Fall detection system and method based on channel state information of Wi-Fi router

Abstract

The invention discloses a fall detection system and method based on channel state information of a Wi-Fi router. The fall detection system comprises a CSI data acquisition module, a preprocessing module, a behavior detection module and a real-time alarm module, the CSI data acquisition module comprises two Wi-Fi routers which are provided with OpenWrt systems and provided with Atheros AR9580 network cards, the two routers are respectively used as a transmitting end and a receiving end, are respectively connected with a computer, and are respectively used for transmitting and receiving CSI data and acquiring channel state information of the Wi-Fi routers; the preprocessing module is used for carrying out noise reduction processing on the collected CSI data; the behavior detection module is used for carrying a bidirectional long-short-term memory network model; the CSI data acquisition module adjusts a signal propagation path through a multipath reflection optimization technology, and the real-time alarm module triggers an alarm in combination with a dynamic threshold adjustment technology. The method is used for solving the problems of high invasiveness, high cost, low accuracy and insufficient privacy protection of the existing fall detection technology in an electric power working environment.

Description

Technical Field

[0001] The present invention relates to a fall detection system and method based on channel state information of a Wi-Fi router. Background Art

[0002] In the power industry, fall accidents are one of the main causes of casualties among inspection and maintenance personnel, seriously threatening personnel safety and affecting work efficiency. Traditional fall monitoring methods have obvious shortcomings: Manual monitoring: It relies on the subjective judgment of managers and is limited by the complex environment of the power site, making it difficult to monitor comprehensively and effectively. Wearable device solutions: Workers are required to actively wear sensors, which increases the burden and affects work efficiency. The weight and discomfort of the equipment may cause health risks. At the same time, there are detection errors caused by individual differences, and workers are less willing to wear them. Non-wearable technology based on cameras: There is a risk of privacy infringement, high installation and maintenance costs, and it is affected by field of view, light and occlusion, and the detection accuracy and coverage are limited.

[0003] In recent years, activity recognition technology based on wireless signals has gained increasing attention due to its non-invasive and privacy-preserving properties. Wi-Fi channel state information (CSI) can capture changes in the signal propagation path. When a person moves between the transmitter and receiver, the amplitude and phase of the CSI signal are disturbed, making it possible to detect falls without contact. Summary of the Invention

[0004] The purpose of the present invention is to provide a fall detection system and method based on the channel state information of Wi-Fi routers, so as to solve the problems of existing fall detection technology in power operation environments, such as high intrusion, high cost, low accuracy and insufficient privacy protection.

[0005] In order to solve the above problems, the technical solution of the present invention is: A fall detection system based on the channel status information of Wi-Fi routers includes: a CSI data acquisition module, a pre-processing module, a behavior detection module and a real-time alarm module; The CSI data acquisition module includes two Wi-Fi routers running OpenWrt and equipped with Atheros AR9580 network cards. The two routers serve as transmitters and receivers, respectively, and are connected to a computer. The two routers are used to send and receive CSI data, collecting channel status information from the Wi-Fi routers. The pre-processing module is used to perform noise reduction on the collected CSI data; The behavior detection module is used to carry a bidirectional long short-term memory network model to identify falling behavior; The CSI data acquisition module adjusts the signal propagation path through multipath reflection optimization technology, and the real-time alarm module triggers an alarm in combination with dynamic threshold adjustment technology. Furthermore, the Wi-Fi router is configured in “injection monitoring mode” to collect channel state information including amplitude, phase and RSSI, and simultaneously transmit the data to MATLAB for storage via a computer.

[0006] Furthermore, the bidirectional long short-term memory network model learns the signal characteristics of body tilt before falling and body smoothness after falling. Furthermore, the CSI data acquisition module adjusts the signal propagation path through multipath reflection optimization technology, and the real-time alarm module triggers an alarm in combination with dynamic threshold adjustment technology. Furthermore, the Wi-Fi router is connected to the computer via a network cable. Furthermore, the preprocessing module and behavior detection module implement data processing and model training through the MATLAB platform.

