Method and system for classifying strides crossing human body in wireless local area network based on BILSTM

Abstract

The invention discloses a BILSTM-based wireless local area network human body crossing stride classification method and system, and the method comprises the steps: transmitting and receiving a signal under an 802.11 protocol frame structure through a pair of radio devices disposed at two sides of a channel, and carrying out the unpacking of a signal received by a receiving end device, and obtaining the channel state information (CSI) of a certain time period; estimating the received CSI to obtain a shortest arrival time delay parameter sequence; using an interference sub-path separation method to obtain CSI after interference multipath removal, and separating the CSI received by the horizontal polarized antenna and the vertical polarized antenna; first principal components of CSI are obtained through principal component analysis, the obtained first principal components of CSI sequences in the horizontal direction and the vertical direction are input into a BILSTM model for learning, and strides crossing a human body in a certain period of time are classified through the trained model. The human body stride sensing function can be achieved through communication signals, and assistance is provided for human body sensing in the everything interconnection era.

Description

Technical Field

[0001] The present invention belongs to the technical field of integrated communication and perception, and relates to a method and system for classifying the stride of a person walking indoors by using a wireless local area network. Background Art

[0002] In view of the 6G vision, the International Mobile Telecommunications 2030 (IMT-2030) has listed communication and perception integration as one of the six major application scenarios of communication technology in the next decade. Communication and perception integration refers to combining communication and perception, two necessary conditions for building an intelligent environment, and using various signals in communication to assist in the perception of various targets and activities in the environment. At the same time, with the advent of the era of the Internet of Everything, the application of wireless LANs has flourished, and the application prospects of communication and perception integration technology based on wireless LANs are broad.

[0003] As the core of perception research, human perception has always attracted the attention of researchers. Stride characteristics refer to the characteristics of the coordination between the two feet of pedestrians in a set of footprints, and are a form of reflection of walking movement characteristics in a set of footprints. Stride mainly includes features such as step length, step width and step angle. Usually, the vertical distance between the same part of the left and right footprints in a set of footprints is defined as a normal step length. Stride length has a certain relationship with height. The normal walking stride length of an adult is about 0.37 times the height. In addition, under certain circumstances, stride length also has a certain relationship with weight, age, physical condition, etc., and is an important physiological characteristic. In order to conduct in-depth research on stride length, the problem of stride length classification came into being. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method and system for classifying human stride across a wireless local area network based on BILSTM, which utilizes communication signals to realize the function of human stride perception and provide assistance for human perception in the era of the Internet of Things.

[0005] Technical solution: The technical solution adopted by the present invention is: a BILSTM-based wireless local area network stride classification method for human body, comprising the following steps:

[0006] Step 1: Unpack the signal received by the dual-polarized antenna at the receiving end on one side of the channel and sent by the dual-polarized antenna at the transmitting end on the other side of the channel to obtain the channel state information (CSI) when the human body passes through a certain period of time;

[0007] Step 2: Estimate the delay parameters of CSI and construct the shortest delay parameter sequence;

[0008] Step 3: Use the interference sub-path separation method to obtain the CSI after removing the interference multipath and separate it into the CSI in the horizontal polarization direction and the CSI in the vertical polarization direction;

[0009] Step 4: Use principal component analysis (PCA) to obtain the first principal component of the CSI sequence, and input the first principal components of the CSI sequence in the horizontal direction and the vertical direction into the bidirectional long short-term memory network BILSTM model for learning;

[0010] Step 5: Use the model trained in step 4 to classify the strides of a person crossing the human body in a certain period of time.

[0011] Preferably, in step 1, one transmitting end corresponds to multiple receiving ends, and signals from the multiple receiving ends are unpacked to obtain multiple groups of CSI.

[0012] Preferably, in step 2, the CSI data received by the dual-polarized antenna at the receiving end is subjected to delay parameter estimation using a frequency domain forward and backward smoothed Multiple Signal Classification (MUSIC) algorithm, and the delay parameter sequence is constructed using the MUSIC algorithm.

[0013] Preferably, in step 3, the multipath components in the CSI are decomposed by trilinear parallel factors, and after removing the interference multipath, the CSI sequence is extracted and separated into CSI in the horizontal polarization direction and CSI in the vertical polarization direction.

[0014] Preferably, in step 4, the data of each receiving end is taken as an independent sample, and after obtaining the first principal component of the CSI in the horizontal polarization direction and the vertical polarization direction through principal component analysis, they are respectively input into the BILSTM network model corresponding to the horizontal polarization direction and the vertical polarization direction for learning and classification.

