People flow counting method, people flow counting model generation method, people flow counting model generation device, people flow counting equipment and medium

Abstract

The invention provides a people flow number statistical method, a people flow number statistical model generation method, a people flow number statistical model generation device, equipment and a medium, and aims to solve the problem of low people flow number statistical accuracy in the prior art. The method comprises the following steps: unpacking a received wireless signal to obtain an initial channel state information matrix; performing multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix; performing amplitude extraction according to the target channel state information matrix to obtain amplitude information; performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information; processing the first principal component information by adopting a short-time Fourier transform algorithm to obtain a time-frequency spectrogram; and training a target people flow number statistical model based on a data set to obtain a pre-trained target people flow number statistical model, the data set comprising the time-frequency spectrogram and a pre-acquired label. According to the invention, the accuracy of people flow counting can be improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method for counting the number of people flow and a method, device, equipment and medium for generating a model thereof. Background Art

[0002] The existing technology for counting the number of people is mainly based on computer vision. Crowd estimation is an application of crowd estimation in special scenarios. The more mature solution is to use computer vision-based methods to estimate crowd density. The mainstream research method of this solution is to detect or track human bodies in images captured by cameras by building neural networks. The specific implementation method is to perform foreground segmentation on the crowd in the image captured by the camera.

[0003] However, when the crowd density is high, the image overlaps, which makes it more difficult to segment people. There is a high probability that the number of people counted will be wrong, especially in scenes with high crowd density and serious overlap of image overlaps. The current technology has a low accuracy rate in counting the number of people. Summary of the invention

[0004] The embodiment of the present invention provides a method for counting the number of people and a method, device, equipment and medium for generating a model thereof, so as to solve the problem of low accuracy of counting the number of people in the prior art.

[0005] To solve the above technical problems, the present invention is achieved as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for training a pedestrian flow statistics model, comprising:

[0007] Unpacking the received wireless signal to obtain an initial channel state information matrix;

[0008] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0009] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0010] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0011] Using a short-time Fourier transform algorithm to process the first principal component information to obtain a time-frequency spectrum diagram;

[0012] The target pedestrian flow number statistical model is trained based on the data set to obtain a pre-trained target pedestrian flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0013] Optionally, filtering out multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes:

[0014] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0015] constructing a parallel factorization model according to the delay parameters;

[0016] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0017] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0018] Optionally, the training process of the target pedestrian flow number statistical model based on the data set to obtain a pre-trained target pedestrian flow number statistical model includes:

[0019] Dividing the data set into a training set and a validation set according to a preset ratio;

[0020] Inputting the training set and the verification set into a target pedestrian flow number statistical model;

[0021] Training the target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0022] Extracting training data from the initial pedestrian flow number statistical model to determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0023] The initial pedestrian flow number statistical model is adjusted according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

[0024] In a second aspect, an embodiment of the present invention provides a method for counting the number of people, including:

[0025] Acquire wireless signals in a preset area within a preset time period;

[0026] Processing the wireless signal to obtain a time-frequency spectrum diagram;

[0027] The time-frequency spectrum diagram is input into a pre-trained target crowd number statistical model to obtain the crowd number in a preset area within a preset time period. The pre-trained target crowd number statistical model is obtained based on the crowd number statistical model training method described in the first aspect above.

[0028] Optionally, the processing the wireless signal to obtain a time-frequency spectrum diagram includes:

[0029] Unpacking the wireless signal to obtain an initial channel state information matrix;

[0030] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0031] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0032] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0033] The first principal component information is processed using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0034] Optionally, filtering out multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes:

[0035] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0036] constructing a parallel factorization model according to the delay parameters;

[0037] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0038] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0039] In a third aspect, an embodiment of the present invention provides a pedestrian flow statistical model training device, including:

[0040] An unpacking module, used to unpack the received wireless signal to obtain an initial channel state information matrix;

[0041] A filtering module, used to filter out multipath components from the initial channel state information matrix to obtain a target channel state information matrix;

[0042] An extraction module, used to perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0043] An analysis module, used for performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0044] A processing module, used for processing the first principal component information by using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram;

[0045] A training module is used to train a target pedestrian flow statistical model based on a data set to obtain a pre-trained target pedestrian flow statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0046] Optionally, the filtering module includes:

[0047] A processing unit, configured to process the initial channel state information matrix using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0048] A construction unit, configured to construct a parallel factor decomposition model according to the delay parameters;

[0049] The decomposition unit is used to iteratively decompose the initial channel state information matrix using the parallel factor decomposition model to obtain a target channel state information matrix.

