Target detection method, integrated circuit, sensor, device, and medium

Abstract

A target detection method, integrated circuit, sensor, device, and medium, the method including: storing frame data for 1D-FFT data, the 1D-FFT data being acquired by performing range-dimensional FFT processing on echo signals; performing inter-frame FFT processing based on the stored frame data to acquire an RD spectrum; and detecting a target in a region of interest based on the RD spectrum.

Description

[Technical Field]

[0001] This application claims priority to a Chinese patent application filed on July 24, 2023, with application number 202310913653.8, and the title of the invention is "Target detection method and system, integrated circuit, sensor and apparatus," and its contents should be understood to be incorporated into this application by means of reference. The embodiments of this disclosure relate to, but are not limited to, the technical field of electromagnetic wave sensors, and more specifically to target detection methods and systems, integrated circuits, sensors and devices. [Background technology]

[0002] When performing target detection in a sealed or relatively sealed spatial area, such as target detection inside a car cabin, there are technical challenges such as low detection rates, a high number of false alarms, and inaccurate angle estimation. [Overview of the project]

[0003] The following is an overview of the topic, which will be explained in detail in the main text. This overview is not intended to limit the scope of protection of the patent claims.

[0004] Embodiments of the present invention provide a target detection method, and the method is Frame data is accumulated for 1D-FFT data, and the 1D-FFT data is obtained by performing distance-dimensional FFT processing on the echo signal. Based on the accumulated frame data, an FFT process is performed between frames to obtain the RD spectrum, and This may also include detecting a target in the region of interest based on the aforementioned RD spectrum.

[0005] As an alternative, the method may include accumulating frame data from the 1D-FFT data, accumulating data up to a predetermined amount, and then performing FFT processing between frames to obtain the RD spectrum. As an alternative, the method may include, if the data is cached using a Ring FIFO, reading the data from the Ring FIFO and rearranging the read data. As an option, the noise floor estimation result for a single transmit / receive channel consists of the noise floor estimation results for each distance unit in the single transmit / receive channel, and the constant false alarm probability detection result for the single transmit / receive channel consists of the constant false alarm probability detection results for each Doppler unit of each distance unit in the single transmit / receive channel. Based on the noise floor estimation result of the aforementioned single transmit / receive channel, detecting a fixed false alarm probability for the corresponding transmit / receive channel independently is achieved by the following equation: JPEG0007847889000001.jpg1373 Here, TIFF0007847889000002.tif6150 This is the result of detecting a constant false alarm probability for the v-th Doppler unit of the r-th distance unit of the c-th channel. TIFF0007847889000003.tif6150 This is the echo energy in the v-th Doppler unit of the r-th distance unit of the c-th channel, TIFF0007847889000004.tif6150 This is the noise floor estimation result for the r-th distance unit of the c-th channel, TIFF0007847889000005.tif6160 These are preset parameters, TIFF0007847889000006.tif6150 That is the case. As an option, based on the RD spectrum, noise floor estimation is performed independently for each transmit / receive channel that performs a fixed false alarm probability processing independently within a preset range, and the upper limit of the preset range is determined based on the noise floor estimation result of the furthest distance unit. As an option, detecting a certain false alarm probability based on the currently acquired candidate target data and preset thresholds is achieved by the following formula: TIFF0007847889000007.tif851 JPEG0007847889000008.jpg1159 Here, TIFF0007847889000009.tif6150 The candidate target data for the v-th Doppler unit of the r-th distance unit of the acquired c-th channel is as follows: TIFF0007847889000010.tif6150 This is the total number of channels, TIFF0007847889000011.tif6150 This is the preset threshold, TIFF0007847889000012.tif6150 is the final target data for the v-th Doppler unit of the r-th distance unit. The options are to verify the targets in each preset region based on the occupancy rate of the detected targets within each preset region, or Based on the signal-to-noise ratio of the target that falls within each preset region, the target in each preset region is verified. As an option, during the processing of a certain false alarm probability, one or more operations from the following are performed on candidate target detection points: global maximum value filtering, first threshold filtering, and second threshold filtering to obtain the target detection result. Global maximum value filtering is used to filter whether the global maximum value is a target detection point. The first threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the noise floor estimate. The second threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the global maximum value. The overall maximum value filtering includes determining whether the difference between the digital beamforming spectral amplitude value corresponding to the current overall maximum value and the noise floor estimate is within a first preset range, and if it is within the first preset range, the candidate filtering result corresponding to the overall maximum value is the target detection point. The first threshold filtering includes determining whether the difference between the current candidate filtering result and the noise floor estimate is within a second preset range, and if it is within the second preset range, the current candidate filtering result is a target detection point. The second threshold filtering includes determining whether the difference between the power value of the current candidate filtering result and the maximum power value across the entire range is within a third preset range, and if it is within the third preset range, the current candidate filtering result is the target detection point.

[0006] As an alternative, for multi-channel application scenarios, achieving target detection in the region of interest based on the RD spectrum may include performing non-coherent integration on the RD spectrum of each channel, followed by constant false alarm probability processing based on noise estimation, wave direction estimation, target determination, positioning, and recognition operations.

[0007] As an alternative, the constant false alarm probability processing may be performed by applying DAE CFAR processing to the azimuth angle and elevation angle, respectively.

[0008] As an alternative, the detection of targets in the region of interest may further include performing target recognition operations based on extracting at least a portion of the accumulated data from multiple frames, the interframe Fourier transformed data, the non-coherent integrated data, and / or the data for detecting a constant false alarm probability.

[0009] As an alternative, for multi-channel application scenarios, achieving target detection in the region of interest based on the RD spectrum may include performing a certain false alarm probability processing on the RD spectrum of each channel, followed by binary aggregation processing of the channel domain, wave arrival direction estimation, and further target determination, positioning, and recognition operations.

[0010] As an option, detecting a target in a region of interest based on the RD spectrum includes performing two-dimensional digital beam synthesis on the RD spectrum, followed by target classification processing, thereby enabling target determination, positioning, and recognition operations.

[0011] As an option, performing deep learning target detection, deep learning target positioning, and / or deep learning target recognition on the 1D-FFT data may include acquiring target point cloud data based on the 1D-FFT data, and performing ML-based false alarm suppression processing, clustering processing, and / or ML-based target classification processing on the target point cloud data, thereby realizing target determination, positioning, and / or recognition operations in the region of interest.

[0012] As an option, the method may further include performing two-dimensional digital beam synthesis on the 1D-FFT data, then accumulating frame data, accumulating up to a predetermined amount of data, performing FFT processing between frames to obtain an RD spectrum, and performing target classification processing based on DL based on the RD spectrum, thereby realizing target determination, positioning, and recognition operations.

[0013] Embodiments of the present invention further provide a target detection method, which may include: performing distance-dimensional FFT processing based on an echo signal to acquire 1D-FFT data; performing two-dimensional digital beam synthesis on the 1D-FFT data, followed by frame data storage; and after storing up to a predetermined amount of data, performing target classification processing based on DL, thereby realizing target determination, positioning, and recognition operations.

[0014] Embodiments of the present invention further provide an integrated circuit comprising: a signal transmitting module configured for use with electromagnetic waves for target detection; a signal receiving module configured for use with receiving echoes formed by the reflection and / or scattering of the electromagnetic waves; and a processing module configured for use with performing signal and data processing on the echoes based on the method of any of the embodiments to achieve target detection.

[0015] As an option, the processing module may comprise a baseband unit and an MCU unit, wherein the baseband unit may be configured to perform distance-dimensional FFT processing and interframe FFT processing as described in any embodiment of the present application, and the MCU unit may be configured to perform frame data storage and target detection in a region of interest as described in any embodiment of the present application.

[0016] As an alternative, if the method includes digital beam synthesis and frame data storage, the baseband unit may be configured to be used to implement the digital beam synthesis, and the MCU unit may be configured to be used to implement the frame data storage.

[0017] As an option, the integrated circuit may be a millimeter-wave chip or a sensor chip.

[0018] Embodiments of the present application further provide an electromagnetic wave sensor which may comprise a carrier, an integrated circuit described in any embodiment of the present application installed on the carrier, and an antenna, wherein the antenna is installed on the carrier, or the antenna is integrated with the integrated circuit as an integrated device and installed on the carrier, and the integrated circuit is connected to the antenna and used to transmit the electromagnetic wave signal and / or receive the echo signal.

[0019] Embodiments of the present invention further provide a terminal device which may comprise a main body of the device and an electromagnetic wave sensor installed on the main body of the device as described in any embodiment, wherein the electromagnetic wave sensor is used for target detection and / or communication, thereby providing reference information for the operation of the main body of the device.

[0020] Embodiments of the present disclosure further provide a non-temporary computer-readable storage medium in which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the processor is instructed to perform the method described in any embodiment of the present application.

[0021] After reading and understanding the attached drawings and detailed explanations, other aspects can be understood.

[0022] The above and other objectives, features, and advantages of the present application will become clearer by referring to the description of embodiments of the present application in the following drawings. [Brief explanation of the drawing]

[0023] [Figure 1] This is a schematic flowchart illustrating how target detection is achieved based on SISO-combine in an embodiment of the present invention. [Figure 2] This is another schematic flowchart illustrating how target detection is achieved based on SISO-combine in an embodiment of the present invention. [Figure 3] This is a flowchart of the target detection method according to the embodiment of the present invention. [Figure 4] This is a schematic diagram of the radar operating cycle according to an embodiment of the present invention. [Figure 5] This is a schematic diagram of the waveform of the detection signal of an FMCW radar according to an embodiment of the present invention. [Figure 6] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 7] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 8] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 9] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 10] This is a schematic diagram of data readout related to the target detection method according to an embodiment of the present invention. [Figure 11] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 12] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 13] This is a schematic diagram of data transfer related to the target detection method according to an embodiment of the present invention. [Figure 14] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 15] This is a schematic diagram illustrating the division of a preset region related to the target detection method according to an embodiment of the present invention. [Figure 16] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 17] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 18] This is a schematic diagram illustrating the distribution of target points in a preset region related to the target detection method according to an embodiment of the present invention. [Figure 19] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 20] This is a schematic flowchart illustrating how target detection is achieved based on per-channel data in an embodiment of the present invention. [Figure 21] This is a flowchart of the target detection method according to the embodiment of the present invention. [Figure 22] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 23] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 24] This is a schematic flowchart of a target detection method combining the target detection method according to the embodiment of the present invention with a multi-frame federated processing scheme. [Figure 25] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 26] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 27] This is another flowchart of the target detection method according to the embodiment of the present invention. [Figure 28] This is a schematic diagram of the output related to the target detection method according to an embodiment of the present invention. [Figure 29] This is a schematic flowchart illustrating how target detection is achieved based on a combination of frame-level FFT and DAE CFAR in an embodiment of the present invention. [Figure 30]Figure 3 is a schematic diagram of the azimuth or elevation CFAR based on the target detection flow shown. [Figure 31] This is a flowchart of the target detection method according to the embodiment of this disclosure. [Figure 32] This is a schematic flowchart illustrating how target detection is achieved based on a combination of frame-level FFT and DAE CFAR in an embodiment of the present disclosure. [Figure 33] This is a schematic diagram of the result after noncoherent integration. [Figure 34] This is a partial figure of the noncoherent integration result. [Figure 35] This is a schematic diagram of the target points in the six regions in the embodiment of the present disclosure. [Figure 36A] This shows the processing results for a single-person scene (baby in corridor C). [Figure 36B] This shows the processing results for a single-person scene (baby in corridor C). [Figure 36C] This shows the processing results for a single-person scene (baby in corridor C). [Figure 36D] This shows the processing results for a single-person scene (baby in corridor C). [Figure 37A] This shows the processing results for a scene with two people (baby in seat B, adult in seat A). [Figure 37B] This shows the processing results for a scene with two people (baby in seat B, adult in seat A). [Figure 37C] This shows the processing results for a scene with two people (baby in seat B, adult in seat A). [Figure 37D] This shows the processing results for a scene with two people (baby in seat B, adult in seat A). [Figure 38] This is a schematic flowchart illustrating the post-processing performed on the target in an embodiment of the present invention. [Figure 39] This is a schematic flowchart illustrating the post-processing applied to the target in combination with a DL algorithm in an embodiment of the present invention. [Figure 40]This is a schematic flowchart of a target detection method combined with a DL algorithm in an embodiment of the present invention. [Figure 41] This is a schematic flowchart of another target detection method combined with a DL algorithm in an embodiment of the present invention. [Figure 42] This is a schematic flowchart of yet another target detection method combined with a DL algorithm in an embodiment of the present invention. [Modes for carrying out the invention]

[0024] To facilitate understanding of this application, the application will be described in more detail below with reference to the relevant drawings. The drawings show preferred embodiments of the application. However, the application is not limited to the embodiments described herein and can be realized in different forms. On the other hand, the purpose of providing these embodiments is to provide a more complete understanding of the content of the application.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. The terms used in this specification are for illustrative purposes only and are not intended to limit the application.

[0026] The embodiments of this disclosure provide a target detection method, including S1 to S3. S1 is obtained by accumulating frame data for 1D-FFT data, and the 1D-FFT data is acquired by performing distance-dimensional FFT processing on the echo signal. S2 performs inter-frame FFT processing based on the accumulated frame data to obtain the RD spectrum. S3, based on the RD spectrum, detection of the target in the region of interest is achieved.

[0027] For example, all of the 1DFFT data after the 1D-FFT processing may be inter-frame stored and then inter-frame FFT processing may be performed, or some of the 1D-FFT data may be inter-frame stored. For example, if the region of interest is smaller than the range of the echo data, distance-dimensional FFT processing may be performed on only some of the echo signals to obtain 1DFFT data, and then subsequent inter-frame storage and inter-frame FFT processing may be performed, or inter-frame storage and inter-frame FFT processing may be performed on some of the 1DFFT data.

[0028] For example, distance-dimensional FFT processing and inter-frame FFT processing may be performed simultaneously, or for instance, distance-dimensional FFT processing may be performed continuously while inter-frame FFT processing is being performed.

[0029] When target detection is performed in a sealed or relatively sealed spatial region, in several selectable embodiments, target detection may be achieved by performing region determination based on single-frame high and low-speed time processing and CFAR (Constant False-Alarm Rate) detection, angle measurement, or by employing an algorithm such as Capon after single-frame high-speed time processing to perform operations such as DoA (Wave Arrival Direction Estimation), thereby forming an RA (Range-Azimuth) diagram, and then extracting and recognizing / determining features such as preset region energy to achieve target detection, or by filtering the phase of the received echo and then detecting the respiratory heart rate of the human body. For example, when target detection is achieved by performing region determination based on single-frame high and low-speed time processing and CFAR detection, angle measurement, 1D-FFT (e.g., distance-dimensional FFT) processing on the echo signal, followed by temporal correlation analysis based on multiple echo data, may be performed to determine the human body target, and then the position may be determined by referring to point cloud information, ultimately achieving the objective of target detection within the cabin. For example, when target detection is achieved by performing region determination based on single-frame high / low speed time processing, CFAR detection, and point clouds acquired by angle measurement, 1D-FFT (e.g., distance-dimensional FFT) processing may be performed on the echo signal, followed by temporal correlation analysis based on multiple echo data to achieve human target identification. Subsequently, position determination may be performed by referring to point cloud information, etc., to ultimately achieve the objective of detecting targets inside the cabin.

[0030] In some selectable embodiments of the present invention, further alternative target detection and / or recognition schemes are proposed, namely, schemes that realize target detection in enclosed spaces such as cabins, rooms, and factory buildings by interframe storage, thereby effectively improving the detection rate, reducing the number of false alarm targets, and significantly improving the accuracy of angle estimation, etc., thereby enabling accurate detection of special or weak targets such as infants in a cabin, and realizing applications such as CPD (Child Presence Detection) and SBR (Safety Belt Reminder). In the implementation process, accurate detection of targets in the cabin may be achieved based on a combination of one or at least two schemes from among multi-frame federated processing technology, deep learning-based detection, positioning, and recognition technology, etc.

[0031] The multi-frame combined processing scheme may involve performing a sliding window FFT on the multi-frame data to obtain a distance-Doppler spectrum, or after FIR (Finite Impulse Response) or other complex time-frequency transformation processing, performing processing operations such as CFAR and DOA (Direction of Arrival) on the region of interest, and then, by referring to the set region determination logic and region parameters, detecting and positioning biological targets such as adults, children, pets, or other non-biological targets. The CFAR may be Doppler-dimensional NR-CFAR, RD-CFAR, or DAE (Doppler-Azimuth-Elevation)-CFAR, etc. Simultaneously, post-processing such as clustering, false alarm suppression, and multiple point cloud-related processing may be employed after CFAR and DoA. The region parameters may be determined by at least one or a combination of at least two operations, such as clustering point clouds detected after a sliding window multiframe processing operation, or detecting and removing outliers from point clouds detected after a sliding window multiframe processing operation.

[0032] When applied to a closed or relatively closed spatial region, the spatial region may be pre-divided to enable the definition and detection of different areas of interest (divided regions). For example, in the case of target detection inside a vehicle cabin, the area monitored by the radar may be divided into seating areas and aisle areas, and then appropriate parameter types and thresholds may be pre-set for different types of regions. These settings may then be combined with corresponding processing steps to achieve accurate detection of the target of interest in a specific region.

[0033] In several selectable embodiments, in the case of a family car, the interior space may generally be simply divided into head space, rear space, and trunk space. If the rear space is a priority monitoring area, the rear space may be further divided into seating sections and aisle sections, etc. Simultaneously, the seating sections may be divided into a corresponding number of seating section units based on the seats. Similarly, the aisle sections may be divided into a corresponding number of aisle section units corresponding to the above-mentioned seating section units. For example, in the case of a five-seat family car, the three seats in the rear space may be divided into three seating section units and three corresponding aisle section units. At the same time, corresponding parameter types and thresholds may be pre-set for different types of areas (or sections or section units), and adaptive signal data processing methods and steps may be employed to achieve accurate detection of targets of interest (specific targets) in areas such as specific sections or section units. Adjacent section units may have a partial overlapping area, or they may have adjacent or preset gaps.

