CN116008984A - Millimeter wave radar human body action recognition method based on mode contour restriction - Google Patents
Millimeter wave radar human body action recognition method based on mode contour restriction Download PDFInfo
- Publication number
- CN116008984A CN116008984A CN202211547678.2A CN202211547678A CN116008984A CN 116008984 A CN116008984 A CN 116008984A CN 202211547678 A CN202211547678 A CN 202211547678A CN 116008984 A CN116008984 A CN 116008984A
- Authority
- CN
- China
- Prior art keywords
- human body
- millimeter wave
- frequency
- wave radar
- radar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000009471 action Effects 0.000 title claims abstract description 35
- 230000033001 locomotion Effects 0.000 claims abstract description 50
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000010586 diagram Methods 0.000 claims abstract description 15
- 230000002159 abnormal effect Effects 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000010606 normalization Methods 0.000 claims abstract description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000002592 echocardiography Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 210000002569 neuron Anatomy 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 238000011160 research Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
A millimeter wave radar human body action recognition method based on mode contour limitation designs human body actions and radar parameters, builds a millimeter wave radar system platform, and changes human body actions by standing in front of a platform acquisition position; and transmitting a linear frequency modulation signal by utilizing a millimeter wave radar, receiving an echo signal containing motion information, carrying out frequency mixing processing on the signal to obtain an intermediate frequency signal, selecting and processing data along a slow time axis by using short-time Fourier transform after preprocessing to generate Doppler time (DT diagram), and carrying out normalization processing. And (3) using a threshold value method to find out a high-power density region, drawing upper and lower boundaries, using a Hampel filter to remove boundary abnormal values, and performing operation with the original DT to obtain an improved DT with a mode profile limit. The invention has wide application prospect and strong practicability, and can be used in the fields of intelligent home, auxiliary driving, game entertainment and the like.
Description
Technical Field
The invention relates to the technical field of man-machine interaction, in particular to a millimeter wave radar human body action recognition method based on mode contour limitation.
Background
The recognition of human actions is a research hotspot for a long time, has research value, and has application prospects in a great number of key fields, such as intelligent human-computer interaction, intelligent home, driving assistance, game entertainment and the like. When sensing and recognizing human body actions, the use of radar as a terminal sensor has many advantages not available from other sensors such as cameras and kinematic sensors. The human body action recognition mechanism based on the radar is that the radar echo signal contains information of human body target movement, and the characteristics of human body actions can be extracted by carrying out necessary signal processing on the echo signal.
In human motion recognition, the whole body motion range is larger, the radar resolution requirement is lower, but the echo structure is complex, and because the trunk motion involves the motion of a plurality of parts of the human body, echo signals of the trunk motion are often signal combinations of a plurality of parts of the human body. Therefore, it is important to process the motion signal to improve the accuracy of the signal and reduce the outlier.
In the technical field of radar human motion recognition, the traditional algorithm sometimes cannot meet the motion recognition performance, and the machine learning and deep learning algorithm breaks through the limitations of the traditional signal processing algorithm and recognition algorithm, so that the method has the advantages which the traditional algorithm does not have in a complex background environment. At present, the human body action recognition method of the millimeter wave radar is mainly based on a deep learning algorithm, a deeper network structure is often needed in a high-performance recognition network, the larger the parameter scale of a model is, the higher the calculation complexity is brought by more required training samples, and the larger the overfitting risk of the model is. Therefore, how to improve recognition accuracy and reduce redundant information in radar human motion recognition, and to achieve lighter weight and high performance in recognition networks is a direction of research.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a millimeter wave radar human body action recognition method based on mode contour restriction, and the proposed contour restriction map can reduce redundant information and remove abnormal points; and the Doppler characteristics are learned and classified by adopting a lightweight multi-branch convolutional neural network of an initial-pooling module and a residual-pooling module. Experimental results show that the method can remarkably reduce the parameter quantity and the calculated quantity while maintaining higher detection precision.
