Static human body detection system and method based on feature ratio

Through a static human body detection system based on feature ratios, using filtering processing and feature ratio calculation, a human body detection model was constructed, which solved the problem of distinguishing static and unmanned states in WiFi-CSI research, improved the stability and accuracy of CSI data, and realized Accurate identification of the human body's stillness and unmanned state.

CN118761018BActive Publication Date: 2024-11-12NANCHANG HANGKONG UNIVERSITY +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411252238.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-12
Estimated Expiration
2044-09-09

AI Technical Summary

Technical Problem

The existing WiFi-CSI research is difficult to effectively distinguish between static and unmanned states in the field of human perception, and the impact of changes in the time domain and spatial domain on CSI data is not fully considered, resulting in the situation without crossing the time domain and spatial domain. The human body has low accuracy in perception.

Method used

A static human body detection system based on feature ratio is proposed. Through data preprocessing, feature extraction, feature ratio calculation and feature fusion model training, human body detection model is constructed, and filtering processing, abnormal point recognition, feature ratio calculation and convolutional neural network are used. Techniques such as reducing time-changing interference, improving the stability and reliability of CSI data, and realizing the identification of the human body's static and unmanned state.

Benefits of technology

Effectively reduce the interference caused by time changes, improve the stability and accuracy of CSI data, realize accurate identification of the human body's static and unmanned states, and improve detection performance in different time domains and environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761018B_ABST
    Figure CN118761018B_ABST
Patent Text Reader

Abstract

The present invention provides a static human body detection system and method based on feature ratio, the system comprising: a data preprocessing module, used to perform data preprocessing on the acquired CSI raw data to obtain preprocessed data; a feature extraction module, used to calculate the feature ratio of the preprocessed data based on the extracted feature data; a model training module, used to perform model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model; a feature ratio calculation module, used to input the feature ratio of the data to be identified into the data fusion model to construct a feature fusion model; a state detection module, used to construct a human body detection model according to the feature fusion model, and use the human body detection model to detect the human body state. The present invention constructs multiple features through the feature ratio method and establishes a feature fusion module, performs detailed fusion and extraction of features, and identifies static and unmanned states through the human body detection model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of human body detection, and in particular to a static human body detection system and method based on feature ratio. Background Art

[0002] Due to the wide application of WiFi signals, a large number of studies have been conducted in the literature. According to different research methods, these studies can be mainly divided into the following three categories:

[0003] (1) Manual features combined with classification algorithms: In early WiFi-CSI research, the main approach was to manually extract features from CSI data and use algorithms such as support vector machines (SVMs) and threshold partitioning for classification. Machine learning methods have significant advantages over simple threshold partitioning when processing WiFi-CSI data. First, these algorithms can adapt to changes in the data and better adapt to the changes in the characteristics of WiFi signals in different environments. Second, machine learning models can handle complex nonlinear relationships, which is crucial for describing the behavior of WiFi signals in different environments and interference. Finally, trained machine learning models usually have good generalization capabilities, can handle unseen data and remain robust in the face of noise and changes, which makes them more reliable and effective in practical applications. The initial approach focused on manual feature extraction combined with threshold partitioning methods. Over time, subsequent approaches have chosen to use SVM to classify the extracted features.

[0004] (2) Combining manual features with deep learning: Deep learning performs well in feature extraction and classification tasks due to its powerful learning ability. Through multi-level nonlinear transformations, deep learning models can learn higher-level and more abstract feature representations from raw data. This ability makes deep learning particularly effective in processing complex data such as images, speech, and text. It also shows potential in processing fields with complex structures and nonlinear features such as WiFi-CSI data. The wide application of deep learning includes convolutional neural networks (CNN) for image processing, recurrent neural networks (RNN) for sequence data, and more complex models such as Transformer for natural language processing. These models can not only improve the accuracy of classification and recognition, but also handle large amounts of data and complex features in changing environments. This has brought great convenience to intelligent research and avoided the need to spend a lot of time designing thresholds and selecting algorithms after manual feature extraction.

[0005] (3) Deep extraction: Traditional machine extraction methods often rely on manually designed feature extractors, and these features are usually manually selected and based on experience. In contrast, deep extraction models can automatically extract more advanced and abstract feature representations through multi-level neural network structures. These features often better reflect the complex patterns and associations in the data. Therefore, deep extraction can not only extract key features from the data, but also use these features for accurate classification and prediction, which makes deep extraction an important research method in the field of WiFi-CSI research. Taking WiVi as an example, it is based on the multimodal idea and uses the CNN model to extract features from CSI data and image data respectively, and then fuses these features through the DNN model for classification.

