A self-adapting multi-modal collaborative scheduling method and system based on a radio frequency channel sensing hub

Abstract

This invention discloses an adaptive multimodal collaborative scheduling method and system based on a radio frequency channel sensing hub. The system uses the radio frequency channel sensing hub module as the main sensing module, responsible for primary anomaly detection and event classification, outputting anomaly event types and capability level identifiers. An environment-adaptive scheduling engine dynamically wakes up a group of auxiliary sensing modules for cross-verification based on anomaly event types. Core innovations include: body pose compensation to eliminate false alarms due to device shaking; radio frequency fingerprint-guided optical retroreflection for verification against surreptitious filming; macroscopic feature downsampling and graded wake-up to reduce power consumption; and capability level detection and dynamic adaptation to achieve technological evolution compatibility. The system does not collect optical images throughout the process, achieving non-visual privacy protection, and is suitable for scenarios such as home intrusion prevention, hotel anti-surveillance, and warehouse security.

Description

Technical Field

[0001] This invention relates to the field of IoT security technology, specifically to an environment-adaptive multimodal collaborative sensing method and system based on Channel State Information (CSI) as the central hub. The system uses WiFi channel state information as the main sensing module, combined with an auxiliary sensing module group including an inertial measurement unit (IMU), an active optical retroreflective module, infrared thermal imaging, vibration detection, and millimeter-wave radar. Through an environment-adaptive scheduling engine, it achieves multimodal cross-verification and is suitable for scenarios such as home intrusion prevention, hotel anti-spy camera measures, warehouse security, commercial building security, and industrial park perimeter security. Background Technology

[0002] Existing technology analysis

[0003] In the field of security and anti-spy camera surveillance, sensing technology has mainly evolved along the following routes, and the representative literature for each route is as follows:

[0004] Optical / PIR Intrusion Detection Baseline. A representative patent is Hikvision CN115527328B "An Intrusion Detection Method and Device" (authorization announcement date 2026-07-10) [1], which uses a single passive infrared (PIR) sensor in conjunction with a reflector array and a fixed threshold to trigger an alarm by detecting changes in human body thermal radiation. This approach is purely optical sensing and does not involve radio frequency channel status information (CSI) sensing, inertial measurement unit (IMU) body pose compensation or auxiliary module coordinated scheduling. It can only detect moving heat sources, fails to detect stationary targets, and is easily interfered with by air conditioning hot air, sunlight, pets, etc.

[0005] CSI single-modal intrusion detection. The representative patent is CN116168499B of Nankai University, “Indoor Passive Sensing and Intrusion Detection Method Based on CSI Amplitude Stabilization Analysis” (authorization announcement date 2026-02-03) [2]. It collects Wi-Fi CSI data and identifies the presence status of personnel through amplitude stabilization analysis, and uses a fixed threshold for judgment. This scheme is the closest existing technology to the present invention; its objective limitations are: lack of body pose compensation mechanism, CSI is extremely sensitive to the shaking of the device itself, and is prone to generating a large number of false alarms; no environment adaptive scheduling engine is set, and each sensing unit works independently; no optical retroreflection mechanism is used for anti-spy camera verification; no classification of sensing capabilities is performed, and no three-classification of abnormal events as micro-motion abnormalities, radio frequency abnormalities, and stationary foreign objects is performed.

[0006] Visual fusion approach. A representative patent is CN121410835A from Zhejiang University of Technology entitled "A UAV detection system based on WIFI-CSI and YOLO fusion" (application publication date 2026-01-27, legal status: under review) [3], which fuses CSI signals with camera images (YOLO target detection). This approach relies on cameras to collect optical images, which poses a risk of privacy leakage in private scenarios such as hotels and homes, making it difficult to popularize. This invention differentiates itself by adopting "non-visual privacy protection" - without collecting optical images throughout the process, only using CSI to guide active red LEDs and photoelectric sensors to perform reverse reflection confirmation, thereby achieving the detection purpose under the premise of protecting privacy.

[0007] Prior research on intrusion detection using CSI in the same domain. AR-Alarm (LI S et al., ICOST 2017) [4] has achieved adaptive robust intrusion detection based on commercial Wi-Fi CSI, which proves the technical feasibility of using CSI as the main sensing method for intrusion detection and constitutes prior research in the same domain as this invention.

[0008] The feasibility of CSI and IMU fusion has been verified. The IMFi system proposed by GUO et al. (IEEE Access, 2024) [5] and SONG et al. (MSN 2021) [6] has verified that multimodal fusion of CSI and inertial measurement unit (IMU) can improve perception robustness: [5] uses CSI and IMU for human activity recognition, and [6] uses CSI and IMU (waist / ankle) for cross-environment gait identity recognition. The IMUs in the above literature all serve the purpose of "identifying human activity / gait identity", which is different from the technical purpose of this invention of "using IMU body pose compensation to solve CSI false alarms caused by device shaking". The two are not simply pieced together, but are aimed at different technical problems.

[0009] Privacy compliance with the principle of minimum necessity. GB / T 35273-2020 Information Security Technology Personal Information Security Specification [7] establishes the principle of "minimum necessity" for personal information processing. The design of this invention, which uses CSI as the main sensing, wakes up the auxiliary module on demand, and only uses IMU for body pose compensation (without collecting identity information), conforms to the principle of minimum necessity, and reduces the collection of personal information to a minimum while achieving the purpose of security.

[0010] Based on the above existing technologies, it can be seen that although the CSI single-modal solution [2] can achieve non-visual perception, it is limited by false positives and lack of scheduling; although the visual fusion route [3] improves accuracy, it introduces privacy risks; neither of them can achieve the synergistic effect of "stationary foreign object detection + low power consumption + multi-scene adaptation" under the premise of "protecting privacy (non-visual)". This is the technical problem that this invention aims to solve.

[0011] The shortcomings of existing technology

[0012] Based on the above analysis of existing technologies, current security and anti-spy camera systems mainly have the following shortcomings:

[0013] (1) Common problems of false alarms of CSI body: CSI is extremely sensitive to relative motion. Even a slight shaking of the device itself will cause a drastic change in the CSI matrix, resulting in a large number of false alarms. Existing technologies lack an effective body pose compensation mechanism.

[0014] (2) Inefficient blind scanning for anti-spy camera: Traditional anti-spy camera measures rely on manual red light blind scanning or full-frequency scanning of radio frequency signals, which cannot accurately locate hidden devices and are extremely inefficient.

[0015] (3) Privacy risks: Camera solutions involve privacy leaks and are difficult to popularize in private scenarios.

[0016] (4) Outdated technology: PIR sensors can only detect moving heat sources and fail to detect stationary targets. This technology is being phased out.

[0017] (5) Lack of coordinated scheduling: The sensors in the existing security system work independently and lack dynamic scheduling and cross-verification mechanisms based on abnormal event types.

[0018] (6) Power consumption and constant operation conflict: CSI full-band constant acquisition has high power consumption, which conflicts with the long-term deployment requirements.

[0019] (7) Lack of a compatible architecture for the evolution of sensing technologies: Existing multimodal fusion systems are typically "hard-bonded" to specific sensor hardware. As CSI technology evolves from the current "mobile sensing (Level 1)" to the future "morphological recognition (Level 2)" or "semantic understanding (Level 3)," existing systems cannot automatically adjust the scheduling strategy of upper-layer sensors according to the improvement of underlying sensing capabilities, resulting in high system upgrade costs and an inability to transition smoothly. (Level 1 mobile sensing, Level 2 morphological sensing, and Level 3 semantic sensing are all custom capability classifications defined in this application) Summary of the Invention

[0020] In view of the above-mentioned deficiencies of the prior art, the present invention provides an adaptive multimodal cooperative scheduling method and system based on the radio frequency channel sensing center. Under the premise of protecting privacy (non-visual), it overcomes the problems of false alarms of CSI body, inefficient blind scanning to prevent surreptitious shooting, lack of cooperative scheduling, and excessive resident power consumption, and achieves the cooperative effect of static foreign object detection, low power consumption and multi-scenario adaptation.