[0007] The method for detecting a power worker's fall based on channel state information of a Wi-Fi router includes the following steps: Step 1: CSI data acquisition: Continuously send Wi-Fi data packets through a Wi-Fi router configured in "injection monitoring mode" to collect CSI data and transfer it to MATLAB for storage; Step 2: CSI data preprocessing: using a median filter algorithm to reduce noise on the CSI amplitude information to retain the signal characteristics related to human activity; Step 3: Fall behavior recognition: The preprocessed CSI data is input into the BLSTM model to analyze the time series features to distinguish falls from other actions; Step 4: Real-time alarm triggering. When a fall is detected, security personnel are notified via SMS and alarm lights and event information is recorded. Logic is optimized to reduce false alarms and delays.

[0008] Furthermore, in step 1, the router adjusts the data packet transmission rate, the number of data packets sent, the number of transmitting antennas, and the long / short guard interval through the "injection monitoring mode". Furthermore, the BLSTM model in step 3 captures the state differences before and after the falling behavior through bidirectional time series analysis. Furthermore, the alarm triggering in step 4 is combined with dynamic threshold adjustment technology to adapt to environmental changes.

[0009] The beneficial effects of the present invention are: 1. Non-invasiveness and privacy protection: This invention uses WiFi signals for fall detection. It does not require workers to wear any equipment or install cameras, which will not interfere with workers' freedom of movement and work efficiency. At the same time, it avoids the privacy leakage problems that may be caused by image monitoring, and meets the current requirements for privacy protection on construction sites.

[0010] 2. High Precision and Real-Time: By leveraging a BLSTM (bidirectional long short-term memory) model, this invention fully captures characteristic information before and after a fall and effectively distinguishes actions similar to falls (such as lying down or sitting down), achieving a fall detection accuracy of up to 99%. The system also features low latency and can instantly issue an alarm upon detecting a fall, shortening response time, facilitating timely rescue, and minimizing the severity of injuries.

[0011] 3. Low Cost and High Adaptability: This invention leverages existing commercial Wi-Fi equipment, eliminating the need for expensive dedicated equipment and significantly reducing system deployment costs. Furthermore, the system is capable of operating in complex and changing construction site environments, effectively adapting to environmental noise interference and signal variations through multipath reflection optimization and dynamic threshold adjustment technology.

[0012] 4. Easy to deploy and scalable: This system leverages existing Wi-Fi infrastructure, making the hardware easy to deploy and the algorithm highly versatile, enabling widespread application in diverse environments, including power operations and factory workshops. Future expansion will allow for simultaneous monitoring of multiple workers and the addition of detection capabilities for other dangerous behaviors, such as slips and imbalances, to further enhance construction site safety management.

[0013] 5. Efficient Data Utilization and Continuous Optimization: This invention fully exploits the behavioral information in wireless signals by preprocessing, classifying, and modeling CSI signals. The system incorporates a comprehensive data processing and training process, enabling rapid model updates as new data is collected, continuously improving detection accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the accompanying drawings: Figure 1 It is a schematic diagram of the principle of the present invention, Figure 2 It is a structural schematic diagram of the present invention, Figure 3 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0016] Example 1: like Figures 1 to 3 As shown, the fall detection system based on the channel state information of the Wi-Fi router includes: a CSI data acquisition module, a preprocessing module, a behavior detection module and a real-time alarm module; The CSI data acquisition module consists of two Wi-Fi routers running OpenWrt and equipped with Atheros AR9580 network cards. These routers serve as transmitters and receivers, respectively. Each router is connected to an Intel Core i5 processor-based laptop, which serves as the data transmitter and receiver, respectively. The laptops are equipped with Windows 10, Xshell6, and WinSCP for data processing and transmission. At the actual construction site, the routers and laptops are positioned to ensure a clear signal path, maintaining a distance of approximately 4 meters between the transmitter and receiver to optimize the signal-to-noise ratio. The routers are configured in Injector Monitor Mode, which adjusts the packet transmission rate, number of packets sent, number of transmit antennas, and long / short guard intervals. Activating monitor mode captures standardized CSI data, supporting subsequent model training.

[0017] The two routers are used to send and receive CSI data, respectively. The receiving computer extracts the Wi-Fi router's channel state information; the extracted information includes the signal's amplitude, phase, RSSI (received signal strength indicator), and data rate. To overcome the limitations of the router's limited storage capacity, the data transmission process is optimized, including: (1) The transmitter transmits data to the receiver, which calculates the CSI in real time. (2) The raw CSI data is transmitted to a laptop via a receiver router and then processed, saved, and visualized using MATLAB R2021a. Data processing also includes preprocessing steps, such as noise filtering, to improve data quality and accuracy.

[0018] The system architecture is designed based on the "transmitter-receiver" mode (Tx-Rx Communication), which builds a real-time monitoring system through continuous Wi-Fi data packet transmission and reception.

[0019] The preprocessing module is used to perform noise reduction on the collected CSI data; CSI data consists of two components: amplitude and phase. In this study, to better identify falls, only amplitude information is selected for analysis because it is more sensitive to changes in human activity. However, due to variations in transmission power and rate, the raw CSI data often contains a large amount of noise unrelated to worker behavior, which can interfere with data accuracy. To retain valid information in the data and eliminate significant noise, this study used a median filtering method for preprocessing. Median filtering can effectively smooth sharp changes in the data, reduce noise interference on the signal, and thus improve data quality.

[0020] The behavior detection module is used to carry a bidirectional long short-term memory network (BLSTM) model to identify falling behaviors. The bidirectional long short-term memory network (BLSTM) can comprehensively understand the before and after features of falling behaviors and has the following advantages: First, LSTM can automatically extract features without the need for complex data preprocessing, which meets the research objectives. Second, the long short-term memory network (LSTM) can preserve the temporal state information of activities and effectively distinguish between similar actions such as "lying down" and "falling." Although these actions are similar in speed changes, their before and after states are different. The memory capacity of the long short-term memory network (LSTM) enables the system to correctly identify these activities, thereby achieving accurate detection of falling behaviors. The real-time alarm module is used to trigger an alarm when a fall behavior is detected. CSI data propagation model and its relationship with falling behavior: The propagation model of the CSI signal is based on the path attenuation model (e.g. Figure 2 As shown in Figure 1.1, Wi-Fi signals are reflected by various objects when they propagate through physical space. Physical space limits wireless signal propagation, causing the received signal to carry information about its propagation path. When construction workers are present in the environment, scattering from their bodies introduces additional signal paths. Specifically, the signal's transmission power is , receiving power , transmission and reception gain ,wavelength , propagation distance and worker volume disturbance Together they determine the signal propagation quality, as shown in the following formula:

[0021] This shows that there is a direct connection between CSI signals and workers' falling behaviors, and accurate detection of behaviors can be achieved by analyzing signal changes. Furthermore, the Wi-Fi router is configured in "injection monitoring mode" to collect channel state information, including amplitude, phase, and RSSI. This data is then transferred to MATLAB via a computer for storage. MATLAB is used to analyze and store the collected CSI data, providing high-quality data support for subsequent model training. Furthermore, MATLAB generates CSI data waveforms, visually displaying changes in signal amplitude and phase, ensuring that the collected CSI data meets analysis requirements.

[0022] Furthermore, the bidirectional long short-term memory network model learns the signal characteristics of body tilt before falling and body smoothness after falling. Furthermore, the CSI data acquisition module uses multipath reflection optimization technology to adjust the signal propagation path, and the real-time alarm module triggers alarms in conjunction with dynamic threshold adjustment technology. By adjusting router parameters (such as transmission rate and number of antennas) and deployment location, signal reflection paths caused by human activity are enhanced and interference from irrelevant objects in the environment is suppressed, enabling the CSI signal to more accurately reflect the characteristics of human fall behavior. The implementation process includes real-time CSI data collection and analysis of the signal strength, noise level, and multipath reflection characteristics in the current environment. Based on real-time monitoring data, combined with dynamic threshold adjustment technology, the fall detection threshold is automatically calculated and updated. For example, when environmental noise increases, the threshold is raised to avoid false alarms; when signal strength changes, the threshold is adjusted to ensure detection sensitivity. The dynamically adjusted threshold is used as one of the input parameters of the Bidirectional Long Short-Term Memory Network (BLSTM) model, helping the model to more accurately distinguish falls from other actions (such as lying down or sitting down).

[0023] The principle is that environmental changes (such as people moving around or equipment operating) can cause baseline fluctuations in the CSI signal. Fixed thresholds can easily lead to false positives (e.g., misidentifying normal movement as a fall) or false negatives (e.g., failing to detect a true fall). By analyzing the statistical characteristics of CSI data (such as mean, variance, and signal fluctuation amplitude) in real time, a dynamic threshold model is established. This model adaptively adjusts the threshold based on environmental changes, aligning detection criteria with current signal characteristics and improving recognition accuracy. This prevents false alarms caused by environmental noise or non-falling movements (e.g., rapidly lying down), thereby improving system reliability. Furthermore, dynamic threshold adjustment, combined with the BLSTM model, enables rapid response to fall events, reduces alert latency, and meets emergency rescue needs.

[0024] Furthermore, the Wi-Fi router is connected to the computer via a network cable, forming a complete data collection and storage process.

[0025] Furthermore, the preprocessing module uses the MATLAB platform's median filtering algorithm to reduce noise in CSI data, while the behavior detection module uses the MATLAB platform's deep learning toolbox to complete BLSTM model training and fall behavior identification. Using the MATLAB platform's median filtering algorithm to reduce noise in CSI data effectively removes noise unrelated to worker behavior from the raw data, retaining signal characteristics related to human activity and improving data quality. Using the MATLAB deep learning toolbox to complete BLSTM model training and fall behavior identification, bidirectional time series analysis can be used to capture differences in signal characteristics before and after a fall, accurately distinguishing falls from similar actions such as lying down and sitting down, improving detection accuracy. Furthermore, relying on MATLAB's integrated environment, the data processing and model training processes are efficiently integrated, enhancing the system's overall detection performance.