[0006] Although many research results have emerged in the field of WiFi-CSI research, most of them are aimed at the problem itself. These methods focus on solving problems such as human perception, indoor positioning, and activity recognition, but do not consider the impact of changes in the time and space domains on CSI data. Some methods can even achieve 100% accuracy in detecting some categories in the human perception neighborhood without crossing the time and space domains.

[0007] Time domain problem: Data collected in different time periods may show certain differences after feature extraction and visualization.

[0008] In the field of human perception, human activities are generally divided into three categories: motion, stillness, and absence. Among these three categories, motion is relatively easy to detect, and in many cases such as fall detection and intrusion detection, good results have been achieved, while stillness and absence are relatively difficult to distinguish. Summary of the invention

[0009] Based on this, an object of the present invention is to provide a static human body detection system and method based on feature ratio, so as to at least solve the deficiencies in the above-mentioned technology.

[0010] The present invention proposes a static human body detection system based on feature ratio, comprising:

[0011] A data preprocessing module, used to perform data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data;

[0012] A feature extraction module, used for extracting features from the preprocessed data, and calculating a feature ratio of the preprocessed data based on the extracted feature data;

[0013] A model training module, used for constructing a data classification model, and performing model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model;

[0014] A feature ratio calculation module, used for performing data preprocessing on the acquired data to be identified to calculate the feature ratio of the data to be identified, and using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model;

[0015] The state detection module is used to construct a human body detection model according to the feature fusion model, and use the human body detection model to detect the human body state.

[0016] Furthermore, the data preprocessing module includes:

[0017] A filtering processing unit, configured to filter the CSI raw data using a filtering algorithm, and perform data calibration and correction on the filtered CSI raw data to obtain preliminary processed data;

[0018] The outlier identification unit is used to identify outliers in the preliminary processed data and perform data processing on the outliers to obtain corresponding pre-processed data.

[0019] Furthermore, the feature extraction module includes:

[0020] A feature extraction unit, used for performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data;

[0021] The data comparison unit is used to perform data comparison according to each of the sub-feature data to calculate the feature ratio of the pre-processed data.

[0022] Furthermore, the characteristic ratio calculation module includes:

[0023] A model optimization unit, used to construct a preliminary fusion model and optimize model parameters of the preliminary fusion model to obtain a fusion optimization model;

[0024] The feature fusion unit is used to input the feature ratio of the data to be identified into the fusion optimization model for feature fusion to construct a feature fusion model.

[0025] Furthermore, the state detection module includes:

[0026] A convolution processing unit, used for performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data;

[0027] The state detection unit is used to introduce an attention module and use the attention module to process the feature output data to realize the detection of human body state.

[0028] The present invention also proposes a static human body detection method based on feature ratio, comprising:

[0029] Performing data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data;

[0030] Extracting features from the preprocessed data, and calculating a feature ratio of the preprocessed data based on the extracted feature data;

[0031] Constructing a data classification model, and performing model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model;

[0032] Performing data preprocessing on the acquired data to be identified to calculate the feature ratio of the data to be identified, and using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model;

[0033] A human body detection model is constructed according to the feature fusion model, and the human body detection model is used to detect the human body state.

[0034] Further, the acquired CSI raw data is preprocessed to remove abnormal data in the CSI raw data, and the step of obtaining the preprocessed data includes:

[0035] The CSI raw data is filtered using a filtering algorithm, and the filtered CSI raw data is calibrated and corrected to obtain preliminary processed data;

[0036] An outlier in the preliminary processed data is identified, and data processing is performed on the outlier to obtain corresponding pre-processed data.

[0037] Furthermore, the step of extracting features from the preprocessed data and calculating a feature ratio of the preprocessed data based on the extracted feature data includes:

[0038] Performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data;

[0039] Data comparison is performed based on each of the sub-feature data to calculate the feature ratio of the pre-processed data.

[0040] Furthermore, the step of using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model includes:

[0041] Constructing a preliminary fusion model, and optimizing model parameters of the preliminary fusion model to obtain a fusion optimization model;

[0042] The feature ratio of the data to be identified is input into the fusion optimization model for feature fusion to construct a feature fusion model.

[0043] Furthermore, the steps of constructing a human body detection model according to the feature fusion model and realizing the detection of human body state by using the human body detection model include:

[0044] Performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data;

[0045] An attention module is introduced, and the attention module is used to process the feature output data to realize the detection of human body status.