[0021] The technical solution for implementing the present invention is as follows:

[0022] This invention provides an adaptive multimodal cooperative scheduling method based on a radio frequency channel sensing center, comprising the following steps:

[0023] S1: The radio frequency channel sensing central module collects environmental radio frequency channel status data, performs primary anomaly detection, and outputs the anomaly event type, which includes micro-motion anomaly, radio frequency anomaly, and stationary foreign object; the radio frequency channel sensing central module also outputs capability level identifiers, which include L1 level motion sensing, L2 level morphological sensing, and L3 level semantic sensing.

[0024] S2: The environment adaptive scheduling engine dynamically wakes up the corresponding auxiliary perception module for cross-validation based on the abnormal event type and capability level identifier; wherein, the scheduling engine dynamically adjusts the scheduling strategy of the auxiliary perception module according to the capability level identifier: when at L1 level, the auxiliary perception module participates in fusion verification at high frequency; when at L2 or L3 level, the auxiliary perception module only participates in redundant verification when the confidence level is lower than the threshold.

[0025] S3: The fusion judgment unit performs fusion analysis on the data from the radio frequency channel sensing center module and the auxiliary sensing module to confirm abnormal events. The fusion judgment adopts a weighted voting mechanism, and the comprehensive confirmation score S = wCSI × CCSI + Σ(wi × ci × fi), where wCSI is the weight of the main module, wi is the weight of the auxiliary module, ci is the confirmation confidence level, and fi is the confirmation result coefficient. When the system identifies a high-precision sensing center at level L2 or L3, it automatically increases wCSI and decreases wi.

[0026] S4: Generate and execute response instructions based on the fusion determination results and user interaction settings.

[0027] The present invention also provides an adaptive multimodal cooperative scheduling system based on a radio frequency channel sensing center, comprising:

[0028] The radio frequency channel sensing hub module, as the main sensing module, collects radio frequency channel status information and performs primary anomaly detection and event classification, while outputting capability level identifiers, which include L1 level mobile sensing, L2 level morphological sensing and L3 level semantic sensing.

[0029] The auxiliary sensing module group includes one or more of the following: body pose sensing module, active optical retroreflection module, infrared thermal imaging, vibration detection, millimeter-wave radar, lidar, acoustic detection, door and window magnetic sensors, gas sensing, fiber optic vibration, and RFID asset tags, which can be tailored and combined according to the application level.

[0030] The environment adaptive scheduling engine dynamically adjusts the scheduling strategy of the auxiliary perception module and wakes up the corresponding module for cross-validation based on the type of abnormal event and capability level identifier. When it is at L1 level, the auxiliary perception module participates in fusion verification at a high frequency. When it is at L2 or L3 level, the auxiliary perception module only participates in redundancy verification when the confidence level is lower than the threshold.

[0031] The fusion judgment unit uses a weighted voting mechanism to perform fusion analysis and verification of the main module data and the auxiliary sensing module data. The comprehensive verification score S = wCSI × CCSI + Σ(wi × ci × fi), where wCSI is the main module weight, wi is the auxiliary module weight, ci is the verification confidence level, and fi is the verification result coefficient. When a high-precision sensing center of L2 or L3 level is identified, wCSI is automatically increased and wi is decreased.

[0032] The response execution module executes the response based on the fusion judgment result and user interaction settings.

[0033] Furthermore, the radio frequency channel sensing hub module is a data acquisition unit based on Wi-Fi channel state information (CSI), which acquires the amplitude and phase matrix of multiple OFDM subcarriers.

[0034] Furthermore, corresponding auxiliary sensing modules are scheduled for confirmation based on different abnormal event types: when a micro-motion abnormal event is output, one or more modules are selected from the body pose sensing, vibration detection, acoustic detection, door / window magnetic detection, and fiber optic vibration sensing modules for activation; when a stationary foreign object event is output, one or more modules are selected from the infrared thermal imaging, millimeter-wave radar, lidar, industrial gas sensing, and RFID asset tag modules for activation; when a radio frequency abnormal event is output, the presence of the hidden camera is confirmed by sniffing abnormal MAC addresses or data streams through cross-layer radio frequency fingerprinting, combined with the active optical retroreflection module.

[0035] Furthermore, when a micro-motion abnormality event is output, the body pose perception module is activated to perform body pose compensation: if the body pose data shows that the device body is moving, it is determined to be environmental noise and the baseline is updated adaptively; if the body pose data shows that the device body is stationary, it is determined to be external intrusion; the body pose perception module includes an accelerometer, a gyroscope, a magnetometer or a combination thereof.

[0036] Furthermore, when an abnormal radio frequency event or a stationary foreign object event is output, the abnormal MAC address or data stream is sniffed through cross-layer radio frequency fingerprinting to estimate the topology of the abnormal source area, guide the active red light LED array to directionally illuminate the area, and use a photoelectric sensor to capture the retroreflection of the lens coating to confirm the hidden camera.

[0037] Furthermore, the radio frequency channel sensing hub module adopts a graded wake-up mechanism based on downsampling of the macroscopic features of the radio frequency channel: normally, it monitors the macroscopic variance of the amplitude at a low sampling rate of 1~10Hz, and when the macroscopic variance exceeds a preset threshold, it instantly wakes up to collect full data at 50~200Hz.

[0038] Furthermore, this invention does not collect optical images throughout the entire process, but only extracts the physical layer features of the radio frequency channel and fuses them with the scalar data of the auxiliary sensing module to achieve non-visual privacy protection.

[0039] Furthermore, it also includes a trend analysis and early warning step: by calculating the slope of the change in the amplitude of the radio frequency channel signal, an early warning is triggered when the slope exceeds a preset threshold.

[0040] Furthermore, the scheduling engine controls the wake-up of the auxiliary sensing module by outputting a hardware trigger signal with specific pulse width or frequency characteristics through a hardware interrupt pin. The timing characteristics of this signal are synchronized with the frame header of the radio frequency channel sensing data acquisition at the microsecond level.

[0041] Compared with the prior art, the present invention has the following beneficial effects (the following effect descriptions are all theoretical predictions and expected effects based on published literature, module datasheets, and general signal processing theory, not experimental test data):

[0042] (1) Significantly reduce the false alarm rate of CSI: By using the body pose compensation mechanism to distinguish between the device’s own shaking and environmental intrusion, the problem of false alarms due to device shaking that has long existed in the field of CSI security has been solved.

[0043] (2) Significantly improved anti-spy camera efficiency: The traditional blind red light scanning is transformed into CSI-guided directional verification, which can accurately locate hidden spy cameras.

[0044] (3) Privacy and security protection: No optical images or videos are collected throughout the process. Only CSI subcarrier features are fused with IMU scalar data to completely eliminate the risk of visual privacy leakage. It is suitable for high privacy scenarios such as hotels and homes.

[0045] (4) The system power consumption is significantly reduced: On-demand scheduling based on abnormal event types avoids all auxiliary modules running resident.

[0046] (5) Graded wake-up further reduces power consumption: Low power mode monitors with an extremely low sampling rate and wakes up instantly when the threshold is exceeded to collect full data. Combined with intermittent hibernation, it can greatly extend the battery life.

[0047] (6) Proactive early warning capability: identify potential risks in advance through signal trend analysis, and transform passive response into proactive early warning.

[0048] (7) Wide protection scope and strong scalability: It is summarized as “environmentally adaptive multimodal collaborative sensing” rather than limited to specific sensors, covering future expansion such as millimeter wave and infrared thermal imaging, which is conducive to the evolution of B-end. Attached Figure Description

[0049] Figure 1 This is a system architecture diagram of the present invention, showing the four-layer architecture and data flow of the perception layer (CSI main module, IMU six-axis, active optical retroreflection, infrared thermal imaging / millimeter wave), the central processing layer (data preprocessing, scheduling engine, CSI-IMU fusion judgment, CSI-optical verification, CSI-infrared thermal imaging verification, trend analysis), the response layer (response execution, local storage, cloud server), and the user interaction layer (user settings, mobile app).