[0034] When combining deep learning detection, positioning, and recognition strategies, the implementation may involve performing a 1D-FFT (e.g., distance-dimensional FFT) on the echo signal, followed by a multi-frame sliding window FFT, then a 2D-DBF (double-byte frame) processing of N points to form an N-channel RD (Range-Doppler) spectrum, and then using a deep learning model to perform monitoring, positioning, and recognition operations based on the N-channel RD spectrum; or performing a 1D-FFT on the echo signal, followed by a 2D-DBF processing of N points, and then using a deep learning model to perform monitoring, positioning, and recognition operations based on the N-channel RD spectrum; or first performing a 1D-FFT on the echo signal, followed by a 2D-DBF processing of N points, followed by a multi-frame sliding window FFT to form an N-channel RD spectrum, and then using a deep learning model to perform monitoring, positioning, and recognition operations based on the N-channel RD spectrum. The above N may also be the number of section units. For example, if a rear section with three rear seats is divided into three seat section units and three corresponding aisle section units, then N may be 6.

[0035] According to the above plan, after testing with actually collected data, it is possible to effectively achieve a relatively high detection rate and an extremely low detection failure rate and false alarm rate for application scenarios of the top-mount radar mounting method. For example, when detecting targets inside a cabin, it is possible to achieve a detection rate of over 99%, a detection failure rate of less than 0.5%, and a false alarm rate of less than 1%.

[0036] The present invention will be described in detail below with reference to the drawings, using a frame-level Fourier transform as an example, namely, the chirp in the echo signal is subjected to a fast time-dimension Fourier transform, then at least two frames of fast time-dimension Fourier transform data are accumulated, and then a frame-dimension Fourier transform is performed on the data of these at least two frames to obtain a range-Doppler (RD) spectrum, thereby performing an estimation operation of the target distance and / or velocity based on the RD spectrum.

[0037] Figure 1 is a schematic flowchart illustrating the implementation of target detection based on SISO-combine in an embodiment of the present invention. As shown in Figure 1, after performing operations such as ADC conversion (analog-to-digital conversion) and sampling on the echo signal, range DC removal and range Fourier transform (Range FFT, 1D-FFT, or fast time-dimension FFT) are performed sequentially. Then, multiple frames are stored (for example, 128 frames of data are stored), followed by Doppler DC removal, inter-frame Fourier transform (Frame FFT), and non-coherent integration. Furthermore, based on the noise obtained by noise estimation (for example, using a noise variance estimator), a constant false alarm probability detection (for example, CFAR (detection with peak selection) based on peak values) is performed. Finally, angle detection and target classification are performed to obtain target distance, velocity, and / or angle information. For example, azimuthal and elevation angles may be estimated first based on CFAR results, and this may be achieved using techniques such as DBF (digital beam forming) and DoA (direction of arrival estimation).

[0038] In several selectable embodiments, for an electromagnetic wave sensor having a BB (Baseband) unit and an MCU (Microcontroller Unit) unit, operations such as frame data storage (store 128 frames) and target classification (Region HIST) may be performed in the MCU module based on the target detection flowchart shown in Figure 1. On the other hand, range DC removal and Doppler DC removal may be performed in either the BB module or the MCU module. The flowchart shown in Figure 1 illustrates the example of performing DC filtering on downlink ADC data in the BB module.

[0039] As shown in Figure 1, after performing a 1D-FFT in the BB module to obtain distance dimension information, the 1D-FFT data may be cached in the MCU module. After caching up to a predetermined amount of data (e.g., 128, 56, or 32 frames), DC removal and frame-level FFT may be performed on the data cached in the predetermined number of frames, thereby obtaining the Range-Doppler (RD) spectrum.

[0040] If the system includes multichannels, non-coherent integration may be performed on the multichannel RD spectrum, as shown in Figure 1, and the noise floor of each distance unit (range bin) in the accumulated data may be estimated using the NVE module. For example, the noise floor may be estimated based on a pre-configured formula, and subsequent NR (Noise Reference)-CFAR processing may be performed based on the estimated noise floor value. The above pre-configured formula is n i =min(n i ,α*n g ) may also be n iThis is the noise floor estimate for the i-th range bin, and n g α is the noise floor estimate for the last range bin of interest, and α is a coefficient that may be set and updated based on demand or process data, experience, etc.

[0041] For angle estimation, as shown in Figure 1, DBF and DoA may be performed on the azimuthal dimension and elevation / depression dimension, respectively, for the CFAR detection point data, thereby obtaining estimated azimuthal and elevation / depression angle data for each CFAR detection point. Alternatively, target points detected in each frame (or a predetermined amount of data, i.e., each time) may be counted using a predetermined (or divided) region unit design, and the count result may be determined using a predetermined rule, thereby obtaining whether a physical target of interest (e.g., an adult, child, infant, or pet) exists in that processing time and determining its location region.

[0042] In the above embodiment, by employing frame-level FFT instead of conventional chirp-level Doppler FFT, the target velocity can be determined to achieve accurate detection of the target of interest. Furthermore, by performing sliding window FFT processing over multiple frames, the frequency of result updates can be further improved, thereby enhancing the real-time capabilities of the system. Additionally, the amount of data processed can be reduced by processing only the distance and / or Doppler region of interest. Moreover, when estimating background noise, if the background noise is obtained by taking the minimum value by combining the estimation of the last range bin of interest with the estimation of the current range bin, it can be more adapted to application scenes in relatively enclosed environments such as inside a car or cabin. When performing region logic determination, if the detected target points in each frame are counted according to a pre-set region design, and the count results are determined according to a pre-set rule, accurate determination of whether a target exists within each preset region can be achieved.

[0043] Figure 2 shows an example of digital signal processing achieved by another target detection method applied to the SISO-combine radar.

[0044] The radar receives an echo signal via an antenna, performs analog-to-digital conversion, sampling, and other processing to acquire a digital signal. This digital signal is then processed by a master control module (which can be implemented based on, for example, a Microcontroller Unit (MCU)) and a baseband (BB) module (which can be implemented based on, for example, a baseband chip) as shown in Figure 2 (using target detection of a Frequency Modulated Continuous Wave (FMCW) radar as an example) to achieve target detection. For example, in Figure 2, the digital signal undergoes sequential range DC removal and range Fourier transform (Range FFT, 1D-FFT, or fast time-dimension FFT) on the baseband module. The baseband module may then perform further processing on the data, for example, dagging using a complex averaging method on multiple chirp data in one frame, or it may not have to perform any further processing.Subsequently, the signal processed by the baseband module is sent to the master control module where multiple frames are accumulated. After the data is accumulated up to a preset number of frames, for example, 128 frames, the master control module moves the accumulated data from the CPU's static random-access memory (SRAM) to the baseband module. Then, based on the data from the baseband module, it performs Doppler DC removal, frame FFT, and inter-channel accumulation (non-coherent integration processing, channel accumulation, etc.). Furthermore, based on the noise obtained by noise estimation (for example, using a noise variance estimator) in the baseband module, it performs constant false alarm probability detection (for example, CFAR (detection with peak selection) based on peak values). Finally, it performs processing such as angle detection and target classification to obtain information such as the target's distance, velocity, and / or angle. When performing angle detection, azimuthal and elevation angles may be estimated first based on CFAR results, and this may be achieved, for example, by techniques such as digital beam forming (DBF) and wave arrival direction estimation.

[0045] The implementation of the digital signal processing shown in Figure 2 will be explained below.

[0046] Distance-dimensional DC component removal targets the data corresponding to each chirp in each frame data. In this case, distance-dimensional DC component removal involves averaging the data collected on each receiving (RX) channel of each chirp along the high-speed time dimension, and then subtracting the DC component from all sampling points of each RX channel.

[0047] Distance-dimensional FFT may also be implemented by performing a windowed operation on the data after removing the distance-dimensional DC component and then performing a 1D-FFT.

[0048] Furthermore, during the target detection process, each chirp data in each frame data acquired by the radar through processing such as ADC or sampling based on the echo signal is sent to the master control module after the above operations. Sending data to the master control module may also be achieved by moving data from direct memory access (DMA) to the CPU SRAM.

[0049] Doppler-dimensional DC component removal may be achieved by calculating a complex number average along the frame dimension for multi-frame data stored in the master control module (i.e., averaging data on the same distance unit (bin) of the same channel in different frames), obtaining the average value of different distances in different channels as the DC component of the corresponding distance unit in the corresponding channel, and then subtracting each DC component from each distance unit of each channel.

[0050] The inter-frame Fourier transform may also be performed by windowing the data after removing the DC component in Doppler dimension along the frame dimension, and then performing a 2D-FFT along the frame dimension, thereby obtaining distance-Doppler information for each transmit and receive channel.

[0051] Inter-channel storage may be implemented by the following equation, using non-coherent integration as an example. JPEG0007847889000013.jpg980 or, JPEG0007847889000014.jpg980 Here, TIFF0007847889000015.tif7150 are each TIFF0007847889000016.tif6150th distance unit and TIFF0007847889000017.tif6150 Power and amplitude in the 50th Doppler unit, TIFF0007847889000018.tif6150 is TIFF0007847889000019.tif6150th distance unit TIFF0007847889000020.tif6150th Doppler unit, TIFF0007847889000021.tif6150th transmission channel and TIFF0007847889000022.tif6150 is a complex value obtained by performing an inter-frame Fourier transform on the signal echo in the 50th receiving channel.

[0052] The detection of a constant false alarm probability based on the noise floor obtained through noise estimation has already been described in conventional CFAR technology and will not be explained in further detail here.

[0053] Angle detection may be achieved by performing azimuthal-dimensional digital beamforming (DBF) and DOA, and elevation-dimension-dimensional DBF and DOA, on target points acquired by CFAR. DBF and DOA have already been described in conventional DBF and DOA techniques and will not be explained in further detail here.

[0054] Target classification may be achieved by counting the target points detected in each frame (or a predetermined amount of data, i.e., each time) using a predetermined (or segmented) region unit design, and then determining the count result using predetermined rules, thereby obtaining whether a physical target of interest (e.g., an adult, child, infant, or pet) exists in that processing run and determining its location region.

[0055] Furthermore, for radar systems having a baseband module and a master control module, Figure 2 is only an example; that is, the master control module performs operations such as frame data storage and target classification (which can be achieved using Region HIST, classification, etc.), while the baseband module performs the remaining operations, such as distance-dimensional DC component removal and Doppler-dimensional DC component removal. In other embodiments, frame data storage may be performed in the baseband module, and Doppler-dimensional DC component removal may be performed in the master control module.

[0056] Although Figure 2 illustrates only the caching of 128 frames of data as an example, other embodiments may cache 56 frames of data or 32 frames of data, etc., and the embodiments of this application are not limited thereto, and may be determined based on demand, hardware support capabilities, etc. In this way, the cached multi-frame data is subsequently subjected to Doppler-dimensional DC component removal and inter-frame FFT, thereby obtaining a range-Doppler (RD) spectrum.

[0057] In the flow shown in Figure 2, some steps, such as distance-dimensional DC component removal or distance-dimensional DC component removal or windowing, may be skipped, some steps may be simplified, for example, noncoherent integration may be skipped and CFAR detection performed on only one of the channels, or noncoherent integration may be performed on only some channels and CFAR detection may be performed only on these channels.

[0058] To help those skilled in the art better understand the flow shown in Figure 2, the following explanation will illustrate the flow of different target detection methods.

[0059] In some embodiments, the target detection method flow includes the following steps 201-203, as shown in Figure 3.

[0060] In step 201, a distance-dimensional FFT is performed on the frame data.

[0061] In step 202, the target data is extracted from the frame data obtained by the distance-dimensional FFT process.

[0062] In step 203, windowing is performed on the target data based on a sliding window, and digital signal processing is performed on the data after windowing.

[0063] The length of the sliding window corresponds to a time length of the same magnitude as the period of the target periodic motion.

[0064] To enable those skilled in the art to better understand the target detection method described in the embodiment shown in Figure 3, the steps are described below.

[0065] In step 201, frame data refers to data corresponding to the echo signal formed when a detection signal for one frame is reflected by the target. To understand this, radar target detection operations typically have a fixed period, as shown in Figure 4. Within one detection period, the radar first transmits a detection signal for one frame. Taking detection by frequency-modulated continuous wave (FMCW) radar as an example, as shown in Figure 5, one frame of the detection signal consists of multiple chirp signals (as shown in Figure 5, there is no idle time between chirp signals, but this is just an example, and in some cases, there may be some idle time between chirp signals). Simultaneously with the start of transmission of the detection signal, the radar begins preparing to receive the echo signal that is reflected back by the target. The echo signal corresponding to one frame of the detection signal undergoes the aforementioned analog-to-digital conversion, distance-dimensional DC component removal, and other processing, resulting in the acquisition of the frame data described above. Furthermore, each frame data is subjected to distance-dimensional FFT as described above to realize step 201, which will not be explained in further detail here.

[0066] In step 202, the embodiment of this application does not impose any limitations on the target data. The target data may be any data content corresponding to the frame data. For example, if storage resources and computing power are sufficient, the extraction of target data may involve extracting all data. Alternatively, if the goal is to maximize the real-time nature of target detection, at least some data, or at least some pre-processed data (such as data obtained through pre-processing like a complex number average, a complex number weighted average, or a complex number sum), may be extracted. The target data may be extracted according to the demand and hardware conditions. To facilitate understanding, different implementation methods for extracting some data as target data will be described below.

[0067] In some embodiments, as shown in Figure 6, extracting target data from frame data obtained by distance-dimensional FFT processing is achieved by the following step 2021.

[0068] In step 2021, from each frame data obtained by distance-dimensional FFT processing, data corresponding to any one or a specified index of chirp (e.g., the second chirp of each frame) is extracted as target data. If it is decided to adopt data corresponding to a preset rule chirp period as target data, the rule for selecting chirp for each frame is the same.

[0069] In other words, by using data corresponding to one chirp period as frame data, the processing of target data becomes simpler and easier to implement, the selected data is more realistic, target detection can be performed more efficiently, and this is advantageous for improving the real-time capabilities of target detection.

[0070] In some embodiments, as shown in Figure 7, extracting target data from frame data obtained by distance-dimensional FFT processing may be further achieved by the following step 2022.

[0071] In step 2022, based on the frame data obtained by the distance-dimensional FFT process, the parameters corresponding to each distance unit in each frame data are determined, thereby generating the target data.

[0072] In other words, by compressing multiple data points for each distance unit in the frame data into corresponding parameters, the overall characteristics of the frame data can be preserved, thereby providing more comprehensive information to refer to when performing target detection later based on target data, and resulting in more accurate results.

[0073] Naturally, the above embodiments are merely examples provided to reduce the resources occupied by target data, and in some embodiments, target data extraction may be performed considering data effectiveness. For example, radar signals can usually cover a relatively wide range, but not all areas covered by radar are of interest, and processing data corresponding to these areas would lead to resource waste. Therefore, in some embodiments, as shown in Figure 8, extracting target data from frame data acquired by distance-dimensional FFT processing may be achieved by the following step 2023.

[0074] In step 2023, data within a preset distance range is extracted from the frame data obtained by distance-dimensional FFT processing as target data.

[0075] In other words, the range of interest in the frame data (represented by a pre-defined distance range) is used as the target data. This allows for accurate target detection information to be maintained, while reducing the storage and computing resources occupied by the target data, thus achieving a balance between the accuracy and real-time nature of target detection.

[0076] Furthermore, data within the scope of interest is retained in the manner shown in Figure 8. In some embodiments, as shown in Figure 9, windowing is performed on the target data based on a sliding window, and digital signal processing is performed on the data after windowing, which is achieved by step 2031 below.

[0077] In step 2031, windowing is performed on the target data based on a sliding window, and digital signal processing is performed on the data within the preset Doppler range after windowing.

[0078] This also achieves the same effects as the embodiment shown in Figure 8. However, the dimensions of the two data sets are different; one is the distance dimension, and the other is the Doppler dimension. Therefore, the method for selecting the range of interest is different. Naturally, it is also possible to extract target data using the distance dimension, as in the embodiment shown in Figure 7, and to extract data using the Doppler dimension, as in the embodiment shown in Figure 9, and then perform digital signal processing, etc., but this will not be explained in further detail here.

[0079] Of course, the above is merely an example of how to implement step 202, and in some embodiments it may be implemented in other ways, for example, by modulo calculation of distance units in pairs and using the larger of these distance units (similar to max-pooling) as the target data, which will not be explained in further detail here.

[0080] Step 203 does not limit the windowing method; for example, the window may be a sliding window, and the sliding window may be a fixed-length window, or a window of variable length, etc. Further details will not be explained here, and for the sake of easier understanding later, a fixed-length sliding window may be used as an example. However, this does not mean that only a fixed-length sliding window can be used; for example, a window of variable length that changes in response to the periodic motion of the target, such as changes in respiratory frequency, may also be used.

[0081] In some embodiments, as shown in Figure 10, if the window length of the sliding window is 128 frames for the 1st to 129th frame data generated after radar activation, the data acquired by the sliding window the first time is the 1st to 128th frame data (the shaded portion of the first row in Figure 10), and the data acquired the second time is the 2nd to 129th frame data (the shaded portion of the second row in Figure 9), meaning that the sliding step length of the sliding window is 1 frame.

[0082] Of course, the above are merely examples, and in other embodiments, the window length and sliding step length of the sliding window may be replaced with other parameters, which will not be listed here.

[0083] Furthermore, in step 203, we will not limit ourselves to data signal processing, but rather explain that different needs require different data to be acquired through processing. For example, in some embodiments, simply detecting the target point may be sufficient to achieve the effect of target detection. In other embodiments, as shown in Figure 2, it may be necessary to perform target classification to obtain more accurate, reliable, and referential targets. We will not list any further examples here. The following explanation will use the target detection flow shown in Figure 2 as an example, but this does not mean that target classification must always be performed after processing such as CFAR as shown in Figure 2.

[0084] As shown in FIG. 2, the digital signal processing process for target detection usually includes constant false alarm rate detection, and one typical constant false alarm rate detection is the constant false alarm rate detection based on the noise floor. As can be understood, at this time, the more accurate the estimation of the noise floor is, the better the constant false alarm rate detection effect is, and the more accurate the target detection result is. Therefore, in some embodiments, as shown in FIG. 11, performing window processing on the target data based on a sliding window and performing digital signal processing on the data after window processing may be realized by the following steps 2032 and step 2034.