[0046] The static human body detection system and method based on feature ratio in the present invention performs data preprocessing on the CSI raw data to remove abnormal data in the CSI raw data, and constructs a corresponding feature ratio based on feature extraction of the preprocessed data. The feature ratio is used to effectively reduce the interference caused by time changes, thereby improving the stability and reliability of the CSI data; multiple features are constructed through the feature ratio method and a feature fusion module is established to perform detailed fusion and extraction of features, and a human body detection model is constructed through the feature fusion model to realize the recognition of the static and unmanned state of the human body. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a structural schematic diagram of a static human body detection system based on feature ratio in a first embodiment of the present invention;

[0048] Figure 2 is the standard deviation of a subcarrier in data of two different days in the first embodiment of the present invention;

[0049] Figure 3 It is a structural diagram of a feature fusion module in the first embodiment of the present invention;

[0050] Figure 4 It is a schematic diagram of the structure of the network bottleneck layer in the first embodiment of the present invention;

[0051] Figure 5 is a flow chart of a static human body detection method based on feature ratio in a second embodiment of the present invention;

[0052] Figure 6 This is a scene diagram of a conference room experiment in the second embodiment of the present invention;

[0053] Figure 7 A distribution diagram showing the relationship between sample size and accuracy in the second embodiment of the present invention;

[0054] Figure 8This is a comparison diagram of the experimental environment in the second embodiment of the present invention;

[0055] Fig. 9 FIG. 4 is a structural block diagram of a computer in a third embodiment of the present invention.

[0056] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0057] In order to facilitate understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this embodiment. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0058] Unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention in this embodiment are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used in this embodiment includes any and all combinations of one or more related listed items.

[0059] Embodiment 1

[0060] Please refer to Figure 1 , which is a static human body detection system based on feature ratio in a first embodiment of the present invention, and the system comprises:

[0061] The data preprocessing module 11 is used to perform data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data;

[0062] Furthermore, the data preprocessing module 11 includes:

[0063] A filtering processing unit, configured to filter the CSI raw data using a filtering algorithm, and perform data calibration and correction on the filtered CSI raw data to obtain preliminary processed data;

[0064] The outlier identification unit is used to identify outliers in the preliminary processed data and perform data processing on the outliers to obtain corresponding pre-processed data.

[0065] In specific implementation, CSI (Channel State Information) raw data plays a key role in wireless communication and perception systems. However, its direct use may be affected by various problems and requires effective preprocessing before use.

[0066] First, since CSI data is often affected by various interferences, such as multipath effects, background noise, and interference between devices, these factors cause a lot of noise in the data, reducing the reliability and accuracy of the data. To solve this problem, this embodiment uses a filtering algorithm, such as low-pass filtering or median filtering, in the preprocessing stage to remove noise and extract cleaner and more reliable channel information.

[0067] Secondly, due to the instability of CSI data, which may be affected by environmental changes, device movement, and time factors, directly using these unstable data will lead to inconsistent and unreliable results. In order to meet this challenge, this embodiment requires data calibration and correction in the preprocessing stage, such as path loss correction and phase drift compensation, to ensure the consistency and stability of data under different times and conditions.

[0068] Finally, there may be some abnormal points in the CSI data, which may be due to factors such as data packet loss, interference of wireless signals, etc. To ensure data quality, this embodiment removes or reassigns abnormal points to ensure the accuracy and reliability of data analysis and application.

[0069] Specifically, ESP-S3 is used to collect data at a frequency of 100 packets per second. Each packet contains 52 complex values, each of which represents a subcarrier. Due to the noise and some abnormal points in the data, the preprocessing stage is mainly divided into the following steps:

[0070] (1) Calculate the amplitude based on the collected CSI data.

[0071] (2) Noise removal: Use a Butterworth low-pass filter to remove high-frequency noise, with the cutoff frequency set to 0.1.

[0072] (3) Remove DC component: The DC component is the mean of the data, so just subtract the mean.

[0073] A feature extraction module 12, configured to extract features from the preprocessed data and calculate a feature ratio of the preprocessed data based on the extracted feature data;

[0074] Furthermore, the feature extraction module 12 includes:

[0075] A feature extraction unit, used for performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data;

[0076] The data comparison unit is used to perform data comparison according to each of the sub-feature data to calculate the feature ratio of the pre-processed data.

[0077] In the specific implementation, the feature performance of data from different days is compared to analyze the possible problems of CSI across time domains. This embodiment proposes a feature ratio method and constructs multiple features to alleviate the challenges that may be brought about by such time span:

[0078] See also Figure 2 Through the analysis of various statistical features, it can be seen that the features extracted from the two time periods (set as data set A and data set B), whether data set B is static or unmanned, the fluctuation and value range of the features are different from those of data set A. At this time, if the features extracted from data sets A and B are compared together, it will be found that there are differences between the two.