[0051] Figure 2 This is a flowchart of the anti-intrusion process of the present invention, which shows the complete decision-making chain from system startup → CSI low power monitoring → macro variance threshold judgment → full CSI acquisition → preprocessing and micro-motion feature extraction → IMU comparison and judgment. It includes the loop path of adaptively updating the CSI baseline when the "body is moving" and the alarm path of judging external intrusion when the "body is stationary".

[0053] Figure 3 This is a flowchart illustrating the anti-spy camera process of the present invention, showing the complete process from starting the anti-spy camera mode → CSI scanning of the environmental channel → establishing a legitimate device baseline → abnormal signal detection → MAC / data stream sniffing → CSI estimation of the abnormal source area topology → directional illumination by the red LED array → photoelectric sensor capturing backlight → confirmation of the spy camera.

[0055] Figure 4 The hierarchical wake-up state machine diagram of this invention shows the complete state transition of the system from low power monitoring (1~10Hz) → macroscopic variance threshold judgment → full CSI acquisition (50~200Hz) → anomaly detection and event classification → scheduling engine wake-up auxiliary module → fusion judgment weighted voting → response execution, including two loop paths: returning to low power mode after confirmation and adaptively updating the baseline when the body is in motion.

[0057] Figure 5The present invention provides a weighted voting fusion logic diagram, illustrating the weighted voting mechanism of the fusion judgment layer: the CSI main module weight wCSI, together with the weights wi of each auxiliary module, the confirmation confidence level ci, and the confirmation result coefficient fi, jointly calculate the comprehensive confirmation score S = wCSI × CCSI + Σ(wi × ci × fi), and the complete logic of adaptively and dynamically adjusting the weights according to the L2 / L3 level perception center.

[0059] Figure 6 This is a hardware interrupt synchronization timing diagram of the present invention, which shows the microsecond-level synchronization timing relationship between the CSI acquisition frame header, the hardware interrupt pin trigger signal, and the auxiliary module wake-up status, including a hierarchical scheduling strategy of L1 level triggering every frame, L2 level triggering every 5 frames, and L3 level triggering only when the confidence level is low.

[0061] Figure 7 This is a flowchart of the CSI signal processing of the present invention, showing the complete signal processing chain from raw CSI data (56 OFDM subcarrier amplitude + phase matrix) → preprocessing (Hampel filtering for noise reduction) → feature extraction (amplitude macro variance, phase change slope, subcarrier correlation) → event classification (L1 / L2 / L3 levels) → capability level identification output → scheduling engine decision, as well as the power consumption optimization path of graded downsampling. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0064] The system architecture of this invention is as follows: Figure 1 As shown.

[0065] Module Function Description

[0066] Perception layer module

[0067] Main sensing module (sensing center):

[0068] • Radio Frequency Channel Sensing Central Module (Main Module): As the core sensing hub of the entire system, it includes, but is not limited to, channel state information acquisition units based on Wi-Fi CSI, millimeter-wave radar, UWB, or future communication protocols. It acquires channel state data via the radio frequency interface, including the amplitude and phase matrices of multiple OFDM subcarriers. The radio frequency channel sensing central module is responsible for primary anomaly detection and event classification, outputting the anomaly event type (micro-motion anomaly, radio frequency anomaly, stationary foreign object), and simultaneously outputting capability level identifiers (L1 / L2 / L3), driving the environment adaptive scheduling engine. It supports hierarchical sampling: 5Hz macroscopic variance monitoring in low-power mode, and 100Hz full-dimensional acquisition in full-scale mode. The event classification criteria are as follows: Micro-motion anomaly – if the macroscopic variance σ²_CSI of the CSI amplitude exceeds the downsampling wake-up threshold and is accompanied by periodic micro-Doppler modulation, it is determined to be a micro-motion anomaly; Radio frequency anomaly – if cross-layer radio frequency fingerprint sniffing detects an unknown MAC address or abnormal data stream (throughput, protocol characteristics, or transmission timing deviating from the environmental baseline), it is determined to be a radio frequency anomaly; Stationary foreign object – if the CSI static multipath baseline shows a continuous shift (change in spatial scattering characteristics) but without micro-motion modulation components, it is determined to be a stationary foreign object. The above criteria can be used in combination.

[0069] Assisted sensing module group (existing traditional technologies, scheduled by the CSI main module, classified according to application level):

[0070] Consumer-facing (current development focus):

[0071] • Body pose sensing module (auxiliary module): Used to output the motion state vector of the device body, including but not limited to accelerometers, gyroscopes, magnetometers, or combinations thereof (currently, a six-axis inertial measurement unit (IMU) is preferred, but the specific model is not limited). In existing conventional technologies, when the radio frequency channel sensing center outputs a micro-motion anomaly, the scheduling engine wakes up the body pose sensing module to perform cross-verification, eliminating false alarms caused by drastic changes in the CSI matrix due to the device's own shaking.

[0072] • Active optical retroreflective module (auxiliary module): Existing conventional technology, including a red LED array and photoelectric sensor. When the CSI main module outputs a "RF anomaly" or "stationary foreign object" event, the scheduling engine wakes up the optical module, which is guided by the CSI to directionally illuminate the suspicious area, and confirms the hidden camera by capturing the retroreflection of the lens coating.

[0073] • Infrared thermal imaging module (auxiliary module): Utilizes existing conventional technology (specific model not limited) to acquire images of the environmental infrared thermal radiation distribution. When the CSI main module outputs a "micro-motion anomaly" or "stationary foreign object" event, the scheduling engine wakes up the infrared thermal imaging module to distinguish human bodies from interference sources such as pets and electrical appliances based on heat source characteristics, thus assisting in the detection of stationary intrusions.

[0074] • Vibration Detection Module (Auxiliary Module): Utilizes existing conventional technology (specific model not limited) to detect vibration signals on the equipment mounting surface. When the CSI main module outputs a "micro-motion anomaly" event, the scheduling engine activates the vibration detection module to help distinguish between door / window / wall vibrations and airborne CSI fluctuations, improving intrusion location accuracy.

[0075] • Millimeter-wave radar module (auxiliary module): Based on existing conventional technology (HLK-LD2410, etc., specific models are not limited), it operates in the 24GHz band and detects micro-motions and breathing signals. When the CSI main module outputs a "stationary foreign object" event, the scheduling engine wakes up the millimeter-wave radar module to confirm the presence of a stationary human body through breathing detection, thus compensating for the CSI's deficiency in detecting stationary targets.

[0076] • Environmental parameter sensor (auxiliary module): Existing traditional technology collects environmental parameters such as temperature and humidity to assist in interference identification and dynamic threshold adjustment.

[0077] B2B Business Tiers (Recently Expanded):

[0078] • LiDAR Module (Auxiliary Module): Utilizes existing traditional technology to acquire 3D point cloud data of the environment through laser scanning. When the CSI main module outputs a "stationary foreign object" event, the scheduling engine wakes up the LiDAR module to construct a 3D topological change comparison of the environment and accurately detect newly added objects (such as suspicious packages or illegal devices).

[0079] • Acoustic Detection Module (Auxiliary Module): Utilizes existing traditional technologies to collect environmental acoustic signature features. When the CSI main module outputs a "micro-motion anomaly" event, the scheduling engine activates the acoustic detection module. Through acoustic signature recognition, it distinguishes specific event types such as glass breakage and footsteps, improving event classification accuracy.

[0080] • Door / Window Magnetic Detection Module (Auxiliary Module): Utilizes existing traditional technology to detect the opening and closing status of doors and windows. When the CSI main module outputs a "micro-motion anomaly" event, the scheduling engine wakes up the door / window magnetic module, correlates the door / window status with CSI fluctuations, and confirms the intrusion path.

[0081] Industrial grade (long-term expansion):

[0082] • Industrial Gas Sensing Module (Auxiliary Module): Utilizes existing traditional technology to detect the concentration of harmful gases in the environment. When the CSI main module outputs a "stationary foreign object" event, the scheduling engine wakes up the gas sensing module to detect the presence of hazardous / contraband materials through gas characteristics. Suitable for warehouse security scenarios.