[0085] In step 2032, window processing is performed on the target data based on a sliding window, and noise floor estimation is performed within a preset range on the data after window processing. The upper bound of the preset range is determined based on the noise floor estimation result of the distance unit with the farthest distance.

[0086] In step 2034, constant false alarm rate detection is performed based on the estimated noise floor.

[0087] That is, by setting the preset range, the noise floor estimation is restricted, and thereby an upper bound is constructed based on the noise floor estimation result of the distance unit with the farthest distance, which is beneficial to improving the accuracy of noise estimation.

[0088] In some examples, performing noise floor estimation within a preset range is realized by the following formula. n i ’=min(n i ,α*n g ) n i ’ is the noise floor estimation result within the preset range of the i-th distance unit, n i is the initial noise floor estimation result of the i-th distance unit, α is a preset parameter, TIFF0007847889000023.tif7150, and n gThis is the initial noise floor estimation result for the furthest distance unit.

[0089] Furthermore, the preset parameter α may be set and updated based on demand or process data, experience, etc.

[0090] Furthermore, the above is merely an illustrative explanation of noise floor estimation. In some embodiments, the initial noise floor estimation result of the i-th distance unit may be directly selected as the noise floor result used in the case of CFAR. Alternatively, the minimum, maximum, quantile, mean, median, etc., of the initial noise floor estimation results of the last few regions of interest may be selected as the noise floor estimation. Or, when estimating the noise floor, only some Doppler units of a certain distance unit may be selected instead of all Doppler units. We will not list any further noise floor estimation methods here.

[0091] Furthermore, although the radar shown in Figure 2 is an SISO-combine system, if the radar includes multiple channels, inter-channel accumulation, such as non-coherent integration or association accumulation, may be performed on the multi-channel RD spectrum, and the noise floor of each distance unit (range bin) in the accumulated data may be estimated using an NVE module. Any of the aforementioned noise floor estimation methods may be used, and this will not be explained in further detail here.

[0092] In step 203, during windowing processing of the data, the data has a certain order. If the order is disrupted, the temporal relationship of the data is destroyed, and the accumulation of motion over a certain time length is no longer reflected, which is detrimental to subsequent target detection. On the other hand, as shown in Figure 2, the data is stored in the master control module, and once a certain amount of data has been stored, the baseband module reads the data and performs further processing. In this process, the order in which the data is written and read may differ depending on the storage method, so in some embodiments, it is necessary to rearrange the data to ensure the correct order. For example, in some embodiments, it is assumed that the data is cached using a Ring FIFO, in which case the target detection method further includes step 204, as shown in Figure 12.

[0093] Step 204 reads data from the Ring FIFO and rearranges the read data.

[0094] This improves the efficiency of subsequent processing by rearranging the data.

[0095] To enable those skilled in the art to better understand the above data storage and rearrangement, the following explanation will be given with reference to Figure 13.

[0096] As shown in Figure 13, assuming that the data acquired after the baseband module (indicated as BB in the figure) performs a distance-dimensional FFT on the frame data, and that the data within the range of interest (for example, from the 5th distance unit to the 20th distance unit), belongs to the target data, then the master control module (indicated as MCU in the figure) needs to accumulate a two-dimensional matrix as shown in Figure 13, where the rows of the matrix correspond to the frame numbers and the columns correspond to the distance units.

[0097] Assuming that the master control module employs a Ring FIFO to store the above two-dimensional matrix, the storage process may be as follows.

[0098] Initialize one CPU SRAM, N f It has columns, and each column is N r ×N r x×N tx It can store numbers such that N f N is the maximum number of frames designed to perform the interframe FFT. r , N rx , N tx These represent the number of units of interest, the number of transmit (Tx) channels, and the number of receive (Rx) channels, respectively. The cache may be initialized to all 0s during initialization, and the FIFO head pointer (p_head) points to the beginning of the 0th column of the Ring-FIFO.

[0099] At frame 0, the corresponding data (for each distance unit, each Tx channel, and each Rx channel) is cached in column 0 of the Ring-FIFO, and then p_head points to column 1.

[0100] Similarly, for the k-th frame data (k=1, 2, ..., 126), the corresponding data is cached in the k-th column of the Ring-FIFO, and then p_head points to the (k+1)th column.

[0101] At the 127th frame, the corresponding data is cached in the 127th column of the Ring-FIFO. At this point, the data for 128 frames has been stored, so it is necessary to move the data in the Ring-FIFO to the baseband module and perform the aforementioned inter-frame FFT processing. In this case, the data from the column pointed to by p_head to the 127th column is moved first, and then the data from the 0th column to the column pointed to by p_head is moved. Clearly, at this time, the order in which data is read and the order in which data is written are different, and it is necessary to recover, that is, to rearrange the data. This causes the data to be distributed continuously in the frame dimension, thereby improving the computational efficiency of the baseband module.

[0102] Once the transfer is complete, the Ring-FIFO pointer p_head will again point to column 0. Similarly, in the new data storage and transfer process, the data for frame 128 is stored in column 0 of the Ring-FIFO. During the transfer, the data for frames 1-127 is first moved to the baseband module, then the data in column 0 is moved to the baseband module, and so on for subsequent frames, following the same rules, which will not be explained in further detail here.

[0103] In the example above, interframe processing is performed after the Ring-FIFO is fully filled with valid data. However, in other embodiments, some frame data, for example, frames 0-63, may be stored, and the data in the Ring-FIFO may be moved to the baseband module in a similar manner to the above to perform an interframe FFT, etc. In this case, since columns 64-127 are all initial values ​​given during initialization, the baseband module can perform an interframe FFT by adding zeros to the beginning of the data sequence. Of course, the above is merely an illustrative explanation and does not mean that it is always necessary to use 128 frames of data to perform an interframe FFT, or that it is always necessary to use and store the data in a Ring-FIFO. Further details will not be explained here.

[0104] Furthermore, as shown in Figure 2, if the user desires to obtain more intuitive and accurate results, in some embodiments, steps 205-206 are also performed after obtaining the target, as shown in Figure 14.

[0105] In step 205, the targets in each preset region are verified based on the targets that entered each preset region, and the preset regions are obtained by dividing the detection space.

[0106] In step 206, the detection results for each preset region are output based on the goals that passed the verification.

[0107] In other words, considering that multiple target points are usually detected for targets with a certain volume and occupying a certain space, the currently detected targets are used to further combine region segmentation and perform target verification, thereby improving the detection results. Based on the segmentation of the detection space in the application scene, the detection results for each preset region are output, so the distribution of targets within the preset regions is revealed, making it more intuitive and accurate, which is advantageous for the user to make decisions based on the output results, and resulting in a better user experience.

[0108] To help those skilled in the art better understand the embodiment shown in Figure 14, the following steps will be explained.

[0109] In step 205, the detection space and preset area are not limited and may vary depending on different application scenes and application needs. For example, in an automotive radar application, the detection space may be the interior of a vehicle, and in this case, the preset area may include at least one of the areas corresponding to seats and the area corresponding to aisles (footrests), thereby allowing the driver or adults to better perceive the situation inside the vehicle. Also, for example, in a factory work scene, the detection space may be the factory building, and in this case, the preset area may include each employee's workspace, thereby avoiding production risks such as the inability to monitor the operating status of machines because employees are not in their workspaces. This will not be explained in further detail here. In the following section, for ease of understanding, the above-mentioned interior of a vehicle will be used as an example, but this does not mean that the corresponding proposal can only be implemented inside a vehicle.

[0110] Taking the interior of a vehicle as the detection space as an example, the interior space is abstracted into a coordinate system as shown in Figure 15, where the horizontal coordinate of the coordinate system represents the azimuth angle relative to the radar installed in the vehicle, and the horizontal coordinate of the coordinate system represents the elevation angle relative to the radar installed in the vehicle. In this case, the interior of the vehicle is divided into a total of six preset areas, i.e., different filling areas as shown in Figure 15: the three seats in the rear row and the three aisles in front of the three seats. Depending on the needs, as shown in Figure 15, the different preset areas may overlap with each other or not, and a part of the interior may belong to multiple preset areas simultaneously or not belong to any area.

[0111] Figure 15 is merely one example of a method for abstracting the detection space. In some embodiments, the detection region may be abstracted from all or any one of the dimensions of distance, azimuth, and elevation / depression, or from all or any one of the dimensions along the xyz dimensions of the Cartesian coordinate system, and the preset region may be further divided. This will not be explained in further detail here.

[0112] In Step 205, the verification method is not limited; for example, verification may be performed based on the number of target points, occupancy rate, etc.

[0113] For example, in some embodiments, as shown in Figure 16, verifying the targets in each preset region based on the targets that have entered each preset region may be achieved by the following step 2051. Step 2051 verifies the targets in each preset region based on the occupancy rate of the detected targets within each preset region.

[0114] Furthermore, in some embodiments, for example, as shown in Figure 17, verifying the targets in each preset region based on the targets that have entered each preset region may be achieved by the following step 2052. In step 2052, the target for each preset region is verified based on the target signal-to-noise ratio within each preset region.

[0115] Of course, the above is merely an example, and in some cases, the target points at other locations are further examined based on the situation where the target point is located at a known target location such as the driver's seat, but this will not be explained in further detail here.

[0116] In step 206, the output method is not limited; point cloud output may be performed directly for the detection results of each preset region, or the divided preset regions may be represented and point clouds for each preset region may be output simultaneously. We will not list them all here.

[0117] To help those skilled in the art better understand the above embodiments, the following explanation will use occupancy rate verification as an example.

[0118] In one target detection process, a total of 11 target points are detected, and the division of the preset region is as shown in Figure 15. Simultaneously, the distribution of target points within the preset region is as shown in Figure 18, i.e., assuming that the region consists of seat A (6 target points), seat B (2 target points), seat C (0 target points), aisle A (0 target points), aisle B (1 target point), and aisle C (0 target points), the target points are represented by solid circles in Figure 18.

[0119] In this case, the algorithm obtained according to the above embodiment is as follows. The flag bit flag_region_i, which indicates whether there are people in each of the six regions, is initialized to 0, meaning that none of them are present (i=0, 1, ..., 5). If the total number of valid targets (total_valid_tgt_num) is 0, it is determined that there are no people inside the vehicle and the process ends; otherwise, the following operations are performed. Iterate through all regions, and for region i, if the statistical value tgt_num_region_i of the number of valid targets belonging to that region exceeds 25% of the total number of valid targets total_valid_tgt_num, set the flag bit flag_region_i for that region i to 1. If there is a region where flag_region_i is 1, flag_region_empty is set to 0, meaning that there is a person inside the vehicle. The system iterates through the seating areas, and if any seating area flag_region_i is 1, it determines that there are no people in the corresponding aisle area (for example, if it determines that there are people in seat A, it determines that there are no people in aisle A). Output the results.

[0120] At this time, as shown in Figure 18, the output shows six target points in seat A of the area, and there are no people in the remaining preset areas.

[0121] It should be noted that the 25% threshold in the above example is merely an example, and other thresholds or other criteria may be used in other examples, and further details will not be provided here.

[0122] Furthermore, as can be understood, the detection process may require the observation of a combination of different periodic motions, thereby allowing for more accurate target detection results. Based on this, in some embodiments, the target detection method includes the following steps 1801-1803, as shown in Figure 19.

[0123] In step 1801, a distance-dimensional FFT is performed on the frame data.

[0124] In step 1802, the target data is extracted from the frame data obtained by the distance-dimensional FFT process.

[0125] In step 1803, windowing is performed on the target data based on sliding windows corresponding to the periods of different biological characteristic parameters of the organism, and the windowed data corresponding to sliding windows of different lengths are then digitally processed.

[0126] The periods of the biocharacteristic parameters corresponding to different sliding windows are different, and the length of each sliding window has a time length of the same magnitude as the period of the biocharacteristic parameter.

[0127] Steps 1801 and 1802 are almost identical to steps 201 and 202 described above, but the difference lies in the subsequent steps 1803 and 203, where one performs detection based on multiple different periodic motions, while the other describes only one. As can be understood, when detection is based on multiple different periodic motions, data accumulation continues even after the accumulation of short-period data is complete, and processing of the data is completed when the accumulation of long-period data is finished. For example, assuming that one periodic motion corresponds to 32 frames and another periodic motion corresponds to 128 frames, and both need to be observed, the target detection process would begin outputting one detection result when the acquisition of the target data corresponding to the 32nd frame is complete, then output another new detection result after acquiring the target data corresponding to the 33rd frame, and so on... until the acquisition of the target data corresponding to the 128th frame is complete, at which point two detection results (one corresponding to the 32 frames and the other to the 128 frames), or a combined result of these two results, would be output.

[0128] In this way, by acquiring the periodicity of the biological characteristic parameters of a living organism, the length of the sliding window can be adjusted in a timely manner. This allows for better detection of biological targets, better monitoring of the organism's (e.g., children, pets, etc.) state, and timely processing corresponding to the organism's state.

[0129] Furthermore, the biometric parameters in the embodiments of this application are not limited. For example, the biometric parameters may include respiration. Also, for example, the biometric parameters may include heart rate and / or pulse rate. Of course, the above are merely examples, and other biometric parameters that can be embodied as motor characteristics may be used depending on the needs, and these will not be listed here one by one.

[0130] Thus, in the above embodiment, by employing frame-level FFT instead of conventional chirp-level Doppler FFT, the kinetic energy of the target is effectively stored to achieve accurate detection and measurement. Furthermore, by performing sliding window FFT processing with multiple frames, the frequency of result updates is further improved. Finally, by performing post-processing such as area statistics on target points detected by radar, it is possible to determine whether there are people in each area.

[0131] In the above embodiment, by employing a frame-level FFT instead of a conventional chirp-level Doppler-dimensional FFT, the target velocity can be determined to achieve accurate detection of the target of interest. Furthermore, by performing sliding window FFT processing over multiple frames, the frequency of result updates can be further improved, thereby enhancing the real-time capabilities of the system. Additionally, the amount of data processed can be reduced by processing only the distance of interest and / or the Doppler region. Moreover, when estimating background noise, if the background noise is obtained by taking the minimum value based on the estimation of the last distance unit of interest in combination with the estimation of the current distance unit (bin), it can be more adapted to application scenes in relatively enclosed environments such as inside a car or cabin. During region logic determination, by counting the detected target points in each frame using a pre-set region design and determining the count result using a pre-set rule, accurate determination of whether a target exists within each preset region can be achieved.

[0132] Figure 20 is a schematic flowchart illustrating how target detection is implemented per-channel in an embodiment of the present invention. As shown in Figure 20, compared to the approach of performing non-coherent integration on a multi-channel RD spectrum to obtain single-channel data and then performing CFAR based on the flow structure shown in Figure 1, this embodiment performs CFAR processing on each channel, followed by binary integration in the channel domain, and then estimates the horizontal / azimuth angle and elevation angle. For example, for a system with M channels, after performing the interframe FFT shown in Figure 1 on multiple channels, CFAR processing is performed on each of the multiple channels to obtain M bit masks, each element of which may be of type boolean, and it is determined whether a target exists in each distance unit (range bin) and / or Doppler unit (Doppler bin) (1 is defined as existing, and 0 is defined as not existing). Then, these M bit masks may be cumulatively added, and the range of each element may be defined as 0 to (M-1). Subsequently, secondary detection may be performed on the cumulatively added mask, that is, each element of the cumulatively added mask may be compared with a single pre-set threshold M0, if the value of the unit to be detected (e.g., an element of the mask after cumulative addition) x i >If M0, it outputs that a target exists in the detected unit. If the filename is TIFF0007847889000024.tif6150, it may output that no target exists for the detected unit. i M and M0 are positive integers.

[0133] The detection method flow shown in Figure 20 effectively reduces performance loss due to channel imbalances by performing CFAR processing on each channel independently, and also provides better robustness in actual applications.

[0134] Embodiments of the present disclosure further provide another target detection method, which includes the following steps 101-102, as shown in Figure 21.

[0135] In step 101, a constant false alarm probability processing is performed on at least two transmit / receive channels individually based on the distance-Doppler spectrum to obtain candidate target data for at least two transmit / receive channels.

[0136] In step 102, processing is performed based on candidate target data for at least two transmit / receive channels to obtain the final target data.

[0137] In this way, by performing a fixed false alarm probability detection on at least two transmit / receive channels independently, the results of the at least two transmit / receive channels do not interfere with each other and do not affect other channels. Therefore, even if an abnormal channel or abnormal data exists between the at least two transmit / receive channels, it does not affect the processing of other normal channels. Furthermore, processing is then performed based on the candidate target data of the at least two transmit / receive channels to obtain the final target data, thereby supporting detection on the two transmit / receive channels, ensuring the normal processing of the channel to which the abnormal channel or abnormal data belongs, and avoiding interference of the channel to which the abnormal channel or abnormal data belongs to other channels. Problems of imbalance between channels, delay differences due to mismatched wiring lengths from the transmit / receiver to the antenna and inaccurate compensation between channels, and phase differences are reduced, and detection accuracy is improved.

[0138] To help you better understand the target detection method described in the above embodiment, the following explains the steps involved.

[0139] In step 101, the number of transmit / receive channels that perform a fixed false alarm probability processing independently is not limited; for example, there may be two, three, five, or all transmit / receive channels.

[0140] For example, in some embodiments, a constant false alarm probability processing may be performed on each transmit / receive channel individually, thereby completely avoiding mutual interference between channels and improving accuracy. In this case, as shown in Figure 22, performing a constant false alarm probability processing on at least two transmit / receive channels based on the distance-Doppler spectrum is achieved by the following method: that is, step 201, performing a constant false alarm probability processing on each transmit / receive channel based on the distance-Doppler spectrum. Correspondingly, processing based on candidate target data for at least two transmit / receive channels is achieved by the following method: that is, step 202, performing processing on the candidate target data for each transmit / receive channel.