[0015] Preferably, in step 5, the data in the actual test is input into the model trained in step 4, and the final classification is determined based on the principle of minority obeys majority and horizontal polarization as the main principle.

[0016] A BILSTM-based human stride classification system for wireless local area network crossing, comprising:

[0017] The CSI acquisition module is used to unpack the signal received by the dual-polarized antenna at the receiving end on one side of the channel and sent by the dual-polarized antenna at the transmitting end on the other side of the channel to obtain the channel state information CSI when the human body passes through a certain period of time;

[0018] The delay estimation module is used to estimate the delay parameters of CSI and construct the shortest delay parameter sequence;

[0019] An interference removal module is used to obtain the CSI after removing the interference multipath by using an interference sub-path separation method and separate it into CSI in the horizontal polarization direction and CSI in the vertical polarization direction;

[0020] A model training module is used to obtain the first principal component of the CSI sequence by principal component analysis, and input the first principal component of the CSI sequence in the horizontal direction and the vertical direction into the bidirectional long short-term memory network BILSTM model for learning;

[0021] The stride classification module is used to classify the stride length of a person crossing a human body in a certain period of time through a trained model.

[0022] A computer system comprises a memory, a processor and a computer program / instruction stored in the memory and executable on the processor, wherein when the computer program / instruction is executed by the processor, the steps of the method for classifying human stride in a wireless local area network based on BILSTM are implemented.

[0023] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for classifying human stride in a wireless local area network based on BILSTM are implemented.

[0024] A computer program product comprises a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the steps of the method for classifying human stride length in a wireless local area network based on BILSTM are implemented.

[0025] Beneficial effects: The present invention proposes a method for classifying stride of human body based on BILSTM neural network, which has the following advantages: 1. The method of the present invention uses a radio wave-based method for human body perception. Compared with the computer vision-based solution, it can avoid dependence on light and protect privacy, and is more conducive to practical application; 2. The method of the present invention can use a dual-polarized antenna at the transmitting end and multiple dual-polarized antennas at the receiving end. The antennas are placed side by side along the channel direction, which can improve the data acquisition capability; 3. The method of the present invention can effectively separate the interference sub-paths through the data processed by the forward and backward smoothing MUSIC algorithm in the frequency domain and the trilinear parallel factor decomposition, enhance the characteristics of the signal, and perform principal component analysis and dimensionality reduction processing on the data to reduce the data complexity; 4. Considering that the process of human body passing through the communication link of the transceiver device is a time-related continuous problem, the present invention regards this crossing process as a time series problem, and adopts the BILSTM network, combining the outputs of the forward and backward LSTM networks to capture the bidirectional dependency in the sequence, which has high classification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1The figure is a schematic diagram of the arrangement of the transceiver equipment in the embodiment of the present invention.

[0027] Figure 2 The figure is a flow chart of the method steps of an embodiment of the present invention.

[0028] Figure 3 This is a diagram of the training verification accuracy results of an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0030] In the embodiment of the present invention, the sending end device controls the Universal Software Radio Peripheral (USRP) through the Labview program to use a dual-polarized antenna to send a signal under the 802.11 frame structure, and the receiving end uses the same dual-polarized antenna to collect signals, such as Figure 1 As shown, the transceiver devices are placed on both sides of the channel, and the parameters are shown in Table 1.

[0031] Table 1 Transceiver Parameters

[0032] Carrier frequency Signal bandwidth Send and receive interval Signal transmission rate 5.4GHz 80MHz 2m 100Hz

[0033] like Figure 2 As shown, a method for classifying human stride in a wireless local area network based on BILSTM is disclosed in an embodiment of the present invention, and includes the following steps:

[0034] Step 1: Unpack the signal received by the dual-polarized antenna at the receiving end on one side of the channel and sent by the dual-polarized antenna at the transmitting end on the other side of the channel to obtain the channel state information CSI when the human body passes through a certain period of time.

[0035] In this embodiment, a transmitting device for transmitting signals and a receiving device for receiving signals are placed on both sides of the channel respectively. The transmitting device continuously sends a waveform under the 802.11 frame structure protocol to the antenna with a center frequency of 5.4 GHz as a communication perception signal. The receiving device uses four identical dual-polarized antennas with a center frequency of 5.4 GHz and a bandwidth of 80 MHz to collect signals. The transceiver devices are connected through a rubidium clock and a clock distributor to ensure synchronization of the transceiver devices. The receiving device unpacks the received data according to the 802.11 protocol frame structure preamble code to obtain the CSI matrix.

