Real-time through-wall human body posture estimation method based on WIFI device

Abstract

The invention relates to the field of infinite perception and human body posture estimation, and discloses a real-time through-wall human body posture estimation method based on WIFI equipment, which comprises the following steps: system deployment: a transmitting end and a receiving end carry an inter5300 network card and a plurality of transmitting antennas, a transmitter and a receiver are respectively arranged on two sides of a wall body, a synchronous camera is arranged at the transmitting end, and the synchronous camera is arranged at the receiving end; the method comprises the steps that a receiving end receives CSI which passes through indoor human bodies and is reflected by surrounding walls, a camera synchronously collects video data, and a data set is manufactured, specifically, according to collected channel state information CSI signals and video frames, CSI data frames and video frames are aligned through the timestamps of the CSI and the video frames. The WiFi transmitting end and the receiving end are arranged on the two sides of the wall body, and the corresponding artificial neural network model is carried, so that non-contact human body posture estimation is realized, and the effect of through-wall human body recognition and posture estimation through wireless signals is achieved.

Description

Technical Field

[0001] The present invention relates to the field of wireless perception and human body posture estimation, and in particular to a real-time through-wall human body posture estimation method based on WIFI equipment. Background Art

[0002] Human posture monitoring usually relies on visible light imaging technology, such as cameras or lidar. These devices can capture human posture with high precision and achieve clear image recognition under good lighting and unobstructed conditions. However, solutions that rely on visible light have obvious deficiencies in privacy protection and environmental adaptability, especially when through-wall monitoring is required or used in privacy-sensitive scenarios (such as bedrooms, bathrooms, etc.). The potential risk of privacy leakage exposed to visible light cameras is worrying. In addition, in low-light environments, complex occlusions or multi-room monitoring scenarios, traditional visible light devices are difficult to operate effectively, which limits their application scenarios.

[0003] Existing posture estimation methods mainly rely on cameras and wearable devices. These methods have limitations in the fields of privacy protection and wall penetration detection. WiFi signals have low cost and strong penetration ability, and can become an ideal posture estimation method. The posture estimation method based on WiFi can detect human behavior in non-line-of-sight environments. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a real-time through-wall human posture estimation method based on WIFI devices, which solves the problem that the existing posture estimation methods mainly rely on cameras and wearable devices, and these methods have limitations in the fields of privacy protection and through-wall detection.

[0005] To achieve the above objectives, the present invention is implemented by the following technical solutions: A real-time through-wall human posture estimation method based on WIFI equipment, comprising the following steps:

[0006] System deployment: The transmitter and receiver are equipped with inter5300 network cards and multiple transmitting antennas. A transmitter and a receiver are placed on both sides of the wall, and a synchronous camera is placed on the transmitter.

[0007] Data collection: The receiving end receives the CSI reflected by the human body in the room and the surrounding walls, and the camera synchronously collects video data;

[0008] Dataset creation: Based on the collected channel state information CSI signals and video frames, the CSI data frames and video frames are aligned using the timestamps of the CSI and video frames, and then packaged into a training set;

[0009] Network training: The preprocessed data set is sent to the artificial neural network for training. The human posture estimation is obtained based on the amplitude and phase feature information in the CSI signal. The real posture skeleton of the image of the aligned video frame is used as an annotation. The neural network is trained with cross entropy loss until the network converges to obtain a trained human posture estimation model.

[0010] Real-time detection: During the detection phase, the transmitter continuously transmits WiFi signals. When a human body enters the room, the CSI signal will be reflected, refracted, and penetrate the human body. The receiver receives the CSI signal and transmits it to the server through UPD. The trained human posture estimation model displays the human posture, achieving the real-time purpose.

[0011] Preferably, the channel state information includes amplitude information and phase information.

[0012] Preferably, the preprocessing step includes applying circularity denoising to the channel state information using a Mobius transform.

[0013] Preferably, the artificial neural network includes a convolutional neural network UNET and a recurrent neural network GRU.

[0014] Preferably, the posture estimation model is trained with a large amount of CSI data of different human postures to improve the accuracy and robustness of posture estimation.

[0015] Preferably, the posture estimation includes a time series action for processing dynamically changing to achieve real-time continuous tracking of human body movements.

[0016] Preferably, the method can be widely used in the fields of elderly care and smart home.

[0017] The present invention provides a real-time through-wall human body posture estimation method based on WIFI equipment.