[0141] Of course, the above is merely an example, and in some embodiments, a certain false alarm probability detection may be performed independently for some transmit / receive channels, thereby avoiding adverse effects from channels with a relatively low signal-to-noise ratio or severe transmit / receive leakage (TRX leakage), which will not be explained in further detail here.

[0142] In step 101, the method for processing a constant false alarm probability for a single transmit / receive channel is not limited. For example, in some embodiments, it may be implemented based on existing constant false alarm probability processing, simply by changing the conventional data accumulation sum of multiple transmit / receive channels to data accumulation sum of a single transmit / receive channel.

[0143] In some embodiments, as shown in Figure 23, a constant false alarm probability processing for at least two transmit / receive channels based on the distance-Doppler spectrum may be achieved by the following steps 1011-1012.

[0144] In step 1011, noise floor estimation is performed independently for each transmit / receive channel that performs a constant false alarm probability processing based on the distance-Doppler spectrum, and the noise floor estimation result for a single transmit / receive channel is obtained.

[0145] In step 1012, based on the noise floor estimation result for a single transmit / receive channel, a constant false alarm probability detection is performed for the corresponding transmit / receive channel individually, and candidate target data corresponding to each transmit / receive channel is obtained.

[0146] In other words, when performing a constant false alarm probability processing for a single transmit / receive channel, noise floor estimation is performed on the single transmit / receive channel as a single independent whole, and noise floor-based CFAR is realized based on the noise floor obtained from each estimation.

[0147] In step 1011, the noise floor estimation method is not limited. For example, in some embodiments, estimating the noise floor independently for a transmit / receive channel that performs a constant false alarm probability processing independently based on the distance-Doppler spectrum may be achieved by the following method: estimating the noise floor independently for a transmit / receive channel that performs a constant false alarm probability processing independently within a preset range based on the distance-Doppler spectrum, and obtaining the noise floor estimation result for a single transmit / receive channel, where the upper limit of the preset range is determined based on the noise floor estimation result of the furthest distance unit. That is, the noise floor estimation is constrained by setting a preset range, thereby constructing an upper limit based on the noise floor estimation result of the furthest distance unit, which is advantageous for improving the accuracy of noise estimation.

[0148] In some examples, noise floor estimation for each transmit / receive channel can be achieved using the following formula. n i '=min(n i ,α*n g ) Here, n i ' represents the noise floor estimation result within the preset range of the i-th distance unit of a single transmit / receive channel, and n i is the initial noise floor estimate for the i-th distance unit in a single transmit / receive channel, determined based on the distance-Doppler spectrum, where α is a preset parameter. TIFF0007847889000025.tif7150, n g This is the initial noise floor estimation result for the furthest distance unit in a single transmit / receive channel.

[0149] The preset parameter α may be set and updated based on, for example, demand or process data, experience, etc.

[0150] It should be noted that the above is merely an illustrative explanation of noise floor estimation. In some embodiments, the initial noise floor estimation result of the i-th distance unit in a single transmit / receive channel may be directly selected and used as the noise floor result when performing CFAR processing on that transmit / receive channel. Alternatively, the minimum, maximum, quantile, mean, median, etc., of the initial noise floor estimation results of the last few regions of interest in a selectable single transmit / receive channel may be selected and used as the noise floor estimation. Furthermore, when estimating the noise floor, only some Doppler units of a certain distance unit may be selected instead of all Doppler units. No further details on noise floor estimation methods will be listed here.

[0151] In step 1012, the detection of a constant false alarm probability may be implemented by a threshold detection method. For example, the noise floor estimation result for a single transmit / receive channel consists of the noise floor estimation results for each distance unit in the single transmit / receive channel, and the constant false alarm probability detection result corresponding to a single transmit / receive channel consists of the constant false alarm probability detection results for each Doppler unit of each distance unit in the single transmit / receive channel.

[0152] Based on the noise floor estimation results for a single transmit / receive channel, detecting a fixed false alarm probability for the corresponding transmit / receive channel independently can be achieved using the following formula. JPEG0007847889000026.jpg1375 TIFF0007847889000027.tif6150 is the result of detecting a constant false alarm probability for the v-th Doppler unit of the r-th distance unit of the c-th channel. TIFF0007847889000028.tif6150 is the echo energy at the v-th Doppler unit of the r-th distance unit of the c-th channel. TIFF0007847889000029.tif6150 is the noise floor estimation result for the r-th distance unit of the c-th channel. TIFF0007847889000030.tif7160 is a preset parameter, The filename is TIFF0007847889000031.tif8150.

[0153] As can be seen from the above formula, If TIFF0007847889000032.tif6150 is 1, it indicates that the target is located at the v-th Doppler unit of the r-th distance unit. If TIFF0007847889000033.tif6150 is 0, it indicates that there is no target at the v-th Doppler unit of the r-th distance unit. In other words, by binarizing the detection results of a single transmit / receive channel, the detection process is not interfered with by specific numerical values ​​(especially extreme data), and the accuracy of the detection is improved.

[0154] The above formula is an example that only shows the process of obtaining candidate target data by performing a binarization process on the results. In some embodiments, the results may be quantified using quantification methods other than binarization, which can reduce interference from extreme data to some extent and do not necessarily have to be achieved by the binarization method, but these will not be listed here.

[0155] The above is merely an example of a constant false alarm probability processing for a single transmit / receive channel provided in combination with CFAR based on noise floor estimation. However, this does not mean that the constant false alarm probability processing for a single transmit / receive channel in this embodiment can only be implemented in this way. For example, it may be implemented in combination with CA (Cell Averaging)-CFAR, OS-CFAR, GO-CFAR, etc. In implementation, the processing of multiple transmit / receive channels in conventional algorithms can be considered as processing of a single transmit / receive channel, and this will not be explained in further detail here.

[0156] In step 102, processing candidate target data for at least two transmit / receive channels can actually be considered as performing target verification again based on candidate target data for the transmit / receive channels, after having performed target verification based on data for a single transmit / receive channel in step 101. This can improve the accuracy and reliability of the target. Therefore, step 102 can be implemented using the concept of detecting a constant false alarm probability.

[0157] For example, in some embodiments, as shown in Figure 23, processing the candidate target data for each transmit / receive channel to obtain the final target data is as follows: Step Therefore, it may be implemented.

[0158] In that step, currently Based on the acquired candidate target data and preset thresholds, a certain false alarm probability is detected, and the final target data is obtained.

[0159] In other words, following the conventional process of determining a threshold based on the results of a non-coherent integral, a similar thresholding process is applied to the candidate target data to obtain the final detection result. This approach is easy to implement, highly efficient, improves detection accuracy, offers good real-time capabilities, and enhances the user experience.

[0160] In this case, detecting a certain false alarm probability based on the currently acquired candidate target data and preset thresholds is achieved by the following formula. JPEG0007847889000034.jpg958 JPEG0007847889000035.jpg1470 TIFF0007847889000036.tif6150 is the result of detecting a constant false alarm probability for the v-th Doppler unit of the r-th distance unit of the c-th channel. This is the total number of channels in TIFF0007847889000037.tif6150. TIFF0007847889000038.tif6150 is a preset threshold, TIFF0007847889000039.tif6150 is the result of detecting a constant false alarm probability for the v-th Doppler unit of the r-th distance unit.

[0161] Specifically, candidate target data for the transmission and reception channels are cumulatively added together, and then it is determined whether the result of the cumulative addition exceeds a preset threshold to determine the final target data. This final target data indicates whether a target exists at the v-th Doppler unit of the r-th distance unit, thereby avoiding monitoring errors and detection omissions due to multipath or other factors, and improving the accuracy and reliability of target detection.

[0162] Of course, the above is merely an explanation of an example of obtaining the final target data using a threshold method. In some embodiments, if the detection result of a certain false alarm probability for the v-th Doppler unit of the r-th distance unit of the c-th channel is expressed as a numerical value, the acquisition of the final target data may be determined by determining how many channels the detection result of a certain false alarm probability for the v-th Doppler unit of the r-th distance unit exceeded a preset value. Alternatively, the above cumulative addition process may be replaced with processes such as weighted addition or averaging, and these will not be explained in further detail here.

[0163] Furthermore, in the above embodiment, only the target verification in the target detection process (i.e., obtaining results similar to those of conventional CFAR processing) is limited, and other data processing processes in the target detection process are not limited. To ensure clarity, the embodiments shown in Figures 21 to 23 may be implemented in combination with other data processing implementation methods, such as a multi-frame federated processing method.

[0164] To facilitate understanding of the target detection method and the multi-frame federated processing method described above, the multi-frame federated processing method will be explained first below.

[0165] As shown in Figure 2, a radar (using a frequency-modulated continuous wave (FMCW) radar as an example) receives an echo signal via an antenna, performs analog-to-digital conversion and sampling, and then acquires a digital signal. This digital signal is then processed by a master control module (which can be implemented based on, for example, a microcontroller unit (MCU)) and a baseband (BB) module (which can be implemented based on, for example, a baseband chip) as shown in Figure 2, to achieve target detection. For example, in Figure 2, the digital signal undergoes sequential range DC removal and range Fourier transform (Range FFT, 1D-FFT, or fast time-dimension FFT) on the baseband module. The signal processed by the baseband module is then sent to the master control module for storage of multiple frames. After the data is stored up to a preset number of frames (for example, 128 frames), the master control module moves the stored data from the CPU's static random-access memory (SRAM) to the baseband module. Subsequently, the baseband module performs Doppler DC removal, inter-frame Fourier transform (Frame FFT), and inter-channel storage (non-coherent integration, related storage, etc.). Furthermore, based on the noise obtained by noise estimation (e.g., using a noise variance estimator) in the baseband module, a constant false alarm probability detection (e.g., CFAR detection with peak value) is performed. The system performs selection, and finally, processes such as angle detection and target classification are carried out to obtain information such as the target's distance, speed, and / or angle.When performing angle detection, azimuthal and elevation angles may be estimated first based on CFAR results, and this may be achieved, for example, by techniques such as digital beam forming (DBF) and wave arrival direction estimation.

[0166] The implementation of the digital signal processing shown in Figure 2 is as described above.

[0167] Therefore, as can be seen from the above, by combining the above multi-frame federated processing scheme, that is, by employing multi-frame federated technology when acquiring the RD spectrum, or in other words, by replacing the non-coherent integration and CFAR processing in the above multi-frame federated technology scheme with the target detection method according to the embodiment shown in Figures 21 to 23, that is, by combining the method according to the embodiment with the flow shown in Figure 2, the target detection flow becomes as shown in Figure 24. In this case, after performing CFAR processing on each channel, binary integration in channel-domain processing is performed, and then the horizontal / azimuth angle and elevation / depression angle are estimated. For example, for a system with M channels, after performing the inter-frame FFT shown in Figure 2 on each channel, CFAR processing is performed on each channel to obtain M bit masks, each element of which may be of type bool, and then it is determined whether a target exists in each range bin and / or Doppler bin (1 is defined as existing, and 0 is defined as not existing). The M bit masks may then be cumulatively added, and the range of each element may be defined as 0 to (M-1). Subsequently, a secondary detection may be performed on the cumulatively added mask, that is, each element of the cumulatively added mask may be compared with a pre-set threshold M0, and if the value of the unit to be detected (e.g., an element of the mask after cumulative addition) x i >If M0, it may be output that a target exists in the detected unit (i.e., the final target data indicates that a target exists). If the filename is TIFF0007847889000040.tif6170, it may output that no target exists for the detected unit (i.e., the final target data indicates that no target exists). i , M and M0 are positive integers. In the detection method flow shown in Figure 24, by performing CFAR processing on each channel individually, performance loss due to channel imbalance can be effectively reduced, and it has better robustness in actual application.

[0168] To help those skilled in the art better understand the flow shown in Figure 24, flows using different target detection methods are described below.

[0169] In some embodiments, the target detection flow further includes the following steps 103-105, as shown in Figure 25.

[0170] In step 103, a distance-dimensional FFT is performed on the frame data.

[0171] In step 104, the target data is extracted from the frame data obtained by the distance-dimensional FFT process.

[0172] In step 105, a sliding window process is performed on the target data, and a 2D FFT is performed on the windowed data to obtain the distance-Doppler spectrum.

[0173] The length of the sliding window corresponds to a time length of the same magnitude as the period of the target periodic motion.

[0174] To enable those skilled in the art to better understand the target detection method described in the embodiment shown in Figure 25, the steps are described below.

[0175] In step 103, frame data refers to data corresponding to the echo signal formed when a detection signal for one frame is reflected by the target. To understand this, radar target detection operations typically have a fixed period, as shown in Figure 4. Within one detection period, the radar first transmits a detection signal for one frame. Taking detection by frequency-modulated continuous wave (FMCW) radar as an example, as shown in Figure 5, one frame of the detection signal consists of multiple chirp signals. Simultaneously with the start of transmission of the detection signal, the radar begins preparing to receive the echo signal that is reflected back by the target. The echo signal corresponding to one frame of the detection signal undergoes the aforementioned analog-to-digital conversion, distance-dimensional DC component removal, and other processing, resulting in the acquisition of the frame data described above. Furthermore, each frame data is subjected to a distance-dimensional FFT as described above to realize step 103, which will not be explained in further detail here.

[0176] In step 104, the embodiment of this application does not impose any limitations on the target data. The target data may be any data content corresponding to the frame data. For example, if storage resources and computing power are sufficient, the extraction of target data may involve extracting all data. Alternatively, if the aim is to maximize the real-time nature of target detection, some data may be extracted. The target data may be extracted according to the demand and hardware conditions. To facilitate understanding, different implementation methods for extracting some data as target data will be described below.

[0177] In some embodiments, extracting target data from frame data obtained by distance-dimensional FFT processing is achieved by extracting data corresponding to one chirp period from each frame data obtained by distance-dimensional FFT processing as target data.

[0178] In other words, by using data corresponding to one chirp period as frame data, the processing of target data becomes simpler and easier to implement, the selected data is more realistic, target detection can be performed more efficiently, and this is advantageous for improving the real-time capabilities of target detection.

[0179] In some embodiments, extracting target data from frame data obtained by distance-dimensional FFT processing may further be achieved by determining parameters corresponding to each distance unit in each frame data based on each frame data obtained by distance-dimensional FFT processing, thereby generating target data.

[0180] In other words, by compressing multiple data points for each distance unit in the frame data into corresponding parameters, the overall characteristics of the frame data can be preserved, thereby providing more comprehensive information to refer to when performing target detection later based on target data, and resulting in more accurate results.

[0181] Naturally, the above embodiments are merely examples provided to reduce the resources occupied by target data, and in some embodiments, target data extraction may be performed considering data effectiveness. For example, radar signals can usually cover a relatively wide range, but not all areas covered by radar are of interest, and processing data corresponding to these areas would lead to a waste of resources. Therefore, in some embodiments, extracting target data from frame data acquired by distance-dimensional FFT processing may be further achieved by the step of extracting data within a preset distance range from the frame data acquired by distance-dimensional FFT processing as target data.

[0182] In other words, the range of interest in the frame data (represented by a pre-defined distance range) is used as the target data. This allows for accurate target detection information to be maintained, while reducing the storage and computing resources occupied by the target data, thus achieving a balance between the accuracy and real-time nature of target detection.

[0183] Furthermore, while data within the region of interest can be retained using the method described above, in some embodiments, windowing may be performed on the target data based on a sliding window, and digital signal processing may be performed on the data within the preset Doppler range in the windowed data. This also achieves the effect of retaining data in the region of interest. However, the dimensions of the two sets of data are different; one is the distance dimension, and the other is the Doppler dimension. Therefore, the method of selecting the region of interest is different. Naturally, it is also possible to extract target data using the distance dimension while extracting data using the Doppler dimension and then perform digital signal processing, etc., but this will not be explained in further detail here.

[0184] Of course, the above is merely an example of how to implement step 104, and in some embodiments it may be implemented in other ways, for example, by modulo calculation of distance units in pairs and using the larger of these distance units (similar to max-pooling) as the target data, which will not be explained in further detail here.

[0185] Step 105 does not limit the type of sliding window; for example, the sliding window may be a fixed-length window, or even a window of variable length. We will not explain this in further detail here, but to make it easier to understand later, we may use a fixed-length sliding window as an example. However, this does not mean that only a fixed-length sliding window can be used. For example, a window of variable length that changes in response to the periodic movement of the target, such as changes in respiratory frequency, may also be used.

[0186] In some embodiments, as shown in Figure 10, if the window length of the sliding window is 128 frames for the 1st to 129th frame data generated after radar activation, the data acquired by the sliding window the first time is the 1st to 128th frame data (the data in the shaded area of ​​the first row in Figure 9), and the data acquired the second time is the 2nd to 129th frame data (the data in the shaded area of ​​the second row in Figure 10), meaning that the sliding step length of the sliding window is 1 frame.

[0187] Of course, the above are merely examples, and in other embodiments, the window length and sliding step length of the sliding window may be replaced with other parameters, which will not be listed here.

[0188] In step 105, when windowing the data, the data must have a certain order. If the order is disrupted, the temporal relationship of the data is destroyed, and the accumulation of motion over a certain time period is not reflected, which is detrimental to subsequent target detection. On the other hand, as shown in Figure 24, the data is stored in the master control module, and once a certain amount of data has been stored, the baseband module reads the data and performs further processing. In this process, the order in which the data is written and read may differ depending on the storage method, so in some embodiments, it is necessary to rearrange the data to ensure the correct order. For example, in some embodiments, it is assumed that the data is cached using a Ring FIFO, in which case the target detection method further includes step 106, as shown in Figure 26.

[0189] Step 106 reads data from the Ring FIFO and rearranges the read data.

[0190] This improves the efficiency of subsequent processing by rearranging the data.

[0191] To enable those skilled in the art to better understand the above data storage and rearrangement, the following explanation will be given with reference to Figure 13.

[0192] As shown in Figure 13, assuming that the data acquired after the baseband module (indicated as BB in the figure) performs a distance-dimensional FFT on the frame data, and that the data within the range of interest (for example, from the 5th distance unit to the 20th distance unit), belongs to the target data, then the master control module (indicated as MCU in the figure) needs to store a two-dimensional matrix as shown in Figure 13, where the rows of the matrix correspond to the frame numbers and the columns correspond to the distance units. The detailed storage process can be found in the previously mentioned content and will not be repeated here.