[0083] • Fiber Optic Vibration Sensing Module (Auxiliary Module): This module utilizes existing traditional technology, employing the principle of fiber optic interferometry to detect vibrations over long distances. When the CSI main module outputs a "micro-motion anomaly" event, the scheduling engine activates the fiber optic vibration module to locate perimeter intrusions, making it suitable for perimeter security in industrial parks / warehouses.

[0084] • RFID Asset Tag Module (Auxiliary Module): Utilizes existing traditional technology to read asset tag status via RFID. When the CSI main module outputs a "stationary foreign object" event, the scheduling engine wakes up the RFID module, correlates asset displacement with CSI anomalies, and confirms asset theft or abnormal movement.

[0085] Central processing layer module

[0086] • Data preprocessing module: Performs bandpass filtering, PCA dimensionality reduction, and wavelet transform on the data from the CSI main module to extract human micro-motion feature vectors; performs integral calculations on the data from the IMU auxiliary module to extract the body pose change.

[0087] • Environmental Adaptive Scheduling Engine (Core Innovation): This is the core innovation of this invention. Based on the abnormal event types output by the CSI main module, corresponding auxiliary sensing modules are organically combined for cross-validation. Scheduling logic: Micro-motion anomaly → wakes up the IMU for body pose compensation verification, wakes up the vibration detection module for auxiliary positioning, wakes up the infrared thermal imaging module to distinguish heat sources; Radio frequency anomaly / stationary foreign object → wakes up the optical module for retroreflection verification, wakes up the millimeter-wave radar module to detect breathing. This avoids all auxiliary modules running resident, achieving on-demand scheduling.

[0088] Scheduling engine decision algorithm:

[0089] The scheduling engine employs a priority-based weighted voting decision-making mechanism, with the specific algorithm as follows:

[0090] (1) Abnormal event classification: The CSI main module outputs the abnormal event type E∈{micro-motion abnormality, radio frequency abnormality, stationary foreign object}, and outputs the abnormal confidence level CCSI (0~1).

[0091] (2) Module wake-up scheduling: Based on the abnormal event type E, query the scheduling mapping table to determine the set of auxiliary sensing modules that need to be woken up, M={m1, m2, ..., mk}. When multiple modules need to be woken up concurrently, they are sorted according to the preset priority: IMU (highest, lowest power consumption) > vibration detection > infrared thermal imaging > millimeter wave radar > optical retroreflection (lowest, highest power consumption).

[0092] (3) Validation results of auxiliary modules: Each auxiliary module mi outputs the validation result ri∈{validation anomaly, exclusion anomaly, uncertainty} and the validation confidence ci (0~1).

[0093] (4) Fusion determination decision: A weighted voting mechanism is adopted to calculate a comprehensive confirmation score:

[0094] S = wCSI × CCSI + Σ(wi × ci × fi)

[0095] Where wCSI is the weight of the CSI main module, with a theoretical value range of 0.3 to 0.8 (expected set value, not experimentally calibrated) and a default value of 0.4, wi is the weight of the auxiliary module i (equally divided by default as 0.6 / k), and fi is the confirmation result coefficient (confimed anomaly = +1, anomaly excluded = -1, uncertain = 0). When S ≥ threshold T (default 0.6), it is determined as a real anomaly and a response is triggered; when S ≤ -T, it is determined as a false positive, and the CSI baseline is updated; when |S| < T, low-power monitoring is maintained, and re-determination in the next cycle is awaited.

[0096] (5) Adaptive weight update: The system dynamically adjusts the weight wi of each module according to the historical determination accuracy, and adaptively increases the weight of the module with high accuracy to achieve long-term optimization.

[0097] Sensing capability level detection and dynamic adaptation mechanism (core innovation, covering future sensing capability evolution):

[0098] The environment adaptive scheduling engine is further configured with a sensing capability level detection and dynamic adaptation mechanism. The system is pre-configured with multi-level sensing capability interface definitions, including:

[0099] - Level L1 (motion sensing): Only output the spatial moving / static state;

[0100] - Level L2 (morphology sensing): On the basis of L1, output the target contour or category label;

[0101] - Level L3 (semantic sensing): On the basis of L2, output behavior semantics or posture skeleton.

[0102] It should be noted that the aforementioned L1 / L2 / L3 hierarchical system (mobility-aware L1 level / morphological-aware L2 level / semantic-aware L3 level) is a custom capability hierarchy defined in this application and is not attributed to any prior art literature. This invention defines a three-level capability identification interface (L1 / L2 / L3) to drive the adaptive strategy switching of the scheduling engine. The identification criteria for each level are as follows: L1 level determines whether the target is moving by using the macroscopic variance of CSI amplitude and micro-Doppler features; L2 level, based on L1, further outputs the target contour / category label through subcarrier correlation, multipath pattern matching, or a lightweight model; L3 level, based on L2, combines a temporal behavior model to output behavioral semantics or attitude skeleton information. The above hierarchy is implemented through the capability level identification field output by the RF channel sensing central module, and the scheduling engine selects a high-auxiliary-dependency strategy or a low-auxiliary-dependency strategy accordingly.

[0103] Scheduling logic: During system initialization or runtime, the scheduling engine automatically reads the capability level identifier output by the radio frequency channel sensing hub module.

[0104] - When identified as L1, the scheduling engine adopts a high-assistance-dependency strategy, frequently waking up the ontology pose perception module and optical module for cross-validation;

[0105] - When identified as L2 or L3, the scheduling engine automatically switches to a low auxiliary dependency strategy, reducing the frequency of auxiliary module wake-up and only calling auxiliary modules for redundancy verification in high-confidence requirement scenarios.

[0106] Initial weight preset based on perception capability level: When the system recognizes that the access is a high-precision perception center of level L2 / L3, it automatically increases the initial weight coefficient wCSI of the center in the fusion formula (for example, wCSI=0.6 for L2 and wCSI=0.8 for L3) and decreases the weight coefficient wi of the auxiliary module.

[0107] Hardware synchronous triggering and characteristics of infringement evidence collection:

[0108] The scheduling engine controls the wake-up of the auxiliary sensing module via a hardware interrupt pin or a synchronization signal with a specific timing. Specifically, when it is determined that the auxiliary module needs to be woken up, the scheduling engine outputs a hardware trigger signal with a specific pulse width or frequency characteristic. This signal can be generated by the GPIO interrupt mechanism of the main control chip (such as ESP32-S3) and kept in microsecond-level synchronization with the frame header of the RF channel sensing data acquisition via DMA (Direct Memory Access) or timer interrupts, so as to facilitate subsequent infringement evidence collection by monitoring hardware pin levels or power consumption waveforms. The above-mentioned microsecond-level synchronization is a theoretical design parameter, and the specific accuracy depends on the clock configuration of the selected main control chip.

[0109] • CSI-IMU Fusion Judgment Unit: When the CSI main module outputs a micro-motion anomaly, it compares the data with the IMU auxiliary module: if the IMU shows that the device itself is moving, it is determined to be environmental noise and the CSI baseline is updated adaptively; if the IMU is stationary, it is determined to be external intrusion. This achieves adaptive filtering of CSI environmental noise.

[0110] • CSI-Optical Confirmation Unit: When the CSI main module outputs radio frequency anomalies or stationary foreign objects, it locks the suspicious area by sniffing abnormal MAC / data streams or multipath anomalies through cross-layer radio frequency fingerprints, guides the active red LED array to illuminate in a specific direction, and uses photoelectric sensors to capture the reverse reflection of the lens coating to confirm the hidden camera.

[0111] • CSI-Infrared Thermal Imaging Confirmation Unit: When the CSI main module outputs micro-motion anomalies or stationary foreign objects, it wakes up the infrared thermal imaging module to distinguish human bodies from interference sources such as pets and electrical appliances through the distribution of heat source characteristics, thus assisting in the detection of stationary intrusions.