[0050] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0051] Optionally, the training module includes:

[0052] A division unit, used for dividing the data set into a training set and a validation set according to a preset ratio;

[0053] A learning unit, used for inputting the training set and the verification set into a target passenger flow number statistical model;

[0054] A training unit, used for training a target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0055] A determination unit, used to extract training data from the initial pedestrian flow number statistical model, and determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0056] An adjustment unit is used to adjust the initial pedestrian flow number statistical model according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

[0057] In a fourth aspect, an embodiment of the present invention provides a device for counting the number of people, including:

[0058] An acquisition module, used to acquire wireless signals in a preset area within a preset time period;

[0059] A processing module, used for processing the wireless signal to obtain a time-frequency spectrum diagram;

[0060] A determination module is used to input the time-frequency spectrum diagram into a pre-trained target crowd number statistical model to obtain the number of people in a preset area within a preset time period. The pre-trained target crowd number statistical model is obtained based on the crowd number statistical model training method described in the first aspect above.

[0061] Optionally, the processing module includes:

[0062] An unpacking unit, configured to unpack the wireless signal to obtain an initial channel state information matrix;

[0063] A filtering unit, configured to filter the multipath components of the initial channel state information matrix to obtain a target channel state information matrix;

[0064] an extraction unit, configured to perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0065] an analyzing unit, configured to perform principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0066] The processing unit is used to process the first principal component information by adopting a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0067] Optionally, the filtering unit is specifically used for:

[0068] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0069] constructing a parallel factorization model according to the delay parameters;

[0070] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0071] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0072] In a fifth aspect, an embodiment of the present invention provides an electronic device, including a processor, wherein the processor is configured to:

[0073] Unpacking the received wireless signal to obtain an initial channel state information matrix;

[0074] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0075] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0076] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0077] Using a short-time Fourier transform algorithm to process the first principal component information to obtain a time-frequency spectrum diagram;

[0078] The target pedestrian flow number statistical model is trained based on the data set to obtain a pre-trained target pedestrian flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0079] Optionally, the processor is specifically configured to:

[0080] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0081] constructing a parallel factorization model according to the delay parameters;

[0082] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0083] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0084] Optionally, the processor is specifically configured to:

[0085] Dividing the data set into a training set and a validation set according to a preset ratio;

[0086] Inputting the training set and the verification set into a target pedestrian flow number statistical model;

[0087] Training the target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0088] Extracting training data from the initial pedestrian flow number statistical model to determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0089] The initial pedestrian flow number statistical model is adjusted according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

[0090] In a sixth aspect, an embodiment of the present invention provides an electronic device, including a processor, wherein the processor is configured to:

[0091] Acquire wireless signals in a preset area within a preset time period;

[0092] Processing the wireless signal to obtain a time-frequency spectrum diagram;

[0093] The time-frequency spectrum diagram is input into a pre-trained target crowd number statistical model to obtain the crowd number in a preset area within a preset time period, wherein the pre-trained target crowd number statistical model is obtained based on the crowd number statistical model training method described in any one of claims 1-4.

[0094] Optionally, the processor is specifically configured to:

[0095] Unpacking the wireless signal to obtain an initial channel state information matrix;

[0096] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0097] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0098] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0099] The first principal component information is processed using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0100] Optionally, the processor is specifically configured to:

[0101] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0102] constructing a parallel factorization model according to the delay parameters;

[0103] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0104] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0105] In a seventh aspect, an embodiment of the present invention provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for training a model of pedestrian flow statistics as described in the first aspect are implemented; or, when the program is executed by the processor, the steps of the method for counting the number of pedestrian flows as described in the second aspect are implemented.

[0106] In an eighth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for training a model of pedestrian flow statistics as described in the first aspect are implemented; or, when the computer program is executed by a processor, the steps of the method for counting the number of pedestrian flows as described in the second aspect are implemented.

[0107] In an embodiment of the present invention, the above-mentioned crowd flow statistical model training method can extract amplitude information from the channel state information matrix of the wireless signal after filtering out the multipath component, use principal component analysis to reduce the dimension of the amplitude information to obtain the first principal component information, and use short-time Fourier transform to obtain the time-frequency spectrum of the crowd flow, and then train the target crowd flow statistical model based on the time-frequency spectrum and the pre-acquired labels, and finally obtain the pre-trained target crowd flow statistical model, so as to realize the feature extraction of the crowd flow from the channel state information matrix of the wireless signal, and because the radio electromagnetic waves are penetrable and less affected by the environment, the pre-trained target crowd flow statistical model can be used in the scene of crowd density to improve the accuracy of crowd flow statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0109] Figure 1 It is a flow chart of a method for training a pedestrian flow statistics model provided by an embodiment of the present invention;