The technical scheme provided by the invention is as follows:
a millimeter wave radar human body action recognition method based on mode contour limitation comprises the following steps:
step 1: designing human body action gestures and radar parameters, and building a millimeter wave radar experimental system platform;
step 2: the human body stands in front of the platform acquisition position, performs five motion changes of standing, walking, lying to sitting and falling, transmits signals by utilizing a millimeter wave radar, then receives echo signals containing motion information, and performs mixing processing on the transmitted signals and the received signals to obtain intermediate frequency signals of human body motion echoes;
step 3: clutter pre-processing the intermediate frequency signal and then selecting and processing the data along the slow time axis using a Short Time Fourier Transform (STFT) to generate a doppler time plot (DT plot) for frequency-time analysis of the doppler features characterizing the different motions;
step 4: the characteristic data set of human body actions is normalized, the frequency and time range are set to be maximum, so that the DT image is not distorted, and the training complexity is reduced;
step 5: the normalized DT-graph may be divided into an upper part with positive frequency and a lower part with negative frequency, respectively, delineating the boundaries of the high power density regions of the upper and lower parts. A thresholding method is used to find the high power density region and delineate the upper and lower boundaries.
Step 6: removing boundary outliers by using a Hampel filter, and performing operation with the original DT to obtain an improved DT with mode profile limitation;
step 7: inputting the DT pattern with limited pattern contour into an improved lightweight multi-branch network for human motion recognition training;
step 8: and calling the trained network model to classify and identify the verification set.
Further, in the step 1, the millimeter wave radar experiment system platform deploys a millimeter wave Frequency Modulation Continuous Wave (FMCW) radar module IWR1442BOOST developed by texas instruments, and the working range of the FMCW radar is from 76GHz to 81GHz, and the parameter settings are shown in table 1.
TABLE 1
A transmitting antenna and four receiving antennas are adopted, and the equivalent bandwidth is as follows:
B eq ＝β×N/f s (1)
wherein B is eq Refer to the equivalent bandwidth, β is the chirp rate, N is the number of effective ADC samples in the chirp, f s For the ADC sampling rate, the resulting range resolution R is about 9.8 cm, calculated as follows
Where c is the speed of light. Maximum speed V max Can be calculated as
The signal emitted by the FMCW radar whose operating frequency varies within the chirp period can be represented as x T (t)＝A T cos(2πf c t+πβt 2 ) (4)
Wherein A is T Is the amplitude of the transmitted wave, f c Is the initial frequency of chirp and β is the slope of chirp.
In step 2, the millimeter wave radar is used to transmit signals, the signal parameters are as described in step 1, the human body performs five motion changes of standing, walking, lying, sitting and falling in front of the radar, each motion lasts for about 1-2s, and the received intermediate frequency radar human body motion echo data is transmitted to a computer for storage by using an ADC1000EVM acquisition board and stored in a complex form.
Further, in the step 3, the intermediate frequency signal is subjected to clutter preprocessing, and then Short Time Fourier Transform (STFT) is used to select and process data along the slow time axis to generate a doppler time map (DT map). The Short Time Fourier Transform (STFT) is mathematically defined as
Where z (t) is the signal to be transformed, w (t) is a sliding window function, f d Representing the Doppler frequency;
the process is as follows:
(a) Dividing original intermediate frequency signal data acquired by an acquisition card into a plurality of channel data according to set radar parameters, storing the channel data into I/Q data in a complex form, namely obtaining a bin file of an intermediate frequency echo, reading data of each antenna from the bin file, and arranging the bin file into a three-dimensional array of [ n_sample, n_chirp and n_frame ], wherein n_frame represents the number of sampled frames, n_chirp represents the total number of chirp contained in each frame, n_sample represents the number of sampling points contained in each chirp, performing frame difference processing on the bin file, and finally outputting two-dimensional echo sequences of I paths and Q paths;
(b) Filtering each echo based on the obtained two-dimensional echo sequence, and performing FFT (fast Fourier transform) on the data to obtain distance distribution information which is accumulated into time-distance characteristics along with time;
(c) Applying a Short Time Fourier Transform (STFT) to the time-distance distribution matrix using window functions of different durations to select and process data along the slow time axis to generate a doppler time map (DT map);
in the step 4, the normalization processing is performed on the characteristic data set of the human motion, and the frequency and time ranges are set to be large values, so that the DT graphs are not distorted, the frequency and time ranges in different DT graphs remain the same, and the input of the classifier must be formatted into a fixed form.