[0193] As will be understood in step 105, the detection process may require multi-frame federated detection to observe a combination of different periodic motions, thereby obtaining more accurate target detection results. Based on this, performing sliding window processing on target data and 2D FFT on the windowed data may be implemented by performing sliding window processing on target data based on the periods of different bio-characteristic parameters of the organism, and then digitally processing the windowed data corresponding to sliding windows of different lengths.

[0194] The periods of the biofeature parameters corresponding to different sliding windows are different, and the length of each sliding window is of the same magnitude as the period of the biofeature parameter. To understand this, when detection is based on multiple different periodic motions, data accumulation continues even after the accumulation of short-period data is complete, and processing of the data is completed when the accumulation of long-period data is complete. For example, assuming that one periodic motion corresponds to 32 frames and another periodic motion corresponds to 128 frames, and both need to be observed, the target detection process will start outputting one detection result from the point when the acquisition of target data corresponding to the 32nd frame data is complete, then output one new detection result after acquiring the target data corresponding to the 33rd frame data, and so on, until the acquisition of target data corresponding to the 128th frame data is complete, at which point two detection results (one corresponding to the 32 frame data and the other to the 128 frame data), or a result of combining the two results, will be output.

[0195] In this way, by acquiring the periodicity of the biological characteristic parameters of a living organism, the length of the sliding window can be adjusted in a timely manner. This allows for better detection of biological targets, better monitoring of the organism's (e.g., children, pets, etc.) state, and timely processing corresponding to the organism's state.

[0196] Furthermore, the biometric parameters in the embodiments of this application are not limited. For example, the biometric parameters may include respiration. Also, for example, the biometric parameters may include heart rate and / or pulse rate. Of course, the above are merely examples, and other biometric parameters that can be embodied as motor characteristics may be used depending on the needs, and these will not be listed here one by one.

[0197] Furthermore, if users desire more intuitive and accurate results, they may perform additional processing after acquiring target data, such as target classification or further validation.

[0198] Based on this, in some embodiments, after obtaining the target, the following steps 107-108 are performed, as shown in Figure 27.

[0199] In step 107, the targets in each preset region are verified based on the targets that entered each preset region, and the preset regions are obtained by dividing the detection space.

[0200] In step 108, the detection results for each preset area are output based on the goals that passed the verification.

[0201] In other words, considering that multiple target points are usually detected for targets with a certain volume and occupying a certain space, the currently detected targets are used to further combine region segmentation and perform target verification, thereby improving the detection results. Based on the segmentation of the detection space in the application scene, the detection results for each preset region are output, so the distribution of targets within the preset regions is revealed, making it more intuitive and accurate, which is advantageous for the user to make decisions based on the output results, and resulting in a better user experience.

[0202] To help those skilled in the art better understand the embodiment shown in Figure 27, the following steps will be explained.

[0203] In step 107, the detection space and preset area are not limited and may vary depending on different application scenes and application needs. For example, in an automotive radar application, the detection space may be the interior of a vehicle, and in this case, the preset area may include at least one of the areas corresponding to seats and the area corresponding to footrests, thereby allowing subjects such as drivers and adults to better perceive the situation inside the vehicle. Also, for example, in a factory work scene, the detection space may be the factory building, and in this case, the preset area may include each employee's workspace, thereby avoiding production risks such as being unable to monitor the operating status of machines because employees are not in their workspaces. This will not be explained in further detail here. In the following section, for ease of understanding, the above example of the interior of a vehicle will be used, but this does not mean that the corresponding proposal can only be implemented inside a vehicle.

[0204] Taking the interior of a vehicle as the detection space as an example, the interior space is abstracted into a coordinate system as shown in Figure 15, where the horizontal coordinate of the coordinate system represents the azimuth angle relative to the radar installed in the vehicle, and the horizontal coordinate of the coordinate system represents the elevation angle relative to the radar installed in the vehicle. In this case, the interior of the vehicle is divided into a total of six preset areas, i.e., different filling areas as shown in Figure 15: the three seats in the rear row and the three aisles in front of the three seats. Depending on the needs, as shown in Figure 15, the different preset areas may overlap with each other or not, and a part of the interior may belong to multiple preset areas simultaneously or not belong to any area.

[0205] Figure 15 is merely one example of a method for abstracting the detection space. In some embodiments, the detection region may be abstracted from all or any one of the dimensions of distance, azimuth, and elevation / depression, or from all or any one of the dimensions along the xyz dimensions of the Cartesian coordinate system, and the preset region may be further divided. This will not be explained in further detail here.

[0206] In Step 108, the verification method is not limited; for example, verification may be performed based on the number of target points, occupancy rate, etc.

[0207] For example, in some embodiments, verifying the targets in each preset region based on targets that have entered each preset region may be achieved by a step of verifying the targets in each preset region based on the occupancy rate of the detected targets of the targets that have entered each preset region.

[0208] Furthermore, in some embodiments, for example, verifying the targets in each preset region based on the targets that have entered each preset region may be achieved by a step of verifying the targets in each preset region based on the signal-to-noise ratio of the targets that have entered each preset region.

[0209] Of course, the above is merely an example, and in some cases, the target points at other locations are further examined based on the situation where the target point is located at a known target location such as the driver's seat, but this will not be explained in further detail here.

[0210] In step 108, the output method is not limited; point cloud output may be performed directly for the detection results of each preset region, or the divided preset regions may be represented and point clouds for each preset region may be output simultaneously. We will not list them all here.

[0211] To help those skilled in the art better understand the above embodiments, the following explanation will use occupancy rate verification as an example.

[0212] In one target detection process, a total of 11 target points are detected, and the division of the preset region is as shown in Figure 15. Simultaneously, the distribution of target points within the preset region is as shown in Figure 18, i.e., assuming that the region consists of seat A (6 target points), seat B (2 target points), seat C (0 target points), aisle A (0 target points), aisle B (1 target point), and aisle C (0 target points), the target points are represented by solid circles in Figure 18.

[0213] In this case, the algorithm obtained according to the above embodiment is as follows: That is, The flag bit flag_region_i, which indicates whether there are people in each of the six regions, is initialized to 0, meaning that none of them are present (i=0, 1, ..., 5). If the total number of valid targets (total_valid_tgt_num) is 0, it is determined that there are no people inside the vehicle and the process ends; otherwise, the following operations are performed. Iterate through all regions, and for region i, if the statistical value tgt_num_region_i of the number of valid targets belonging to that region exceeds 25% of the total number of valid targets total_valid_tgt_num, set the flag bit flag_region_i for that region i to 1. If there is a region where flag_region_i is 1, flag_region_empty is set to 0, meaning that there is a person inside the vehicle. The system iterates through the seating areas, and if any seating area flag_region_i is 1, it determines that there are no people in the corresponding aisle area (for example, if it determines that there are people in seat A, it determines that there are no people in aisle A). Output the results.

[0214] At this time, as shown in Figure 28, the output shows that only seat A in the area has six target points, and the remaining preset areas are empty.

[0215] It should be noted that the 25% threshold in the above example is merely an example, and other thresholds or other criteria may be used in other examples, and further details will not be provided here.

[0216] In this way, by performing a fixed false alarm probability processing on a single transmit / receive channel independently, interference from abnormal channels or abnormal data is avoided. Furthermore, by adopting frame-level FFT instead of conventional chirp-level Doppler FFT, the kinetic energy of the target is effectively stored, enabling accurate detection and measurement. Multi-frame sliding window FFT processing further improves the frequency of result updates. Post-processing such as area statistics is performed on target points detected by radar to determine whether there are people in each area. For example, by adopting frame-level FFT instead of conventional chirp-level Doppler-dimensional FFT, the target's velocity is determined, enabling accurate detection of the target of interest. Multi-frame sliding window FFT processing further improves the frequency of result updates, thereby improving the real-time performance of the system. In addition, the amount of data processed can be reduced by processing only the distance of interest and / or the Doppler region. Furthermore, when estimating background noise, if the background noise is obtained by taking the minimum value by combining the estimation of the last unit of interest distance based on the estimation of the current distance unit (bin), it can be more adapted to application scenes in relatively sealed environments such as inside a car or cabin. When performing region logic determination, the detected target points in each frame are counted according to a pre-configured region design, and the count results are then evaluated according to pre-configured rules. This allows for accurate determination of whether a target exists within each preset region.

[0217] Figure 29 is a schematic flowchart illustrating an embodiment of the present invention that realizes target detection based on a combination of frame-level FFT and DAE CFAR. As shown in Figure 29, on the flow structure shown in Figure 1, the azimuth angle and elevation angle are estimated by employing DAE CFAR (Constant False Alarm Rate) on the CFAR data based on frame-dimensional FFT (frame-FFT). For example, after performing Doppler-dimensional CFAR detection on inter-frame FFT data, azimuth-dimensional DBF (Azimuth DBF) is performed and combined with NVE (noise variance estimator) noise floor estimation, followed by azimuth CFAR processing (Azimuth CFAR). Similarly, elevation-dimensional DBF (Elevation DBF) is performed on the azimuth CFAR data and combined with NVE noise floor estimation, followed by elevation CFAR processing (Elevation CFAR), thereby obtaining target azimuth and elevation information.

[0218] Figure 30 is a schematic diagram of the azimuth or elevation CFAR based on the target detection flow shown in Figure 29. As shown in Figure 30, the azimuth / elevation energy data (power data) may be calculated using the HIST method by the NVE engine, and the noise floor of the azimuth / elevation DBF may be estimated by selecting a certain rank value (e.g., mean, median, etc.) from this data. When performing DBF, all peaks in the DBF spectrum may be acquired, and it may be decided whether to output each peak point based on the SNR of all peak points and the estimated noise floor.

[0219] For example, when performing DBF, obtain the amplitudes of all peak points in the DBF spectrum, that is, including the peaks of the global maximum (Global max) and the local maximum (Local max). For any peak, if its SNR exceeds a preset threshold, and at the same time the difference value between its amplitude value (Local max) and the global maximum amplitude value (Global max) is within a preset range, and the number of peaks already output is less than the preset number, then output this peak, that is, output it as the true target point. Compared with the solution of directly using the peak value in the Doppler spectrum to search for the target, there may be noise in the Doppler spectrum, and then the peak value cannot reflect the true target. Therefore, it is more accurate to search for the true target point through the above solution. Moreover, there may be multiple targets in the Doppler spectrum. Searching for the target only by the peak value will lead to missing target recognition. On the other hand, when performing target detection using this implementation method, the relationship between the candidate target and the global maximum amplitude value is considered, and accurate multi-target detection can be realized.

[0220] Embodiments of the present disclosure provide a target detection method. As shown in FIG. 31, it includes the following steps 10 to 20.

[0221] In step 10, after performing distance-dimensional FFT processing based on the echo signal, obtain 1D-FFT data, perform multi-frame association processing on the 1D-FFT data to obtain an RD spectrum. The RD spectrum is, for example, a range-doppler two-dimensional spectrum, and its spectrum includes range-doppler data, that is, distance-velocity two-dimensional data.

[0222] In step 20, realize the detection of the target in the region of interest based on the RD spectrum.

[0223] The detection of the target in the region of interest includes at least one of operations such as determination, positioning, and recognition.

[0224] The target detection method and related device according to the embodiments of the present disclosure can realize a target detection solution in a sealed space area such as inside a cabin, indoors, a factory building, etc. by means of multi-frame association processing technology, that is, by means of inter-frame accumulation method, thereby effectively improving the detection rate, reducing the number of false alarm targets, and greatly improving the accuracy of angle estimation, etc. Thereby, special targets or weak targets such as infants inside the cabin can be accurately detected, and applications such as CPD (Child Presence Detection), SBR (Safety Belt Reminder), etc. can be realized.

[0225] The embodiments of the present disclosure further provide a target detection method, including performing distance-dimensional FFT within a chirp and velocity-dimensional FFT between frames on an echo signal to obtain distance-velocity-dimensional data, performing first constant false alarm probability processing on the distance-velocity-dimensional data to obtain candidate target detection points, and performing second constant false alarm probability processing on the candidate target detection points, and performing target detection based on the second constant false alarm probability processing result.

[0226] In the embodiments of the present application, false alarms can be effectively removed by performing constant false alarm probability processing twice, and the detection and measurement performance can be effectively improved. Thereby, special targets or weak targets such as infants inside the cabin can be accurately detected, and applications such as CPD, SBR, etc. can be realized.

[0227] In an exemplary embodiment, the step of obtaining the distance-velocity-dimensional data includes performing distance-dimensional FFT processing within a chirp on an echo signal to obtain 1D-FFT data, performing frame data accumulation on the 1D-FFT data, and after accumulating to a preset data volume, performing velocity-dimensional FFT processing between frames to obtain an RD spectrum, that is, the above-mentioned distance-velocity-dimensional data. For example, the sliding window method is adopted to read 1D-FFT data of a preset number of frames each time and perform FFT processing between frames.

[0228] In the implementation process, accurate target detection within the cabin may be achieved based on multi-frame federated processing technology. The multi-frame federated processing scheme may involve performing a sliding window FFT on the multi-frame data to obtain a distance-Doppler spectrum, or after FIR (Finite Impulse Response) or other complex time-frequency transformation processing, performing processing operations such as CFAR and DOA on the region of interest, and then, by referring to the set region determination logic and region parameters, achieving detection and positioning operations for biological targets such as adults, children, pets, or other non-biological targets. The CFAR may be Doppler-dimensional NR-CFAR, RD-CFAR, or DAE (Doppler-Azimuth-Elevation)-CFAR, etc. Post-processing such as clustering, false alarm suppression, and multiple point cloud-related processing may be employed after CFAR and DoA. The region parameters may be determined by at least one or a combination of at least two operations, such as clustering point clouds detected after a sliding window multiframe processing operation, or detecting and removing outliers from point clouds detected after a sliding window multiframe processing operation.

[0229] In an exemplary embodiment, the step of performing the first constant false alarm probability processing includes performing a noncoherent integration on a multi-channel RD spectrum, and performing a first constant false alarm probability processing based on noise estimation based on the noncoherent integration results to obtain candidate target detection points.

[0230] As an option, the first constant false alarm probability processing includes performing noncoherent integration on a multi-channel RD spectrum, estimating the noise floor of each distance unit after noncoherent integration, obtaining the estimated noise floor value for each distance unit, and performing noncoherent constant false alarm probability processing using the estimated noise floor value, wherein when estimating the noise floor of each distance unit after noncoherent integration, the global noise floor is used to adjust the noise floor of each distance unit.

[0231] One selectable adjustment method is for the noise floor of each distance unit, This includes performing adjustments using the format TIFF0007847889000041.tif6150, where, TIFF0007847889000042.tif6150 is TIFF0007847889000043.tif6150th distance unit is the original noise floor estimate. TIFF0007847889000044.tif6150 is the average of the noise floor estimates for multiple distance units. TIFF0007847889000045.tif6150 is TIFF0007847889000046.tif6150 is the adjusted noise floor estimate for the 50th distance unit. The filename is TIFF0007847889000047.tif7150. By adjusting the noise floor, it is possible to avoid the target not being detected due to an excessively high noise floor.

[0232] In an exemplary embodiment, the second constant false alarm probability processing includes performing DAE (Doppler-Azimuth-Elevation)CFAR processing for the azimuth angle and elevation angle, respectively.

[0233] In an exemplary embodiment, the step of performing the second constant false alarm probability processing includes performing digital beamforming on candidate target detection points, and then performing a second constant false alarm probability processing based on noise estimation to obtain a selection result that filters out false target points.

[0234] One feasible method is to perform azimuth-dimension DBF on the candidate target detection point, perform constant azimuth false alarm probability processing by referring to the noise floor estimation result, obtain the target selection result for the azimuth dimension, and / or perform elevation-depression-dimension DBF on the candidate target detection point, perform constant elevation-depression false alarm probability processing by referring to the noise floor estimation result, and obtain the target selection result for the elevation-depression dimension.

[0235] Another feasible approach involves performing a two-dimensional DBF (Digital Beam Filtering) on ​​the candidate target detection points using azimuth and elevation, performing a constant false alarm probability processing on the azimuth and elevation angles based on the noise floor estimation results, and obtaining a two-dimensional target selection result using azimuth and elevation.

[0236] The above noise floor estimation results may also be obtained by statistically analyzing the DBF spectrum of the current dimension and selecting quantiles (e.g., median or mean) as the noise floor estimation results for the current dimension.

[0237] For example, the above-mentioned constant azimuth angle false alarm probability processing, constant elevation / depression angle azimuth angle false alarm probability processing, or constant azimuth angle / elevation / depression angle false alarm probability processing may all be performed using the following methods. During the processing of a constant false alarm probability, one or more operations from the following are performed on the candidate target detection points: global maximum value filtering, first threshold filtering, and second threshold filtering to obtain the target detection result. Global maximum value filtering is used to filter whether the global maximum value is a target detection point. The first threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the noise floor estimate, and is used to filter out false targets. The second threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the global maximum value, and is used to realize multi-target detection.

[0238] The above-mentioned full-range maximum value filtering includes determining whether the difference between the digital beamforming spectral amplitude value corresponding to the current full-range maximum value and the noise floor estimate is within a first preset range, and if it is within the first preset range, the candidate filtering result corresponding to the full-range maximum value is the target detection point; the first threshold filtering includes determining whether the difference between the current candidate filtering result and the noise floor estimate is within a second preset range, and if it is within the second preset range, the current candidate filtering result is the target detection point; the second threshold filtering includes determining whether the difference between the power value of the current candidate filtering result and the power of the full-range maximum value is within a third preset range, and if it is within the third preset range, the current candidate filtering result is the target detection point.

[0239] In an exemplary embodiment, when detecting a target in a region of interest based on the RD spectrum, the region of interest is pre-divided into multiple target sub-regions. After target detection, the number and position of target points in each sub-region are determined based on the target detection point positions acquired. It is then determined whether there is a target to be detected in the current sub-region based on whether the ratio of the number of target points in each sub-region to the total number of targets is greater than a pre-set proportional value.