[0112] • CSI-Vibration Confirmation Unit: When the CSI main module outputs a micro-motion anomaly, it wakes up the vibration detection module to help distinguish between physical vibrations of doors / windows / walls and CSI fluctuations propagated through the air, thereby improving the accuracy of intrusion location.

[0113] • CSI-Millimeter Wave Confirmation Unit: When the CSI main module outputs a stationary foreign object, it wakes up the millimeter wave radar module to confirm the presence of a stationary human body through respiratory detection, thus compensating for the CSI's deficiency in detecting stationary targets.

[0114] • Trend Analysis Module: Calculates the slope of signal changes based on data from the CSI main module, identifies abnormal trends, and provides early warnings.

[0115] Response layer module

[0116] • Response Execution Module: Generates response commands based on the fusion judgment results and user interaction settings, including local sound and light alarms, cloud push notifications, etc.

[0117] • Local storage module: Stores detection data and alarm records, supporting offline operation.

[0118] • Cloud server: Sends dynamic threshold parameters through a TLS encrypted channel, receives alarm information, and performs model training and policy optimization.

[0119] User interaction layer module

[0120] • User interaction settings module: Provides users with configurable security mode settings (such as home mode and away mode), which is a regular interaction function and not a core innovation.

[0121] • Mobile App: Receives alert push notifications, displays confirmation results, and supports remote settings.

[0122] System Flow Description

[0123] Intrusion prevention scenario process as follows Figure 2 As shown.

[0124] Anti-spy camera scenario process as follows Figure 3 As shown.

[0125] Symbols and Formulas

[0126] Symbol definition

[0127] Core Formula

[0128] CSI signal acquisition formula:

[0129] S_i,j(t) = A_i,j(t) · e^jφ_i,j(t)

[0130] Where i is the subcarrier index, j is the antenna index, and t is the time.

[0131] IMU body pose change (dimensionless fusion determination):

[0132] Δ p(t) = α · ‖ ∫_t-T_w^ta(τ) dτ ‖ + β · ‖ ∫_t-T_w^t ω(τ) dτ ‖

[0133] Where α is the acceleration integral weighting coefficient, β is the angular velocity integral weighting coefficient, α + β = 1, preferably α = 0.6, β = 0.4. The first term is the displacement component norm (dimension m), and the second term is the angular component norm (dimension rad), which are normalized to the dimensionless decision quantity Δp through the weighting coefficients.

[0134] CSI-IMU fusion decision logic:

[0135] Judgment conditions:

[0136] • If ||Δp|| ≥ δ_p, it is determined to be environmental noise, triggering an adaptive update of the CSI baseline;

[0137] • If ||Δp|| < δ_p and c ≥ θ, then it is determined to be an external intrusion;

[0138] • If ||Δp|| < δ_p and c < θ, then it is determined to be an interference source.

[0139] Where Δp represents the change in body pose obtained by IMU integration (vector norm, dimensionless, 0≤Δp≤1); δ_p represents the preset threshold for body pose change (e.g., 0.05 rad / m); θ represents the fusion confidence threshold, with a theoretical range of 0.7~0.95 (e.g., 0.85); the above δ_p and θ are theoretical settings, not experimental calibrations.

[0140] Downsampling graded wake-up conditions:

[0141] σ^2_CSI ≥ σ^2_th

[0142] Trend warning conditions:

[0143] |k| ≥ k_th

[0144] Key technical parameters

[0145] (Technical route)

[0146] Phased description

[0147] Phase 1: Data Sensing and Collection

[0148] • Key technologies: ESP32-S3 acquires CSI (supports 1~10Hz downsampling and 50~200Hz full switching), IMU six-axis inertial measurement unit acquires body pose (power supply and data synchronization, specific model not limited).

[0149] • Expected verification items: CSI signal stability, consistency between IMU data and device body shaking.

[0150] Phase Two: Multimodal Fusion and Intelligent Analysis

[0151] • Objective: To fuse CSI and IMU / optical data to identify valid events and filter out interference.

[0152] • Key technologies: CSI-IMU fusion determination (body pose compensation), CSI-optical verification (anti-spy camera), environment adaptive scheduling engine, machine learning interference recognition.

[0153] • Expected verification item: The false alarm rate after IMU compensation is lower than the preset threshold.

[0154] Phase Three: Secure Transmission and Compliance Response

[0155] • Objective: Secure data transmission and trigger compliance responses.

[0156] • Key technologies: TLS encrypted transmission, two-way authentication, and user interaction setting trigger logic.

[0157] • Expected verification item: System's ability to resist unauthorized access.

[0158] Phase Four: System Deployment and Maintenance

[0159] • Objective: To complete deployment, debugging, and long-term maintenance.

[0160] • Key technologies: OTA remote upgrade, low-power tiered wake-up, cloud platform monitoring.

[0161] • Expected verification items: Long-term operational stability. The MTBF index depends on the specific device and deployment environment and is not limited.

[0162] Key logical relationships

[0163] Data flow:

[0164] • CSI acquisition → IMU-assisted compensation → Fusion analysis → Secure transmission → Response execution.

[0165] • CSI acquisition → RF fingerprint sniffing → Optical verification → Response execution.

[0166] Technological dependence:

[0167] • Phase 2 depends on the data output of Phase 1, and Phase 3 depends on the identification results of Phase 2.

[0168] • The CSI-IMU fusion decision module relies on the CSI and IMU data synchronization mechanism of Phase 1.

[0169] Iterative optimization:

[0170] • Optimize model parameters and strategies dynamically by leveraging data feedback from the cloud platform.

[0171] • The model is updated every 7 days, and the weights of the fusion algorithm are adjusted based on historical false alarm data.

[0172] (Key technical points)

[0173] Key technical points

[0174] (1) Hierarchical architecture of CSI main module + auxiliary module group: CSI, as the main sensing module (sensing center), is responsible for primary anomaly detection and event classification; the auxiliary sensing module group is divided into C-end civilian (IMU, optical, infrared thermal imaging, vibration detection, millimeter-wave radar), B-end commercial (LiDAR, acoustic detection, door and window magnetic sensors), and industrial (gas sensing, fiber optic vibration, RFID asset tags) according to the application level. All of them are existing traditional technologies, which are awakened by the scheduling engine on demand for cross-validation. The core innovation is not in a single sensor, but in the organic combination of collaborative architecture.

[0175] (2) Environmental Adaptive Scheduling Engine (Core Innovation): Based on the abnormal event types (micro-motion anomaly, radio frequency anomaly, stationary foreign object) output by the CSI main module, the corresponding auxiliary sensing modules are organically combined for cross-validation. The scheduling logic covers all scenarios of C-end / B-end / industrial level, avoiding the need for all auxiliary modules to run continuously.

[0176] (3) CSI and IMU body pose compensation mechanism: When the CSI main module outputs micro-motion anomaly, it compares the data with the IMU auxiliary module. If the body is moving, it is determined to be environmental noise and the CSI baseline is updated adaptively. If the body is stationary, it is determined to be external intrusion. This is the core distinguishing technical feature for solving common CSI security problems.

[0177] (4) CSI-guided optical retroreflection verification mechanism: When the CSI main module outputs radio frequency abnormality or stationary foreign object, it locks the suspicious area by sniffing abnormal MAC / data stream or multipath abnormality through cross-layer radio frequency fingerprint, guides the active red light LED array to directional irradiation, and captures the lens coating retroreflection through photoelectric sensor to verify the hidden camera.

[0178] (5) CSI-infrared thermal imaging collaborative verification mechanism: When the CSI main module outputs micro-motion abnormality or stationary foreign object event, the infrared thermal imaging module is awakened. The human body is distinguished from interference sources such as pets and electrical appliances by heat source characteristics, which assists in the detection of stationary intrusion.

[0179] (6) CSI-Vibration Co-verification Mechanism: When the CSI master module outputs a micro-motion abnormal event, the vibration detection module is activated to help distinguish between door / window / wall vibration and CSI fluctuations propagated by air, thereby improving the accuracy of intrusion location.