[0018] It has the following beneficial effects:

[0019] 1. The present invention realizes non-contact human posture estimation by arranging WiFi transmitters and receivers on both sides of the wall and carrying corresponding artificial neural network models, achieving the technical effect of through-wall human body recognition and posture estimation through wireless signals. Compared with the solutions in the prior art that rely on visible light devices such as cameras, it solves the shortcomings in privacy protection and environmental restrictions.

[0020] 2. The present invention uses the amplitude and phase information in the CSI (channel state information) signal as the basic data for human posture estimation, and combines the UNET convolutional neural network and GRU recurrent neural network models for training to achieve high-precision posture recognition. Compared with the single CSI feature or static network architecture solution in the prior art, it solves the problem of inaccurate and unstable real-time response to human posture changes in dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of a transmitting end device of the present invention;

[0022] Figure 2 It is a schematic diagram of a receiving end device of the present invention;

[0023] Figure 3 It is a schematic diagram of the network structure of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0025] Please refer to the attached Figure 1 -Attached Figure 3 The embodiment of the present invention provides a real-time through-wall human body posture estimation method based on WIFI equipment, comprising the following steps:

[0026] System deployment: The transmitter and receiver are equipped with inter5300 network cards and multiple transmitting antennas. A transmitter and a receiver are placed on both sides of the wall, and a synchronous camera is placed on the transmitter.

[0027] Data collection: The receiving end receives the CSI reflected by the human body in the room and the surrounding walls, and the camera synchronously collects video data;

[0028] Dataset creation: Based on the collected channel state information CSI signals and video frames, the CSI data frames and video frames are aligned using the timestamps of the CSI and video frames, and then packaged into a training set;

[0029] Network training: The preprocessed data set is sent to the artificial neural network for training. The human posture estimation is obtained based on the amplitude and phase feature information in the CSI signal. The real posture skeleton of the image of the aligned video frame is used as an annotation. The neural network is trained with cross entropy loss until the network converges to obtain a trained human posture estimation model.

[0030] Real-time detection: During the detection phase, the transmitter continuously transmits WiFi signals. When a human body enters the room, the CSI signal will be reflected, refracted, and penetrate the human body. The receiver receives the CSI signal and transmits it to the server through UPD. The trained human posture estimation model displays the human posture, achieving the real-time purpose.

[0031] Specifically, system deployment: the transmitter and receiver are equipped with inter5300 network cards and multiple transmitting antennas. A transmitter and a receiver are placed on each side of the wall. A synchronous camera is placed on the transmitter. The system uses the penetration characteristics of WiFi signals to capture human activities on the other side of the wall. Compared with traditional visible light monitoring solutions, WiFi signals do not need to look directly at the target and can effectively penetrate the wall, realizing the detection of human activities in hidden environments. Data acquisition: The receiver receives CSI that passes through the human body in the room and is reflected by the surrounding walls. The camera synchronously collects video data. By processing the received CSI signal, the reflection and transmission characteristics of the signal around the human body are captured, and the feature information reflecting the change of human posture is extracted. At the same time, the video data obtained by the synchronous camera provides the reference data of the real posture skeleton for the annotation of posture estimation. Data set production: According to the collected channel state information CSI signal and video frame, the CSI data frame and the video frame are aligned using the timestamp of the CSI and video frame, and then packaged into a training set. In this process, the characteristics of the WiFi signal are aligned with the real posture information in the video to form a data set, which ensures the accuracy of data annotation and provides high-quality samples for subsequent model training. Network training :The preprocessed data set is sent to the artificial neural network for training, and the human posture estimation is obtained according to the amplitude and phase feature information in the CSI signal. The real posture skeleton of the image of the aligned video frame is used as an annotation, and the cross entropy loss is used to train the neural network until the network converges to obtain a trained human posture estimation model. Through the amplitude and phase feature learning during the training process, the network can accurately extract the human posture information in the WiFi signal; Compared with the traditional solution that only relies on a single signal feature, the present invention improves the accuracy of posture estimation through multi-dimensional feature analysis, real-time detection: In the detection stage, the transmitter always transmits the WiFi signal. When the human body enters the room, the CSI signal will be reflected, refracted and penetrate the human body. The receiving end receives the CSI signal and transmits it to the server through UDP at the same time. The trained human posture estimation model displays the human posture to achieve the real-time purpose. In the real-time detection stage, the system can quickly respond to the entry of the human body. The CSI signal transmitted from the receiving end is instantly analyzed by the trained model to generate the current human posture display, realizing real-time posture tracking that does not rely on visual input, with the advantages of privacy protection and all-weather monitoring, effectively filling the deficiencies of traditional visual monitoring in privacy, real-time and environmental adaptability.