[0079] (1) If Figure 2 As shown (taking standard deviation as an example), although there is a gap between stillness and absence in A and B, there is also a gap between A and B. This gap will make the visualization of absence in B tend towards stillness in A.

[0080] (2) If Figure 2 As shown (taking the mean of the standard deviation as an example), the difference between no one in B and no one in A is 0.045, while the difference between no one in B and stillness in A is only 0.002.

[0081] Specifically, CSI data presents the following characteristics: within the same dataset (same time period), there are significant differences between different categories; there are also significant differences between different datasets (different time periods). Therefore, if the model trained with dataset A is used to predict dataset B, the prediction effect may not be satisfactory. Taking the standard deviation as an example, if the threshold method is used to divide different categories, the threshold trained with dataset A should fall within the range of [0.289, 0.336] (see Table 1). However, when this threshold is applied to predict dataset B, there is a high probability that the unmanned state in dataset B will be incorrectly predicted as a stationary state.

[0082] Table 1 Average values ​​of the first 100 sample eigenvalues

[0083]

[0084] Furthermore, in view of the differences in CSI features across time periods, this embodiment reduces the differences between different data sets (different time periods) while retaining the differences within the same data set (same time period), and selects two features to divide to obtain a new feature (feature ratio). Compared with the original feature, the feature ratio is a relative value, which enhances the comparability of data in different time periods. This method improves the accuracy of capturing real changes and trends in data in different time domains, while reducing the impact of time changes.

[0085] The feature construction process in this embodiment mainly includes the following three steps:

[0086] 1) Selection of sub-features: One problem that needs to be avoided when selecting features is information loss. Information loss mainly comes from two aspects: On the one hand, in each sample (each sample contains 400 data packets, and each data packet contains 52 subcarriers), if only characteristic values ​​such as variance are extracted, this operation will highly compress the data information, resulting in a large amount of information loss. On the other hand, as the degree of feature abstraction continues to increase, the amount of lost information will also increase accordingly. According to Table 2, this embodiment selects the autocorrelation sequence as a sub-feature, and its calculation formula is as follows.

[0087]

[0088] Table 2 Feature matrix size

[0089]

[0090] 2) Construction of feature ratio: In order to solve the above problems in the time domain, this embodiment proposes a feature ratio. By calculating the ratio, the difference between features in different time periods is reduced, making the features more stable and comparable. As shown in Table 3, the autocorrelation sequence also has the same problem as the above cross-time domain, that is, the no one in data set B is closer to the static in data set A; while the autocorrelation sequence divided by the average does not have this problem, and the no one in data set B is closer to the no one in data set A. In order to verify the effectiveness of the feature ratio, this embodiment gives the results of the comparative experiment on the autocorrelation sequence and the feature ratio in Table 4.

[0091] Table 3 The average values ​​of the first 100 sample eigenvalues:

[0092] Regarding the calculation of autocorrelation series, the average value is used instead of

[0093]

[0094] 3) Construct multiple features: In order to further reduce information loss, this embodiment uses the feature ratio as a sub-feature and combines the standard deviation, mean value and variance to construct a 3*400*52 feature matrix. The calculation formula is as follows:

[0095]

[0096] in, α 1 、α 2 、α 3 is a hyperparameter, which is mainly used to avoid the gap between the three eigenvalues ​​being too large, and is determined based on the standard deviation, variance, and mean. In order to verify the effectiveness of the feature ratio and multiple features, this embodiment conducted a comparative experiment, and the experimental results are shown in Table 4.

[0097] Table 4 Comparative experiment

[0098]

[0099] A model training module 13 is used to construct a data classification model and perform model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model;

[0100] The feature ratio calculation module 14 is used to perform data preprocessing on the acquired data to be identified so as to calculate the feature ratio of the data to be identified, and use the feature ratio of the data to be identified to input into the data fusion model so as to construct a corresponding feature fusion model;

[0101] Furthermore, the characteristic ratio calculation module 14 includes:

[0102] A model optimization unit, used to construct a preliminary fusion model and optimize model parameters of the preliminary fusion model to obtain a fusion optimization model;

[0103] The feature fusion unit is used to input the feature ratio of the data to be identified into the fusion optimization model for feature fusion to construct a feature fusion model.

[0104] In specific implementation, the feature fusion module is responsible for effectively integrating and fusing features from different sources to improve the representation ability and generalization performance of the model; the manual features extracted from the original data represent different meanings. Therefore, it is necessary to build a fusion module for manual features. The design of this module can link unrelated or low-related manual features together to achieve the fusion of multi-level features. By fusing features from different levels, the seamless integration of abstract and specific features is achieved, thereby improving the modeling ability of the model for complex data.