[0026] Example 2: The method for detecting a power worker's fall based on channel state information of a Wi-Fi router includes the following steps: Step 1: CSI data acquisition: Two Wi-Fi routers installed with OpenWrt system and equipped with Atheros AR9580 network cards are used as the transmitter and receiver, respectively connected to a computer (such as a laptop with Intel Core i5 processor and Windows 10 system), and the routers are configured to "injection monitoring mode". In this mode, the packet transmission rate, number of transmissions, number of transmitting antennas and long / short protection intervals can be adjusted (for example, the distance between the transmitter and the receiver is set to approximately 4 meters to optimize the signal-to-noise ratio), and Wi-Fi packets are continuously sent. The receiving router extracts CSI data including amplitude, phase, and RSSI in real time, transmits it to the MATLAB R2021a platform via a network cable for storage, and completes preliminary visualization and format standardization of the raw data at the same time; Step 2: CSI data preprocessing. MATLAB's built-in median filter algorithm (such as the medfilt2 function) is used to reduce noise on the collected CSI amplitude information (because amplitude is more sensitive to human movement). This algorithm calculates the signal median using a sliding window, effectively filtering out noise caused by transmission power fluctuations, environmental reflections, and other factors, while retaining signal disturbances caused by human motion (such as amplitude changes due to body tilt and movement). During preprocessing, the data is simultaneously normalized and segmented to ensure consistency and identifiability of the signal sequence input to the model, laying a foundation for high-quality data for fall behavior analysis. Step 3: Fall behavior recognition: The preprocessed CSI data is fed into a bidirectional long short-term memory (BLSTM) model built with the MATLAB Deep Learning Toolbox. This model uses bilstmLayer to define a bidirectional LSTM layer and trains it with the trainNetwork function. This model leverages bidirectional time series analysis to capture the differences between the pre- and post-fall states (e.g., the sudden change in the signal indicating body tilt before a fall and the smoothness of the signal indicating the body's smooth contact with the ground after a fall). During model training, CSI data from actions such as falling, lying down, sitting, and walking are used as samples. Backpropagation is used to optimize weights and automatically extract time series features. This allows the model to distinguish falls from other similar actions and avoid misjudgments. Step 4: Real-time alarm triggering. When the BLSTM model detects a fall, the real-time alarm module triggers an alarm using dynamic threshold adjustment (adaptively updating the detection threshold based on ambient noise and signal strength) and multipath reflection optimization (enhancing human signal reflection and suppressing environmental interference). This alert is sent to security personnel via SMS, illuminating an on-site alarm light, and recording the event time, location, and CSI feature data in the system backend. To reduce false alarms, the alarm logic incorporates lightweight algorithms (for example, using a sliding window to determine the probability of a fall within consecutive frames, e.g., "a window size of 10 frames triggers an alarm if 5 consecutive frames are identified as a fall"). This reduces interference from non-falling actions (such as rapidly lying down), and maintains response latency within milliseconds, ensuring timely initiation of rescue measures.

[0027] Furthermore, in step 1, the router uses "injection monitoring mode" to adjust the packet transmission rate, number of packets sent, number of transmit antennas, and long / short guard intervals. By adjusting the router's packet transmission rate, number of packets sent, number of transmit antennas, and long / short guard intervals in "injection monitoring mode," the quality of CSI data acquisition can be precisely optimized. Adjusting the transmission rate and number of packets ensures the temporal continuity and sampling density of the CSI signal, preventing the omission of fall signatures due to data loss. Adjusting the number of transmit antennas enhances the spatial diversity of multipath signals, highlighting differences in signal reflections caused by human activity and suppressing interference from irrelevant environmental reflections. Configuring long / short guard intervals adapts to multipath delays in complex construction site environments, reducing inter-symbol interference (ISI). This ensures that the collected CSI data (including amplitude, phase, and RSSI) more accurately reflects posture changes during a fall, providing raw data with a high signal-to-noise ratio for subsequent median filtering and BLSTM model recognition. Ultimately, this improves fall detection accuracy and ensures low-latency alarm responses.

[0028] Furthermore, the BLSTM model in step 3 captures the state differences before and after a fall through bidirectional time series analysis. This bidirectional time series analysis captures the dynamic feature sequences of a fall from both the forward and reverse time dimensions. It accurately identifies the sudden changes in the CSI signal (such as amplitude fluctuations and phase shifts) caused by body tilt before a fall and the smooth signal changes after the body touches the ground. This effectively distinguishes a fall from similar actions such as lying down or sitting down. For example, the CSI signal changes associated with lying down lack the pre-fall tilt warning feature. However, by memorizing the state differences before and after (such as the persistent signal perturbations during the tilt phase and the smooth signal transition after contact), the BLSTM avoids misclassifying regular actions as falls, improving fall detection accuracy. Furthermore, leveraging the parallel computing capabilities of bidirectional time series, it achieves millisecond-level response latency, ensuring real-time and reliable alarm triggering.