[0240] When applied to a closed space region or a relatively closed space region, by pre-dividing the space region, it is also possible to realize the definition and detection for different target regions (divided regions). For example, in the case of target detection inside a vehicle cabin, the region monitored by the radar can be divided into a seat section, a passage section, etc. Then, corresponding parameter types, thresholds, etc. are preset for different types of regions, and combined with corresponding processing steps, accurate detection of the target of interest in a specific region may be realized.

[0241] In some selectable embodiments, in the case of a household vehicle, generally, the space region inside the vehicle may be simply divided into a headspace, a rear space, a trunk space, etc. If the rear space is the key monitoring region, the rear space may be further divided into a seat section, a passage section, etc. The seat section may be divided into a corresponding number of seat section units based on the seats. Similarly, the passage section may also be divided into a corresponding number of passage section units corresponding to the above seat section units. For example, in the case of a 5-seat household car, for the three seats in the rear space, it may be divided corresponding to three seat section units and three corresponding passage section units. For different types of regions (or sections, section units), corresponding parameter types, thresholds, etc. may be preset, and an adaptive signal data processing method and steps may be adopted, thereby realizing accurate detection of the target of interest (specific target) in a region such as a specific section or section unit. Adjacent section units may have a partial overlap region, or may have a gap with an adjacent or preset width.

[0242] The embodiments of the present disclosure further provide a target detection method applied to target detection in a specific target region. The method includes performing a first constant false alarm probability processing on distance-Doppler data, and then further performing a second constant false alarm probability processing in the angle dimension to obtain target data.

[0243] In the above embodiment, by employing frame-level FFT instead of conventional chirp-level Doppler FFT, the target velocity can be determined to achieve accurate detection of the target of interest. Furthermore, by performing sliding window FFT processing over multiple frames, the frequency of result updates can be further improved, thereby enhancing the real-time capabilities of the system. Additionally, the amount of data processed can be reduced by processing only the distance and / or Doppler region of interest. Moreover, when estimating background noise, if the background noise is obtained by taking the minimum value by combining the estimation of the last range bin of interest with the estimation of the current range bin, it can be more adapted to application scenes in relatively enclosed environments such as inside a car or cabin. When performing region logic determination, if the detected target points in each frame are counted according to a pre-set region design, and the count results are determined according to a pre-set rule, accurate determination of whether a target exists within each preset region can be achieved. The purpose detection method of this disclosure will be described in detail below using one application example, and the processing process may be as shown in Figure 32.

[0244] In step 1, if the target detection system detects a target within its detectable area, the transmitting antenna in the target detection system may transmit a frequency-modulated continuous wave (FMCW) signal containing several chirps, which, after refraction and / or reflection by the target, forms an echo signal that can be received by the receiving antenna in the target detection system.

[0245] The target detection system may be an antenna array system or radar system having at least one transmit / receive channel, such as a Multiple Input Multiple Output (MIMO) radar system or a Single-Input Multi-Output (SIMO) system. The target detection system typically has N transmit antennas (TX) and M receive antennas (RX), where N and M are both positive integers of 1 or more, and a virtual antenna array of N × M transmit / receive channels can be formed by rationally arranging the positions of each antenna.

[0246] In step 2, distance-dimensional FFT (1D-FFT) processing is performed on each chirp data of the echo signal for each received frame to obtain distance-dimensional data, i.e., 1D-FFT data.

[0247] Before performing 1D-FFT processing, one option is to first perform analog-to-digital conversion (ADC) on the chirp data, and then perform a first DC filtering process on the acquired ADC data. The first DC filtering process involves calculating the average value of the data collected on each RX channel of each chirp along the fast time dimension and subtracting its DC component from all sampling points of each RX channel. The fast time dimension refers to the data sequence formed by sampling the radar echo signal in the distance (or angle) direction. DC filtering removes the DC offset in the signal, making the signal's centerline zero.

[0248] Alternatively, before performing the 1D-FFT process, the DC-filtered data may be first subjected to windowing, and then the 1D-FFT process may be performed to obtain distance dimension information. Windowing can reduce spectral leakage and improve the performance of spectral estimation. The windowing operation involves multiplying the original signal by a window function, which may be a weight function with a specific shape, such as 64, 96, or 128. Common window functions include the rectangular window, Hamming window, Hanning window, and Blackman window.

[0249] 1D-FFT data may be moved from the broadband (BB) to the SRAM in the CPU or MPU via DMA and cached.

[0250] In step 3, a distance-dimensional FFT (2D-FFT) is performed on the 1D-FFT data of the preset number of frames to obtain distance-dimension minus velocity-dimension data, i.e., 2D-FFT data (or referred to as the RD spectrum).

[0251] The velocity dimension may also be called the Doppler dimension, and the distance dimension-velocity dimension information described above may also be called the distance dimension-Doppler information.

[0252] Unlike the slow inter-chirp time processing of radar signal processing in related technologies, this step improves the frequency of result updates by accumulating multiple frame data and performing inter-frame 2D-FFT processing on a preset number of 1D-FFT data using a sliding window method each time, thereby improving the real-time performance of the system. The 2D-FFT processing in this step may also be called inter-frame FFT processing. As an example, with a preset number of frames of 128 and a sliding window step of 1, the 2D-FFT processing is performed on frames 0-127 in the first iteration, and on frames 1-128 in the second iteration. Here, the sliding window step is explained as an example of 1, but the sliding window step may be set to a positive integer of 1 or more, and the preset number of frames may also be set.

[0253] Before performing 2D-FFT processing, an option is to first perform a second DC filtering process on the 1D-FFT data. This second DC filtering process involves calculating a complex numerical average along the frame dimension for a preset number of 1D-FFT data, obtaining the average value for different channels and different distances as its DC component, and then subtracting this DC component from each channel and each distance unit (range bin). A distance unit refers to a small distance interval divided along the radial direction (i.e., the direction in which signals are transmitted and echoes are received) by the radar system. Distance units are used to represent discrete distance intervals used by the radar system when detecting and tracking targets.

[0254] Alternatively, before performing the 2D-FFT processing, one may first perform windowing on the data after the second DC filtering, and then perform 2D-FFT processing along the frame dimension to obtain distance-velocity dimension data. Windowing can reduce spectral leakage and improve the performance of spectral estimation. The windowing operation involves multiplying the original signal by a window function, which may be a weight function with a specific shape, such as 64, 96, or 128.

[0255] For each chirp data in each frame, when the data cached in SRAM reaches a certain number of frames, for example 128 frames, the data for those 128 frames may be moved from SRAM to BB by DMA, and then the following steps may be performed.

[0256] In step 4, non-coherent integration (NCI) is performed on the multi-channel 2D-FFT data to acquire the stored data.

[0257] You may perform noncoherent integration by adopting one of the following formulas. JPEG0007847889000048.jpg3777 TIFF0007847889000049.tif6150 is TIFF0007847889000050.tif6150th distance unit and TIFF0007847889000051.tif6150 is the power in the 50th Doppler unit, TIFF0007847889000052.tif6150 is TIFF0007847889000053.tif6150 The second distance unit and TIFF0007847889000054.tif6170 is the amplitude in the 170th Doppler unit, TIFF0007847889000055.tif6150 is TIFF0007847889000056.tif6150th distance unit TIFF0007847889000057.tif6150th Doppler unit, TIFF0007847889000058.tif6150th transmission channel and The echo signal in the 150th received channel of TIFF0007847889000059.tif6 is the complex value after undergoing the above 2D-FFT processing.

[0258] After noncoherent integration, the result shown in Figure 33 is obtained, where the horizontal coordinate represents the Doppler, i.e., velocity, and the vertical coordinate represents distance, and the 0th distance unit in the figure includes all Doppler values ​​where the vertical coordinate is 0.

[0259] In step 5, the noise floor for each distance unit in the accumulated data, i.e., the background noise, is estimated.

[0260] In this example, The noise floor estimation method TIFF0007847889000060.tif6150 was adopted. TIFF0007847889000061.tif6150 is TIFF0007847889000062.tif6150th distance unit is the original noise floor estimate. TIFF0007847889000063.tif6150 is the average of the noise floor estimates for multiple distance units, or the noise floor estimate for the last distance unit of interest. TIFF0007847889000064.tif6150 is TIFF0007847889000065.tif6150 is the adjusted noise floor estimate for the 50th distance unit. The filename is TIFF0007847889000066.tif6160. As shown in Figure 34, each cell in the figure represents the original noise floor estimate for one distance unit. For example, the median or mean value of each distance unit may be calculated to obtain the original noise floor estimate. For noise floor estimates of multiple distance units, for example, the median of the multiple original noise floor estimates shown in the dashed box (1) in the figure may be calculated. These noise floor estimates of multiple distance units can be used as a global noise floor estimate to help adjust the noise floor (or is called noise floor saturation). When a target (e.g., an adult) moves significantly, the overall Doppler spectral value increases, which increases the original noise floor estimate and can lead to the target not being detected during subsequent target selection. However, by adjusting the noise floor with the global noise floor estimate, subsequent target detection can be made more accurate. The number and position of distance units used in calculating the global noise floor estimate are both configurable. For example, the noise floor estimates of multiple distance units shown in the dashed box (2) in the figure may be used as the global noise floor estimate.

[0261] In step 6, NR-CFAR (Non-Coherent Constant False Alarm Rate) is performed to obtain candidate target detection points.

[0262] In Step 7, the first target selection is performed on the candidate target detection points.

[0263] Each candidate target detection point The following operations are performed on TIFF0007847889000067.tif6150. Specifically, 2D-FFT data corresponding to the detection point is extracted, an azimuthal dimension antenna is selected, azimuthal dimension DBF (Digital Beam Synthesis) is performed based on the azimuthal dimension guide vector, and the DBF spectrum is obtained. The DBF spectrum shows the situation in which the signal power or intensity in different directions changes with frequency. The noise floor is estimated for the DBF spectrum, azimuthal dimension CFAR (may also be called Az-CFAR (Azimuth CFAR)) is performed, the results of the azimuthal dimension CFAR are judged according to pre-set conditions, and candidate target points that match the pre-set conditions are formed as the first target selection result to form the first candidate target point set, which includes all candidate target points that satisfy the pre-set conditions. TIFF0007847889000068.tif6150 and its corresponding azimuth This includes TIFF0007847889000069.tif6150.

[0264] The azimuthal DBF spectrum is statistically analyzed to obtain the azimuthal-dimensional CFAR result (similar to the elevation-depression-dimensional CFAR) shown in Figure 3. In the figure, the x-axis represents the azimuthal direction, and the y-axis represents the amplitude of the DBF spectrum. If only the DBF amplitude peak is used as the target point, both the global max and local max in the figure will be considered target points, which could lead to false alarms. Therefore, to suppress false targets, the following operations are performed.

[0265] The azimuthal DBF spectrum is statistically analyzed, and a certain quantile (e.g., median or mean) is selected to estimate the noise floor of the azimuthal DBF, i.e., Az_HIST as shown in the figure. The file is named TIFF0007847889000070.tif6150, and the global max value in the DBF spectrum is found to be... It is written as TIFF0007847889000071.tif6150, if TIFF0007847889000072.tif6150, If it is determined to be TIFF0007847889000073.tif6150, the candidate target point corresponding to the maximum value across the entire DBF spectrum is determined to be a false target point, and processing for that target point is terminated. TIFF0007847889000074.tif6150 is one of the threshold values ​​for the azimuthal dimension CFAR, and the above formula can be used to determine whether the maximum value across the entire area is a target detection point, if If the filename is TIFF0007847889000075.tif6150, perform the following steps. The system iterates through all points in the DBF spectrum, and if any point satisfies the following conditions, it is determined to be a candidate target point that meets the pre-defined criteria. In the JPEG0007847889000076.jpg28160 formula, TIFF0007847889000077.tif7160 is the number of the candidate target point in the azimuthal dimension, The filename is TIFF0007847889000078.tif6150. TIFF0007847889000079.tif6150 is the number of points in the azimuthal dimension DBF spectrum, TIFF0007847889000080.tif6150 is the power of the candidate target point, TIFF0007847889000081.tif6150 is a configurable parameter, TIFF0007847889000082.tif6150, The filename is TIFF0007847889000083.tif6150. Equation 1 above shows that the ratio of the power to the noise floor of the candidate target point is a preset threshold. This indicates that it must be greater than or equal to TIFF0007847889000084.tif6150, which corresponds to setting a first limit, which is the noise floor estimate + first threshold, and by setting the first limit, filtering of false target points can be achieved, and Equation 2 shows that the ratio of the power of the candidate target point to the maximum power across the range is the preset threshold This indicates that it must be greater than or equal to TIFF0007847889000085.tif7160, which corresponds to setting a second limit, where the total maximum power value minus the second threshold, and setting a second limit enables multi-target detection, which is advantageous for recognizing targets such as children and babies. If the value is TIFF0007847889000086.tif6150, then equations 3 and 4 above indicate that the point is the local maximum.

[0266] In Step 8, we will perform a second round of target selection.

[0267] Each candidate's target points The azimuth angle for TIFF0007847889000087.tif6150 Based on TIFF0007847889000088.tif6150 and the array arrangement, their respective elevation and depression guide vectors. Generate TIFF0007847889000089.tif6150 and each candidate target point For TIFF0007847889000090.tif6150, a DBF (Depth-Level Filter) is performed to obtain the DBF spectrum. The noise floor is then estimated from this DBF spectrum. A second CFAR (referred to as El-CFAR) is then performed, and the results of the CFAR are evaluated based on pre-defined conditions. Candidate target points that meet the pre-defined conditions are selected as the second target selection result to form a second set of candidate target points, which includes all candidate target points that satisfy the conditions. TIFF0007847889000091.tif6150 and its corresponding elevation / depression angles This includes TIFF0007847889000092.tif6150. The second target selection process is similar to the first target selection process. First, the elevation-to-depression DBF is statistically analyzed to obtain the elevation-to-depression CFAR result, and this result is evaluated using the formula. TIFF0007847889000093.tif6150 is used to determine if the maximum value across the entire area is the target detection point, and if If the data type is TIFF0007847889000094.tif6150, the system iterates through all points in the DBF spectrum, and if any point satisfies equations 1-4 above, it is determined to be a candidate target point that meets the pre-defined conditions.

[0268] In the embodiments of this disclosure, false targets are suppressed and multi-target identification at the same distance and speed is achieved to some extent by performing a second CFAR on the azimuth and elevation DBF after the initial NR-CFAR.

[0269] In the El-CFAR described above, when generating the DBF spectrum of the elevation dimension, an elevation-depression guide vector is generated again based on the azimuth angle information obtained by Az-CFAR. In the exemplary embodiment, an elevation-depression guide vector with a pre-set azimuth angle (e.g., 0° azimuth angle) may be directly adopted.

[0270] The order of the first target selection in step 7 and the second target selection in step 8 is interchangeable. In some embodiments, only step 7 or only step 8 may be performed.

[0271] Step 9 involves extracting target information.

[0272] All candidate target points that meet the conditions Information is extracted from TIFF0007847889000095.tif6150, including, but not limited to, the range bin index (distance unit index) and Doppler bin index (Doppler unit index) of the target point. It may also include the SNR (signal-to-noise ratio) of RD-CFAR (Range-Doppler CFAR), Az-CFAR, El-CFAR (Elevation CFAR), etc. The above SNR can be used later for target person determination, and for example, weights may be assigned to corresponding target detection points based on the SNR value, with the weights being related to the SNR value.

[0273] In step 10, all target points detected in each frame are counted using a preset region design, and the count results are determined by pre-set rules to determine whether personnel are present in that processing round. The location of the personnel is then determined, i.e., target information is obtained.

[0274] One embodiment of the preset regions, as shown in Figure 35, has the horizontal coordinate in the figure as the X direction (vehicle width direction) and the Z direction as the rear-front direction. The interior space of the vehicle is divided into regions that need to be determined along all or some of the XYZ dimensions, or along all or some of the distance-azimuth-elevation dimensions of the polar coordinate system. As shown in the example in Figure 35, it is divided into a total of six regions: the three rear seats and the three aisles in front of the three seats. These regions may or may not overlap, and some regions of the interior may belong to multiple preset regions simultaneously, or may not belong to any region.

[0275] Assume that a total of 11 target points were detected in a given processing run, and the number of these target points in these preset regions is statistically calculated. In this example, the counts for the regions of these target points—seat A, seat B, seat C, aisle A, aisle B, and aisle C—are 0, 2, 6, 0, 1, and 0, respectively.

[0276] Using the above area division and count results as an example, this process determines whether personnel are present, and then determines the location of the personnel, including the following steps. In step 10.1, the flag bit flag_region_i, which indicates whether there are people in each of the six regions, is initialized to 0, meaning that none of them are present (i=0, 1, ..., 5). In step 10.2, it is determined whether the total number of valid targets, total_valid_tgt_num, is 0. If it is 0, it is determined that there are no people in the vehicle and the process ends. If it is not 0, step 10.3 is executed. In step 10.3, all regions are iterated through, and for region i, if it is determined that the statistical value tgt_num_region_i of the number of valid targets belonging to that region exceeds a preset proportional value (for example, 25%, which can be set based on experience), the flag bit flag_region_i for that region i is set to 1. In step 10.4, if there is a region where flag_region_i is 1, flag_region_empty is set to 0, meaning that there is a person inside the vehicle. In step 10.5, the system iterates through the seating areas, and if any seating area flag_region_i is 1, it determines that there are no people in the corresponding aisle area (for example, if it determines that there are people in seat A, it determines that there are no people in aisle A). Step 10.6 outputs the results.

[0277] By dividing the area and setting preset proportional values, the target number of personnel can be determined more accurately, and misjudgments can be prevented.