In step 5, the motion doppler feature in the DT graph typically has a continuous high power density point and is centered around the zero frequency point. The specific process of extracting the mode profile limiting graph with the high power density part comprises the following steps: in the DT-plot, positive and negative values of the doppler frequency correspond to the direction of motion away from and towards the radar, respectively. By taking the zero frequency as a boundary, the DT graph can be divided into an upper part with positive frequency and a lower part with negative frequency, drawing the upper and lower boundaries of the high power density region, respectively, finding the high power density region using a threshold method and drawing its boundary, the points with a power density greater than the threshold being classified in the high power density region, the threshold P th Calculated as
Where α ε (0, 1) is a predefined parameter, max { P }, and
is the maximum and average of all power densities.
In step 6, there may be some noise or outliers with high power density outside the high power density region, and when some machine learning algorithm is used for classification, outliers or outliers may cause detection failure, and when the contour is extracted by using the thresholding method, some outliers outside the high power density region may be generated, and in the proposed method, the outliers are removed by using the Hampel filter. Where the median is the median for a given sequence, contour boundaries are delineated in the DT-graph using a threshold extraction method, sharp portions are removed using a Hampel filter, and for a given sequence x1, x2, x3, xn and half-window length K, the response of the Hampel filter is calculated using the following equation:
where the median is the median of a given sequence and n is the decision threshold. Points greater than the resulting value are changed to the median value, the remainder remaining the original value. The method using the Hampel filter can effectively remove sharp parts, effectively reduce abnormal points, and calculate with the original DT diagram to obtain an improved DT diagram with mode profile limitation, and the obtained DT diagram for extracting the mode profile is input into the lightweight multi-branch convolutional neural network provided by the invention.
In the step 7, the feature map with limited mode profile is input into an improved lightweight multi-branch network for human motion recognition to perform human motion recognition. The specific process is as follows: the DT diagram with the mode profile limitation is obtained as a data set and is divided into a training set and a testing set, the training set is input into the lightweight multi-branch convolutional neural network provided by the invention for training, and the network structure is shown in figure 6. Compared with the traditional convolutional neural network, the lightweight multi-branch convolutional neural network provided by the invention is added with a residual layer and an initial layer module. The convolution-pooling module consists of a convolution kernel with a convolution kernel size of 3 and a step size of 1, and a pooling layer with a pool size of 2 and a step size of 2. The residual layer is formed by replacing the convolution kernel size in the convolution layer with 1, the initial layer module comprising four convolution branches followed by a max pooling layer, the convolution branches consisting of one or two convolution layers with different convolution kernel sizes. The number of neurons of the two fully connected layers FC is 1024 and N class All convolution layers are followed by one batch normalization layer and one activation function layer of the ReLU, all pool layers being the same size max-pool layers.
The invention optimizes the two-dimensional feature map generated by radar echo signal processing by using threshold detection and filtering to generate the two-dimensional map with mode contour limitation, and extracts and classifies the features of human actions by adopting an optimized lightweight multi-branch CNN model to improve the recognition performance and achieve higher action recognition rate.
The beneficial effects of the invention are as follows:
1. the invention uses a mode contour limiting method to extract the contour of the input Doppler time chart, remove abnormal values, effectively reduce redundancy and improve the effectiveness and recognition accuracy of the input image.
2. The invention mainly applies the lightweight multi-branch convolutional neural network for improving the convolutional neural network, and obviously reduces the parameter quantity and the calculated quantity on the premise of keeping the identification accuracy.