[0110] Figure 2 It is a structural schematic diagram of a target passenger flow number statistical model provided by an embodiment of the present invention;

[0111] Figure 3 It is a schematic diagram of the accuracy of a target passenger flow number statistical model in an iterative process provided by an embodiment of the present invention;

[0112] Figure 4 It is a schematic diagram of loss values ​​of a target passenger flow number statistical model in an iteration process provided by an embodiment of the present invention;

[0113] Figure 5 is a flow chart of a method for counting the number of people provided by an embodiment of the present invention;

[0114] Figure 6 is a flow chart of another method for counting the number of people provided by an embodiment of the present invention;

[0115] Figure 7 It is a structural schematic diagram of a pedestrian flow statistical model training device provided by an embodiment of the present invention;

[0116] Figure 8 It is a structural schematic diagram of a device for counting the number of people provided by an embodiment of the present invention;

[0117] Fig. 9 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention;

[0118] Fig.10 It is a schematic diagram of the structure of another electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0119] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0120] The embodiment of the present invention provides a method for counting the number of people and a method, device, equipment and medium for generating a model thereof, so as to solve the problem of low accuracy of counting the number of people in the prior art.

[0121] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a method for training a pedestrian flow statistics model provided by an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0122] Step 101: unpack the received wireless signal to obtain an initial channel state information matrix.

[0123] Specifically, the wireless signal may be a WIFI signal sent by a commercial access point (AP) or a router, or may be a wireless signal simulated by a transceiver device.

[0124] Exemplarily, the transmission waveform of the transceiver link may be an IEEE 802.11ac format waveform, and the receiving end may be a unit wavelength uniform array commonly found in commercial APs or routers. In some embodiments, the transceiver devices may be placed on both sides of a typical indoor corridor environment, and the configuration of each parameter is shown in Table 1:

[0125] Table 1

[0126] Carrier frequency Signal bandwidth Receiving antenna spacing Send and receive interval Signal transmission rate 5.4GHz 80MHz 5.4cm 2.5m 100Hz

[0127] The transmitter for transmitting the sensing signal and the receiver for receiving the sensing signal are placed on both sides of the indoor corridor to be tested. The wireless signal transmitter continuously sends 802.11ac waveforms, which serve as both communication and sensing signals, to the wireless LAN access point antenna array at a certain signal transmission rate.

[0128] In some embodiments, the receiving end can use an array form with a spacing of about one wavelength commonly seen in existing commercial APs and routers to simulate the communication capacity, performance and architecture of existing WIFI signals to achieve the integration of communication and perception. After receiving the wireless signal, the receiving antenna array can unpack the preamble code in the physical layer frame structure of the 802.11ac protocol to obtain the channel state information matrix.

[0129] Step 102: Filter out multipath components of the initial channel state information matrix to obtain a target channel state information matrix.

[0130] Specifically, the multipath component may refer to a signal component generated by multiple reflections, attenuations and refractions during the propagation of a signal from one location to another. The target channel state information matrix is ​​a channel state information matrix obtained by filtering out the multipath component from the initial channel state information matrix.

[0131] Step 103: Perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix.

[0132] Specifically, the target channel state information matrix may include amplitude information, phase information, etc., and the amplitude information may be used to reflect the attenuation of the signal during the transmission process.

[0133] Step 104: Perform principal component analysis on the amplitude information to obtain first principal component information of the amplitude information.

[0134] Specifically, the first principal component information is obtained by using the principal component analysis (PCA), which can retain the characteristics of the original information to the greatest extent possible, and realize denoising of the original data, reduce the computational burden, and improve the data processing speed and the iteration speed of the neural network.

[0135] Step 105: Use a short-time Fourier transform algorithm to process the first principal component information to obtain a time-frequency spectrum diagram.

[0136] Specifically, since the entry and exit activities vertically crossing the transceiver link in any direction will cause continuous fluctuations in the antenna response amplitude, the frequency domain information of the antenna response amplitude contains a large number of features. The short-time Fourier transform (STFT) can be used to analyze the frequency characteristics of the timing antenna response amplitude.

[0137] The short-time Fourier transform algorithm is essentially a windowed Fourier transform. The traditional Fourier transform can only analyze the frequency characteristics of the entire signal, but abandons the time domain characteristics. When the frequency characteristics of the signal change over time, the traditional Fourier transform cannot effectively analyze its frequency characteristics. The short-time Fourier transform can slide the time window and analyze the spectrum of the signal in the window, and then splice them in the time domain to obtain the time-frequency spectrum, that is, on the one hand, the frequency domain characteristics of the signal are extracted, and on the other hand, the changes in the frequency characteristics of the signal in the time domain are retained. The time-frequency spectrum diagram includes the time domain, frequency domain and real and imaginary parts of the STFT matrix.