[0278] In an exemplary embodiment, a direct azimuthal-elevation two-dimensional DBF may be performed on the first CFAR result (i.e., the result of step 6), and a second CFAR may be performed directly on the azimuthal-elevation two-dimensional DBF spectrum, the main steps of which include the following:

[0279] Extract 2D-FFT data, select an antenna, perform a 2D DBF based on azimuth and elevation 2D guide vectors, obtain the DBF spectrum, estimate the noise floor from the DBF spectrum (see step 5 above for the noise floor estimation method), perform a second CFAR (referred to as AE-CFAR), and find all candidate target points that satisfy the conditions. TIFF0007847889000096.tif6150 and its corresponding azimuth TIFF0007847889000097.tif6150 and elevation / depression angle Obtain TIFF0007847889000098.tif6150, and one possible method is to iterate through all points in the DBF spectrum and determine that any point is a candidate target point if it satisfies the following conditions. In the JPEG0007847889000099.jpg45160 format, TIFF0007847889000100.tif7160 is the number of the point in the azimuthal dimension, The filename is TIFF0007847889000101.tif6150. TIFF0007847889000102.tif6150 This is the number of points in the azimuthal dimension DBF spectrum, TIFF0007847889000103.tif7160 is the number of that point in the elevation dimension, TIFF0007847889000104.tif6150 And, TIFF0007847889000105.tif6150 is the number of points in the elevation-depression dimension DBF spectrum. TIFF0007847889000106.tif6150 is the power of the point, TIFF0007847889000107.tif6150 is a configurable parameter, TIFF0007847889000108.tif6150, The filename is TIFF0007847889000109.tif6150. Equation 5 above states that the ratio of the power to the noise floor of the candidate target point is a preset threshold. This indicates that it must be greater than or equal to TIFF0007847889000110.tif6150, which corresponds to setting a first limit, which is the noise floor estimate + first threshold, and by setting the first limit, filtering of false target points can be achieved, and Equation 6 states that the ratio of the power of the candidate target point to the maximum power across the range is the preset threshold This indicates that it must be greater than or equal to TIFF0007847889000111.tif7160, which means a second limit is set, and this second limit is equal to the maximum value of the entire range minus the second threshold. Setting a second limit enables multi-target detection, which is advantageous for recognizing targets such as children and babies. If the value is TIFF0007847889000112.tif11170, then equations 7-10 above indicate that the point is the local maximum value.

[0280] Figures 36A to 36D show the processing results for a single-person scene (baby in corridor C). A total of 500 frames of data were used, with 128 frames of interframe FFT performed. The sliding window length for the sliding window processing was 1, and a total of 373 processing iterations were performed. Figures 36A and 36C show the results when only NR-CFAR in RD dimensions was used, followed by azimuthal-elevation DoA (wave arrival angle estimation). Figures 36B and 36D show the results when NR-CFAR in RD dimensions was used, followed by azimuthal-elevation DBF, and then a second CFAR processing. Figures 36A and 36B show the elevation-azimuthal results, while Figures 36C and 36D show the azimuthal-processing number results. As can be seen from this, the method of this embodiment (using NR-CFAR in RD dimensions, then performing azimuth / elevation DBF, and then performing a second CFAR) can effectively reduce the number of invalid false target points, concentrate the target points, effectively reduce the difficulty of post-processing, and enhance the detection effect.

[0281] Figures 37A to 37D show the processing results for a two-person scene (baby in seat B, adult in seat A). Figures 37A and 37C show the results of using only NR-CFAR in the RD dimension, followed by azimuth-elevation DoA, while Figures 37B and 37D show the results of using NR-CFAR in the RD dimension, followed by azimuth-elevation DBF, and then a second CFAR processing. Figures 37A and 37B show the elevation-azimuth results, while Figures 37C and 37D show the azimuth-processing number results. As can be seen, the method of this embodiment can effectively reduce the number of invalid false target points, concentrate target points, effectively reduce the difficulty of post-processing, and enhance detection effectiveness.

[0282] The embodiments of this disclosure can be applied not only to human target detection and positioning scenes inside vehicles, but also to other similar application scenes such as personnel detection indoors and personnel detection in factory buildings.

[0283] The above operations may be performed on the MCU, on the baseband accelerator, or partially on the MCU and partially on the baseband accelerator. For example, 1D-FFT, 2D-FFT, CFAR, and DBF may be performed on the baseband accelerator, while target detection and positioning are performed on the MCU.

[0284] Compared to conventional technologies, the CPD processing method according to the embodiments of this disclosure mainly employs a completely new interframe storage method, coherently storing data based on the frequency of human respiration, increasing the signal-to-noise ratio between the personnel target and static noise, thereby improving detection performance. Furthermore, it employs a second CFAR detection on the DBF spectrum generated in the azimuth / elevation dimensions of the CFAR results, thereby effectively eliminating false alarms and effectively improving detection and measurement performance. This enables accurate detection of special or weak targets, such as infants in a cabin, and realizes applications such as in-vehicle child detection (CPD) and seatbelt warning devices (SBR).

[0285] Performance evaluations conducted through numerous field tests revealed that the method described herein achieves a detection rate of over 99%, detection failure rates of 0.5% and 1% or less, and false alarm rates, demonstrating a clear advantage over conventional technologies.

[0286] Embodiments of the present disclosure further provide a target detection method applicable to target detection in a specific target area, the method comprising performing a first constant false alarm probability processing on distance-Doppler data, followed by a second constant false alarm probability processing in the angular dimension, thereby obtaining target data. The specific target area is, for example, a closed area and / or a semi-closed area such as a cabin or room. The first constant false alarm probability processing and the second constant false alarm probability processing can be described with reference to the above embodiments.

[0287] Figure 38 is a schematic flowchart illustrating post-processing for a target in an embodiment of the present invention. As shown in Figure 38, after azimuth / elevation-depression estimation is performed on the flow structure shown in Figure 1, processing may be performed on the output target point cloud data in combination with a machine learning model, thereby enabling accurate target detection in the area of ​​interest. For example, on a single target point cloud data acquired after azimuth / elevation-depression estimation, suppression processing may be performed on non-ideal data such as noise, static interference, or false alarms due to target multipath in combination with ML (Machine Learning, e.g., SVM or RF algorithm) false alarm suppression technology. That is, for example, on the target point cloud data output by elevation-depression DBF&DoA, operations such as ML-based false-alarm suppression, clustering, and / or ML-based target classification may be performed, and the clustering process may be an algorithm such as agglomerative clustering or DBSCAN. Simultaneously, the point cloud data after clustering may be input into a trained machine learning model (e.g., SVM or RF) to determine whether there are people in the current processing area. If there are people, their location may be determined, and operations such as distinguishing and identifying physical targets such as adults, children, and infants may be performed.

[0288] In the embodiment shown in Figure 38, the effects of undesirable elements such as noise, static interference, and target multipath are effectively reduced, and the difficulty of selecting domain parameters and designing judgment logic for complex domains is effectively reduced. This allows for more effective use of point cloud information, and better target detection and discrimination performance can be achieved, especially when applied to complex scenes.

[0289] Figure 39 is a schematic flowchart illustrating the post-processing of a target in combination with a DL algorithm in an embodiment of the present invention. As shown in Figure 39, after extracting data acquired through at least one of the following operational steps on the flow shown in Figure 1, such as storing multiple frames, inter-frame Fourier transform (Frame FFT), non-coherent integration, and constant false alarm probability detection (CFAR), a deep learning algorithm (DL) may be used to determine, position, and recognize the target, for example, whether the physical target is an adult or a child (Adult / Child Classification).

[0290] For example, the following methods may be employed to achieve the distinction and determination of subsequent targets: using the original 1D-FFT data as the input to DL (Access 1D-FFT data as raw input of DL), using the original 2D-FFT data between frames as the input to DL (Access 2D-FFT data as raw input of DL), using the non-coherent integrated 2D-FFT data as the input to DL (Access non-coherent integrated 2DFFT data as raw input of DL), and / or using SNR data as the input to DL (Access SNR data as raw input of DL). The original data here may be complex number data, the modulo value of complex number data, the module square of complex number data, or other conversion formats, and may be values ​​in the linear domain, values ​​in the dB(log) domain, or other calculation formats.

[0291] Based on the flow shown in Figure 39, by employing at least one of the above methods, multi-frame 1D-FFT data, multi-channel data after multi-frame FFT, and energy (power) or SNR data after multi-frame FFT and SISO-combine are extracted. The extracted data is then input into a constructed deep learning model (e.g., CNN or Transformer) to determine whether a target exists in the current process and to determine the type of the current target (e.g., whether the target is an adult or a child).

[0292] Furthermore, the flow steps and related descriptions shown in Figure 39 are compatible with the flow steps and related descriptions shown in Figure 38. For example, the Region HIST and subsequent modules in the flow shown in Figure 39 are compatible with some or all of the ML-based false alarm suppression and subsequent modules in the flow shown in Figure 38.

[0293] Figure 40 is a schematic flowchart of a target detection method combined with a DL algorithm in an embodiment of the present invention. As shown in Figure 40, the operations of two-dimensional digital beam synthesis (2D-DBF) and DL-based target classification may be performed directly after the inter-frame FFT (Frame FFT) operation on the flow shown in Figure 1. That is, instead of performing CFAR processing after the inter-frame FFT, a 2D-DBF operation with azimuth and elevation dimensions is performed on each region of interest (or unit interval), thereby obtaining a number of RD spectra corresponding to the region of interest, and these RD spectra are then input into the constructed deep learning model (DL) to perform target determination. In some selectable embodiments, there may be a one-to-many relationship between the region of interest and the RD spectra, that is, at least two RD spectra can be obtained based on one region of interest, and the detailed number may be adjusted according to actual needs.

[0294] The DL input described above may be a linear or dB-range amplitude or power, and may also perform operations such as normalization simultaneously. The constructed deep learning model may then classify based on the input, and the types of classifications may include whether or not there is a target, the attributes of the target, the specific unit interval location (i.e., region location) where each target is located, etc.

[0295] For example, in an application scenario where the rear area inside a car is referred to as the region of interest or target area, and it is divided into three seat areas (section units) and corresponding three aisle areas, after interframe FFT, a 2D-DBF of azimuth and elevation is performed on the center positions of the six regions of interest (the region of interest may also be set to three seat areas) to obtain six RD spectra (or three RD spectra) corresponding to the region of interest. At least a portion of these six RD spectra (or three RD spectra) can then be input into a constructed deep learning model to perform operations such as target determination and distinction. For example, the constructed deep learning model can classify based on the input to determine whether there is a biological target in the region of interest. If there is a biological target, it can further determine whether the biological target is an adult, child, infant, or pet, and can also determine specific location information such as which seat or aisle area the biological target is in.

[0296] In the embodiment shown in Figure 40, the target detection method is a dual-drive processing flow based on the model and data, which has relatively low requirements for signal processing. This effectively avoids steps such as selecting signal processing parameters, domain parameters, and designing CFAR logic and domain determination logic, thereby significantly reducing the difficulty of implementing and designing the solution.

[0297] In some selectable embodiments, all FFT processing in all of the aforementioned target detection methods may be windowed FFTs, and may also be implemented in combination with SVA and FFT.

[0298] Figure 41 is a schematic flowchart of another target detection method combined with a DL algorithm in an embodiment of the present invention. As shown in Figure 41, in the flow shown in Figure 40, a 2D-DBF may be performed first after the distance-dimensional Fourier transform (Range FFT, or possibly 1D-FFT), and then operations such as frame data accumulation and inter-frame FFT processing may be performed. That is, before the inter-frame FFT, a 2D-DBF of the azimuth and elevation / depression angles is first performed on the center position of the region of interest (e.g., three seat regions), and this embodiment realizes a parallel processing scheme for data processing and transmission time, thereby effectively reducing processing time.

[0299] In some selectable embodiments, the multi-frame sliding window FFT may be further replaced in all of the aforementioned target detection methods with other time-frequency transformation processes, such as short-time Fourier transforms or fractional Fourier transforms, and similarly effective detection of targets of interest can be achieved.

[0300] In some selectable embodiments, the multi-frame sliding window FFT may be further replaced in all of the aforementioned target detection methods with methods such as high-pass FIR, Comb FIR, or filters optimized and integrated by an optimization function, thereby achieving performance similar to or better than frame FFT processing with fewer hardware resources.

[0301] As shown in Figures 40 and 41, the combination of 2D-FFT and DL-based classification can effectively improve the accuracy of target recognition. The steps of the 2D-FFT and DL-based classification can be flexibly set between each step in signal processing according to the actual needs, for example, by setting the 2D-FFT step after Range FFT or after Frame FFT.

[0302] Figure 42 is a schematic flowchart of yet another target detection method combined with a DL algorithm in an embodiment of the present invention. As shown in Figure 42, on the flow shown in Figure 40 or Figure 41, i.e., after the distance-dimensional Fourier transform (Range FFT, which may also be called 1D-FFT), a 2D-DBF is performed, and then the accumulation of frame data is completed, and a DL-based target classification operation (Complex DL-based Classification) is performed directly. That is, in the entire target detection flow, inter-frame FFT processing is not performed, and after the 1D-FFT, a 2D-DBF is performed directly on the center position of the region of interest (e.g., three seat regions) to obtain a corresponding number of distance-frame spectra (e.g., three range-frames), and then the above distance-frame spectra are input to a constructed (e.g., complex number) deep learning model to perform operations such as target distinction and determination.

[0303] In the embodiment shown in Figure 42, the FFT processing step between frames is avoided, thereby effectively reducing the computational complexity. Furthermore, by employing a complex number deep learning model, a higher processing gain than FFT can be obtained, thereby achieving the objective of improving target detection performance. In addition, this embodiment effectively shortens the processing hierarchy, making it convenient for efficient scheduling operations.

[0304] In some selectable embodiments, the DBF in the target detection method shown in Figures 40-42 may be replaced with Capon, MUSIC, ESPRINT, or other derived algorithms, and optimization integration may be performed on beam synthesis at the central position of the region of interest and / or on the antenna arrangement, thereby further improving the target detection performance of the system.

[0305] Furthermore, the embodiments of this application are mutually referenced and interchangeable as long as they do not conflict, and the order and configuration of each functional module can be adjusted according to the needs. In the case of a system having a BB module and an MCU module, if the system performs target detection by transmitting electromagnetic waves and receiving corresponding echo signals, each step in each target detection method of the embodiments of this application may be arranged and executed in accordance with the BB module and / or MCU module, taking into consideration the actual needs and the data processing capacity and time efficiency of the system execution, and the relevant examples shown in the figures can be used as reference for some of these options.

[0306] In several selectable embodiments, all of the target detection methods described above can be applied to a TDM-MIMO system. However, for systems that do not employ TDM-MIMO, different phase shifts may be pre-set for different Tx transmissions, and then direct transmission waveform digital synthesis (phase-controlled array) may be performed on the region of interest. Subsequently, target detection and determination operations may be performed in combination with the processing flow shown in Figures 40-42.

[0307] In some other embodiments, for systems employing pulsed, pulse-compressed, or ultra-wideband methods, target detection may be performed directly using the method flow shown in Figures 40 to 42, and the performance of target detection may be further improved by employing transmitted beam digital synthesis technology. For pulse-compressed systems, 1D-FFT processing is not required, but the pulse compression step must be added.

[0308] Furthermore, the processing methods described above for different types of systems can also be applied to hybrid systems. For example, in a system with six transmitting antennas (i.e., six Tx), three of the transmitting antennas (three Tx) may use the TDM method, while the other three may use other methods. Subsequent processing should then correspond to the processing methods described above.

[0309] After inter-frame FFT processing, two-dimensional digital beam synthesis (2D-DBF) processing and DL-based target classification processing are performed directly. That is, instead of performing CFAR processing after inter-frame FFT, 2D-DBF operations for azimuth and elevation are performed on each region of interest (or unit interval), thereby obtaining a number of RD spectra corresponding to the region of interest. These RD spectra are then input into the constructed deep learning model (DL) to perform target determination. In some selectable embodiments, if processing is performed using methods such as FIR or STFT (Short-Time Fourier Transform) instead of the inter-frame FFT method, the above-mentioned RD spectra are one-dimensional spectral outputs corresponding to FIR, while if time-frequency transformation is used corresponding to STFT, they are three-dimensional spectra.

[0310] Furthermore, electromagnetic waves in this application may include radio waves and light waves. Radio waves include short waves, medium waves, long waves, microwaves, etc. Microwaves include centimeter waves (i.e., electromagnetic waves from 3 GHz to 30 GHz, e.g., electromagnetic waves from 3.1 GHz to 10.6 GHz, electromagnetic waves in the 24 GHz band, etc.) and millimeter waves (i.e., electromagnetic waves from 30 GHz to 300 GHz, e.g., electromagnetic waves in the 60 GHz band, electromagnetic waves in the 77 GHz band (e.g., 77 GHz to 81 GHz, etc.)). Light waves include ultraviolet light, visible light, infrared light, laser light, etc. The electromagnetic wave band for laser light is (3.846 to 7.895) * 10^5 GHz, meaning that laser light is included in the frequency bands of ultraviolet light and some visible light.

[0311] Embodiments of the present disclosure further provide an integrated circuit comprising: a signal transmitting module configured for use with electromagnetic waves for target detection; a signal receiving module configured for use with receiving echoes formed by the reflection and / or scattering of the electromagnetic waves; and a processing module configured for use with respect to the echoes, thereby achieving target detection.

[0312] As an option, the target detection is, for example, at least one of the following operations: determination, positioning, and / or recognition of a target in a region of interest.

[0313] In exemplary embodiments, the processing module may include at least a baseband unit, which may be configured to perform distance-dimensional FFT processing, velocity-dimensional FFT, a first constant false alarm probability processing, and a second constant false alarm probability processing as described in any embodiment of the present disclosure.

[0314] As an option, in an exemplary embodiment, the processing module may further comprise an MCU unit, and if the method includes digital beam synthesis and frame data storage, the baseband unit is configured to be used to implement the digital beam synthesis, and the MCU unit is configured to be used to implement the frame data storage.

[0315] In exemplary embodiments, the integrated circuit may be a millimeter-wave radar chip (chip or die).

[0316] Embodiments of the present invention further provide an integrated circuit which may comprise a sequentially connected radio frequency module, an analog signal processing module, and a digital signal processing module. The radio frequency module is used to generate a radio frequency transmission signal and receive an echo signal. The analog signal processing module is used to down-convert the echo signal to obtain an intermediate frequency signal. The digital processing module is used to convert the intermediate frequency signal from analog to digital to obtain a digital signal and to process the digital signal based on the target detection method in the embodiment of the present invention to achieve the objective of target detection. For example, the integrated circuit may be a millimeter-wave radar chip (chip or die). The digital processing module may comprise subunits such as a BB unit and an MCU unit, each subunit being configured to perform each corresponding step in the target detection method in the embodiment.