[0036] Step 2: Estimate the delay parameters of CSI and construct the shortest delay parameter sequence.

[0037] In this embodiment, after obtaining the CSI matrix, in order to obtain the shortest delay parameter sequence, the frequency domain forward and backward smoothing MUSIC algorithm is used to estimate the delay parameters, search the signal delay spectrum to extract the shortest arrival delay, and construct the shortest delay parameter sequence.

[0038] Step 3: Use the interference sub-path separation method to obtain the CSI after removing the interference multipath and separate it into the CSI in the horizontal polarization direction and the CSI in the vertical polarization direction.

[0039] The shortest delay sequence obtained after processing in step 2 still has many multipath components. In this embodiment, the multipath components in the CSI are decomposed by trilinear parallel factors, and the CSI sequence after filtering out multipath interference is extracted and divided into CSI sequences in the horizontal polarization direction and the vertical polarization direction.

[0040] Step 4: Use principal component analysis to obtain the first principal component of the CSI sequence, and input the first principal components of the CSI sequence in the horizontal direction and the vertical direction into the bidirectional long short-term memory network BILSTM model for learning.

[0041] In this embodiment, the principal component analysis method is used to obtain the first principal component components of the CSI sequences in the horizontal polarization direction and the vertical polarization direction respectively. The principal component components perform noise reduction on the original data while retaining the original data information to the greatest extent, thereby reducing the computational burden of the neural network and improving the iteration speed of the neural network.

[0042] When the human body passes through the transceiver equipment, the signal received by the receiving device will change continuously. The process of the human body passing through the transceiver equipment can be regarded as a time series problem.

[0043] The BILSTM network used in this step is a neural network containing three bilstm layers, which is very powerful for time series data. Bilstm is a bidirectional recurrent neural network that can combine the forward and backward LSTM networks to capture the bidirectional dependencies in the sequence. The LSTM layer contains an input gate, a forget gate, and an output gate. The input gate consists of a sigmoid activation function plus a tanh activation function to determine which new information should be added to the memory unit. The forget gate consists of only a sigmoid activation function to determine which old information should be forgotten or removed from the memory unit. The output gate also consists of a sigmoid activation function plus a tanh activation function to determine which information in the memory unit should be output to the hidden state of the current time step. The processed CSI sequence is input into the BILSTM network with 128 hidden units for learning. A dropout layer with a dropout rate of 50% is added after each bilstm layer to prevent overfitting.

[0044] During training, the data from each receiving end is taken as an independent sample, and the first principal components of the CSI in the horizontal polarization direction and the vertical polarization direction are respectively input into the BILSTM network models corresponding to the horizontal polarization direction and the vertical polarization direction for learning and classification.

[0045] Step 5: Use the model trained in step 4 to classify the strides of people in a certain period of time. During the test, the data in the actual test is input into the model trained in step 4, and the final classification is determined based on the principle of minority obeys majority and horizontal polarization as the main principle.

[0046] Experiments were conducted in an indoor environment, with people of different heights walking at 55 cm, 60 cm, 65 cm, 70 cm, and 75 cm steps, respectively. A total of 1,279 sets of data were collected, of which 3 / 4 were used for training and 1 / 4 for validation. In order to evaluate the difference between the model prediction results and the actual results, the cross entropy function was used as the loss function L, which is expressed as follows:

[0047]

[0048] Where N refers to the number of samples in the validation set, L n is the loss of the nth sample, c is the category index, y nc is the indicator symbol, s nc is the probability that sample n is predicted to be category c, and 5 refers to the 5-category stride.

[0049] like Figure 3 As shown, in this embodiment, the model is trained and then verified, and the final accuracy is 92.30%.

[0050] Based on the same inventive concept, an embodiment of the present invention discloses a BILSTM-based stride classification system for a wireless local area network crossing a human body, comprising: a CSI acquisition module, used to unpack a signal received by a dual-polarized antenna at a receiving end on one side of a channel and sent by a dual-polarized antenna at a transmitting end on the other side of a channel, to obtain channel state information CSI when a human body passes through a certain time period; a delay estimation module, used to estimate delay parameters of the CSI and construct a shortest delay parameter sequence; an interference removal module, used to obtain the CSI after removing interference multipaths by using an interference sub-path separation method and separate it into CSI in a horizontal polarization direction and CSI in a vertical polarization direction; a model training module, used to obtain the first principal component of the CSI sequence by using principal component analysis, and input the first principal components of the obtained CSI sequences in the horizontal direction and the vertical direction into a bidirectional long short-term memory network BILSTM model for learning; and a stride classification module, used to classify the stride of a human body passing through a certain time period by using a trained model.

