the assumed target, the distance interval may include one or more distance bins in the compact data matrix. The target may be a person or a body part of a person for which gesture recognition is to be performed. The assumption that the measured distance belongs to the target of interest may depend on the specific application of the device 100. For example, the compact data matrix may indicate a plurality of mobile objects, such as leaving a signature with a sufficiently high signal amplitude in the compact data matrix. Then, the processing circuit device 110 may select one or more, a reduced subset, or all detected mobile objects as targets of interest for determining the distance interval.

[0037] Processing of a compact data matrix as proposed herein can enable "early" target detection of range profiles, which can reduce power consumption, memory, and computing resources compared to conventional gesture recognition. This may be particularly useful or even necessary when implementing gesture recognition on a deeply embedded target platform with limited resource availability. Conventionally, targets and their distances are determined by creating a range Doppler map (or image) to detect, for example, the nearest target in front of a radar sensor. The creation of such a range Doppler map can prevent the reduction of the original data matrix. Conventional methods may include the following steps: depending on the velocity resolution used (the number of chirps per frame), the radar sensor must transmit and receive back a certain number of chirps; the digital representation of the chirp (radar data) must be stored in a memory such as RAM (random access memory) and Fourier transformed to obtain a range profile; and another Fourier transform is performed for each bin in the range profile of each chirp in a frame of radar data. In contrast, the device 100 can have significantly lower use of RAM, CPU (central processing unit) time, and power consumption. With the proposed apparatus 100, implementation on smaller processor classes can be realized.

[0038] In some examples, the processing circuit device 110 can be configured to determine the distance interval of the closest target and / or the second closest target to the radar sensor based on the compact data matrix. For example, when it is assumed that the potential interferer is closest to the radar sensor, the processing circuit device 110 can only process the second closest target or process both the closest target and the second closest target to search for the posture of the target. Alternatively or additionally, any higher-order target can be selected as the target of interest, such as the third closest target, etc. High-order target selection can also be used to identify the posture of more than one person around the radar sensor.

[0039] The processing circuit device 110 can select a target of interest, such as the closest target and / or the second closest target, and determine the distance interval of the target of interest. For example, the processing circuit device 110 can determine the distance interval (and / or select the target) by determining a local peak in a compact data matrix, such as by running a local peak search in the compact data matrix. One of the determined local peaks can be selected as a target, and the distance interval can be determined accordingly, such as to include the distance of the selected target. Additionally or alternatively, the processing circuit device 110 can determine the target and its distance interval by, for example, thresholding, pattern recognition, etc.

[0040] In some examples, the determination of targets and range intervals may be performed without applying any transforms, such as Fourier transforms or Laplace transforms, to the compact data matrix.

[0041] In addition, the compact data matrix can be pre-processed before selecting a target or determining a distance interval. In some examples, the processing circuit device 110 is also configured to filter the compact data matrix and determine the distance interval based on the filtered compact data matrix. For example, the processing circuit device 110 can be configured to filter the compact data matrix by performing Gaussian smoothing on the compact data matrix. Gaussian smoothing can be applied to the distance of the compact data matrix to reduce noise, suppress small-scale changes, or enhance the visibility of targets in radar data. Additionally or alternatively, the processing circuit device 110 can filter the compact matrix using any other filtering technique such as median filtering.

[0042] For gesture recognition of hand gestures, it can be assumed for a specific application that a person will perform a gesture towards the radar sensor, and likewise the person's hand will be assumed to be the closest moving target. The person's body movement can cause a higher signal amplitude in a compact data matrix, but it is farther from the radar sensor than the hand, e.g. Figure 2 shown.

[0043] Figure 2 A distance spectrum 200 is shown showing an example of binned distances 210 of a compact data matrix and an example of binned distances 220 of a smoothed compact data matrix. The distances 210 and 220 are arranged as a distance profile, i.e., the values ​​of the distances 210, 220 are binned in the distance spectrum 200, thus emphasizing the main distance bins indicating the presence of a target.

[0044] exist Figure 2 In the example of , distance 220 crosses threshold 230 twice, resulting in a first peak 240 and a second peak 250. Threshold 230 can be set so that weak local maxima are suppressed to avoid false targets. The smaller first peak 240 at range bin 17 is caused by the person's moving hand, and the higher second peak 250 at range bin 23 is caused by the person's moving body. The distance interval can be determined to include range bin 17, and optionally include adjacent range bins, for example so that the distance difference of the hand during its gesture can be captured. Peak 250 can be discarded in further processing for identifying the gesture of the hand. This means that instead of selecting the target with the highest signal strength corresponding to the body, the closest target corresponding to the hand is selected. This can result in effective but stable target detection that is more immune to random body movements of the person performing the gesture.