[0138] Exemplarily, the formula of the short-time Fourier transform algorithm is as follows:

[0139]

[0140] Among them, g(n) is the window function, R represents the sampling point interval of the window sliding, fs is the sampling rate of the signal, and the specific parameters of the short-time Fourier transform are shown in Table 2:

[0141] Table 2

[0142] t g(n) W R NDFT fs 300 Blackman Window 28 4 256 100HZ

[0143] Step 106: training the target crowd flow number statistical model based on the data set to obtain a pre-trained target crowd flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0144] Specifically, the data set may be a data set for training a target crowd number statistical model, and the data set may include the time-frequency spectrum and pre-acquired labels. The labels may be generated by calling a camera for the time period and number of people crossing the scene using the target detection algorithm YOLOv5 while receiving wireless signals, for neural network learning and verification.

[0145] The target crowd number statistical model can be an untrained convolutional neural network (CNN) model. The CNN model can include a convolutional layer, a pooling layer, a fully connected layer and a classification layer. The classification layer can use a soft maximization softmax function. The target crowd number statistical model is trained based on the data set, and the time domain, frequency domain and real and imaginary parts of the STFT matrix in the data set and the pre-acquired labels are respectively regarded as the length, width, channel and label of the two-dimensional CNN input picture for learning. The pre-trained target crowd number statistical model can be a trained crowd number statistical model.

[0146] For example, Figure 2 is a schematic diagram of a target passenger flow number statistical model provided by an embodiment of the present invention. Figure 2 As shown in the figure, the CNN network structure contains two convolutional layers to fully extract the time-frequency spectrum features. The maximum pooling layer is used to achieve downsampling and provide nonlinearity for the network together with the Rectified Linear Unit (ReLU) layer. After two layers of convolution and pooling, the output is transferred to two fully connected layers through the flat layer. The output size of the last fully connected layer is the number of categories. Finally, it can be classified after passing through the soft maximization (Softmax) layer. In order to unify the matrix size, the input STFT matrix size is specified to be 129×65. The specific size parameters of the CNN network structure are shown in Table 3;

[0147] Table 3

[0148]

[0149] In an embodiment of the present invention, the above-mentioned crowd flow statistical model training method can extract amplitude information from the channel state information matrix of the wireless signal after filtering out the multipath component, use principal component analysis to reduce the dimension of the amplitude information to obtain the first principal component information, and use short-time Fourier transform to obtain the time-frequency spectrum of the crowd flow, and then train the target crowd flow statistical model based on the time-frequency spectrum and the pre-acquired labels, and finally obtain the pre-trained target crowd flow statistical model, so as to realize the feature extraction of the crowd flow from the channel state information matrix of the wireless signal, and because the radio electromagnetic waves are penetrable and less affected by the environment, the pre-trained target crowd flow statistical model can be used in the scene of crowd density to improve the accuracy of crowd flow statistics.

[0150] Optionally, filtering out multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes:

[0151] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0152] constructing a parallel factorization model according to the delay parameters;

[0153] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0154] Specifically, the multiple signal classification (MUSIC) algorithm can be used to estimate the delay parameter, and the delay parameter can be the difference in relative delay between the characteristic signal and the interference signal. The parallel factor decomposition model can be used to separate the signal and accurately extract the characteristic signal and the interference signal.

[0155] In an embodiment of the present invention, the above-mentioned crowd flow statistical model training method can obtain the target channel state information matrix by iteratively decomposing the initial channel state information matrix using the MUSIC algorithm and the parallel factor decomposition model, thereby filtering out the multipath components of the initial channel state information matrix, and then effectively eliminating multipath interference, thereby improving the accuracy of counting the number of people using the crowd flow statistical model.

[0156] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0157] Specifically, the frequency domain forward and backward smoothed MUSIC (FBS-MUSIC) algorithm performs forward and backward smoothing on the frequency domain estimation model based on the existing MUSIC algorithm, which can effectively suppress the cross-correlation between signals, improve the accuracy of the forward and backward smoothing estimation model, and make the delay parameters more accurate.

[0158] In an embodiment of the present invention, the above-mentioned crowd flow statistical model training method can process the initial channel state information matrix by adopting the MUSIC algorithm with forward and backward smoothing in the frequency domain to obtain the delay parameters, thereby improving the accuracy of delay parameter estimation, thereby further improving the accuracy of counting the number of people flow using the crowd flow statistical model.