[0029] Furthermore, the alarm triggering in step 4 incorporates dynamic threshold adjustment technology to adapt to environmental changes. By analyzing environmental characteristics such as the noise level and signal strength fluctuations of the CSI signal in real time, the fall detection threshold is dynamically adjusted to avoid false alarms or missed alarms due to environmental changes. For example, when factors such as movement of personnel and equipment operation in a construction site environment cause the CSI signal baseline to fluctuate, the dynamic threshold can automatically increase or decrease with the noise level, ensuring that the detection standard matches the current environment. This effectively distinguishes true falls from non-falling movements such as rapid lying down or squatting, reducing false alarm rates while maintaining detection accuracy (e.g., 99%). Furthermore, this technology, combined with a lightweight algorithm, can rapidly update the threshold in the event of sudden environmental changes, ensuring that the alarm triggering delay is controlled to the millisecond level, enabling real-time response to fall events and improving the timeliness and reliability of safety rescue in power operation scenarios.

[0030] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The scope of protection of the present invention should not be regarded as limited to the specific forms described in the embodiments. The scope of protection of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A fall detection system based on Wi-Fi router channel state information, characterized in that: include: CSI data acquisition module, pre-processing module, behavior detection module and real-time alarm module; The CSI data acquisition module consists of two Wi-Fi routers running OpenWrt and equipped with Atheros AR9580 network cards. The two routers serve as transmitters and receivers, respectively, and are connected to a computer. The two routers are used to send and receive CSI data, while the receiving computer extracts the Wi-Fi routers' channel status information. The pre-processing module is used to perform noise reduction on the collected CSI data; The behavior detection module is used to carry a bidirectional long short-term memory network model to identify falling behavior; The real-time alarm module is used to trigger an alarm when a fall behavior is detected.

2. The fall detection system based on Wi-Fi router channel state information according to claim 1, characterized in that: The Wi-Fi router is configured in “injection monitoring mode” to collect channel state information including amplitude, phase and RSSI, and transmit the data to MATLAB via a computer for storage.

3. The fall detection system based on Wi-Fi router channel state information according to claim 2, characterized in that: The bidirectional long short-term memory network model learns the signal characteristics of body tilt before falling and body smoothness after falling.

4. The fall detection system based on Wi-Fi router channel state information according to claim 2, characterized in that: The CSI data acquisition module adjusts the signal propagation path through multipath reflection optimization technology, and the real-time alarm module triggers an alarm in combination with dynamic threshold adjustment technology.

5. The fall detection system based on Wi-Fi router channel state information according to claim 2, characterized in that: The Wi-Fi router is connected to the computer via a network cable.

6. The system according to claim 2, wherein: The preprocessing module and behavior detection module realize data processing and model training through the MATLAB platform.

7. A method for detecting a power worker fall using the system based on channel state information of a Wi-Fi router according to any one of claims 2 to 6, characterized in that: The following steps are involved: Step 1: CSI data acquisition: Use a Wi-Fi router configured in "injection monitoring mode" to continuously send Wi-Fi data packets, collect CSI data, and transfer it to MATLAB for storage. Step 2: CSI data preprocessing: using a median filter algorithm to reduce noise on the CSI amplitude information to retain the signal characteristics related to human activity; Step 3: Fall behavior recognition: The preprocessed CSI data is input into the BLSTM model to analyze the time series features to distinguish falls from other actions; Step 4: Real-time alarm triggering. When a fall is detected, security personnel are notified via SMS and alarm lights and the event information is recorded.

8. The method for detecting power worker falls based on channel state information of Wi-Fi routers according to claim 7, characterized in that: In step 1, the router adjusts the data packet transmission rate, the number of data packets sent, the number of transmitting antennas, and the long / short guard interval through the "injection monitoring mode".

9. The method for detecting power worker falls based on channel state information of Wi-Fi routers according to claim 7, characterized in that: In step 3, the BLSTM model captures the state differences before and after the falling behavior through bidirectional time series analysis.

10. The method for detecting power worker falls based on channel state information of Wi-Fi routers according to claim 7, characterized in that: The alarm triggering in step 4 is combined with dynamic threshold adjustment technology to adapt to environmental changes.

Images