[0105] See also Figure 3, the structural design of the preliminary feature fusion module mainly refers to multimodal fusion. The basic structure of the feature fusion module is Encode-Fusion-Decode, where the main parts of Encode and Decode are three convolution blocks (Conv block). Each convolution block contains two convolution operations, the first is a 1*1 convolution, and the second is a 3*3 convolution. In order to pay more attention to important features, the channel attention mechanism (SEModule) is selected as the Fusion layer. After experiments, it was found that adding a Fusion layer to Decode can better help the module extract features. In Decode, this embodiment uses the maximum pooling operation instead of the activation function. Maximum pooling can better extract significant features in terms of timing while reducing the amount of calculation. The sample size adopted in this embodiment is 400*52, which becomes a feature matrix of 50*52 (each subcarrier retains 50 features in timing) after three pooling operations.

[0106] The state detection module 15 is used to construct a human body detection model according to the feature fusion model, and use the human body detection model to detect the state of the human body.

[0107] Furthermore, the state detection module 15 includes:

[0108] A convolution processing unit, used for performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data;

[0109] The state detection unit is used to introduce an attention module and use the attention module to process the feature output data to realize the detection of human body state.

[0110] In the specific implementation, the human detection model mainly consists of two key parts, the network bottleneck layer (Bottlenecklayer) and the attention module (CBAM, Convolutional Block Attention Module).

[0111] (1) Constructing the network bottleneck layer

[0112] In the early stages of the model, the depthwise convolution in MobileNetV1 and the inverted residual structure in MobileNetV2 were selected to construct the main network bottleneck layer. The depthwise convolution is known for its excellent performance in computational efficiency and parameter quantity, providing the model with the ability to efficiently learn features in resource-constrained environments. The inverted residual structure improves the feature expression capability by introducing lightweight shortcut connections and nonlinear activation functions, thereby enhancing the network's ability to capture complex patterns.

[0113] The overall structure of the network bottleneck layer ( Figure 4 ) First, a 1*1 convolution operation is performed to increase the dimension of the input feature to Exp_size; then a convolution operation with a convolution kernel size of Kernel_size is performed for feature extraction and processing; finally, a 1*1 convolution operation is performed to reduce the feature dimension to Out_c to generate the final output. After each convolution operation, there will be a normalization process and activation function. After the second convolution operation, the CBAM module will be enabled according to whether the Use_CBAM parameter is used, which can selectively further optimize the feature representation capability and the overall performance of the network.

[0114] In addition, if the input and output features have the same dimension and the same size (Stride size is 1), a residual connection will be added between the input and output. In this way, the network can learn how to retain important information after information compression or stretching, while also alleviating the problem of gradient vanishing or gradient exploding.

[0115] (2) Introducing the CBAM attention module

[0116] In this embodiment, a convolutional block attention module (Convolutional Block Attention Module), referred to as CBAM, is introduced. This module is a simple and effective feedforward convolutional neural network attention module. Given a feature map, CBAM infers the attention map in sequence along two separate dimensions, channel and spatial, and then multiplies the attention map with the input feature map for adaptive feature refinement. Because CBAM is a lightweight and universal module, it can be seamlessly integrated into any CNN architecture with negligible overhead and can be trained end-to-end with the basic CNN. The introduction of this module aims to enhance the model's attention to important features, thereby improving the overall performance, while maintaining the versatility and integrability of the model.

[0117] (3) Human body detection model structure design

[0118] Based on the above modules, the designed human detection model is shown in Table 5. The input of the human detection model is the output of the feature fusion module, and the final output result of the human detection model is the number of classified categories. In the table, input is the dimension of the input feature as a sample size, Operator is how the input feature is processed, Exp_size is how many dimensions the input feature dimension is increased to in each Bottleneck for operation (after the operation, it will be reduced to Out_c dimensions), Out_c is the final output dimension of the feature after this operation, Kernel_size is the size of the convolution kernel, Stride is the stride of the convolution kernel movement, and Use_CBAM is whether to use the CBAM module.

[0119] Table 5 Human body detection model (where k is the number of classification categories)

[0120]

[0121] In summary, the static human body detection system based on feature ratio in the above-mentioned embodiment of the present invention performs data preprocessing on the CSI raw data to remove abnormal data in the CSI raw data, and constructs a corresponding feature ratio based on feature extraction of the preprocessed data. The feature ratio is used to effectively reduce the interference caused by time changes, thereby improving the stability and reliability of the CSI data; multiple features are constructed by the feature ratio method and a feature fusion module is established to perform detailed fusion and extraction of features, and a human body detection model is constructed by the feature fusion model to realize the recognition of the static and unmanned state of the human body.