Drawings
FIG. 1 is a flow chart of an implementation process of a human motion recognition method of the present invention;
fig. 2 is a schematic diagram of five human actions defined by an embodiment of the present invention. Wherein, (a) lie to sit, (b) stand, (c) walk, (d) fall, (e) lie;
figure 3 is a doppler time plot of five human actions defined by an embodiment of the present invention. Wherein, (a) lie to sit, (b) stand, (c) walk, (d) fall, (e) lie;
FIG. 4 is a two-dimensional view of the present invention after use of the pattern profile limiting method in an embodiment;
FIG. 5 is a schematic diagram of the radar signal processing and generation of a pattern profile limit DT pattern of the present invention;
FIG. 6 is a block diagram of a lightweight multi-branch convolutional neural network proposed by the human motion recognition method of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making clear and defining the scope of the present invention.
Referring to fig. 1 to 6, a millimeter wave radar human body motion recognition method based on mode profile limitation includes the following steps:
(1) Designing human body action gestures and radar parameters, and building a millimeter wave radar experimental system platform;
(2) The human body stands in front of the platform acquisition position, performs five motion changes of standing, walking, lying to sitting and falling, transmits signals by utilizing a millimeter wave radar, then receives echo signals containing motion information, and performs mixing processing on the transmitted signals and the received signals to obtain intermediate frequency signals of human body motion echoes;
(3) Clutter pre-processing the intermediate frequency signal and then selecting and processing the data along the slow time axis using a Short Time Fourier Transform (STFT) to generate a doppler time plot (DT plot) for frequency-time analysis of the doppler features characterizing the different motions;
(4) The characteristic data set of human body actions is normalized, the frequency and time range are set to be maximum, so that the DT image is not distorted, and the training complexity is reduced;
(5) The normalized DT-graph may be divided into an upper part with positive frequency and a lower part with negative frequency, respectively, delineating the boundaries of the high power density regions of the upper and lower parts. A thresholding method is used to find the high power density region and delineate the upper and lower boundaries.
(6) Removing boundary outliers by using a Hampel filter, and performing operation with the original DT to obtain an improved DT with mode profile limitation;
(7) Inputting the DT pattern with limited pattern contour into an improved lightweight multi-branch network for human motion recognition training;
(8) And calling the trained network model to classify and identify the verification set.
The deployed experimental system platform radar is a millimeter wave FMCW radar developed by Texas instruments, and the working range of the FMCW radar is from 76GHz to 81GHz. In the invention, a transmitting antenna and four receiving antennas are adopted. The start frequency and chirp slope were set to 77GHz and 33MHz/μm, respectively. The ADC sampling rate is 5MHz and the effective number of ADC samples per chirp is 256. In the invention, a transmitting antenna and four receiving antennas are adopted.
The human body actions defined in the embodiment are shown in fig. 2, and include 5 human body actions of standing, walking, lying to sit and falling, wherein the human body stands at a position of about 150cm at the collecting position of the platform, and the action changes, so that radar echo signals are obtained. To ensure data universality, each action arranged multiple personnel to unevenly sample 500 groups at different angles and distances for a total of 2500 groups of data, as shown in table 2.
TABLE 2
In this embodiment, the data along the slow time axis is selected and processed using a short time fourier transform, and the resulting doppler time plot of 5 human actions is shown in fig. 3.
In this embodiment, the profile of the generated DT graph is extracted by using a threshold method, and then an outlier is removed by using a Hampel filter, so as to generate a DT graph with a mode profile limitation, and the specific process is shown in fig. 5.
In the network training stage of the embodiment, the invention mainly adopts the proposed lightweight multi-branch convolutional neural network, and the structure is shown in fig. 6. And inputting the DT diagram with the mode profile limitation generated after the Doppler time diagram is improved into a lightweight multi-branch convolutional neural network for training. Finally, five human actions were tested in this example, 100 samples for each action, and the results are shown in table 3.