[0159] Optionally, the training process of the target pedestrian flow number statistical model based on the data set to obtain a pre-trained target pedestrian flow number statistical model includes:

[0160] Dividing the data set into a training set and a validation set according to a preset ratio;

[0161] Inputting the training set and the verification set into a target pedestrian flow number statistical model;

[0162] Training the target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0163] Extracting training data from the initial pedestrian flow number statistical model to determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0164] The initial pedestrian flow number statistical model is adjusted according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

[0165] Specifically, the training set may be a data set used to train the model, and the validation set may be a data set used to adjust the model. During the iteration of the model, the parameters in the model are adaptively updated to improve the accuracy of the model.

[0166] The accuracy rate can be obtained by comparing the labels obtained by the initial pedestrian flow statistical model during the verification process to the data. The loss value can be obtained by using a loss function to evaluate the difference between the predicted result and the actual result.

[0167] For example, an experiment was conducted in a typical indoor corridor environment, and a total of 2000 sets of data were collected. 3 / 4 of the data were used to train the model, and the remaining 1 / 4 of the data were used for verification. The experimental data included situations where 1, 2, and 3 people crossed at the same time. The cross entropy function was used as the loss function L to evaluate the difference between the predicted results and the actual results. The loss of the model was calculated based on the predicted probability as follows:

[0168]

[0169] Among them, L n is the loss of the nth sample, c is the category index, y ic is the indicator symbol. When the true category of sample n is c, y ic =1, otherwise y ic =0,s nc is the probability that sample n is predicted to be category c. After training the model, the model can be verified to obtain the accuracy and loss value. Figure 3 is a schematic diagram of the accuracy of a target passenger flow number statistical model in an iterative process provided by an embodiment of the present invention, Figure 4 FIG. 4 is a schematic diagram of the loss value of a target passenger flow number statistical model in an iterative process provided by an embodiment of the present invention, such as Figure 3 and Figure 4 As shown, the final accuracy of the model is 95% and the final loss value is 0.2.

[0170] It should be noted that, in the process of training the target pedestrian flow number statistical model based on the data set, the model can be trained and verified multiple times and iterated continuously to improve the accuracy of the pre-trained target pedestrian flow number statistical model.

[0171] In an embodiment of the present invention, the above-mentioned pedestrian flow statistical model training method can divide the data set into a training set and a verification set according to a preset ratio, and train and verify the model respectively, so that the model can better fit the input data, thereby effectively improving the accuracy of model training and making the model more stable.

[0172] See also Figure 5 , Figure 5 FIG. 1 is a flow chart of a method for counting the number of people provided by an embodiment of the present invention. Figure 5 As shown, the method comprises the following steps:

[0173] Step 501: Acquire wireless signals in a preset area within a preset time period.

[0174] The preset time period can be selected according to specific needs, and the time period to be counted can be divided into preset time periods for statistics. The preset area can be a typical indoor corridor environment, a hall entrance, a room entrance, or other areas.

[0175] Step 502: Process the wireless signal to obtain a time-frequency spectrum diagram.

[0176] The wireless signal may be a WIFI signal sent by a commercial access point (AP) or a router, or may be a wireless signal simulated by a transceiver. The time-frequency spectrum may be the time domain, frequency domain and real and imaginary parts of the STFT matrix corresponding to the wireless signal.

[0177] Step 503: input the time-frequency spectrum into a pre-trained target crowd number statistical model to obtain the crowd number in a preset area within a preset time period, wherein the pre-trained target crowd number statistical model is obtained based on the above-mentioned crowd number statistical model training method.

[0178] In an embodiment of the present invention, the above-mentioned method for counting the number of people can obtain wireless signals in a preset area within a preset time period, and process the wireless signals to obtain a time-frequency spectrum of crowd flow, and then input the time-frequency spectrum into a pre-trained target crowd number statistical model to obtain the number of people in the preset area within the preset time period. Since radio electromagnetic waves are penetrating and less affected by the environment, the pre-trained target crowd number statistical model is used to obtain the number of people in a scene with high crowd density, so that the accuracy of crowd number statistics is higher.

[0179] Optionally, the processing the wireless signal to obtain a time-frequency spectrum diagram includes:

[0180] Unpacking the wireless signal to obtain an initial channel state information matrix;

[0181] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0182] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0183] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0184] The first principal component information is processed using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0185] Specifically, the multipath component may refer to a signal component generated by multiple reflections, attenuations, and refractions during the propagation of a signal from one location to another. The amplitude information may be used to reflect the attenuation of the signal during the transmission process. By using the first principal component information obtained after the principal component analysis, the characteristics of the original information can be retained to the greatest extent possible, and the denoising of the original data can be achieved, reducing the computational burden, and at the same time improving the data processing speed and the iteration speed of the neural network. The short-time Fourier transform can analyze the frequency characteristics of the timing antenna response amplitude.