[0317] In some selectable embodiments, the integrated circuit may be an AiP (Antenna-In-Package) chip structure, an AoP (Antenna-On-Package) chip structure, or an AoC (Antenna-On-Chip) chip structure.

[0318] Some other embodiments of the present invention further provide an electromagnetic wave sensor. The electromagnetic wave sensor may include an antenna and an integrated circuit as described above. The integrated circuit is electrically connected to the antenna and used to transmit and receive electromagnetic wave signals. For example, the electromagnetic wave sensor may include a carrier, the integrated circuit described in any of the above embodiments, and an antenna, etc., wherein the integrated circuit may be installed on the carrier, the antenna may be installed on the carrier or integrated with the integrated circuit as an integrated device and installed on the carrier (i.e., in this case the antenna may be an antenna installed in an AiP, AoP, or AoC structure), and the integrated circuit is connected to the antenna (i.e., in this case the antenna is not integrated into the sensor chip or integrated circuit, but is, for example, a normal SoC, etc.) and used to transmit and receive electromagnetic wave signals. The carrier may be a printed circuit board (PCB), and the corresponding transmission lines may be PCB wiring.

[0319] Embodiments of the present invention provide a terminal device which may comprise a main body of the device and an electromagnetic wave sensor as described above installed on the main body of the device, the electromagnetic wave sensor being used for target detection and / or communication, thereby providing reference information for the operation of the main body of the device.

[0320] Embodiments of the present application further provide electronic devices (which may be understood as terminal devices) which may appear in the form of general-purpose computing devices. Components of the electronic devices may include, but are not limited to, at least one processing unit, at least one storage unit, a bus connecting different system components (including the storage unit and the processing unit), a display unit, and the like. Program code is stored in the storage unit, and the program code can be executed by the processing unit, thereby enabling the processing unit to perform methods according to various exemplary embodiments of the present application as described herein. The storage unit may include a readable medium in the form of a volatile storage unit, for example, a random access storage unit (RAM) and / or a cache storage unit, and may further include a read-only storage unit (ROM).

[0321] The storage unit may further include a program / utility having one set (at least one) of program modules, such program modules including, but not limited to, an operating system, one or more application programs, other program modules, and program data, and each of these examples, or any combination thereof, may include the implementation of a network environment.

[0322] The bus may represent one or more of several types of bus structures, including a storage unit bus or storage unit controller, a peripheral bus, a graphics accelerator port, a processing unit, or a local bus that uses one of the bus structures.

[0323] The electronic device may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), and further with one or more devices that allow a user to interact with the electronic device, and / or with any devices (e.g., routers, modems, etc.) that allow the electronic device to communicate with one or more other computing devices. Such communication may be via input / output (I / O) interfaces. The electronic device may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, e.g., the Internet) via a network adapter. The network adapter may communicate with other modules of the electronic device via a bus. For clarity, other hardware and / or software modules may be used in conjunction with the electronic device, although not shown in the diagram, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0324] For example, the electronic device in the embodiment of the present application may further include a main body of the device and an electromagnetic wave sensor installed on the main body of the device, the electromagnetic wave sensor which can be used to realize functions such as target detection and / or wireless communication.

[0325] Specifically, based on the above embodiments, in one selectable embodiment of the present application, the electromagnetic wave sensor may be installed outside the main body of the device or inside the main body of the device, and in another selectable embodiment of the present application, the electromagnetic wave sensor may be further installed partially inside the main body of the device and partially outside the main body of the device. The embodiments of the present application are not limited thereto and may be specifically determined on a case-by-case basis.

[0326] In one selectable embodiment, the device body may be, for example, components and products applied to fields such as smart cities, smart houses, transportation, smart homes, consumer electronics, security monitoring, industrial automation, shipboard detection (e.g., smart cabins), medical devices, and hygiene and health. For example, the device body may be smart transportation devices (e.g., automobiles, bicycles, motorcycles, ships, subways, trains, etc.), security devices (e.g., cameras), liquid level / flow velocity detection devices, smart wearable devices (e.g., bracelets, glasses, etc.), smart home devices (e.g., cleaning robots, door locks, televisions, air conditioners, smart lamps, etc.), various communication devices (e.g., mobile phones, tablets, etc.), and, for example, barrier gates, smart traffic indicator lamps, smart signs, traffic cameras, and various industrial machine arms (or robots). It may also be various devices for detecting biological characteristic parameters and various devices that mount such devices, for example, biological characteristic detection in car cabins, indoor personnel monitoring, smart medical devices, consumer electronics devices, etc.

[0327] Embodiments of the present invention further provide a non-temporary computer-readable storage medium in which computer-readable instructions are stored, and when the instructions are executed by a processor, the processor performs the above-described method for compensating for unequal lengths of power lines.

[0328] As those skilled in the art will understand from the above description of embodiments, the embodiments described herein may be implemented by software, or by a combination of software and necessary hardware. The technical solutions according to the embodiments of the present application may be embodied in the form of a software product, which may be stored on a non-volatile storage medium (such as a CD-ROM, USB disk, removable hard disk, etc.) or on a network, and which includes some instructions for causing a computing device (such as a personal computer, server, or network device, etc.) to perform the above method according to the embodiments of the present application.

[0329] A software product may employ any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (non-exclusive list) include electrical connections with one or more leads, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0330] A computer-readable storage medium may include data signals propagated in the baseband or as part of a carrier, thereby carrying readable program code. The propagated data signals may employ multiple forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may further be any other readable medium, which can transmit, propagate, or transmit programs for use by or in combination with instruction execution systems, apparatus, or devices. The program code contained in the readable storage medium may be transmitted by any suitable medium, including but not limited to wireless, wires, optical cables, RF, or any suitable combination thereof.

[0331] The program code for performing the operations of the present invention may be written in any combination of one or more program design languages, including object-oriented program design languages ​​such as Java and C++, and also including conventional process program design languages ​​such as the "C" language or similar programs. The program code may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. If a remote computer is involved, it may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it may be connected to an external computer (for example, via the Internet using an Internet service provider).

[0332] The computer-readable medium carries one or more programs, and when the one or more programs are executed by one of the devices, the computer-readable medium performs the functions described above.

[0333] Those skilled in the art will understand that each of the above-described modules may be arranged in the apparatus as described in the embodiment, or may be appropriately modified to be located in one or more apparatuses different from those in this embodiment. The modules of the above embodiment may be integrated as a single module, or they may be further divided into multiple submodules.

[0334] According to embodiments of the present application, the above method can be performed by submitting a computer program, which includes a computer program or instructions, and when the computer program or instructions are executed by a processor. In one optional embodiment, the integrated circuit may be a millimeter-wave radar chip. The type of digital function module in the integrated circuit may be determined according to actual needs. For example, in a millimeter-wave radar chip, the data processing module may be used for, for example, distance-dimensional Doppler conversion, velocity-dimensional Doppler conversion, constant false alarm probability detection, wave arrival direction detection, point cloud processing, etc., and is used to acquire information such as the distance, angle, velocity, height, micro-Doppler motion characteristics, shape, dimensions, surface roughness, dielectric properties of a target.

[0335] Furthermore, by transmitting and receiving wireless signals, wireless devices can perform functions such as target detection and / or communication, thereby providing detected target information and / or communication information to the main device, and thereby supporting or controlling the operation of the main device.

[0336] For example, when the above-mentioned device is applied to an advanced driver-assistance system (i.e., ADAS), the wireless device as an in-vehicle sensor (e.g., millimeter-wave radar) can support the ADAS system and enable application scenarios such as adaptive cruise control, automatic emergency braking (i.e., AEB), blind spot detection (i.e., BSD), lane change assist (i.e., LCA), rear cross-traffic alert (i.e., RCTA), parking assist, rear vehicle warning, collision avoidance, pedestrian detection, and in-cabin biometric detection (i.e., CPD).

[0337] As those skilled in the art will understand, functional modules / units in all or some steps, systems, or apparatus of the methods disclosed above can be implemented as software, firmware, hardware, or appropriate combinations thereof. In hardware embodiments, the distinctions between functional modules / units mentioned above do not necessarily correspond to distinctions between physical components; for example, one physical component may have multiple functions, or one function or step may be performed collaboratively by multiple physical components. Some or all components may be implemented as software executed by a processor, e.g., a digital signal processor or a microprocessor, or as hardware, or as an integrated circuit, e.g., a dedicated integrated circuit. Such software may be placed on a computer-readable medium, which may include computer storage media (or non-temporary media) and communication media (or temporary media). As those skilled in the art will know, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technique for storing information (e.g., computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital multifunction disk (DVD) or other optical disc storage devices, magnetic cassettes, magnetic tapes, magnetic disk storage devices or other magnetic storage devices, or any other media used to store desired information and accessible by a computer. As is well known to those skilled in the art, communication media generally include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carriers or other transmission mechanisms, and may include any information distribution media.

[0338] The technical features of the above embodiments may be combined in any way, and for the sake of simplicity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no inconsistency among these combinations of technical features, they should all be considered to fall within the scope described herein.

[0339] The above embodiments represent only preferred embodiments and technical principles used in the present invention, and while their description is relatively specific and detailed, it should not be understood as a limitation on the scope of the patent. Those skilled in the art can make various obvious modifications, readjustments, and substitutions without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in relatively detail by the above embodiments, the present invention is not limited to these embodiments and may include many other equivalent embodiments without departing from the concept of the present invention, and the scope of patent protection of the present invention is determined by the appended claims.

Claims

1. A target detection method, Frame data is accumulated for 1D-FFT data, and the 1D-FFT data is obtained by performing distance-dimensional FFT processing on the echo signal, the frame data refers to data corresponding to the echo signal formed when the frame detection signal is reflected by the target, the frame detection signal includes multiple chirp signals, and the accumulation of frame data for 1D-FFT data includes performing 1D-FFT processing on each chirp data of the echo signal of each frame to obtain the 1D-FFT data, accumulating the 1D-FFT data, and obtaining the accumulated 1D-FFT data of multiple frames. Based on the 1D-FFT data of the multiple frames accumulated, inter-frame FFT processing is performed to obtain the RD spectrum. A target detection method comprising: detecting a target in a region of interest based on the aforementioned RD spectrum.

2. To achieve target detection in the region of interest based on the aforementioned RD spectrum, Performing a first constant false alarm probability processing on the RD spectrum to obtain candidate target detection points, The target detection method according to claim 1, further comprising: performing a second constant false alarm probability processing on the candidate target detection point; and performing target detection based on the result of the second constant false alarm probability processing.

3. Performing a first constant false alarm probability processing on the RD spectrum to obtain candidate target detection points is: The target detection method according to claim 2, comprising performing noncoherent integration on multichannel distance-velocity-dimensional data, and performing a first constant false alarm probability processing based on noise estimation based on the noncoherent integration results to obtain candidate target detection points.

4. Performing a second constant false alarm probability processing on the candidate target detection point is: Perform azimuth angle dimension DBF on the candidate target detection point, perform constant azimuth angle false alarm probability processing by referring to the noise floor estimation result, obtain target selection results for the azimuth angle dimension, and / or perform elevation angle dimension DBF on the candidate target detection point, perform constant elevation angle false alarm probability processing by referring to the noise floor estimation result, obtain target selection results for the elevation angle dimension, The target detection method according to claim 2, comprising: performing a two-dimensional DBF of azimuth and elevation on the candidate target detection point; performing a constant false alarm probability processing of azimuth angle and elevation angle by referring to the noise floor estimation result; and obtaining a two-dimensional target selection result of azimuth and elevation.

5. The target detection method according to claim 4, wherein, during processing of a constant false alarm probability, one or more operations from the range maximum value filtering, first threshold filtering, and second threshold filtering are performed on the candidate target detection point to obtain a target detection result, the range maximum value filtering is used to filter whether the range maximum value is a target detection point, the first threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the noise floor estimate, and the second threshold filtering determines whether the current candidate filtering result is a target detection point based on the relationship between the candidate filtering result and the range maximum value.

6. Performing noncoherent integration on the multi-channel distance-velocity dimension data and then performing a first constant false alarm probability processing based on noise estimation using the noncoherent integration results is: The target detection method according to claim 3, comprising performing noncoherent integration on multichannel distance-velocity-dimensional data, estimating the noise floor of each distance unit after noncoherent integration to obtain the estimated noise floor value for each distance unit, and performing noncoherent constant false alarm probability processing using the estimated noise floor value, wherein when estimating the noise floor of each distance unit after noncoherent integration, the noise floor of each distance unit is adjusted using the global noise floor.

7. Performing target detection based on the second constant false alarm probability processing result is, The target detection method according to claim 2, comprising: pre-dividing a region of interest into multiple target sub-regions; determining the number and location of target points in each sub-region based on the target detection point locations acquired after target detection; and determining whether there is a target to be detected in the current sub-region based on whether the ratio of the number of target points in each sub-region to the total number of targets is greater than a pre-set proportional value.

8. Performing interframe FFT processing based on the accumulated 1D-FFT data of multiple frames, obtaining an RD spectrum, and then detecting a target in the region of interest based on the RD spectrum is: A target detection method according to claim 1, comprising extracting target data from multiple accumulated frame data, performing sliding window processing on the target data, performing digital signal processing on the data after window processing to obtain target information, wherein the length of the sliding window corresponds to a time length of the same magnitude as the period of the periodic motion of the target.

9. Extracting target data from the aforementioned accumulated multiple frame data is This includes extracting at least some data or at least some digitally processed data from each frame data obtained by distance-dimensional FFT processing to obtain the target data, or determining parameters corresponding to each distance unit in each frame data based on each frame data obtained by distance-dimensional FFT processing, and thereby generating the target data. Performing digital signal processing on the data after the aforementioned windowing process to obtain target information is, A target detection method according to claim 8, comprising: performing noise floor estimation within a preset range, the upper limit of the preset range being determined based on the noise floor estimation result of the furthest distance unit; and performing a constant false alarm probability detection based on the estimated noise floor, thereby obtaining target information.

10. Extracting target data from the accumulated multiple frame data is This includes, and / or, extracting data within a preset distance range from frame data obtained by distance-dimensional FFT processing as the target data. Performing digital signal processing on the data after the aforementioned windowing process is: The target detection method according to claim 8, further comprising performing digital signal processing on data within a preset Doppler range from the data after window processing.

11. Performing a sliding window process on the aforementioned target data, and then performing digital signal processing on the data after the windowing process to obtain target information, This includes performing window processing on the target data based on sliding windows corresponding to the periods of different biological characteristic parameters of the living organism, and performing digital signal processing on the windowed data corresponding to sliding windows of different lengths. The target detection method according to claim 8, wherein the periods of the bio-characteristic parameters corresponding to different sliding windows are different, and the length of each sliding window has a time length of the same magnitude as the period of the bio-characteristic parameter.

12. To achieve target detection in the region of interest based on the aforementioned RD spectrum, Based on the RD spectrum, a constant false alarm probability processing is performed on at least two transmit / receive channels individually, thereby obtaining candidate target data for at least two transmit / receive channels. A target detection method according to claim 1, comprising processing candidate target data from at least two transmission / reception channels to obtain final target data.

13. Performing a constant false alarm probability processing on at least two transmit / receive channels individually based on the RD spectrum is: Based on the RD spectrum, perform a certain false alarm probability processing for at least two transmission / reception channels, or, The target detection method according to claim 12, further comprising: performing noise floor estimation independently for a transmit / receive channel to perform a fixed false alarm probability processing independently based on the RD spectrum; obtaining the noise floor estimation result for a single transmit / receive channel; and performing fixed false alarm probability detection independently for the corresponding transmit / receive channel based on the noise floor estimation result for the single transmit / receive channel.

14. Processing based on the candidate target data of the at least two transmission and reception channels is: The target detection method according to claim 12, further comprising performing a certain false alarm probability detection based on the currently acquired candidate target data and preset thresholds to acquire the final target data.

15. After obtaining the target information, the method further: Based on the target that has entered each preset region, the target in each preset region is verified, and the preset region is obtained by dividing the detection space. A target detection method according to any one of claims 8 to 14, comprising outputting the detection results of each preset region based on the targets that have passed verification.

16. For multi-channel application scenarios, achieving target detection in the region of interest based on the RD spectrum is possible. After performing non-coherent integration on a multi-channel RD spectrum, a constant false alarm probability processing based on noise estimation and wave arrival direction estimation are performed, thereby achieving target detection, or The target detection method according to claim 1, further comprising performing a constant false alarm probability processing on each of the multi-channel RD spectra, followed by binary aggregation processing of the channel domains, wave arrival direction estimation, and then performing one of the following operations on the target: determination, positioning, and recognition.

17. To achieve target detection in the region of interest based on the aforementioned RD spectrum, The target detection method according to claim 1, further comprising recognizing a target based on extracting at least a portion of the accumulated data from multiple frames, the interframe Fourier transformed data, the non-coherent integrated data, and / or the data for detecting a constant false alarm probability.

18. It is an integrated circuit, A signal transmission module configured for use in electromagnetic waves for target detection, A signal receiving module configured to be used for receiving echoes formed by the reflection and / or scattering of the aforementioned electromagnetic waves, An integrated circuit comprising: a processing module configured to be used to perform processing on the echo based on the target detection method described in any one of claims 1 to 14, thereby achieving target detection.

19. It is an electromagnetic wave sensor, Career and, The integrated circuit according to claim 18 is installed on the carrier, Equipped with an antenna, The antenna is installed on the carrier, or the antenna is integrated as a device with the integrated circuit and installed on the carrier. An electromagnetic wave sensor, wherein the integrated circuit is connected to the antenna and used to transmit an electromagnetic wave signal and / or receive the echo signal.

20. A terminal device, The main body of the device, The device body is equipped with the electromagnetic wave sensor described in claim 19, The aforementioned electromagnetic wave sensor is used for target detection and / or communication, thereby providing reference information for the operation of the main body of the device, and is a terminal device.

Images