[0122] Embodiment 2

[0123] Please refer to Figure 5 , which shows a static human body detection method based on feature ratio in a second embodiment of the present invention, and the method includes steps S101 to S105:

[0124] S101, performing data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data;

[0125] Furthermore, the step S101 specifically includes:

[0126] The CSI raw data is filtered using a filtering algorithm, and the filtered CSI raw data is calibrated and corrected to obtain preliminary processed data;

[0127] An outlier in the preliminary processed data is identified, and data processing is performed on the outlier to obtain corresponding pre-processed data.

[0128] S102, extracting features from the preprocessed data, and calculating a feature ratio of the preprocessed data based on the extracted feature data;

[0129] Furthermore, the step S102 specifically includes:

[0130] Performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data;

[0131] Data comparison is performed based on each of the sub-feature data to calculate the feature ratio of the pre-processed data.

[0132] S103, constructing a data classification model, and using the feature ratio of the preprocessed data to perform model training on the data classification model to obtain a data fusion model;

[0133] S104, performing data preprocessing on the acquired data to be identified to calculate the feature ratio of the data to be identified, and using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model;

[0134] Furthermore, the step S104 specifically includes:

[0135] Constructing a preliminary fusion model, and optimizing model parameters of the preliminary fusion model to obtain a fusion optimization model;

[0136] The feature ratio of the data to be identified is input into the fusion optimization model for feature fusion to construct a feature fusion model.

[0137] S105, constructing a human body detection model according to the feature fusion model, and using the human body detection model to detect the human body state.

[0138] Furthermore, the step S105 specifically includes:

[0139] Performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data;

[0140] An attention module is introduced, and the attention module is used to process the feature output data to realize the detection of human body status.

[0141] In this embodiment, the above-mentioned method is compared and analyzed with the methods of other researchers in the current field, and a robust experiment is conducted:

[0142] 1. Experimental design

[0143] The experimental scene is a conference room. Figure 6. A commercial router (TP-LINK WDR6500) was used as a signal transmitter and deployed at the end of the table away from the door; ESP32-S3 was used as a receiver and deployed at the end of the table close to the door. The specific experimental equipment parameter configuration is shown in Table 6. When collecting data, only the router, ESP32-S3 and a laptop were kept on the table. In order to reduce the interference of other factors except time, this embodiment keeps the environment stable during the entire period of data collection, and ensures that the positions of the router and ESP32-S3 do not shift significantly.

[0144] Table 6 Experimental settings

[0145]

[0146] In order to study the changes in CSI data in different time periods, this embodiment first collected data for three consecutive days, and then collected data for four consecutive days after an interval of three days. The data collection work was completed by two people, and 45 minutes of data were collected in the unmanned and static states every day, totaling about 540,000 data. All collected data were divided into three parts, including training set, validation set and test set. The specific division method is shown in Table 7:

[0147] Table 7 Dataset collection and data division

[0148]

[0149] 2. Sample size setting

[0150] In this embodiment, a sample size (window size) of 400 is selected as the initial setting of the experiment. Figure 7 As shown in Figure 2, when the sample size is 400, the prediction accuracy of the model reaches 98.12%. Subsequently, this embodiment further explores the impact of larger samples, but the experimental results show that even with a larger sample, the improvement in accuracy is not significant. This shows that under this experimental setting, a sample size of 400 is sufficient to support the model to achieve a high accuracy. Further increasing the sample size has limited effect on improving performance, probably because the stability of the data and the complexity of the model limit the benefits of additional samples.

[0151] 3. Ablation experiment

[0152] In order to evaluate the effectiveness of the fusion module, this embodiment conducted an ablation experiment and gave the experimental results in Table 8. The experimental evaluation criteria are accuracy, Flops and Params, accuracy is the main evaluation criterion of the model, Flops is the time complexity of the model, and Params is the space complexity of the model. The results show that after the introduction of the fusion module, the accuracy of the system is significantly improved by 1.72 percentage points. At the same time, the use of the maximum pooling operation not only significantly reduces the amount of calculation, but also further improves the accuracy by 0.46 percentage points. These results not only confirm the effectiveness and importance of the fusion module in improving system performance, but also reveal its positive impact on model complexity and computational efficiency. Through the precise control and gradual adjustment of the ablation experiment, the study can clearly evaluate the contribution of each component to the final result, thereby ensuring the reliability and scientificity of the experimental results.