[0045] Reference again Figure 1, the processing circuit device 110 is also configured to specifically process the determined range intervals in the data matrix for determining the movement of the target. When the range interval of the target is known, the relevant processing (for example, for extracting scattering characteristics) can be limited to the range interval, which can increase processing efficiency. The device 100 can only process the range bins (range intervals) of interest, rather than performing a slow time FFT across all range bins to generate a range Doppler image as in conventional gesture recognition. For example, the processing circuit device 110 can discard other distances (or range bins) in the data matrix that are different from the distance determined by the range interval. Therefore, the range interval can include a reduced subset of all detected distances or all measurable distances of the moving target, and thereby exclude at least one of the detected distances or at least one of the measurable distances.

[0046] For example, determining the movement of a target may include extracting features (characteristics) of the movement, which may be used to detect a gesture to be identified. The feature may be at least one of the distance, speed, signal amplitude, elevation, and azimuth of the target causing the movement. Alternatively or additionally, the feature may be a phase difference between channels of radar data. For example, the processing circuit device 110 may be configured to specifically process the distance interval determined in the data matrix by determining at least one of the speed, azimuth, and elevation of the target within the determined distance interval. As described above, features such as the distance and signal amplitude of the target may be derived directly from the compact data matrix.

[0047] Specializing in processing the determined distance intervals in the data matrix to extract the characteristics of the movement can increase the power and computational efficiency of the device 100, because irrelevant distances are excluded from the processing. It can be based on any processing method, such as time domain or frequency domain processing. In the following, two examples of such methods will be described in more detail. Since these methods are combinable, they do not have to be considered isolated from each other.

[0048] In a first example, processing circuitry 110 is configured to specifically process a determined range interval in the data matrix by determining a velocity representation (or Doppler representation) of the data matrix within the determined range interval. The velocity representation may be determined by converting the relevant portion of the data matrix within the range interval from the range-time domain to the velocity domain. This may produce a velocity distribution with a reduced size compared to conventional methods of creating a full range Doppler map for all ranges. Each element of the velocity distribution may represent the strength or amplitude of the radar return signal at a specific distance within the range interval and velocity.

[0049] The velocity representation may be determined based on any technique for velocity analysis, such as matched filtering, thereby determining an ambiguity function for varying distances, or compressed sensing. In some examples, the processing circuit device 110 is configured to determine the velocity representation of the data matrix by performing a Fourier transform, such as an FFT, along the slow time of the data matrix within a determined distance interval. The latter may reduce computational complexity compared to any direct calculation of the velocity. In the case of determining the data matrix based on a Fourier transform, the two Fourier transforms may also be combined into a two-dimensional Fourier transform of the radar data.

[0050] The (radial) velocity of the target movement can then be derived from the velocity representation. For example, the processing circuit device 110 can be configured to determine the velocity by determining a peak in the velocity representation of the data matrix. For example, a maximum value (amplitude) can be searched in the velocity representation, and the resulting position within the velocity representation can indicate the radial velocity, while the maximum value can indicate the signal amplitude. As an optional pre-processing, the velocity representation can be integrated over the channels mentioned above, and the velocity can be derived from the resulting integrated velocity representation.

[0051] The first example can increase the power and computational efficiency of gesture recognition on the one hand, and can provide feature extraction with high accuracy on the other hand, since a velocity representation including only relevant values ​​(determined distance intervals) is determined.

[0052] If other features such as azimuth and elevation are to be determined, these features can be derived from the velocity representation. For example, the processing circuit device 110 can be configured to apply a phase comparison single pulse to the velocity representation of the target for determining at least one of the azimuth and elevation. The phase comparison single pulse technique can include comparing the values ​​of different channels with each other. This comparison can include measuring the phase difference between the signals. Various techniques can be used to perform this phase comparison, such as using a phase detector or a phase shifter. The angle of arrival can then be estimated by processing the phase difference obtained from the processed phase comparison. The angle estimate can include a ratio single pulse, a difference single pulse, or a sum and difference single pulse. In order to further reduce the processing workload, the processing circuit device 110 can (specially) apply a phase comparison single pulse to the velocity representation for the peak value determined in the velocity representation. In other examples, any alternative method can be used to extract the angle of the target, such as an amplitude comparison single pulse, a time difference of arrival technique, cross-correlation, etc.

[0053] Figure 3 The specific implementation of the first example is further illustrated. Figure 3 A flowchart of an example of a method 300 for gesture recognition is shown. The method 300 may be executed by a processing circuit device of an apparatus for gesture recognition (eg, apparatus 100).

[0054] The method 300 includes receiving 310 (raw) radar data acquired by a radar sensor. The radar data is received at 30 frames per second (fps). One frame (or burst) can be represented as a three-dimensional array [R×C×S], where R=3 (receive) channels, C=32 is the length of the radar data along the slow time axis (or the number of chirps in the radar data), and S=64 is the length of the radar data along the fast time axis. Figure 3 The frame rates, number of channels and length of the data matrix in the fast-time or slow-time dimension shown in should be understood as examples for illustrative purposes only. Figure 3 In other examples of the examples shown, the frame rate can be any number of n≥1 frames per unit time; the number of channels can be any number of R≥1; the length of the radar data along the fast time axis and the slow time axis can be any number of S≥2 and C≥2 respectively.