[0186] In an embodiment of the present invention, the above-mentioned method for counting the number of people can extract amplitude information from the channel state information matrix of the wireless signal after filtering out the multipath component, use principal component analysis to reduce the dimension of the amplitude information to obtain the first principal component information, and use short-time Fourier transform to obtain the time-frequency spectrum of the crowd flow, and then train the target crowd flow number statistical model based on the time-frequency spectrum and the pre-acquired labels, and finally obtain the pre-trained target crowd flow number statistical model, so as to realize the feature extraction of the crowd flow from the channel state information matrix of the wireless signal. Since the extracted features are prominent, the computational burden of the model can be reduced.

[0187] Optionally, filtering out multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes:

[0188] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0189] constructing a parallel factorization model according to the delay parameters;

[0190] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0191] Specifically, the MUSIC algorithm can be used to estimate a delay parameter, and the delay parameter can be the difference in relative delay between a characteristic signal and an interference signal. The parallel factor decomposition model can be used to separate signals and accurately extract characteristic signals and interference signals.

[0192] In an embodiment of the present invention, the above-mentioned method for counting the number of people can obtain a target channel state information matrix by iteratively decomposing the initial channel state information matrix using a MUSIC algorithm and a parallel factor decomposition model, thereby filtering out the multipath components of the initial channel state information matrix, and further effectively eliminating multipath interference, thereby improving the accuracy of counting the number of people.

[0193] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0194] Specifically, the frequency domain forward and backward smoothing MUSIC algorithm performs forward and backward smoothing on the frequency domain estimation model based on the existing MUSIC algorithm, which can effectively suppress cross-correlation between signals, improve the accuracy of the forward and backward smoothing estimation model, and make the delay parameters more accurate.

[0195] In an embodiment of the present invention, the above-mentioned method for counting the number of people can process the initial channel state information matrix by adopting the MUSIC algorithm with forward and backward smoothing in the frequency domain to obtain the delay parameter, thereby improving the accuracy of delay parameter estimation, thereby further improving the accuracy of counting the number of people.

[0196] For example, Figure 6 A flowchart of another method for counting the number of people provided by an embodiment of the present invention, such as Figure 6 As shown, the method includes:

[0197] Step 1: The transmitters on both sides of the corridor transmit sensing signals, and the receivers receive and unpack the signals to obtain channel state information (CSI), and use the camera to obtain the tags.

[0198] Step 2: According to the CSI, the forward and backward smoothing MUSIC algorithm is used to estimate the delay parameters and the parallel factor model is used to filter out multipath interference and extract the amplitude of the CSI;

[0199] Step 3: Obtain the first principal component of the CSI amplitude sequence by principal component analysis;

[0200] Step 4: Use STFT to analyze the frequency characteristics of the antenna response amplitude for the first principal component of the CSI amplitude sequence;

[0201] Step 5: The time domain, frequency domain, real and imaginary parts and labels of the obtained STFT matrix are regarded as the length, width, channel and label of the two-dimensional CNN input image for learning;

[0202] Step 6: Use the trained model to identify the number of people crossing at a certain moment.

[0203] See also Figure 7 , Figure 7 FIG. 1 is a schematic diagram of a structure of a pedestrian flow statistical model training device provided by an embodiment of the present invention. Figure 7 As shown, the pedestrian flow statistical model training device 700 includes:

[0204] The unpacking module 701 is used to unpack the received wireless signal to obtain an initial channel state information matrix;

[0205] A filtering module 702 is used to filter out multipath components from the initial channel state information matrix to obtain a target channel state information matrix;

[0206] An extraction module 703 is used to perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0207] An analysis module 704 is used to perform principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0208] A processing module 705 is used to process the first principal component information using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram;

[0209] The training module 706 is used to train the target pedestrian flow number statistical model based on the data set to obtain a pre-trained target pedestrian flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0210] Optionally, the filtering module 702 includes:

[0211] A processing unit, configured to process the initial channel state information matrix using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0212] A construction unit, configured to construct a parallel factor decomposition model according to the delay parameters;

[0213] The decomposition unit is used to iteratively decompose the initial channel state information matrix using the parallel factor decomposition model to obtain a target channel state information matrix.

[0214] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0215] Optionally, the training module 706 includes:

[0216] A division unit, used for dividing the data set into a training set and a validation set according to a preset ratio;

[0217] A learning unit, used for inputting the training set and the verification set into a target passenger flow number statistical model;

[0218] A training unit, used for training a target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0219] A determination unit, used to extract training data from the initial pedestrian flow number statistical model, and determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0220] The adjustment unit is used to adjust the initial pedestrian flow number statistical model according to the accuracy, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model. Figure 8 , Figure 8 is a schematic diagram of the structure of a device for counting the number of people provided by an embodiment of the present invention, such as Figure 8 As shown, the pedestrian flow counting device 800 includes:

[0221] An acquisition module 801 is used to acquire wireless signals in a preset area within a preset time period;

[0222] The processing module 802 is used to process the wireless signal to obtain a time-frequency spectrum diagram;

[0223] The determination module 803 is used to input the time-frequency spectrum diagram into a pre-trained target crowd number statistical model to obtain the number of people in a preset area within a preset time period. The pre-trained target crowd number statistical model is obtained based on the above-mentioned crowd number statistical model training method.