[0180] (7) CSI-millimeter wave collaborative verification mechanism: When the CSI main module outputs a stationary foreign object event, the millimeter wave radar module is activated to verify the existence of a stationary human body through breathing detection, thus making up for the shortcomings of CSI in detecting stationary targets.

[0181] (8) CSI-LiDAR Collaborative Authentication Mechanism (B-end Extension): When the CSI main module outputs a stationary foreign object event, the LiDAR module is activated to construct a three-dimensional topological change comparison of the environment and accurately detect newly added objects.

[0182] (9) CSI-Acoustic Collaborative Confirmation Mechanism (B-end Extension): When the CSI main module outputs a micro-motion abnormal event, it wakes up the acoustic detection module and distinguishes specific event types such as glass breakage and footsteps through voiceprint recognition.

[0183] (10) CSI-fiber vibration collaborative verification mechanism (industrial-grade extension): When the CSI master module outputs a micro-motion abnormal event, it wakes up the fiber vibration sensing module to realize long-distance perimeter intrusion location.

[0184] (11) Hierarchical wake-up mechanism based on CSI macroscopic feature downsampling: Normally, the macroscopic variance of CSI amplitude is monitored at an extremely low sampling rate. When the threshold is exceeded, the full data acquisition is woken up instantly, which greatly reduces power consumption.

[0185] (12) Non-visual privacy protection: No optical images are collected throughout the process; only CSI physical layer subcarrier features are extracted and fused with IMU pose scalar data.

[0186] (13) Trend analysis and early warning algorithm: By monitoring the slope of the signal amplitude change of the CSI main module, abnormal trends are identified and early warning is achieved.

[0187] (Example)

[0188] The specific embodiments of the present invention are described below with reference to the accompanying drawings and theoretical analysis. It should be noted that this application is a theoretical research result, and the various technical parameters and effects recorded in the specification (such as sampling rate, main module weight, false alarm rate, power consumption, detection latency, detection distance, battery life, etc.) are theoretical predictions and expected effects, not experimental test data; the capability levels L1 / L2 / L3 are custom classifications defined in this application, used to drive the switching of scheduling engine strategies, and are not attributed to any prior art literature.

[0189] Theoretical Analysis and Explanation

[0190] The performance parameters (including false alarm rate, power consumption, positioning accuracy, and battery life) of each embodiment of this invention are all theoretically calculated based on publicly available device parameters, signal processing theoretical formulas, and system architecture, and are not actual experimental test data. The theoretical calculations are based on the following:

[0191] • CSI acquisition parameters are based on the ESP32-S3 technical reference manual (56 CSI subcarriers, the specific main frequency and wireless standard are not limited to specific models).

[0192] • IMU parameters are based on publicly available datasheets for common six-axis IMU modules (specific models are not limited);

[0193] • Power consumption parameters are calculated based on the typical operating current of each module;

[0194] • The false alarm rate is estimated based on signal detection theory (Bayesian decision theory) and the CSI signal model.

[0195] Example 1: Home Intrusion Prevention Scenario

[0196] Deployment environment: 80-square-meter two-bedroom apartment, the device is placed on the TV cabinet in the living room, and the ESP32-S3 is connected to the home Wi-Fi router.

[0197] Hardware configuration: ESP32-S3 main controller + MPU-6050 six-axis IMU + infrared thermal imaging module (MLX90640) + vibration detection module (SW-420).

[0198] Workflow:

[0199] (1) Normally, the macroscopic variance of CSI amplitude is monitored at a low sampling rate of 5Hz, with a power consumption of about 20-30mA (including Wi-Fi radio frequency reception).

[0200] (2) When the macro variance exceeds the 2dB² threshold, the 100Hz full CSI acquisition is instantly activated.

[0201] (3) Preprocess the CSI data (bandpass filtering 5-15Hz, PCA dimensionality reduction to retain 3 principal components) and extract micro-motion features.

[0202] (4) Synchronously wake up the IMU via the hardware interrupt pin to perform CSI-IMU fusion determination:

[0203] - Step 1: Time synchronization alignment (microsecond precision).

[0204] - Step 2: Calculate the IMU body pose change Δp (integrate the three-axis acceleration to obtain the displacement component, integrate the three-axis angular velocity to obtain the rotation component, and synthesize the body pose change).

[0205] - Step 3: Judgment Logic - If ||Δp|| ≥ δ_p (0.05 rad / m), it is determined to be environmental noise, and the adaptive update of the CSI baseline is triggered (using the EWMA formula S_t = α·S_{t-1} + (1-α)·X_t, where α is 0.95); if ||Δp|| < δ_p and the fusion confidence c ≥ θ, it is determined to be an external intrusion; if ||Δp|| < δ_p and c < θ, it is determined to be an interference source, and monitoring continues.

[0206] (5) If it is determined to be an external intrusion, the infrared thermal imaging module is activated to distinguish between human body and pet / electrical interference source through the heat source feature distribution.

[0207] (6) Trigger response: Local sound and light alarm + cloud push notification to mobile app.

[0208] Theoretical analysis and expected results:

[0209] • False alarm rate: Based on Bayesian decision theory, the false alarm rate is expected to decrease from approximately 15.3% to approximately 2.0% after IMU compensation.

[0210] • Power consumption: Approximately 20-30mA in low power mode, combined with intermittent sleep strategy (deep sleep <10μA), theoretically estimated battery life is approximately 40 days.

[0211] • Response delay: The time from the occurrence of an anomaly to the triggering of an alarm is expected to be less than 3 seconds.

[0212] Example 2: Hotel Anti-Spycam Scenario

[0213] Deployment environment: Standard hotel room (approximately 30 square meters), with the equipment placed on the bedside table and deployed before check-in.

[0214] Hardware configuration: ESP32-S3 main controller + active red LED array (5W, 850nm wavelength) + photoelectric sensor (TSL2591) + Bluetooth sniffing module.

[0215] Workflow:

[0216] (1) After the anti-spy camera mode is activated, CSI scans the Wi-Fi / Bluetooth channels in the environment and establishes a temporary memory whitelist of legitimate devices (MAC addresses of known devices such as routers, TVs, and air conditioners, which are only used for real-time comparison, not for persistent storage or uploading).

[0217] (2) Continuously monitor abnormal signals: When an unknown MAC address or abnormal data stream is detected, proceed to the verification process.

[0218] (3) Sniff out abnormal MAC / data streams and estimate the topology of the abnormal source region by CSI multipath fingerprint mesh matching:

[0219] - Divide the room into a 0.5m × 0.5m grid, with each grid pre-stored with a CSI multipath fingerprint feature vector.

[0220] - Real-time CSI observation vectors are matched with the grid fingerprint database, and the grid with the smallest Euclidean distance is taken as the estimated location of the anomaly source.

[0221] - The theoretical estimated positioning accuracy is 0.5-5m (depending on the multipath complexity of the room).

[0222] (4) Guide the red LED array to directionally illuminate the suspicious area (the illumination angle is adjusted by the position driven servo motor estimated by CSI).

[0223] (5) The photoelectric sensor captures the backlight signal and distinguishes between lens coating reflection (strong directional reflection) and ordinary object scattering (weak diffuse reflection) by the intensity and angle characteristics of the reflection.

[0224] (6) After confirming the hidden camera, trigger the following response: local alarm + cloud notification + record the MAC address and location of the abnormal device.

[0225] Theoretical analysis and expected results:

[0226] • Anti-spy camera detection rate: Theoretically estimated to be over 90% (for spy cameras with a lens diameter > 2mm).

[0227] • False alarm rate: Theoretically estimated to be less than 5% (confirmed by dual differentiation of CSI-guided directional illumination and backlight characteristics).

[0228] • Positioning accuracy: Theoretically estimated at 0.5-5m.

[0229] Example 3: Warehouse Security Scenario (B2B Extension)

[0230] Deployment environment: 500 square meters of warehouse space, with equipment deployed at key nodes of the cargo passage.