[0055] The method 300 also includes obtaining 320 a data matrix indicating distances in slow time based on the radar data (where R=3, C=32, S=32). Obtaining 320 the data matrix includes determining the data matrix by performing a fast Fourier transform on the radar data along fast time. Figure 3 In the example of , the real values ​​of the radar data are converted to complex values ​​of the data matrix, so the length of the S dimension is reduced from 64 to 32. Figure 3 In other examples than the illustrated examples, the size of the data matrix may be the same as the size of (the frame of) radar data, or may be otherwise modified.

[0056] The method 300 also includes determining 330 a data matrix using the moving target indicator. In this way, when the static or quasi-static target is attenuated, the data matrix can indicate the distance of the moving target. Figure 3 In the example shown in Figure 2, after applying a moving target indicator on a data matrix, the dimensions of the data matrix remain unchanged.

[0057] The method 300 also includes determining 340 a compact data matrix of reduced size by compressing the data matrix in slow time (within the chirp). In addition, the data matrix is ​​averaged over the three channels of radar data. The resulting compact data matrix has a size of C=1, S=32.

[0058] The method 300 further includes filtering 350 the compact data matrix by performing Gaussian smoothing on the compact data matrix, and determining 360 the range interval of the target based on the filtered compact data matrix. The range interval is determined by determining local peaks in the compact data matrix (local peak search).

[0059] Method 300 can achieve the determination of the movement characteristics of the target. Therefore, method 300 includes determining 370 at least one of the speed, signal amplitude, azimuth and elevation of the target within a determined distance interval. Details of how to determine the characteristics are given below:

[0060] The method 300 also includes specifically processing the distance intervals determined in the data matrix for use in determining the movement of the target. Figure 3 In an example of , specifically processing the determined range intervals in the data matrix includes determining 380 a velocity representation of the data matrix by performing a Fourier transform along slow time of the data matrix within the determined range intervals. The Fourier transform is applied to the data matrix obtained after using the moving target indicator.

[0061] The method 300 also includes determining 385 a moving speed of the target by determining a peak in the velocity representation of the data matrix.

[0062] Thus, method 300 can achieve "early" target detection on range profiles rather than range Doppler images, which can reduce power consumption, memory, and computing resources compared to conventional gesture recognition. The relevant processing for extracting motion features can be limited to the range interval, which can increase processing efficiency. For example, by performing early detection on the nearest target already on the range profile, and performing velocity conversion only on the relevant range profile bin, the time required to calculate the gesture can be reduced - depending on the application, it can be reduced to less than 1 / 2.

[0063] For further feature extraction, the method 300 comprises applying 390 a phase comparison pulse to the velocity representation of the data matrix for peaks determined in the velocity representation. Thus, the azimuth and elevation angles can be extracted in an efficient manner.

[0064] The method 300 also includes extracting a time sequence of at least one feature of the movement, and recognizing a gesture based on the time sequence of at least one feature of the movement ( Figure 3 not shown).

[0065] Reference again Figure 1 In a second example, the processing circuit device 110 is configured to specifically process the determined distance intervals in the data matrix by determining the phase offset of the slow time along the data matrix within the determined distance interval, and determining the moving speed of the target based on the phase offset. The latter processing can be performed optionally or in addition to the processing of the first example above. The processing circuit device 110 can compare the phases of the chirps with each other and detect the difference in the phases. For example, the phase offset can be converted into an estimate of the target speed using known characteristics or chirp parameters of the radar sensor.

[0066] In addition to the advantages mentioned above, the second example can be based on phase offset determination and can therefore omit a more elaborate determination of the speed representation as in the first example. Therefore, the second example can further reduce the computational complexity.

[0067] In some examples, the phase may exhibit ambiguity. In these cases, the processing circuit device 110 can be configured to determine the movement speed by unfolding the phase offset. Any unfolding technique can be performed to determine the speed, for example, simple unfolding (by adding or subtracting multiples of 2π to the phase value), branch cut unfolding, least squares unfolding, etc.

[0068] For further feature extraction, processing circuitry 110 may be configured to apply a phase comparison single pulse to the phases of multiple channels in the compact data matrix. Processing circuitry 110 may extract azimuth and elevation in this manner.

[0069] In applications with extremely limited resources, the number of chirps can be reduced to save power. For example, the data matrix can include at most two chirps (or at most three, at most four, or at most five chirps). When there are only such a small amount of chirps in the data matrix, conventional methods may not be able to provide acceptable accuracy in gesture detection. In contrast, since the second example is based on phase offset detection, gesture detection can be achieved even with such an extremely reduced number of chirps.