[0224] Optionally, the processing module 802 includes:

[0225] An unpacking unit, configured to unpack the wireless signal to obtain an initial channel state information matrix;

[0226] A filtering unit, configured to filter the multipath components of the initial channel state information matrix to obtain a target channel state information matrix;

[0227] an extraction unit, configured to perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0228] an analyzing unit, configured to perform principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0229] The processing unit is used to process the first principal component information by adopting a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0230] Optionally, the filtering unit is specifically used for:

[0231] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0232] constructing a parallel factorization model according to the delay parameters;

[0233] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0234] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0235] For details, see Fig. 9 As shown, an embodiment of the present invention further provides an electronic device, including a bus 901 , a transceiver 902 , an antenna 903 , a bus interface 904 , a processor 905 and a memory 906 .

[0236] The processor 905 is configured to:

[0237] Unpacking the received wireless signal to obtain an initial channel state information matrix;

[0238] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0239] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0240] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0241] Using a short-time Fourier transform algorithm to process the first principal component information to obtain a time-frequency spectrum diagram;

[0242] The target pedestrian flow number statistical model is trained based on the data set to obtain a pre-trained target pedestrian flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

[0243] exist Fig. 9In the embodiment, the bus architecture (represented by bus 901) may include any number of interconnected buses and bridges, and bus 901 links various circuits including one or more processors represented by processor 905 and memory represented by memory 906. Bus 901 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 904 provides an interface between bus 901 and transceiver 902. Transceiver 902 may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. Data processed by processor 905 is transmitted on a wireless medium via antenna 903, and further, antenna 903 also receives data and transmits the data to processor 905.

[0244] The processor 905 is responsible for managing the bus 901 and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 906 can be used to store data used by the processor 905 when performing operations.

[0245] Optionally, the processor 905 may be a CPU, an ASIC, an FPGA or a CPLD.

[0246] Optionally, the processor 905 is specifically configured to:

[0247] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0248] constructing a parallel factorization model according to the delay parameters;

[0249] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0250] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0251] Optionally, the processor 905 is specifically configured to:

[0252] Dividing the data set into a training set and a validation set according to a preset ratio;

[0253] Inputting the training set and the verification set into a target pedestrian flow number statistical model;

[0254] Training the target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model;

[0255] Extracting training data from the initial pedestrian flow number statistical model to determine the accuracy and loss value of the initial pedestrian flow number statistical model;

[0256] The initial pedestrian flow number statistical model is adjusted according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

[0257] For details, see Fig.10 As shown, an embodiment of the present invention further provides another electronic device, including a bus 1001 , a transceiver 1002 , an antenna 1003 , a bus interface 1004 , a processor 1005 and a memory 1006 .

[0258] The processor 1005 is configured to:

[0259] Acquire wireless signals in a preset area within a preset time period;

[0260] Processing the wireless signal to obtain a time-frequency spectrum diagram;

[0261] The time-frequency spectrum diagram is input into a pre-trained target crowd number statistical model to obtain the crowd number in a preset area within a preset time period, and the pre-trained target crowd number statistical model is obtained based on the above-mentioned crowd number statistical model training method.

[0262] exist Fig.10 In the embodiment, the bus architecture (represented by bus 1001) may include any number of interconnected buses and bridges, and bus 1001 links various circuits including one or more processors represented by processor 1005 and memory represented by memory 1006. Bus 1001 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 1004 provides an interface between bus 1001 and transceiver 1002. Transceiver 1002 may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. Data processed by processor 1005 is transmitted on a wireless medium via antenna 1003, and further, antenna 1003 also receives data and transmits the data to processor 1005.

[0263] The processor 1005 is responsible for managing the bus 1001 and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 1006 can be used to store data used by the processor 1005 when performing operations.

[0264] Optionally, the processor 1005 may be a CPU, an ASIC, an FPGA or a CPLD.