[0231] Hardware configuration: ESP32-S3 main controller + IMU + millimeter-wave radar (HLK-LD2410) + lidar (RPLIDAR A1) + acoustic detection module.

[0232] Workflow:

[0233] (1) The CSI main module monitors low power consumption and wakes up the millimeter-wave radar when a stationary foreign object event is detected.

[0234] (2) Millimeter-wave radar confirms the presence of a stationary human body by detecting breathing (detecting micro-movements of human breathing in the 24GHz band, with a frequency of 0.1-0.5Hz).

[0235] (3) If further precise positioning is required, the lidar is activated to construct a three-dimensional topology of the environment and the newly added objects are detected by comparing it with the baseline topology.

[0236] (4) The acoustic detection module helps to distinguish specific event types such as glass breakage and footsteps.

[0237] (5) The fusion determination adopts a weighted voting mechanism: S = wCSI × CCSI + Σ(wi × ci × fi), where wCSI=0.4, millimeter wave weight w1=0.3, lidar weight w2=0.2, and acoustic weight w3=0.1.

[0238] Theoretical analysis and expected results:

[0239] • Detection rate of stationary human body: theoretically estimated to be higher than 85% (millimeter wave respiration detection and CSI fusion).

[0240] • Improved object detection accuracy: theoretically estimated to be better than 5cm (3D topology comparison with LiDAR).

[0241] • System power consumption: Average power consumption is less than 50mA (intermittent wake-up strategy).

[0242] Parameter examples

[0243] The key parameters involved in this invention and their value ranges are as follows (all are theoretical analysis and design parameters):

[0244] Note: The above parameters are theoretical analysis and design parameters. The actual values ​​need to be adjusted according to the specific deployment environment and hardware model.

[0245] Theoretical boundary calculation

[0246] To illustrate the reasonable expectation of the technical effects of this invention, the following theoretical and semi-quantitative calculations of the key performance indicators of this invention are based on verified published literature, competitor patents, and publicly available module datasheets. All values ​​are theoretical predictions and expected effects, not experimental test data; no specific module selection is limited.

[0247] (1) Detection distance through walls: Based on the general theory of non-line-of-sight penetration propagation of radio frequency signals, it is theoretically predicted that the radio frequency channel sensing center of this invention can realize the penetration detection of human body behind non-load-bearing walls (brick walls / plasterboard), with a detection distance of several meters.

[0248] (2) Sampling rate and graded wake-up: According to the main control module technical reference manual, CSI acquisition can be configured by firmware as a graded strategy of macro monitoring at 1~10Hz and full acquisition at 50~200Hz, which is consistent with claim 6 / 16.

[0249] (3) Positioning / resolution: Based on the general theory of radio frequency signal processing and multipath positioning, it is theoretically predicted that the human body positioning resolution of the CSI main module of this invention can reach the decimeter level.

[0250] (4) Velocity estimation: Based on the general model of velocity estimation using the Doppler effect, it is theoretically predicted that this invention can estimate human velocity.

[0251] (5) Privacy compliance: In accordance with GB / T 35273-2020, the non-visual privacy protection design of the present invention, which "does not collect optical images throughout the process", complies with the requirements of personal information security specifications (compliance support).

[0252] (6) Public literature related to IMU ontological pose compensation: GUO et al. and IMFi confirmed that CSI+IMU multimodal fusion can improve perception robustness; AR-Alarm confirmed the feasibility of CSI adaptive intrusion detection. However, none of the above literature disclosed the complete combination of "IMU ontological pose compensation to eliminate false alarms of CSI ontological sway + environment adaptive scheduling engine + non-visual privacy protection".

[0253] (7) Baseline comparison of existing technologies: Hikvision CN115527328B [1] (PIR / optical route) highlights the non-visual CSI advantage of this invention; Nankai University CN116168499B [2] (CSI amplitude stability analysis, no body pose compensation) is the closest to the existing technology, and this invention superimposes IMU compensation and multimodal scheduling on it; Zhejiang University of Technology CN121410835A [3] (under review, CSI+visual fusion) is different from the "non-visual privacy protection" of this invention.

[0254] Upon verification, none of the aforementioned 7 documents disclosed the same combination scheme as this invention: "RF channel sensing hub + environment adaptive scheduling engine + IMU body pose compensation + non-visual privacy protection".

[0255] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention; any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0256] References

[0257] [1] Hangzhou Hikvision Digital Technology Co., Ltd. Infrared Intrusion Detector: CN115527328B[P]. 2026-07-10.

[0258] [2] Nankai University. Indoor Passive Perception and Intrusion Detection Method Based on CSI Amplitude Stationarity Analysis: CN116168499B[P]. 2026-02-03.

[0259] [3] Zhejiang University of Technology Artificial Intelligence Innovation Research Institute, Binjiang District, Hangzhou. A UAV detection system based on WIFI-CSI and YOLO fusion: CN121410835A[P]. 2026-01-27.

[0260] [4] LI S, LI 10.1007 / 978-3-319-66188-9_18.

[0261] [5] GUO W, YAMAGISHI S, JING L. Human activity recognition via Wi-Fiand inertial sensors with machine learning[J]. IEEE Access, 2024, 12: 18821-18836. DOI: 10.1109 / ACCESS.2024.3360490.

[0262] [6] SONG Z, ZHOU H, WANG S, FAN J, GUO K, ZHOU W, WANG X, LI 279-286. DOI: 10.1109 / MSN53354.2021.00052.

[0263] [7] State Administration for Market Regulation of the People's Republic of China, Standardization Administration of China. GB / T35273-2020 Information Security Technology - Personal Information Security Specification [S]. Beijing: China Standards Press, 2020.

[0264] Ethical Compliance Statement

[0265] The Channel State Information (CSI) collection and MAC address sniffing involved in this invention are both based on legally authorized security scenarios (such as user-owned residences or hotel rooms deployed under authorization). The system only extracts the amplitude and phase features of WiFi physical layer subcarriers and does not collect, store, or transmit any personal privacy information (including facial, voiceprint, or biometric information). MAC address sniffing is only used to detect the presence of abnormal devices. A temporary memory whitelist is established for legitimate device MAC addresses for real-time comparison; legitimate device MAC addresses are not persistently stored or uploaded. The system does not collect optical images throughout the entire process, complying with the relevant provisions of the "Personal Information Protection Law of the People's Republic of China" and the "Data Security Law of the People's Republic of China."

Claims

1. An adaptive multimodal cooperative scheduling method based on a radio frequency channel sensing center, characterized in that, Includes the following steps: S1: The radio frequency channel sensing central module collects environmental radio frequency channel status data, performs primary anomaly detection, and outputs the anomaly event type, which includes micro-motion anomaly, radio frequency anomaly, and stationary foreign object; the radio frequency channel sensing central module also outputs capability level identifiers, which include L1 level motion sensing, L2 level morphological sensing, and L3 level semantic sensing. S2: The environment adaptive scheduling engine dynamically wakes up the corresponding auxiliary perception module for cross-validation based on the abnormal event type and capability level identifier; wherein, the scheduling engine dynamically adjusts the scheduling strategy of the auxiliary perception module according to the capability level identifier: when at L1 level, the auxiliary perception module participates in fusion verification at high frequency; when at L2 or L3 level, the auxiliary perception module only participates in redundant verification when the confidence level is lower than the threshold. S3: The fusion judgment unit performs fusion analysis on the data from the radio frequency channel sensing center module and the auxiliary sensing module to confirm abnormal events. The fusion judgment adopts a weighted voting mechanism, and the comprehensive confirmation score S = wCSI × CCSI +Σ(wi × ci × fi), where wCSI is the weight of the main module, wi is the weight of the auxiliary module, ci is the confirmation confidence level, and fi is the confirmation result coefficient. When the system identifies a high-precision sensing center at level L2 or L3, it automatically increases wCSI and decreases wi. S4: Generate and execute response instructions based on the fusion determination results and user interaction settings.

2. The method according to claim 1, characterized in that, The radio frequency channel sensing central module is a data acquisition unit based on Wi-Fi channel state information (CSI), which acquires the amplitude and phase matrix of multiple OFDM subcarriers.