[0070] Figure 4 The specific implementation of the second example is further illustrated. Figure 4 A flowchart of an example of a method 400 for gesture recognition is shown. The method 400 may be executed by a processing circuit device of an apparatus for gesture recognition (eg, apparatus 100).

[0071] Method 400 may include steps 310 to 370 of method 300. Therefore, steps 410 to 470 may correspond to steps 310 to 370. Figure 4 In , the data matrix has only dimensions of R=3, C=2, S=32, and therefore, it includes only two chirps, thereby reducing the power consumption of the radar sensor. However, in other examples, the number of chirps included in the data matrix may be more than two.

[0072] The method 400 also includes specifically processing the determined range intervals in the data matrix by determining 480 a phase offset along the slow time of the data matrix within the determined range interval and determining the moving speed of the target based on the phase offset. The phase offset is determined using a phase subtraction method. Over time, a small displacement ri(T2)-ri(T1) in range bin i can appear as a proportional phase offset Δφi(T1, T2). The phase offset Δφi can be used directly as its approximation instead of calculating the spectrum using FFT along the chirp in the slow time axis to estimate the speed of the target. The measured SNR can be increased by integrating the receive channel R.

[0073] For additional feature extraction, the method 400 further includes applying 490 a phase comparison monopulse to the phases of the multiple channels in the compact data matrix. The phase differences across the received channels can be used in the detected range bins to calculate the horizontal and vertical angles using the phase comparison monopulse.

[0074] Method 400 can achieve "early" target detection on range profiles rather than range Doppler images, which can reduce power consumption, memory, and computing resources compared to conventional gesture recognition. The correlation processing for extracting mobile features can be limited to the range interval, which can increase processing efficiency. In addition, method 400 may be feasible for a very limited number of chirps and may therefore have the potential to further reduce resource consumption. For example, the frame size can be reduced from 32 chirps to only 2 chirps, and method 400 includes a simpler phase subtraction method in the range bin of interest, rather than the previously used fast Fourier transform. After phase unwrapping, the result can be used in the same manner as in the previous method. Depending on the application, this can reduce the amount of RAM and the required computational budget, as well as reduce the overall consumed energy.

[0075] Reference again Figure 1 , the processing circuit device 110 is further configured to extract a time series of at least one feature of the movement, and recognize a gesture based on the time series of at least one feature of the movement. For example, the time series can be extracted by tracking the feature over a specific time. To ensure the alignment of the time series, the processing circuit device 110 can be configured to extract the time series of at least one feature by starting the time series at the data point with the minimum distance. With the latter configuration, the time series of the feature can be more easily associated with the specific gesture that caused the characteristic process of the time series.

[0076] For example, time series can be mapped to corresponding poses. Figure 5 Table 500 illustrates an example of a time series of features 510 - 514 associated with examples of gestures 520 - 524 . Figure 5The time series of the features shown are radial distance (distance (range)) 510, radial velocity 511, horizontal angle (azimuth) 512, vertical angle (elevation) 513, and amplitude (signal amplitude) 514. For each of the features 510-514, five examples of the time series are shown. Each of the five examples corresponds to a specific posture in postures 520-524. Figure 5 Examples of gestures shown are hand gestures, which include push 520 , left swipe 521 , right swipe 522 , upward swipe 523 , and downward swipe 524 .

[0077] Processing circuit device 110 can use any gesture recognition technology to recognize gestures, for example, probabilistic methods such as hidden Markov models, similarity measurements such as dynamic time warping, etc. In some examples, processing circuit device 110 is configured to recognize gestures by using a trained machine learning model. For example, the trained machine learning model can be based on a trained recursive neural network, a convolutional neural network, a support vector machine, etc.

[0078] A trained machine learning model may be a data structure and / or rule set representing a statistical model that the processing circuit device 110 uses to recognize gestures without explicit instructions, but instead relying on models and reasoning. The data structure and / or rule set represents learned knowledge (e.g., based on training performed by a machine learning algorithm). For example, in machine learning, data transformations inferred from analysis of historical and / or training data may be used instead of rule-based data transformations. In the proposed technology, a trained machine learning model (i.e., a data structure and / or rule set representing the model) may be used to analyze the content of a time series of features.

[0079] The machine learning model can be trained in external training using a machine learning algorithm. The term "machine learning algorithm" refers to a set of instructions for creating, training, or using a machine learning model. In order for the machine learning model to analyze the time series of features, the machine learning model can be trained using the training and / or historical time series of features as input and using training content information (e.g., labels indicating corresponding gestures) as output.

[0080] The apparatus 100 may implement a power-saving operating mode of the radar sensor, such as a standby mode, in which the frame rate, number of chirps, sampling, etc. of the radar sensor are reduced, so that the power consumption of the radar sensor can be reduced as well. Alternatively or additionally, the apparatus 100 may implement an operating mode of the radar sensor in which the frame rate of the radar sensor is increased, so that the extracted time series exhibits higher accuracy. For example, the apparatus 100 may be communicatively coupled to the radar sensor, such as via the interface circuit device 120, and send a trigger to the radar sensor to activate one mode or another.