TABLE 3 Table 3
The parameters and model sizes were reduced compared to the other two networks while ensuring verification accuracy, as shown in table 4. The validity of the millimeter wave radar human body action recognition method based on the mode contour limitation is verified.
TABLE 4 Table 4
The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent structures or equivalent processes or direct or indirect application in other related technical fields are included in the scope of the present invention.
Claims (8)
1. The millimeter wave radar human body action recognition method based on mode contour limitation is characterized by comprising the following steps of:
step 1: designing human body action gestures and radar parameters, and building a millimeter wave radar experimental system platform;
step 2: the human body stands in front of the platform acquisition position, five motion changes of standing, walking, lying to sitting and falling are carried out, millimeter wave radar transmitting signals and receiving echo signals containing motion information are utilized for carrying out frequency mixing processing, and intermediate frequency signals of human body motion echoes are obtained;
step 3: clutter pre-processing is performed on the intermediate frequency signal, and then Short Time Fourier Transform (STFT) is used for selecting and processing data along a slow time axis to generate a Doppler time diagram, namely a DT diagram, so that frequency-time analysis is performed on Doppler characteristics representing different motions;
step 4: the characteristic data set of human body actions is normalized, the frequency and time range are set to be maximum, so that the DT image is not distorted, and the training complexity is reduced;
step 5: the normalized DT graph can be divided into an upper part with positive frequency and a lower part with negative frequency, using a thresholding method to find the high power density region and delineate the upper and lower boundaries;
step 6: removing boundary outliers by using a Hampel filter, and performing operation with the original DT to obtain an improved DT with mode profile limitation;
step 7: inputting the DT pattern with limited pattern contour into an improved lightweight multi-branch network for human motion recognition training;
step 8: and calling the trained network model to classify and identify the verification set.
2. The millimeter wave radar human body action recognition method based on mode profile limitation according to claim 1, wherein in the step 1, the radar experiment system platform refers to an IWR1443BOOST millimeter wave FMCW radar developed by texas instruments, the working range is from 76GHz to 81GHz, specific radar parameters are that one transmitting antenna and four receiving antennas are adopted, the initial frequency and chirp slope are respectively set to 77GHz and 33MHz/μs, the ADC sampling rate is 5MHz, the effective ADC sample number of each chirp is 25, and each action is acquired for 50 frames.
3. The millimeter wave radar human body motion recognition method based on mode profile limitation according to claim 1 or 2, wherein in the step 2, the millimeter wave radar is used for transmitting signals, signal parameters are as described in the step 1, a human body performs five motion changes of standing, walking, lying, sitting and falling in front of the radar, each motion lasts for about 1-2s, and the received intermediate frequency radar human body motion echo data is transmitted to a computer for storage by using an ADC1000EVM acquisition board and is stored in a complex form.
4. The millimeter wave radar human motion recognition method based on mode profile limitation according to claim 3, wherein in the step 3, the intermediate frequency signal is subjected to clutter preprocessing, and then Short Time Fourier Transform (STFT) is used to select and process data along a slow time axis to generate a DT graph, and the process is as follows:
(a) Dividing original intermediate frequency signal data acquired by an acquisition card into a plurality of channel data according to set radar parameters, storing the channel data into I/Q data in a complex form to obtain an intermediate frequency echo, then reading data of each antenna from the bin file, and finishing the data into a three-dimensional array of [ n_sample, n_chirp and n_frame ], wherein n_frame represents the number of sampled frames, n_chirp represents the total number of chirp contained in each frame, n_sample represents the number of sampling points contained in each chirp, then carrying out frame difference processing on the sampling points, namely subtracting previous frame data from the next frame data, and finally outputting I-path and Q-path two-dimensional echo sequences after frame difference;
(b) Filtering each echo based on the obtained two-dimensional echo sequence, and performing FFT (fast Fourier transform) on the data to obtain distance distribution information which is accumulated into time-distance characteristics along with time;
(c) A Short Time Fourier Transform (STFT) is applied to the time-distance distribution matrix using window functions of different durations to select and process data along the slow time axis to generate the DT map.