3. The method according to claim 1, characterized in that, In step S2, corresponding auxiliary sensing module groups are scheduled for verification based on different abnormal event types: When a micro-motion abnormality event is output, one or more of the following modules can be selected for wake-up: body pose sensing, vibration detection, acoustic detection, door / window magnetic detection, and fiber optic vibration sensing. When a stationary foreign object event is output, one or more of the following modules can be selected for wake-up: infrared thermal imaging, millimeter-wave radar, lidar, industrial gas sensing, and RFID asset tag. When an abnormal radio frequency event is output, the presence of the hidden camera is confirmed by sniffing the abnormal MAC address or data stream through cross-layer radio frequency fingerprinting and combining it with the active optical retroreflection module. Furthermore, in the weighted voting mechanism, the value range of the main module weight wCSI is 0.3 to 0.

8. When a high-precision sensing center of level L2 or level L3 is identified, wCSI is dynamically increased within this value range and the weights wi of each auxiliary module are reduced accordingly.

4. The method according to claim 3, characterized in that, When a micro-motion abnormality event is output, the body pose perception module is activated to perform body pose compensation: if the body pose data shows that the device body is moving, it is determined to be environmental noise and the baseline is updated adaptively; if the body pose data shows that the device body is stationary, it is determined to be external intrusion; the body pose perception module includes an accelerometer, a gyroscope, a magnetometer or a combination thereof.

5. The method according to claim 3, characterized in that, When an abnormal radio frequency event or a stationary foreign object event is output, the abnormal MAC address or data stream is sniffed through cross-layer radio frequency fingerprinting to estimate the topology of the abnormal source area, guide the active red light LED array to directionally illuminate the area, and use a photoelectric sensor to capture the retroreflection of the lens coating to confirm the hidden camera.

6. The method according to claim 1, characterized in that, In step S1, a graded wake-up mechanism based on downsampling of macroscopic features of radio frequency channel is adopted: normally, the macroscopic variance of amplitude is monitored at a low sampling rate of 1~10Hz, and when the macroscopic variance exceeds a preset threshold, the full acquisition of 50~200Hz is instantly activated.

7. The method according to claim 1, characterized in that, The method does not collect optical images throughout the process, but only extracts the physical layer features of the radio frequency channel and fuses them with the scalar data of the auxiliary sensing module to achieve non-visual privacy protection.

8. The method according to claim 1, characterized in that, It also includes a trend analysis and early warning step: by calculating the slope of the change in the amplitude of the radio frequency channel signal, an early warning is triggered when the slope exceeds a preset threshold.

9. The method according to claim 1, characterized in that, The scheduling engine controls the wake-up of the auxiliary sensing module by outputting a hardware trigger signal with specific pulse width or frequency characteristics through a hardware interrupt pin. The timing characteristics of this signal are synchronized with the frame header of the radio frequency channel sensing data acquisition at the microsecond level.

10. An adaptive multimodal cooperative scheduling system based on a radio frequency channel sensing center, characterized in that, include: The radio frequency channel sensing hub module, as the main sensing module, collects radio frequency channel status information and performs primary anomaly detection and event classification, while outputting capability level identifiers, which include L1 level mobile sensing, L2 level morphological sensing and L3 level semantic sensing. The auxiliary sensing module group includes one or more of the following: body pose sensing module, active optical retroreflection module, infrared thermal imaging, vibration detection, millimeter-wave radar, lidar, acoustic detection, door and window magnetic sensors, gas sensing, fiber optic vibration, and RFID asset tags, which can be tailored and combined according to the application level. The environment adaptive scheduling engine dynamically adjusts the scheduling strategy of the auxiliary perception module and wakes up the corresponding module for cross-validation based on the type of abnormal event and capability level identifier. When it is at L1 level, the auxiliary perception module participates in fusion verification at a high frequency. When it is at L2 or L3 level, the auxiliary perception module only participates in redundancy verification when the confidence level is lower than the threshold. The fusion judgment unit uses a weighted voting mechanism to perform fusion analysis and verification of the main module data and the auxiliary sensing module data. The comprehensive verification score S = wCSI × CCSI + Σ(wi × ci × fi), where wCSI is the main module weight, wi is the auxiliary module weight, ci is the verification confidence level, and fi is the verification result coefficient. When a high-precision sensing center of L2 or L3 level is identified, wCSI is automatically increased and wi is decreased. The response execution module executes the response based on the fusion judgment result and user interaction settings.

11. The system according to claim 10, characterized in that, The auxiliary sensing module group can be customized and combined according to the application level: C-end products can be equipped with body pose sensing, optical, infrared thermal imaging, vibration, and millimeter-wave radar modules; B-end products can be expanded with lidar, acoustic, and door magnetic modules; industrial-grade products can be expanded with gas sensing, fiber optic vibration, and RFID modules.

12. The system according to claim 10, characterized in that, The radio frequency channel sensing central module is a data acquisition unit based on Wi-Fi channel state information (CSI), which acquires the amplitude and phase matrix of multiple OFDM subcarriers.

13. The system according to claim 10, characterized in that, The environment adaptive scheduling engine schedules corresponding auxiliary perception module groups for verification based on different abnormal event types: When a micro-motion abnormality event is output, one or more of the following modules can be selected for wake-up: body pose sensing, vibration detection, acoustic detection, door / window magnetic detection, and fiber optic vibration sensing. When a stationary foreign object event is output, one or more of the following modules can be selected for wake-up: infrared thermal imaging, millimeter-wave radar, lidar, industrial gas sensing, and RFID asset tag. When an abnormal radio frequency event is output, the presence of the hidden camera is confirmed by sniffing the abnormal MAC address or data stream through cross-layer radio frequency fingerprinting and combining it with the active optical retroreflection module. Furthermore, in the weighted voting mechanism, the value range of the main module weight wCSI is 0.3 to 0.

8. When a high-precision sensing center of level L2 or level L3 is identified, wCSI is dynamically increased within this value range and the weights wi of each auxiliary module are reduced accordingly.

14. The system according to claim 13, characterized in that, When a micro-motion abnormality event is output, the system wakes up the body pose perception module to perform body pose compensation: if the body pose data shows that the device body is moving, it is determined to be environmental noise and the baseline is updated adaptively; if the body pose data shows that the device body is stationary, it is determined to be external intrusion; the body pose perception module includes an accelerometer, a gyroscope, a magnetometer or a combination thereof.

15. The system according to claim 13, characterized in that, When an abnormal radio frequency event or a stationary foreign object event is output, the abnormal MAC address or data stream is sniffed through cross-layer radio frequency fingerprinting to estimate the topology of the abnormal source area, guide the active red light LED array to directionally illuminate the area, and use a photoelectric sensor to capture the retroreflection of the lens coating to confirm the hidden camera.

16. The system according to claim 10, characterized in that, When collecting environmental radio frequency channel status data, the radio frequency channel sensing central module adopts a graded wake-up mechanism based on downsampling of macroscopic characteristics of radio frequency channels: normally, it monitors the macroscopic variance of amplitude at a low sampling rate of 1~10Hz, and when the macroscopic variance exceeds a preset threshold, it instantly wakes up to collect full data at 50~200Hz.

17. The system according to claim 10, characterized in that, The system does not collect optical images throughout the process; it only extracts the physical layer features of the radio frequency channel and fuses them with scalar data from the auxiliary sensing module to achieve non-visual privacy protection.

18. The system according to claim 10, characterized in that, The fusion determination unit also includes a trend analysis and early warning function: by calculating the slope of the change in the amplitude of the radio frequency channel signal, an early warning is triggered when the slope exceeds a preset threshold.

19. The system according to claim 10, characterized in that, The scheduling engine controls the wake-up of the auxiliary sensing module by outputting a hardware trigger signal with specific pulse width or frequency characteristics through a hardware interrupt pin. The timing characteristics of this signal are synchronized with the frame header of the radio frequency channel sensing data acquisition at the microsecond level.

Images