[0081] For example, the processing circuit device 110 may also be configured to determine whether a target exists based on the data matrix, and in response to determining that a target exists, trigger the radar sensor to increase at least one of the frame rate of the radar sensor, the number of chirps per frame, and the number of samples per chirp. Therefore, as long as the device 100 does not detect a target, the radar sensor can be in sleep mode, and when a target exists, the device 100 can wake up the radar sensor. In another example, the processing circuit device 110 may be configured to trigger the radar sensor to reduce at least one of the frame rate, the number of chirps, and the number of samples in response to determining that no target exists. For example, the frame rate of the radar sensor can be reduced from 30 frames per second (in operating mode) to 3 frames per second (in energy saving mode). Therefore, gesture detection can only be activated when motion is detected in the field of view of the radar sensor.

[0082] Figure 6 An example of a radar system 600 is illustrated. The radar system 600 includes an apparatus 610 (eg, apparatus 100) for gesture recognition as described herein, and a radar sensor 620. The radar sensor 620 is configured to acquire radar data.

[0083] For example, radar sensor 620 may be a FMCW (frequency modulated continuous wave) radar sensor. Radar sensor 620 may be configured to emit a chirp, for example, spanning from f low =58.5GHz(gigahertz) to f high =62.5 GHz frequency range, resulting in a range resolution Δr=37.5 mm (millimeter). The radar sensor 620 can receive the reflected radio frequency signal, for example, three receiving antennas that can be arranged in an L shape. This can allow the angle of the scattering target in two planes to be estimated. The received chirped signal can be converted into an intermediate frequency signal. Anti-aliasing filtering and digitization can then be performed, for example, where 64 samples are performed at 2 MHz (megahertz), resulting in a maximum resolvable distance r max= 1.2 m (meters). Radar sensor 620 may transmit bursts of chirps, for example at a frame rate of 33.3 Hz (hertz), where individual chirps are separated by a pulse repetition time (TPRT) of 300 μs (microseconds). A burst may include a specific number of chirps, for example 32 or 2. Thus, each burst may be indicated by a three-dimensional array.

[0084] Although the device 610 and the radar sensor 620 are Figure 6 6 is depicted as a separate box, but in other examples, device 610 may be partially or entirely included in radar sensor 620, which radar sensor 620 thus correspondingly includes all or part of the processing circuit device of device 610 (e.g., processing circuit device 110).

[0085] In the case where device 610 is only partially included in radar sensor 620, radar system 600 may include a distributed processing circuit device that implements corresponding parts of the processing steps, which distributed processing circuit device is, for example, in the form of a first processing (sub-)circuit device included in radar sensor 620, and a second processing (sub-)circuit device that is external to the sensor and communicates with the first processing circuit device via an interface circuit device (e.g., interface circuit device 120), for example for exchanging data between the first processing circuit device and the second processing circuit device.

[0086] In the case where device 610 is integrated in radar sensor 620 , processing circuitry and radar sensor 620 may be integrated together in a single semiconductor chip, or in more than one semiconductor chip.

[0087] Where apparatus 610 is not included in radar sensor 620 , the processing circuitry may take the form of circuitry external to radar sensor 620 and may be communicatively coupled thereto via interface circuitry.

[0088] Further details and aspects of the radar system 600 will be explained in conjunction with the proposed technology or one or more examples described above, such as with reference to Figure 1 The radar system 600 may include one or more additional optional features corresponding to one or more aspects of the proposed technology, or one or more examples described above.

[0089] Radar system 600 can achieve "early" target detection on range profiles rather than range Doppler images, which can reduce power consumption, memory and computing resources compared to conventional gesture recognition. The correlation processing for extracting mobile features can be limited to the range interval, which can increase processing efficiency.

[0090] Figure 7An example of an electronic device 700 is illustrated. Electronic device 700 includes a radar system 710 (eg, radar system 600) as described herein, and control circuitry 720. Control circuitry 720 is configured to control operation of electronic device 700 based on the recognized gesture.

[0091] The control circuit device 720 can be a single dedicated processor, a single shared processor, or multiple individual processors (some or all of which can be shared), digital signal processor (DSP) hardware, application specific integrated circuit (ASIC), or field programmable gate array (FPGA). For example, the control circuit device 720 can be selectively coupled to a read-only memory (ROM), random access memory (RAM), and / or non-volatile memory for storing software.

[0092] The electronic device 700 may be any device having a gesture detection function. For example, the electronic device 700 may be a consumer device. For example, the control circuit device 720 may control the operation of the electronic device 700 by activating or disabling a function of the electronic device 700 based on the determined gesture, for example, if it is determined that the user of the electronic device 700 has performed a specific gesture, a function may be activated.