[0265] Optionally, the processor 1005 is specifically configured to:

[0266] Unpacking the wireless signal to obtain an initial channel state information matrix;

[0267] Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix;

[0268] Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix;

[0269] Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information;

[0270] The first principal component information is processed using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

[0271] Optionally, the processor 1005 is specifically configured to:

[0272] The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter;

[0273] constructing a parallel factorization model according to the delay parameters;

[0274] The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

[0275] Optionally, the MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

[0276] An embodiment of the present invention also provides an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, the program implements the above-mentioned crowd flow statistical model training method and each process of the crowd flow statistical method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described here.

[0277] The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the various processes of the above-mentioned human flow number statistical model training method and human flow number statistical method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0278] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0279] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0280] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for training a pedestrian flow statistics model, characterized in that: include: Unpacking the received wireless signal to obtain an initial channel state information matrix; Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix; Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix; Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information; Using a short-time Fourier transform algorithm to process the first principal component information to obtain a time-frequency spectrum diagram; The target pedestrian flow number statistical model is trained based on the data set to obtain a pre-trained target pedestrian flow number statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

2. The method according to claim 1, characterized in that The step of filtering the multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes: The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter; constructing a parallel factorization model according to the delay parameters; The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

3. The method according to claim 2, characterized in that The MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

4. The method according to claim 1, characterized in that: The training process of the target pedestrian flow number statistical model based on the data set to obtain the pre-trained target pedestrian flow number statistical model includes: Dividing the data set into a training set and a validation set according to a preset ratio; Inputting the training set and the verification set into a target pedestrian flow number statistical model; Training the target pedestrian flow statistical model according to the training set to generate an initial pedestrian flow statistical model; Extracting training data from the initial pedestrian flow number statistical model to determine the accuracy and loss value of the initial pedestrian flow number statistical model; The initial pedestrian flow number statistical model is adjusted according to the accuracy rate, the loss value, the training set and the verification set to obtain a pre-trained target pedestrian flow number statistical model.

5. A method for counting the number of people, characterized in that: include: Acquire wireless signals in a preset area within a preset time period; Processing the wireless signal to obtain a time-frequency spectrum diagram; The time-frequency spectrum diagram is input into a pre-trained target crowd number statistical model to obtain the crowd number in a preset area within a preset time period, wherein the pre-trained target crowd number statistical model is obtained based on the crowd number statistical model training method described in any one of claims 1-4.

6. The method according to claim 5, characterized in that The processing of the wireless signal to obtain a time-frequency spectrum diagram includes: Unpacking the wireless signal to obtain an initial channel state information matrix; Perform multipath component filtering on the initial channel state information matrix to obtain a target channel state information matrix; Performing amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix; Performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information; The first principal component information is processed using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram.

7. The method according to claim 6, characterized in that The step of filtering the multipath components of the initial channel state information matrix to obtain a target channel state information matrix includes: The initial channel state information matrix is ​​processed using a multi-signal classification (MUSIC) algorithm to obtain a delay parameter; constructing a parallel factorization model according to the delay parameters; The initial channel state information matrix is ​​decomposed iteratively using the parallel factor decomposition model to obtain a target channel state information matrix.

8. The method according to claim 7, characterized in that The MUSIC algorithm is a frequency domain forward and backward smoothing MUSIC algorithm.

9. A device for training a statistical model of the number of people, characterized in that: include: An unpacking module, used to unpack the received wireless signal to obtain an initial channel state information matrix; A filtering module, used to filter out multipath components from the initial channel state information matrix to obtain a target channel state information matrix; An extraction module, used to perform amplitude extraction according to the target channel state information matrix to obtain amplitude information of the target channel state information matrix; An analysis module, used for performing principal component analysis on the amplitude information to obtain first principal component information of the amplitude information; A processing module, used for processing the first principal component information by using a short-time Fourier transform algorithm to obtain a time-frequency spectrum diagram; A training module is used to train a target pedestrian flow statistical model based on a data set to obtain a pre-trained target pedestrian flow statistical model, wherein the data set includes the time-frequency spectrum diagram and pre-acquired labels.

10. A device for counting the number of people, characterized in that: include: An acquisition module, used to acquire wireless signals in a preset area within a preset time period; A processing module, used for processing the wireless signal to obtain a time-frequency spectrum diagram; A determination module is used to input the time-frequency spectrum diagram into a pre-trained target crowd number statistical model to obtain the number of people in a preset area within a preset time period, and the pre-trained target crowd number statistical model is obtained based on the crowd number statistical model training method described in any one of claims 1-4.

11. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for training a model for counting the number of people flow as described in any one of claims 1 to 4 are implemented; or, when the program is executed by the processor, the steps of the method for counting the number of people flow as described in any one of claims 5 to 8 are implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for training a pedestrian flow statistics model as described in any one of claims 1 to 4; or, when executed by a processor, implements the steps of the method for counting the number of pedestrian flows as described in any one of claims 5 to 8.

Images