[0032] The channel state information includes amplitude information and phase information.

[0033] Specifically, channel state information (CSI) refers to the change information generated after the WiFi signal interacts with the environment during the propagation process. The amplitude information reflects the attenuation of the signal energy after passing through obstacles such as the human body and walls, while the phase information records the phase shift experienced by the signal on the propagation path. By obtaining this information, the characteristics of reflection, transmission and refraction of the WiFi signal around the human body can be captured, thereby obtaining changes in human posture without relying on vision. This technology is based on the sensitivity of wireless signals, that is, when the human posture changes, the amplitude and phase characteristics of the WiFi signal also produce small but detectable fluctuations. These fluctuations are detected by Complex algorithm processing can accurately infer the posture state of the human body. Compared with traditional visual posture estimation solutions, the application of CSI data is not affected by lighting conditions and is not limited by physical obstacles. It can penetrate walls to achieve real-time monitoring in a hidden state, and has stronger privacy protection and environmental adaptability. At the same time, the combination of amplitude and phase features in the CSI signal enables the system to more comprehensively capture subtle changes in human posture. Compared with solutions that use amplitude or phase features alone, this technical solution improves the accuracy and robustness of posture estimation, effectively overcomes the impact of environmental noise and dynamic interference on the signal, and ensures the stability of the system in complex scenarios.

[0034] The preprocessing step includes applying circularity denoising to the channel state information using a Mobius transform.

[0035] Specifically, the Möbius transform is a mathematical operation that maps channel state information to the complex plane. Through this transformation, the noise in the signal can be separated from the target human posture information, and the roundness feature can be used to more accurately extract useful channel information. In the CSI signal, the fluctuations in amplitude and phase information often contain various noises in the environment, such as wall reflections, multipath effects, etc. These noises will interfere with the accuracy of human posture estimation. By performing a Möbius transform on the signal data, the interference part in the signal can be mapped to a position far away from the target feature, and the human posture related information can be gathered in a specific area on the complex plane, so that it is easier to identify signal features related to human movements. The roundness denoising method can significantly enhance the stability of the signal, so that the slight changes in human posture can be more clearly presented in the denoised signal. Compared with the traditional filtering denoising method, it can more effectively handle nonlinear noise interference and improve the system's adaptability to complex dynamic scenes. This preprocessing step not only improves the subsequent neural network model's ability to parse CSI signals, but also reduces the impact of environmental noise on posture estimation, so that the system can achieve high robustness and accuracy in a variety of scenarios to meet the needs of real-time human posture tracking.

[0036] The artificial neural network includes a convolutional neural network UNET and a recurrent neural network GRU.

[0037] Specifically, the neural network architecture combines the advantages of convolutional neural network (CNN) and recurrent neural network (RNN), uses the UNET network to extract the spatial features of CSI signals, and then uses the GRU network to capture the dynamic changes in the time series. UNET, as a typical convolutional neural network structure, can effectively extract multi-level spatial information in the signal with its encoding-decoding structure, so that the system can identify the spatial distribution characteristics of human posture. At the same time, the GRU network, as a recurrent neural network, can process the time dependency in the signal through the design of its gate control unit, thereby tracking the continuous changes of human posture. Compared with the traditional neural network architecture, the solution of the present invention can It can capture subtle changes in human posture in real time and maintain high accuracy and robustness in a noisy environment. The joint use of UNET and GRU provides stronger feature expression capabilities in human posture estimation. UNET captures the complex spatial patterns in CSI signals, while GRU effectively solves the time series feature problems caused by human motion. The multi-level information fusion of this network structure improves the accuracy and real-time performance of posture estimation. Compared with a single architecture that only uses CNN or RNN, it solves the problem of low accuracy in human posture recognition in dynamic scenes, making the system more efficient and stable when processing complex human activities, and is suitable for a variety of intelligent monitoring scenarios, such as smart homes, health monitoring, and security monitoring.

[0038] The pose estimation model is trained with a large amount of CSI data of different human poses to improve the accuracy and robustness of pose estimation.