[0093] More details and aspects of the electronic device 700 will be explained in conjunction with the proposed technology or one or more examples described above, such as reference Figure 1 and Figure 6 The electronic device 700 may include one or more additional optional features corresponding to one or more aspects of the proposed technology, or one or more examples described above.

[0094] The electronic device 700 can implement "early" target detection on the range profile rather than the range Doppler image, which can reduce power consumption, memory and computing resources compared to conventional gesture recognition. The correlation processing for extracting the moving features can be limited to the range interval, which can increase processing efficiency.

[0095] Figure 8 The flowchart of an example of a method 800 for gesture recognition is illustrated. The method 800 may be performed by the apparatus for gesture recognition described herein (eg, apparatus 100). In addition, the method 800 may include one or more steps of the method 300 and / or one or more steps of the method 400.

[0096] The method 800 includes obtaining 810 a data matrix indicating distances in slow time based on radar data acquired by a radar sensor, determining 820 a compact data matrix of reduced size by compressing the data matrix in slow time, and determining 830 a distance interval of a target based on the compact data matrix. The method 800 also includes specifically processing 840 the determined distance intervals in the data matrix for determining movement of the target, extracting 850 a time series of at least one feature of the movement, and recognizing 860 a gesture based on the time series of at least one feature of the movement.

[0097] More details and aspects of the method 800 will be explained in conjunction with the proposed technology or one or more examples described above, such as reference Figure 1 The method 800 may include one or more additional optional features corresponding to one or more aspects of the proposed technology, or one or more examples described above.

[0098] Method 800 can achieve "early" target detection on range profiles rather than range Doppler images, which can reduce power consumption, memory and computing resources compared to conventional gesture recognition. The correlation processing for extracting mobile features can be limited to the range interval, which can increase processing efficiency.

[0099] In the following, some examples of the proposed technique are presented:

[0100] One example (e.g., Example 1) relates to a device for gesture recognition, comprising: a processing circuit device configured to obtain a data matrix indicating distance in slow time based on radar data acquired by a radar sensor, determine a compact data matrix of reduced size by compressing the data matrix in slow time, determine a distance interval of a target based on the compact data matrix, specifically process the determined distance interval in the data matrix for determining movement of the target, extract a time series of at least one feature of the movement, and recognize a gesture based on the time series of at least one feature of the movement.

[0101] Another example (e.g., Example 2) relates to the previous example (e.g., Example 1) or any other example, and also includes: the processing circuit device is configured to determine the distance interval of the target closest to the radar sensor and / or the second closest target based on the compact data matrix.

[0102] Another example (e.g., Example 3) relates to a previous example (e.g., one of Examples 1 or 2) or any other example, and also includes: the processing circuit device is configured to specifically process the distance interval determined in the data matrix by determining at least one of the speed, azimuth, and elevation of the target within the determined distance interval.

[0103] Another example (e.g., Example 4) relates to a previous example (e.g., one of Examples 1 to 3) or any other example, and also includes: the processing circuit device is configured to specifically process the determined distance intervals in the data matrix by determining the velocity representation of the data matrix within the determined distance intervals.

[0104] Another example (e.g., Example 5) relates to the previous example (e.g., Example 4) or any other example, and also includes: the processing circuit device is configured to determine a velocity representation of the data matrix by performing a Fourier transform along a slow time of the data matrix within a determined distance interval.

[0105] Another example (e.g., Example 6) relates to a previous example (e.g., one of Examples 4 or 5) or any other example, and also includes: the processing circuit device is configured to determine the moving speed of the target by determining a peak in the speed representation of the data matrix.

[0106] Another example (eg, Example 7) relates to the previous example (eg, one of Examples 4 to 6) or any other example, further comprising: the processing circuit device is configured to apply a phase comparison single pulse to the velocity representation of the data matrix.

[0107] Another example (eg, Example 8) relates to the previous examples (eg, Examples 6 and 7), further comprising: the processing circuit device is configured to apply a phase comparison single pulse to the speed representation for a peak value determined in the speed representation.

[0108] Another example (e.g., Example 9) relates to a previous example (e.g., one of Examples 1 to 8) or any other example, and also includes: the processing circuit device is configured to specifically process the determined distance intervals in the data matrix by determining the slow time phase offset along the data matrix within the determined distance interval, and determining the moving speed of the target based on the phase offset.

[0109] Another example (eg, Example 10) relates to the previous example (eg, Example 9) or any other example, further comprising: the processing circuit device is configured to determine the movement speed by unfolding the phase offset.

[0110] Another example (e.g., Example 11) relates to a previous example (e.g., one of Examples 9 or 10) or any other example, and also includes: the processing circuit device is configured to apply a phase comparison single pulse to the phases of multiple channels in the compact data matrix.

[0111] Another example (e.g., Example 12) relates to a previous example (e.g., one of Examples 1 to 11) or any other example, and also includes: the processing circuit device is configured to determine the distance interval of the target by determining a local peak in the compact data matrix.

[0112] Another example (e.g., Example 13) relates to a previous example (e.g., one of Examples 1 to 12) or any other example, and also includes: the processing circuit device is configured to determine a data matrix by performing a Fourier transform along a fast time of the radar data.

[0113] Another example (eg, Example 14) relates to a previous example (eg, one of Examples 1 to 13) or any other example, further comprising: the processing circuit device is configured to determine the data matrix using a moving target indicator.

[0114] Another example (e.g., Example 15) relates to a previous example (e.g., one of Examples 1 to 14) or any other example, and also includes: the processing circuit device is configured to determine a compact data matrix by compressing the data matrix across multiple channels of radar data.

[0115] Another example (e.g., Example 16) relates to a previous example (e.g., one of Examples 1 to 15) or any other example, and also includes: the processing circuit device is also configured to filter the compact data matrix and determine the distance interval based on the filtered compact data matrix.

[0116] Another example (eg, Example 17) relates to the previous example (eg, Example 16) or any other example, further comprising: the processing circuit device is configured to filter the compact data matrix by performing Gaussian smoothing on the compact data matrix.

[0117] Another example (eg, Example 18) relates to a previous example (eg, one of Examples 1 to 17) or any other example, further comprising: the data matrix includes at most two chirps of radar data.

[0118] Another example (e.g., Example 19) relates to a previous example (e.g., one of Examples 1 to 18) or any other example, and also includes: the processing circuit device is also configured to determine whether a target exists based on the data matrix, and in response to determining that the target exists, trigger the radar sensor to increase the frame rate of the radar sensor.

[0119] Another example (e.g., Example 20) relates to the previous example (e.g., Example 19) or any other example, and also includes: the processing circuit device is also configured to trigger the radar sensor to reduce the frame rate in response to determining that no target is present.

[0120] Another example (e.g., Example 21) relates to a previous example (e.g., one of Examples 1 to 20) or any other example, and also includes: the feature is at least one of the range, speed, signal amplitude, elevation, and azimuth of the target.

[0121] Another example (e.g., Example 22) relates to a previous example (e.g., one of Examples 1 to 21) or any other example, and also includes: the processing circuit device is configured to recognize gestures by using a trained machine learning model.

[0122] Another example (e.g., Example 23) relates to a previous example (e.g., one of Examples 1 to 22) or any other example, and also includes: the processing circuit device is configured to extract a time series of at least one feature by starting the time series at a data point with a minimum distance.

[0123] One example (e.g., Example 24) relates to a radar system comprising an apparatus according to a previous example (e.g., any of Examples 1 to 23) or any other example, and a radar sensor, wherein the radar sensor is configured to acquire radar data.

[0124] One example (e.g., Example 25) relates to an electronic device that includes a radar system according to a previous example (e.g., Example 24) or any other example, and a control circuit device configured to control the operation of the electronic device based on the identified gesture.

[0125] One example (e.g., Example 26) relates to a method for gesture recognition, comprising: obtaining a data matrix indicating distance in slow time based on radar data acquired by a radar sensor, determining a compact data matrix of reduced size by compressing the data matrix in slow time, determining a distance interval of a target based on the compact data matrix, specifically processing the determined distance interval in the data matrix for determining movement of the target, extracting a time series of at least one feature of the movement, and recognizing a gesture based on the time series of at least one feature of the movement.

[0126] Another example (e.g., Example 27) relates to a non-transitory machine-readable medium having stored thereon a program having program code, which, when executed on a processor or programmable hardware, is used to perform the method of the previous (e.g., Example 26) or any other example.

[0127] Another example (eg, Example 28) relates to a program having a program code, which, when executed on a processor or programmable hardware, is used to perform the method of the previous example (eg, Example 26) or any other example.

[0128] The aspects and features described with respect to a specific example among the foregoing examples may also be combined with one or more other examples to replace the completely identical or similar features of the other examples, or to additionally introduce features into the other examples.

[0129] Examples may also be or relate to (computer) programs, which include program codes that execute one or more methods in the above method when the program is executed on a computer, a processor, or other programmable hardware components. Therefore, the steps, operations, or processes of different methods in the above method may also be performed by a programmed computer, a processor, or other programmable hardware components. Examples may also encompass program storage devices, such as digital data storage media, which are machine, processor, or computer readable, and encode and / or include machine executable, processor executable, or computer executable programs and instructions. For example, program storage devices may include or may be digital storage devices, magnetic storage media such as disks and tapes, hard drives, or optically readable digital data storage media. Other examples may also include computers, processors, control units, (field) programmable logic arrays ((F) PLAs), (field) programmable gate arrays ((F) PGAs), graphics processor units (GPUs), application specific integrated circuits (ASICs), integrated circuits (ICs), or system-on-chip (SoC) systems that are programmed to perform the steps of the above method.

[0130] It should also be understood that, unless explicitly stated in individual cases or necessary for technical reasons, the disclosure of several steps, processes, operations or functions disclosed in the specification or claims should not be interpreted as implying that these operations must rely on the described order. Therefore, the previous description does not limit the execution of several steps or functions to a certain order. In addition, in other examples, a single step, function, process or operation can include and / or be decomposed into several sub-steps, functions, processes or operations.

[0131] If some aspects have been described with respect to a device or system, these aspects should also be understood as descriptions of the corresponding methods. For example, a block, device, or functional aspect of a device or system may correspond to a feature of a corresponding method, such as a method step. Therefore, aspects described with respect to a method should also be understood as descriptions of corresponding blocks, corresponding elements, attributes, or functional features of the corresponding device or corresponding system.

[0132] The following claims are hereby incorporated into the detailed description, where each claim can be independently presented as a separate example. It should also be noted that, although in the claims, a dependent claim refers to a specific combination with one or more other claims, other examples may also include a combination of a dependent claim with the subject matter of any other dependent or independent claim. Unless it is stated in individual cases that a specific combination is not intended, such a combination is explicitly proposed here. In addition, even if a claim is not directly defined as dependent on other independent claims, the features of the claim should also be included in any other independent claim.

Claims

1. A device (100) for gesture recognition, comprising a processing circuit device (110), wherein the processing circuit device (110) is configured to: Obtaining a data matrix indicating a distance in slow time based on radar data acquired by a radar sensor; determining a compact data matrix of reduced size by compressing the data matrix in a slow time; determining a range separation of targets based on the compact data matrix; processing the determined range intervals in the data matrix specifically for use in determining movement of the target; extracting a time series of at least one feature of the movement; as well as A gesture is recognized based on the temporal sequence of the at least one characteristic of the movement.

2. The device (100) according to claim 1, wherein the processing circuit device (110) is configured to: determine the distance interval of the target closest to the radar sensor and / or the second closest target based on the compact data matrix.

3. An apparatus (100) according to any one of the preceding claims, wherein the processing circuit device (110) is configured to specifically process the determined distance intervals in the data matrix by determining at least one of the speed, azimuth and elevation of the target within the determined distance interval.

4. An apparatus (100) according to any one of the preceding claims, wherein the processing circuit means (110) is configured to specifically process the determined distance intervals in the data matrix by determining a velocity representation of the data matrix within the determined distance intervals.

5. The apparatus (100) of claim 4, wherein the processing circuitry (110) is configured to determine the speed of the movement of the target by determining a peak in the velocity representation of the data matrix.

6. The device (100) according to any one of claims 4 or claim 5, wherein the processing circuit device (110) is configured to: apply a phase comparison single pulse to the speed representation for the determined peak value in the speed representation.

7. The apparatus (100) according to any one of the preceding claims, wherein the processing circuit means (110) is configured to exclusively process the determined distance intervals in the data matrix by: determining a phase offset of the slow time along the data matrix within the determined range interval; and A speed of the movement of the target is determined based on the phase offset.

8. The apparatus (100) of claim 7, wherein the processing circuitry (110) is configured to apply a phase comparison single pulse to the phases of a plurality of channels in the compact data matrix.

9. The apparatus (100) of any one of the preceding claims, wherein the processing circuitry (110) is configured to determine the compact data matrix by compressing the data matrix over a plurality of channels of the radar data.

10. The apparatus (100) of any one of the preceding claims, wherein the data matrix comprises at most two chirps of the radar data.

11. The apparatus (100) according to any one of the preceding claims, wherein the processing circuit arrangement (110) is further configured to: determining whether a target exists based on the data matrix; and In response to determining that a target is present, the radar sensor is triggered to increase at least one of a frame rate, a number of chirps per frame, and a number of samples per chirp of the radar sensor.

12. The apparatus (100) of claim 11, wherein the processing circuit device (110) is further configured to, in response to determining that no target is present, trigger the radar sensor to reduce at least one of the frame rate, the number of chirps per frame, and the number of samples per chirp.

13. A radar system (600), comprising: The device (610) according to any one of claims 1 to 12; as well as The radar sensor (620), wherein the radar sensor (620) is configured to acquire the radar data.

14. An electronic device (700), include ： The radar system (710) of claim 13, and A control circuit device (720) is configured to control the operation of the electronic device based on the recognized gesture.

15. A method (800) for gesture recognition, comprising: Based on radar data acquired by the radar sensor, obtaining (810) a data matrix indicating distance in slow time; determining (820) a compact data matrix of reduced size by compressing the data matrix in a slow time; determining (830) a range separation of targets based on the compact data matrix; processing (840) the determined distance intervals in the data matrix specifically for use in determining movement of the target; extracting (850) a time series of at least one feature of the movement; and A gesture is identified (860) based on the temporal sequence of the at least one characteristic of the movement.

Images