5. The method for recognizing human body motion of millimeter wave radar based on mode profile limitation according to claim 4, wherein in the step 4, the characteristic data set of human body motion is normalized, and the frequency and time ranges are set to be large values, so that the DT graphs are not distorted, the frequency and time ranges in different DT graphs remain the same, and the input of the classifier must be formatted into a fixed form.
6. The method for recognizing human body motion of millimeter wave radar based on mode profile limitation according to claim 4, wherein in the step 5, the motion doppler feature in the DT graph generally has continuous high power density points and is concentrated around zero frequency points, and the specific process of extracting the mode profile limitation graph with high power density parts is as follows: in the DT-plot, positive and negative values of the doppler frequency correspond to the direction of motion away from and towards the radar, respectively; by taking the zero frequency as a boundary, the DT graph can be divided into an upper part with a positive frequency and a lower part with a negative frequency, a high power density region is found using a thresholding method and its upper and lower boundaries are delineated, and points with a power density greater than the threshold are classified in the high power density region.
7. The method for recognizing human body actions of millimeter wave radar based on mode profile limitation according to claim 6, wherein in said step 6, there may be some noise or abnormal points with high power density outside the high power density region; when some machine learning algorithms are used for classification, outliers or noise points can lead to detection failures; when the contours are extracted using a thresholding method, some outliers outside the high power density region are produced; in the method, a Hampel filter is used for removing abnormal values, wherein the median is the median of a given sequence, a threshold extraction method is used for drawing a contour boundary in a DT (digital television) graph, a Hampel filter is used for removing sharp parts, and the Hampel filter is used for carrying out operation with the original DT graph to obtain an improved DT graph with a mode contour limitation; inputting the DT diagram of the obtained extraction mode profile; in a lightweight multi-branch convolutional neural network.
8. The millimeter wave radar human body motion recognition method based on the mode profile limitation according to claim 7, wherein in the step 7, the characteristic diagram of the mode profile limitation is input into an improved lightweight multi-branch network for human body motion recognition to perform human body motion recognition training, and the process is as follows: the DT diagram with the mode profile limitation is obtained as a data set and is divided into a training set and a testing set, the training set is input into the lightweight multi-branch convolutional neural network provided by the invention, a residual error layer and an initial layer module are added to the lightweight multi-branch convolutional neural network compared with the traditional convolutional neural network, the convolutional-pooling module consists of a convolutional kernel with the size of 3 and the step length of 1, and a pooling layer with the size of 2, the residual error layer is formed by replacing the size of the convolutional kernel in the convolutional layer with 1, the initial layer module comprises four convolutional branches and a maximum pooling layer, and the convolutional branches consist of one or two convolutional layers with different convolutional kernel sizes; the number of neurons of the two fully connected layers FC is 1024 and N, respectively class All convolution layers are followed by one batch normalization layer and one activation function layer of the ReLU, all pool layers being the largest pooled layer of the same size.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211547678.2A CN116008984A (en) | 2022-12-05 | 2022-12-05 | Millimeter wave radar human body action recognition method based on mode contour restriction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211547678.2A CN116008984A (en) | 2022-12-05 | 2022-12-05 | Millimeter wave radar human body action recognition method based on mode contour restriction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116008984A true CN116008984A (en) | 2023-04-25 |
Family