[0051] An embodiment of the present invention also discloses a computer system, including a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor. When the computer program / instruction is executed by the processor, the steps of the method for classifying human stride in a wireless local area network based on BILSTM are implemented.

[0052] The embodiment of the present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for classifying human stride in a wireless local area network based on BILSTM are implemented.

[0053] The embodiment of the present invention also discloses a computer program product, including a computer program / instruction, which implements the steps of the BILSTM-based wireless local area network crossing human stride classification method when executed by a processor. The program / instruction code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program / instruction codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program / instruction code implements the steps of the method of the present invention when executed by the processor or controller. The program / instruction code can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package, or entirely on a remote machine or server. The parts not described in detail in the present invention are all well-known technologies of those skilled in the art.

Claims

1. A BILSTM-based method for classifying human stride in wireless local area network, characterized in that: The following steps are involved: Step 1: Unpack the signal received by the dual-polarized antenna at the receiving end on one side of the channel and sent by the dual-polarized antenna at the transmitting end on the other side of the channel to obtain the channel state information CSI when the human body passes through a certain period of time; Step 2: Estimate the delay parameters of CSI and construct the shortest delay parameter sequence; Step 3: Use the interference sub-path separation method to obtain the CSI after removing the interference multipath and separate it into the CSI in the horizontal polarization direction and the CSI in the vertical polarization direction; Step 4: Use principal component analysis to obtain the first principal component of the CSI sequence, and input the first principal components of the CSI sequence in the horizontal direction and the vertical direction into the bidirectional long short-term memory network BILSTM model for learning; Step 5: Use the model trained in step 4 to classify the strides of a person crossing the human body in a certain period of time.

2. The BILSTM-based wireless local area network stride classification method for human body according to claim 1, characterized in that: In step 1, one transmitting end corresponds to multiple receiving ends, and the signals of the multiple receiving ends are unpacked to obtain multiple groups of CSI.

3. The BILSTM-based wireless local area network stride classification method for human body according to claim 1, characterized in that: In the step 2, the CSI data received by the dual-polarized antenna at the receiving end is subjected to delay parameter estimation by using a multiple signal classification (MUSIC) algorithm with forward and backward smoothing in the frequency domain, and the delay parameter sequence is constructed using the algorithm.

4. The method for classifying human stride in wireless local area network based on BILSTM according to claim 1, characterized in that: In step 3, the multipath components in the CSI are decomposed by trilinear parallel factors, and the CSI sequence is extracted after removing the interference multipath and separated into the CSI in the horizontal polarization direction and the CSI in the vertical polarization direction.

5. The method for classifying human stride in wireless local area network based on BILSTM according to claim 1, characterized in that: In step 4, the data of each receiving end is taken as an independent sample, and after the first principal components of the CSI in the horizontal polarization direction and the vertical polarization direction are obtained through principal component analysis, they are respectively input into the BILSTM network models corresponding to the horizontal polarization direction and the vertical polarization direction for learning and classification.

6. The method for classifying human stride in wireless local area network based on BILSTM according to claim 1, characterized in that: In step 5, the data in the actual test is input into the model trained in step 4, and the final classification is determined based on the principle of minority obeys majority and horizontal polarization as the main principle.

7. A BILSTM-based wireless local area network stride classification system for human body, characterized in that: include: The CSI acquisition module is used to unpack the signal received by the dual-polarized antenna at the receiving end on one side of the channel and sent by the dual-polarized antenna at the transmitting end on the other side of the channel to obtain the channel state information CSI when the human body passes through a certain period of time; The delay estimation module is used to estimate the delay parameters of CSI and construct the shortest delay parameter sequence; An interference removal module is used to obtain the CSI after removing the interference multipath by using an interference sub-path separation method and separate it into CSI in the horizontal polarization direction and CSI in the vertical polarization direction; A model training module is used to obtain the first principal component of the CSI sequence by principal component analysis, and input the first principal component of the CSI sequence in the horizontal direction and the vertical direction into the bidirectional long short-term memory network BILSTM model for learning; The stride classification module is used to classify the stride length of a person crossing a human body in a certain period of time through a trained model.

8. A computer system comprising a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor, characterized in that: When the computer program / instructions are executed by the processor, the steps of the method for classifying human stride through a wireless local area network based on BILSTM according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for classifying human stride in a wireless local area network based on BILSTM are implemented according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the steps of the method for classifying human stride through a wireless local area network based on BILSTM according to any one of claims 1 to 6 are implemented.

Images