[0039] Specifically, by widely collecting and using CSI data of different human postures for training, the model can learn the change pattern of WiFi signals under various postures, thereby forming a highly generalized ability for different human postures. Each human posture will have different effects on the amplitude and phase of the CSI signal. Therefore, the more postures covered in the training process, the stronger the model's understanding and recognition ability of various complex postures. During the training process, by continuously inputting rich CSI sample data, the model can recognize slight posture changes, and adjust network parameters through repeated training to gradually adapt to different types of dynamic postures. Compared with models that rely on a small number of samples or specific scene training, the posture estimation model of the present invention has higher accuracy and robustness, and can maintain stable posture recognition effects under different environments, different human characteristics and even multiple interference conditions. This extensive training greatly reduces the model's dependence on specific environments in practical applications, and improves the adaptability and reliability of posture estimation under different backgrounds, so that the system can not only perform well in a specific experimental environment, but also achieve stable posture tracking and recognition in daily applications, meeting the real-time monitoring needs of various scenarios, such as elderly care, security monitoring and behavior analysis.

[0040] The posture estimation includes a time series action for processing dynamic changes to achieve real-time continuous tracking of human body movements.

[0041] Specifically, by processing time series actions, the system can continuously track the dynamic posture changes of the human body and achieve coherent capture of the actions. Specifically, the posture estimation model continuously samples the CSI signal in the time dimension, and combined with the powerful parsing ability of the recurrent neural network for time series information, it can accurately identify the complete trajectory of the start, process and end of the action. As the human body's movements continuously change, the amplitude and phase information in the CSI signal will also show corresponding dynamic patterns. By capturing the changing trends of these patterns, the model can predict and identify the real-time movement posture of the human body. Different from traditional static posture estimation, this method can not only capture the posture information at the current moment, but also grasp the action based on the characteristics of the time series. The continuity of the posture estimation can achieve accurate real-time tracking in a rapidly changing posture environment. Compared with the traditional method that shows the shortcomings of delayed or interrupted motion recognition in dynamic scenes, this solution can provide smooth and continuous posture tracking effects in various motion states, and is not affected by changes in motion speed and amplitude. This feature not only improves the real-time performance of posture estimation, but also makes the system highly sensitive to complex movements and can accurately distinguish different types of dynamic behaviors. It is particularly suitable for posture monitoring needs with strong continuity and diverse changes, such as real-time monitoring of possible falls of the elderly in elderly care, or monitoring the continuous behavior trajectory of residents in smart homes, which greatly expands the application scenarios and practical value of WiFi posture estimation systems.

[0042] The method can be widely used in the fields of elderly care and smart home.

[0043] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time through-wall human posture estimation method based on WIFI equipment, characterized in that: The following steps are involved: System deployment: The transmitter and receiver are equipped with inter5300 network cards and multiple transmitting antennas. A transmitter and a receiver are placed on both sides of the wall, and a synchronous camera is placed on the transmitter. Data collection: The receiving end receives the CSI reflected by the human body in the room and the surrounding walls, and the camera synchronously collects video data; Dataset creation: Based on the collected channel state information CSI signals and video frames, the CSI data frames and video frames are aligned using the timestamps of the CSI and video frames, and then packaged into a training set; Network training: The preprocessed data set is sent to the artificial neural network for training. The human posture estimation is obtained based on the amplitude and phase feature information in the CSI signal. The real posture skeleton of the image of the aligned video frame is used as an annotation. The neural network is trained with cross entropy loss until the network converges to obtain a trained human posture estimation model. Real-time detection: During the detection phase, the transmitter continuously transmits WiFi signals. When a human body enters the room, the CSI signal will be reflected, refracted, and penetrate the human body. The receiver receives the CSI signal and transmits it to the server through UPD. The trained human posture estimation model displays the human posture, achieving the real-time purpose.

2. A real-time through-wall human posture estimation method based on WIFI equipment according to claim 1, characterized in that: The channel state information includes amplitude information and phase information.

3. The real-time through-wall human posture estimation method based on WIFI equipment according to claim 1 is characterized in that: The preprocessing step includes applying circularity denoising to the channel state information using a Mobius transform.

4. The real-time through-wall human posture estimation method based on WIFI equipment according to claim 1 is characterized in that: The artificial neural network includes a convolutional neural network UNET and a recurrent neural network GRU.

5. The real-time through-wall human posture estimation method based on WIFI equipment according to claim 1 is characterized in that: The pose estimation model is trained with a large amount of CSI data of different human poses to improve the accuracy and robustness of pose estimation.

6. The real-time through-wall human posture estimation method based on WIFI equipment according to claim 1 is characterized in that: The posture estimation includes a time series action for processing dynamic changes to achieve real-time continuous tracking of human body movements.

7. The real-time through-wall human posture estimation method based on WIFI equipment according to claim 1 is characterized in that: The method can be widely used in the fields of elderly care and smart home.

Images