[0153] Table 8 Ablation experiment

[0154]

[0155] 4. Experimental results

[0156] In this experiment, the data from days 1 and 2 were used as training sets, and the data from days 3, 7, 8, 9, and 10 were used as test sets. It is clear from the data in Table 9 that the prediction accuracy of FRModel remained above 97% in both single-day and five-day comprehensive evaluations. Specifically, in the single-day data evaluation, FRModel performed well, with the best prediction accuracy reaching 99.26%, and even in the worst case, the accuracy was maintained at 97.10%. These results show the robustness and reliability of the model over different time periods.

[0157] When analyzing five days of data, FRModel showed excellent accuracy in predicting unmanned situations, reaching 98.54%; and in predicting stationary situations, the accuracy was as high as 97.71%; the overall average accuracy reached 98.12%.

[0158] Table 9 Experimental results (only keep two decimal places)

[0159]

[0160] 5. Comparison with other methods

[0161] In order to effectively evaluate the performance of the method of the present invention, this embodiment will be compared with methods such as WiVi, FDF-PIHD, DTJD (Dual-Threshold Joint Determination), and CRONOS. Among them, WiVi uses a deep neural network to extract features from images and CSI data respectively, and then sends them to a DNN model for feature fusion and classification prediction. FDF-PIHD extracts feature fingerprints from CSI data, and classifies and predicts the data by matching online fingerprints with offline fingerprints. The dual-threshold joint determination extracts two thresholds, and then uses the two thresholds to jointly predict the data. CRONOS generates a dynamic recursive graph (RP) and a color-coded CSI ratio to distinguish between Move, Static, and None in the room. Among them, the dynamic recursive graph (RP) is used to screen Move samples, and the color-coded CSI ratio is used to distinguish between Static and None.

[0162] Since this embodiment only distinguishes between static and no one, all classifications will be listed when comparing methods. In Table 10, the accuracy of None, Static, and Move are all from the original paper, and the average accuracy is the average of the sum of the three. Although it is impossible to make a direct comparison with the methods in recent years, in the comparison across time domains, the method in this embodiment performs better than other methods in detecting static people, and the average accuracy is also higher than them.

[0163] Table 10 Comparison with other methods

[0164]

[0165] 6. Robustness Experiment

[0166] In order to further simulate real-life scenarios, after collecting the data in the above experimental design in October 2023, the present invention collected three data segments on June 30, 2024 (respectively recorded as Data 1, Data 1, and Data 3). The time interval reached 8 months, and the environment also changed dramatically (such as Figure 8 ). Compared with October 2023, the experimental environment on June 30, 2023 has changed dramatically. First, the table is moved from the middle of the room to near the door. Second, some large objects are added to the room. Finally, some small objects such as sockets and remote controls are also added to the table.

[0167] At the same time, in order to study the influence of the outside of the room on the inside, this embodiment sets additional interference factors as follows:

[0168] (1) When collecting Data 1, there were 2-3 people walking outside the room (near the wall).

[0169] (2) When collecting Data 2, someone walked past the room (two or three times).

[0170] (3) When collecting Data 3, there was no one outside the room.

[0171] In this experiment, the model trained using the above method is used to predict Data 1, Data 2, and Data 3. The experimental results are shown in Table 11.

[0172] Table 11 Robustness experimental results

[0173]

[0174] As can be seen from Table 11, when the interference is very large (Data 1), the collected unmanned data is easily predicted as unmanned, and the accuracy of unmanned data is only 40.79%. When the interference is very small (Data 2), compared with the case of no interference (Data 3), the unmanned data has 4.59% data prediction errors, and the average accuracy rate reaches 93.71%. When there is no interference, the accuracy rate of unmanned data is 93.33%, the accuracy rate of stationary data is 96.0%, and the average accuracy rate is 94.67%.

[0175] Embodiment 3

[0176] The present invention also provides a computer, see Fig. 9 , shown is a computer in the third embodiment of the present invention, including a memory 10, a processor 20, and a computer program 30 stored in the memory 10 and executable on the processor 20. When the processor 20 executes the computer program 30, the above-mentioned static human body detection system based on feature ratio is implemented.

[0177] The memory 10 includes at least one type of readable storage medium, which includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 10 may be an internal storage unit of a computer, such as a hard disk of the computer. In other embodiments, the memory 10 may also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Further, the memory 10 may also include both an internal storage unit of the computer and an external storage device. The memory 10 may be used not only to store application software and various types of data installed in the computer, but also to temporarily store data that has been output or is to be output.

[0178] Among them, in some embodiments, the processor 20 can be an electronic control unit (Electronic Control Unit, abbreviated as ECU, also known as a vehicle computer), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run the program code stored in the memory 10 or process data, such as executing access restriction programs, etc.

[0179] It should be pointed out that Fig. 9 The structure shown does not constitute a limitation on the computer. In other embodiments, the computer may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0180] The embodiment of the present invention further provides a readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the static human body detection system based on feature ratio as described above is implemented.

[0181] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0182] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0183] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0184] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A static human body detection system based on feature ratio, characterized in that: include: A data preprocessing module, used to perform data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data; A feature extraction module, used for extracting features from the preprocessed data, and calculating a feature ratio of the preprocessed data based on the extracted feature data; A model training module, used for constructing a data classification model, and performing model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model; A feature ratio calculation module, used for performing data preprocessing on the acquired data to be identified to calculate the feature ratio of the data to be identified, and using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model; The state detection module is used to construct a human body detection model according to the feature fusion model, and use the human body detection model to detect the human body state.

2. The static human body detection system based on feature ratio according to claim 1, characterized in that: The data preprocessing module comprises: A filtering processing unit, configured to filter the CSI raw data using a filtering algorithm, and perform data calibration and correction on the filtered CSI raw data to obtain preliminary processed data; The outlier identification unit is used to identify outliers in the preliminary processed data and perform data processing on the outliers to obtain corresponding pre-processed data.

3. The static human body detection system based on feature ratio according to claim 1, characterized in that: The feature extraction module comprises: A feature extraction unit, used for performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data; The data comparison unit is used to perform data comparison according to each of the sub-feature data to calculate the feature ratio of the pre-processed data.

4. The static human body detection system based on feature ratio according to claim 1, characterized in that: The characteristic ratio calculation module comprises: A model optimization unit, used to construct a preliminary fusion model and optimize model parameters of the preliminary fusion model to obtain a fusion optimization model; The feature fusion unit is used to input the feature ratio of the data to be identified into the fusion optimization model for feature fusion to construct a feature fusion model.

5. The static human body detection system based on feature ratio according to claim 1, characterized in that: The state detection module comprises: A convolution processing unit, used for performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data; The state detection unit is used to introduce an attention module and use the attention module to process the feature output data to realize the detection of human body state.

6. A static human body detection method based on feature ratio, characterized in that: include: Performing data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data to obtain preprocessed data; Extracting features from the preprocessed data, and calculating a feature ratio of the preprocessed data based on the extracted feature data; Constructing a data classification model, and performing model training on the data classification model using the feature ratio of the preprocessed data to obtain a data fusion model; Performing data preprocessing on the acquired data to be identified to calculate the feature ratio of the data to be identified, and using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model; A human body detection model is constructed according to the feature fusion model, and the human body detection model is used to detect the human body state.

7. The static human body detection method based on feature ratio according to claim 6, characterized in that: The step of performing data preprocessing on the acquired CSI raw data to remove abnormal data in the CSI raw data, and obtaining the preprocessed data includes: Filtering the CSI raw data using a filtering algorithm, and performing data calibration and correction on the filtered CSI raw data to obtain preliminary processed data; An outlier in the preliminary processed data is identified, and data processing is performed on the outlier to obtain corresponding preprocessed data.

8. The static human body detection method based on feature ratio according to claim 6, characterized in that: The steps of extracting features from the preprocessed data and calculating a feature ratio of the preprocessed data based on the extracted feature data include: Performing feature extraction on the preprocessed data to construct a plurality of sub-feature data of the preprocessed data; Data comparison is performed based on each of the sub-feature data to calculate the feature ratio of the pre-processed data.

9. The method for static human body detection based on feature ratio according to claim 6, characterized in that: The step of using the feature ratio of the data to be identified to input into the data fusion model to construct a corresponding feature fusion model includes: Constructing a preliminary fusion model, and optimizing model parameters of the preliminary fusion model to obtain a fusion optimization model; The feature ratio of the data to be identified is input into the fusion optimization model for feature fusion to construct a feature fusion model.

10. The static human body detection method based on feature ratio according to claim 6, characterized in that: The steps of constructing a human body detection model according to the feature fusion model and realizing detection of human body status by using the human body detection model include: Performing three convolution operations on the output data of the feature fusion model to obtain corresponding feature output data; An attention module is introduced, and the attention module is used to process the feature output data to realize the detection of human body state.

Citation Information

Patent Citations

  • People counting method based on channel state information

    CN112381063A

  • Human body activity sensing and identifying method and system based on Wi-Fi signal and storage medium

    CN115134848A

  • Method, apparatus, and system for wireless sleep monitoring

    US20200397365A1