ID=86019981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211547678.2A Pending CN116008984A (en) | 2022-12-05 | 2022-12-05 | Millimeter wave radar human body action recognition method based on mode contour restriction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116008984A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107430444A (en) * | 2015-04-30 | 2017-12-01 | 谷歌公司 | For gesture tracking and the tracking of the micromotion based on RF of identification |
| US20190254544A1 (en) * | 2018-02-22 | 2019-08-22 | Vayyar Imaging Ltd. | Detecting and measuring correlated movement with mimo radar |
| US20190317191A1 (en) * | 2018-04-11 | 2019-10-17 | Infineon Technologies Ag | Human Detection and Identification in a Setting Using Millimiter-Wave Radar |
| CN112051567A (en) * | 2020-09-17 | 2020-12-08 | 中南大学 | Human body target micro Doppler frequency estimation method |
| CN114708663A (en) * | 2022-04-12 | 2022-07-05 | 浙江工业大学 | Millimeter wave radar sensing gesture recognition method based on few-sample learning |
-
2022
- 2022-12-05 CN CN202211547678.2A patent/CN116008984A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107430444A (en) * | 2015-04-30 | 2017-12-01 | 谷歌公司 | For gesture tracking and the tracking of the micromotion based on RF of identification |
| US20190254544A1 (en) * | 2018-02-22 | 2019-08-22 | Vayyar Imaging Ltd. | Detecting and measuring correlated movement with mimo radar |
| US20190317191A1 (en) * | 2018-04-11 | 2019-10-17 | Infineon Technologies Ag | Human Detection and Identification in a Setting Using Millimiter-Wave Radar |
| CN112051567A (en) * | 2020-09-17 | 2020-12-08 | 中南大学 | Human body target micro Doppler frequency estimation method |
| CN114708663A (en) * | 2022-04-12 | 2022-07-05 | 浙江工业大学 | Millimeter wave radar sensing gesture recognition method based on few-sample learning |
Non-Patent Citations (2)
| Title |
|---|
| 尹辉斌;许志猛;: "基于FMCW雷达的人体运动轨迹检测系统", 传感器与微系统, no. 09, 26 August 2020 (2020-08-26) * |
| 袁延鑫;孙莉;张群;: "基于卷积神经网络和微动特征的人体步态识别技术", 信号处理, no. 05, 25 May 2018 (2018-05-25) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113156391B (en) | Radar signal multi-dimensional feature intelligent sorting method | |
| CN110363151A (en) | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm | |
| Liu et al. | Deep learning and recognition of radar jamming based on CNN | |
| CN111461037B (en) | End-to-end gesture recognition method based on FMCW radar | |
| CN113313040B (en) | Human body posture identification method based on FMCW radar signal | |
| CN111880157B (en) | Method and system for detecting target in radar image | |
| CN111813222B (en) | Terahertz radar-based fine dynamic gesture recognition method | |
| CN111476058A (en) | Gesture recognition method based on millimeter wave radar | |
| CN111505632A (en) | Ultra-wideband radar action attitude identification method based on power spectrum and Doppler characteristics | |
| CN115063884B (en) | Millimeter wave radar head action recognition method based on multi-domain fusion deep learning | |
| CN113050797A (en) | Method for realizing gesture recognition through millimeter wave radar | |
| CN113762130B (en) | Millimeter wave radar gesture detection and recognition method | |
| CN107679516A (en) | Lower extremity movement recognition methods based on multiple dimensioned Gauss Markov random field model | |
| CN112580486A (en) | Human behavior classification method based on radar micro-Doppler signal separation | |
| CN113064489A (en) | Millimeter wave radar gesture recognition method based on L1-Norm | |
| CN115877376A (en) | Millimeter wave radar gesture recognition method and recognition system based on multi-head self-attention mechanism | |
| CN114814832A (en) | Millimeter wave radar-based real-time monitoring system and method for human body falling behavior | |
| CN115656958A (en) | Detection method and detection device for real-time track initiation and track classification | |
| Lin et al. | Using CNN to classify hyperspectral data based on spatial-spectral information | |
| CN115273229A (en) | Action recognition method and device, storage medium, sensor and vehicle | |
| CN113900101A (en) | Obstacle detection method and device and electronic equipment | |
| Cinelli et al. | Vessel classification through convolutional neural networks using passive sonar spectrogram images | |
| CN113064133A (en) | Sea surface small target feature detection method based on time-frequency domain depth network | |
| CN116008984A (en) | Millimeter wave radar human body action recognition method based on mode contour restriction | |
| CN115909086A (en) | SAR target detection and identification method based on multistage enhanced network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |