Door lock unlocking intention recognition method and system based on behavior analysis

Abstract

The invention relates to a door lock unlocking intention recognition method and system based on behavior analysis, and the method comprises the steps: obtaining user behavior information, hand action information and door lock handle pressure information collected by a door lock, and carrying out identification calculation to obtain a behavior intention level; according to the behavior intention level, operation intention analysis is conducted on the door lock handle pressure information and the hand action information, and abnormal behavior information is obtained; performing hardware state hierarchical switching on the door lock according to the abnormal behavior information to obtain a hierarchical control instruction; and performing alarm protocol matching based on the hierarchical control instruction and the user behavior information to obtain alarm execution information. According to the method, the limitation of single-dimension judgment can be broken through, the intention misjudgment rate is remarkably reduced, and the accuracy of unlocking intention recognition is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of door locks, and in particular to a door lock opening intention recognition method and system based on behavior analysis. BACKGROUND

[0002] With the diversification of security threats and the increasing demand for intelligence, how to accurately capture the user's opening intention and quickly respond to potential risks has become a key challenge to optimize the security performance of the door lock. Existing recognition methods usually focus on a single behavior dimension, such as relying only on independent analysis of hand movements or pressure signals, while ignoring the dynamic interaction between user behavior, hand movements and handle pressure and their synergistic effect in intention recognition. This simplified process can easily lead to one-sidedness in intention judgment, thereby reducing the accuracy of anomaly detection and affecting the response efficiency and overall security level of the door lock system. SUMMARY

[0003] The main purpose of the present application is to provide a door lock opening intention recognition method and system based on behavior analysis, which can break through the limitations of single-dimensional judgment, significantly reduce the intention misjudgment rate, and improve the accuracy of opening intention recognition.

[0004] To achieve the above purpose, the present application provides a door lock opening intention recognition method based on behavior analysis, comprising: acquiring user behavior information, hand movement information and door handle pressure information collected by the door lock, and performing identification calculation to obtain a behavior intention level; According to the behavior intention level, the door handle pressure information and the hand movement information are analyzed for operation intention, and abnormal behavior information is obtained; According to the abnormal behavior information, the hardware state of the door lock is switched by classification, and a classification control instruction is obtained; Based on the classification control instruction and the user behavior information, an alarm protocol matching is performed to obtain alarm execution information.

[0005] Further, the acquisition of user behavior information, hand movement information and door handle pressure information collected by the door lock, and the identification calculation to obtain the behavior intention level, comprises: extracting the behavior trajectory from the continuous image frame sequence collected by the horizontal view camera of the door lock to obtain the user behavior information; detecting the hand feature points from the continuous image frame sequence collected by the oblique downward view camera of the door lock to obtain the hand movement information; identifying the handle pressure from the original pressure data collected by the sensor of the door lock to obtain the door handle pressure information; Perform modal intention recognition on the user behavior information and the hand action information based on the door lock handle pressure information, to obtain the behavior intention level. Further, the modal intention recognition on the user behavior information and the hand action information based on the door lock handle pressure information, to obtain the behavior intention level, comprises: Perform direction trajectory segmentation on the user behavior information, to obtain approaching direction information; Perform gesture recognition on the hand action information, to output action feature information; Perform gradient analysis on the door lock handle pressure information, to obtain pressure change information; Perform operation intention calculation on the approaching direction information, the action feature information and the pressure change information based on preset intention recognition rules, to obtain the behavior intention level.

[0006] Further, the operation intention analysis on the door lock handle pressure information and the hand action information according to the behavior intention level, to obtain abnormal behavior information, comprises: Perform dynamic allocation on the behavior intention level, to obtain a pressure analysis coefficient and an action analysis coefficient; Perform pressure segmentation on the door lock handle pressure information according to the pressure analysis coefficient, to obtain segmented pressure information; Perform joint trajectory clustering on the hand action information according to the action analysis coefficient, to obtain action intention information; Perform multi-modal fusion on the segmented pressure information and the action intention information, to obtain behavior feature information; Perform abnormal intention recognition on the behavior feature information according to a preset abnormal behavior feature library, to obtain the abnormal behavior information.

[0007] Further, the pressure segmentation on the door lock handle pressure information according to the pressure analysis coefficient, to obtain segmented pressure information, comprises: Perform frequency component decomposition on the door lock handle pressure information, to obtain a high-frequency pressure component and a low-frequency pressure component; Perform mutation point detection on the high-frequency pressure component according to the pressure analysis coefficient, to obtain a candidate pressure point set; Perform trend turning point recognition on the low-frequency pressure component, to obtain a trend pressure point set; Perform conflict resolution on the candidate pressure point set and the trend pressure point set, to obtain an effective pressure point sequence; Perform segment division on the door lock handle pressure information according to the effective pressure point sequence, to obtain the segmented pressure information.

[0008] Further, the hardware state hierarchical switching of the door lock according to the abnormal behavior information obtains a hierarchical control instruction, comprising: The confidence degree of the abnormal behavior information is determined based on a preset abnormal behavior data set, and a determination result is obtained; When the determination result is a high confidence degree determination result, the preparatory state of the actuator of the door lock is activated according to the high confidence degree determination result, and a driving preparation instruction is obtained; The motor driving module of the door lock is controlled in standby mode based on the driving preparation instruction, and a motor standby signal is obtained; The driving preparation instruction and the motor standby signal are controlled in cooperation with the door lock hardware, and the hierarchical control instruction is obtained.

[0009] Further, it also includes: When the determination result is a low confidence degree determination result, the main control unit of the door lock is switched to a sleep mode according to the low confidence degree determination result, and a peripheral sleep instruction set is obtained; The sampling frequency of the door lock is adjusted based on the peripheral sleep instruction set, and an energy-saving monitoring signal is obtained; The torque of the motor driving module is locked according to the energy-saving monitoring signal, and a locked state signal is obtained; The peripheral sleep instruction set and the locked state signal are combined to control the door lock hardware to sleep, and the hierarchical control instruction is obtained.

[0010] Further, the alarm protocol matching is performed based on the hierarchical control instruction and the user behavior information, and alarm execution information is obtained, comprising: The hardware state of the door lock is decoded according to the hierarchical control instruction, and a door lock motor state identifier and a security level code are obtained; The user behavior information is segmented according to the door lock motor state identifier, and a behavior segment set is obtained; The security level code is indexed based on a preset alarm protocol library, and a target protocol type is obtained; The behavior segment set and the target protocol type are matched to obtain the alarm execution information.

[0011] Further, it also includes: The alarm execution information is converted into a protocol format to obtain a standard alarm data packet and an alarm level identifier; The standard alarm data packet is assigned to a priority queue according to the alarm level identifier, and a to-be-transmitted alarm queue is obtained; The to-be-transmitted alarm queue is detected for a transmission path to obtain an optimal transmission path and a redundant backup node; Distribute the standard alarm data packet based on the optimal transmission path to obtain a cloud storage location; Synchronize and check data of the redundant backup node and the cloud storage location to obtain a backup consistency result and a failure recovery strategy.

[0012] The application further provides a door lock opening intention recognition system based on behavior analysis, applied to the door lock opening intention recognition method based on behavior analysis. The acquisition module is configured to acquire user behavior information, hand action information and door handle pressure information collected by the door lock, and perform identification calculation to obtain a behavior intention level. The analysis module is configured to perform operation intention analysis on the door handle pressure information and the hand action information according to the behavior intention level, and obtain abnormal behavior information. The correlation module is configured to perform hardware state hierarchical switching on the door lock according to the abnormal behavior information, and obtain hierarchical control instructions. The processing module is configured to perform alarm protocol matching based on the hierarchical control instructions and the user behavior information, and obtain alarm execution information.

[0013] The door lock opening intention recognition method and system based on behavior analysis have the following advantages: The behavior intention level is established through ternary collaborative analysis of user behavior information, hand action information and door handle pressure information, which breaks through the limitation of single-dimensional judgment, significantly reduces the intention misjudgment rate, and improves the accuracy of opening intention recognition. The door lock hardware state hierarchical switching is driven in real time according to the abnormal behavior information, the safety threat judgment result is directly converted into a physical layer control instruction, a millisecond-level risk blocking response is realized, and the safety protection hysteresis problem is solved. The user behavior data is linked based on the hierarchical control instructions, a multi-dimensional matching mechanism of the alarm protocol is established, the real-time mapping of abnormal behavior characteristics and hardware response state is realized, the on-site risk level is automatically adapted, and the resource utilization efficiency of the security system is improved. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 It is a flowchart of the door lock opening intention recognition method based on behavior analysis provided by the application. Figure 2 It is a structure diagram of the door lock opening intention recognition system based on behavior analysis provided by the application.

[0015] The implementation, functional characteristics and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0017] The present application is further described below in combination with the drawings and specific embodiments.

[0018] The present application provides a door lock opening intention recognition method based on behavior analysis, comprising: Step S1: acquiring user behavior information, hand action information and door handle pressure information collected by the door lock, and performing identification calculation to obtain a behavior intention level; Step S2: according to the behavior intention level, performing operation intention analysis on the door handle pressure information and the hand action information to obtain abnormal behavior information; Step S3: performing hardware state grading switching on the door lock according to the abnormal behavior information to obtain a grading control instruction; Step S4: performing alarm protocol matching based on the grading control instruction and the user behavior information to obtain alarm execution information.

[0019] Based on the steps shown above, the detailed step process is shown as follows: Step S1: The moving trajectory sequence of the user approaching the door lock is captured by a horizontal view camera, and the hand key joint space vector collected by a downward oblique view camera is synchronously acquired. A pressure sensing array deployed on the door handle generates a pressure distribution curve. Time stamp synchronization alignment operation is performed on the above-mentioned multi-source data to establish the time axis consistency association of the moving trajectory, hand action and pressure signal.

[0020] The identification calculation adopts a hierarchical feature fusion architecture. The direction discrete feature and acceleration change feature of the moving trajectory are extracted, the joint angle dynamic change parameter and posture stability measure of the hand action information are analyzed, and the concentration degree feature and gradient change feature of the pressure distribution are calculated. The above-mentioned features are input into a pre-trained behavior intention evaluation model, which analyzes the action context association through a time sequence feature fusion unit and outputs a behavior intention level in the form of continuous numerical value. This level represents the confidence level of the user's opening intention, and the high level interval corresponds to the explicit operation intention, and the low level interval indicates the to-be-confirmed intention state.

[0021] Step S2: The analysis strategy is dynamically configured based on the behavior intention level. When the intention level is high, the fine analysis mode is enabled, and when the intention level is low, the basic monitoring mode is activated. Core indicators are extracted for pressure information, including pressure change rate characteristics, pressure distribution deviation from the standard template, and sustained grip stability parameters. The hand action information is reconstructed in three-dimensional space trajectory, and the deviation of the hand movement path from the standard operation path is calculated. The abnormal acceleration characteristics of joint motion are analyzed synchronously.

[0022] An analysis logic is constructed with the intention level as the decision node. For high intention level, stress mutation and grip stability detection are focused on, and for low intention level, spatial trajectory deviation analysis is strengthened. Pressure features and hand action features are fused, and abnormal behavior information is output through the abnormal decision logic unit, including abnormal state markers, risk classification codes, and confidence evaluation values.

[0023] Step S3: A hardware state decision tree mapping is performed based on the abnormal behavior information. The risk classification code and confidence evaluation value in the abnormal behavior information are analyzed, and the preset safety response level matrix is matched. When the risk code is high (class A) and the confidence is greater than or equal to the preset threshold, the emergency protection state is triggered. When the risk code is medium (classes B and C), the enhanced verification state is activated. When the risk code is low (class D) or the confidence is insufficient, the regular monitoring state is maintained.

[0024] The hardware state switching is realized through the door lock embedded control bus to achieve physical response. In the emergency protection state, pulse instructions are sent to the motor drive module to instantaneously start the electromagnetic lock dead lock mechanism and cut off the power supply of the biometric identification module. In the enhanced verification state, the multi-spectrum living body detection unit and the backup pressure sensor array are activated. In the regular monitoring state, the non-core computing unit is closed and the sensor sampling frequency is reduced. The output hierarchical control instructions include binary state codes (corresponding to three levels of hardware state) and attached parameter sets (such as motor torque value and biometric identification activation flag), ensuring that the instructions can drive the door lock physical execution mechanism.

[0025] Step S4: A multi-dimensional decision engine operation is performed for dynamic alarm strategy. The hardware state code and biometric identification activation flag in the hierarchical control instruction are extracted, the mobile trajectory persistence feature of user behavior information and the behavior intention level history record are bound, and a composite decision feature vector is generated. According to the vector, three-layer retrieval is performed in the preset protocol library: Primary matching: the hardware state code is used as the index to locate the basic protocol cluster; Secondary filtering: the biometric identification activation flag is used to screen the sub-protocol set; Final optimization: the protocol weight parameters are adjusted in combination with the behavior persistence feature.

[0026] The alarm execution information generation process integrates a redundancy check mechanism, performs conflict detection and priority sorting on matched protocols, and selects the protocol version with the highest confidence according to a risk time decay model when multiple protocol conflicts are detected.

[0027] The application provides a door lock opening intention recognition method based on behavior analysis, which breaks through the limitation of single dimension judgment by ternary collaborative analysis of user behavior information, hand movement information and door lock handle pressure information, establishes a behavior intention level, significantly reduces the intention misjudgment rate, and improves the accuracy of opening intention recognition. According to the abnormal behavior information, the door lock hardware state is driven to grade switching in real time, the safety threat judgment result is directly converted into a physical layer control instruction, a millisecond level risk blocking response is realized, and the safety protection hysteresis problem is solved. Based on the hierarchical control instruction, the user behavior data is linked, a multi-dimensional matching mechanism of alarm protocol is established, the real-time mapping of abnormal behavior characteristics and hardware response state is realized, the on-site risk level is automatically adapted, and the resource utilization efficiency of the security system is improved.

[0028] In one embodiment, user behavior information, hand movement information and door lock handle pressure information collected by a door lock are acquired, and identification calculation is performed to obtain a behavior intention level, including: A horizontal perspective camera collects a continuous image frame sequence of a user approaching the door lock, and a target contour tracking technology is used to process the image frame. The user moving body and the static environment are separated by a background difference algorithm to obtain target contour spatial position change data. The target position between continuous frames forms a displacement vector sequence, and the motion speed direction distribution and acceleration change curve are calculated based on the displacement vector. Key parameters of the motion trajectory are extracted, including path curvature change extreme point detection, uniform speed movement duration quantitative analysis and trunk axis angle offset. The trajectory data are normalized and coded to form structured user behavior information, including distance change vector from the door lock reference point, movement path probability distribution graph and acceleration feature matrix, which are used as basic spatial parameters for behavior intention recognition.

[0029] The oblique downward-looking camera captures a sequence of image frames of the handle operating area, and performs multi-stage positioning of the hand biometric feature area. In the initial processing stage, the hand region contour is segmented by fusing skin color and edge features, and in the secondary processing stage, a preset number of joint reference point coordinates are positioned inside the contour. The feature point coordinate data is added to the inter-frame displacement smoothing constraint, and the joint motion velocity parameters are generated according to the pose change vector of adjacent frames. The output data includes a four-dimensional dynamic feature set: a hand key point three-dimensional space coordinate tensor, a joint angle change rate time curve, a holding state stability index, and a hand approaching the handle space acceleration spectrum. This data set completely records the biological kinematics features of operating the handle, and provides a joint-level kinematics evidence chain for intention determination.

[0030] The pressure sensing array continuously acquires a raw pressure data stream at a preset sampling frequency. In the data preprocessing stage, signal smoothing processing is adopted to filter out environmental vibration and temperature drift interference noise. The effective holding area is identified by a touch point density clustering algorithm, and non-operation contact signals are segmented. The pressure center point movement trajectory is calculated, and a pressure gradient distribution change curve is generated. A pressure distribution three-dimensional heat map is constructed, and the core parameters are quantified: pressure center displacement vector sequence, pressure peak change rate curve, holding force balance index, and contact area stability parameter. The pressure data is attached with a space-time synchronization identifier to ensure millisecond-level time axis alignment with user behavior information and hand action information, forming a complete operating mechanics feature matrix, and providing a physical load basis for modal intention recognition.

[0031] The pressure recognition process integrates an operating state classification mechanism. Based on the contact form features, the holding operation is divided into three modes: full-palm wrapping holding, four-finger holding, and single-point pressing contact. Corresponding pressure feature extraction logic is enabled for different holding modes: full-palm holding focuses on pressure distribution uniformity analysis, and single-point pressing emphasizes peak duration detection. The final output of the door lock handle pressure information includes pressure heat distribution map, operation mode classification code, and dynamic calibration parameter set, which completely records the spatial distribution characteristics and time evolution law of human hand force.

[0032] A pressure-dominated multi-modal feature fusion decision is executed. A three-dimensional decision space coordinate system is constructed: the X-axis maps the contact area stability parameter in the pressure information, the Y-axis is associated with the movement direction probability distribution of the user behavior information, and the Z-axis is bound to the joint angular velocity change curve of the hand action information. In the decision space, intention determination areas are divided: high-density pressure distribution areas are associated with operation certainty intentions, and low-balance pressure areas map potential abnormal risk states.

[0033] The modal recognition adopts a phased analysis strategy: Pre-operation stage: when the user behavior information shows that the distance is within the threshold range of the door lock, combined with the approach acceleration spectrum in the hand action information, an intention warming-up level is generated; Contact establishment stage: Based on the holding force balance index in pressure information, superimpose the hand key point holding posture stability parameter, calculate the contact confidence; Pressure execution stage: Fusion pressure gradient change curve and joint motion trajectory deviation, generate the final behavior intention level.

[0034] The intention level output is a composite structure of discrete state coding and continuous confidence value. The discrete coding defines four levels of operation intention: clear operation intention, to be confirmed intention, low-risk unknown intention, and high-risk abnormal intention. The continuous confidence value ranges from 0 to 100, representing the reliability of the judgment. When the confidence value is below the critical threshold, the multi-modal data review mechanism is automatically triggered.

[0035] This embodiment establishes the spatiotemporal synchronization relationship between user movement trajectory and hand joint motion through the cooperation of door lock horizontal view angle and oblique downward view angle dual cameras, realizes the cooperative modeling of limb behavior and operation action, and overcomes the defect that single modal analysis scene coverage is not complete. Through the multi-point holding mode recognition technology of pressure sensing array, three types of operation forms are divided: full palm wrapping, four-finger holding and single-point pressing, providing customized pressure feature extraction logic for different contact states, eliminating the risk of false triggering caused by traditional single threshold judgment. Based on the three-dimensional decision space coordinate system, a fusion analysis mechanism of pressure distribution, movement direction and joint motion is constructed, and the intention level is dynamically calibrated in the pre-operation, contact establishment and pressure execution stages, ensuring the temporal continuity of operation intention judgment and improving the real-time performance of safety protection.

[0036] In one embodiment, based on the door lock handle pressure information, the user behavior information and hand action information are subjected to modal intention recognition to obtain the behavior intention level, including: The user behavior information contains a continuous time stamp movement trajectory sequence. Direction domain segmentation processing is performed on the trajectory sequence: based on the movement speed change characteristics, the motion stages are divided, including the acceleration approach segment, the uniform speed movement segment and the deceleration adjustment segment. Direction vector discretization is implemented in each motion stage to calculate the distribution density of the movement direction angle in the polar coordinate system. The curvature change rate is used to determine the boundary points in the segmentation process, and when the trajectory curvature value exceeds the critical threshold, a segmentation marker point is established to split the continuous trajectory into several sub-paths with consistent direction.

[0037] Extract the core direction features of the sub-paths: calculate the spatial projection components of the direction vectors of each segment, and establish a spatial projection coordinate system with the door lock handle as the origin. The approach direction information contains a three-order data structure: motion stage marker sequence (indicating acceleration / uniform speed / deceleration stage), sub-path direction angle distribution set, and overall movement trend convergence parameter (quantifying the convergence degree of the trajectory to the handle center). This information accurately describes the intention direction and dynamic adjustment mode of the user approaching the door lock, providing a spatial behavior basis for intention recognition.

[0038] The hand action information contains time series data of joint key point spatial coordinates. Hierarchical gesture posture analysis is implemented: seven basic gesture postures are defined through joint angle change matrix, including palm-down holding, lateral gripping, fingertip tapping, etc. The recognition process includes three-level verification logic: Primary contour verification: based on the area and shape index of the hand circumscribed polygon to determine the operation intent category; Intermediate joint topology verification: according to the joint angle equation of the phalanx to calculate the gesture stability parameter; Advanced dynamic trajectory verification: analyze the spatiotemporal continuity features of continuous frame gesture transition.

[0039] Output action feature information constructs a four-dimensional feature tensor: gesture posture classification code (mapping seven basic types), joint motion trajectory curvature spectrum, operation force discretization evaluation value, and gesture transition frequency matrix. This data tensor records the biomechanical feature evolution process of the operation handle in real time, providing dynamic evidence chain for intent analysis of operation posture.

[0040] The three-dimensional pressure distribution thermograph collected by the pressure sensing array inputs the gradient field dynamic analysis engine, and performs pressure center point displacement trajectory tracking. The pressure gradient vector direction change is calculated and a time evolution spectrum is constructed to continuously analyze the pressure field evolution process, including initial contact stage pressure propagation path modeling, stable pressure stage isopleth density extraction, and pressure release stage decay feature recording.

[0041] The generated pressure change information encapsulates a multi-feature set including displacement coordinate sequence and gradient direction change spectrum, and synchronously integrates pressure intensity change rate curve, contact area differential parameter, and balance decay index, to fully characterize the pressure spatiotemporal evolution characteristics. The gradient analysis process is associated with the initial loading point coordinate space mapping, ensuring that the displacement trajectory reflects the operation mechanics evolution law, and providing dynamic mechanics benchmark for intent calculation.

[0042] Multi-dimensional data input implements hierarchical fusion decision-making based on a pre-set rule library, which includes spatial coordination logic, biomechanics matching mechanism, and time sequence evolution alignment strategy. The spatial coordination logic verifies the movement convergence parameters of the approaching direction information and the spatial consistency of the pressure center trajectory; the biomechanics matching mechanism is associated with the action feature gesture code and the pressure contact area parameter; the time sequence evolution strategy aligns the motion phase marker with the pressure gradient curve phase feature.

[0043] The intention calculation process outputs a three-level processing result based on rules: spatial coordination logic generates an initial intention confidence value, biomechanical rules perform dynamic correction operations, and a timing evolution strategy finally determines the discrete intention level. When the confidence value is below the critical threshold, a multi-modal historical data review mechanism is activated, and the behavior intention level is output, including continuous confidence values and four types of discrete risk codes, establishing a causal closed loop of operation intention judgment and risk response. The processing flow implements a hardware resource dynamic adaptation strategy, which closes non-core unit resources in low-risk states and triggers instantaneous recovery of full-frequency monitoring in stress mutation events.

[0044] The embodiment decomposes the user movement trajectory into acceleration, uniform speed and deceleration movement stages through direction trajectory segmentation technology, and quantifies the spatial projection characteristics of direction vectors, overcoming the defect that traditional continuous trajectory analysis is not sensitive to dynamic adjustment behavior, and accurately capturing the intention direction change of the user approaching the door lock. A hierarchical gesture recognition mechanism is used to identify seven types of basic operation postures through joint angle simultaneous equations and contour morphology cooperative verification, eliminating the misclassification risk caused by hand feature point drift in complex environments, and improving the biomechanical reliability of operation intention discrimination. Based on the dynamic analysis technology of pressure gradient field, the pressure center displacement trajectory and intensity change spectrum are constructed, combined with the spatio-temporal evolution characteristics of the three stages of initial contact, stable pressure application and pressure release, the physical linkage modeling of pressure change characteristics and user behavior is realized, and the detection sensitivity of abnormal pressure patterns is enhanced.

[0045] In one embodiment, according to the behavior intention level, the door lock handle pressure information and hand action information are analyzed for operation intention to obtain abnormal behavior information, including: The behavior intention level is dynamically allocated with coefficients, and the dual-track analysis of intention confidence and risk parameters is performed. The intention level is decomposed into two dimensions of discrete state coding and continuous confidence value, the discrete coding maps the preset pressure analysis mode matrix, and the continuous confidence value drives the action analysis intensity curve generation. The four-level operation intention triggers different coefficient combinations: the explicit operation intention state enables the high-sensitivity pressure analysis coefficient matching, and the high-risk abnormal intention state strengthens the action analysis coefficient weight.

[0046] The dynamic allocation process integrates the historical intention review mechanism, and adjusts the coefficient decay rate parameter according to the recent intention level change trend. The output pressure analysis coefficient contains three types of core parameters: pressure segmentation sampling frequency base, pressure gradient detection sensitivity threshold, and holding stability detection window size; the action analysis coefficient encapsulates four-dimensional parameter set: joint trajectory clustering accuracy level, motion trajectory analysis depth, dynamic feature refresh rate setting value and confidence compensation factor. The coefficient allocation result is attached with a time validity identifier to ensure synchronization update with the real-time state of the door lock.

[0047] The hierarchical pressure signal segmentation processing is performed based on a pressure analysis coefficient. A sampling frequency basis determines a pressure time axis cutting density, and a pressure gradient detection sensitive threshold defines an abnormal event judgment boundary in a section. Pressure information is subjected to multi-dimensional segmentation: in the time dimension, the pressure curve is divided into equal or variable length time sections according to the sampling frequency basis; and in the space dimension, a space clustering region is constructed according to the pressure thermal map distribution characteristics, and combined with a holding stability detection window size to filter transient interference signals.

[0048] The segmentation process adopts a double verification mechanism: the time segmentation point needs to meet the pressure gradient value and sensitive threshold ratio constraint, and the space partition needs to pass the pressure distribution uniformity check. The output segmented pressure information encapsulates three types of data entities: a time domain pressure segment feature set (containing start and end time stamps, average pressure intensity, and gradient peak point), a space pressure partition thermal map (identifying a pressure center displacement path), and a time-space coupling verification mark (recording the mapping relationship between the time slice and the space partition). The segmentation result retains the physical characteristics of the original pressure information while reducing the analysis complexity, and provides a structured input for anomaly detection.

[0049] The hierarchical aggregation processing of joint motion trajectory is performed based on a motion analysis coefficient. A joint trajectory clustering accuracy level parameter determines the space division granularity, and a motion trajectory analysis depth parameter defines the feature extraction dimension. The processing process realizes three-stage data aggregation: the primary space grid division establishes a three-dimensional clustering grid according to the joint coordinate space range, and the grid density is dynamically adjusted to adapt to the accuracy level requirement.

[0050] The intermediate motion feature fusion extracts the joint displacement vector features in the grid element, calculates the consistency parameter of the motion direction and the acceleration change spectrum. The high-level behavior mode classification integrates the motion features to form an aggregated motion vector, matches the preset operation mode template to generate an intention classification code. The output action intention information encapsulates a four-dimensional data body: an aggregated motion vector set, an operation mode classification label, a joint group motion coordination index, and a trajectory stability evaluation matrix. The data body realizes the structured representation of the operation intention through space compression and feature fusion.

[0051] The time-space coupling verification processing of pressure and motion features is implemented. The time axis synchronization calibration incrementally aligns the time stamp sequence of the segmented pressure information with the motion intention information trajectory index, and the interpolation compensation mechanism is started when the error exceeds the limit. The space domain feature mapping projects the pressure space partition thermal map to the joint coordinate space, and establishes a three-dimensional position mapping equation of the pressure center point and the metacarpal joint.

[0052] The feature level cross-validation dynamically switches the pressure gradient detection mode by operating mode classification label, and the articulation coordination index real-time corrects the pressure distribution uniformity parameter. The output behavior characteristic information constructs a five-dimensional data body: time and space synchronous verification mark, pressure-action coupling coefficient matrix, modal conflict feature set, multi-source confidence score table, and compressed feature vector, realizing lossless and simplified expression of physical operation characteristics.

[0053] Based on the abnormal behavior feature library, progressive rule application is performed. The physical conflict rule checks the deviation of the pressure-action coupling coefficient from the standard operation parameter, identifies the mechanical contradiction between the force direction and the joint motion. The time and space anomaly rule compares the time and space synchronous verification mark with the normal sequence mode, detects the time sequence dislocation phenomenon of the operation rhythm. The behavior contradiction rule analyzes the triggering frequency of the modal conflict feature set to determine the logical deviation of the limb behavior and the operation intention.

[0054] The rule application adopts a hierarchical response strategy: the physical conflict preliminary screening marks potential risks, the time and space anomaly verification triggers in-depth analysis, and the behavior contradiction final judgment generates abnormal conclusions. The output abnormal behavior information contains a four-element data structure: abnormal type classification code (including A1-C3 levels), risk confidence quantitative value (0-100 interval), impact range space vector, and emergency response control identifier. When the confidence threshold is reached, the segmented pressure original data review mechanism is started. The risk judgment result is directly associated with the hardware response protocol, and the A1 level abnormality immediately activates the electromagnetic dead lock physical protection.

[0055] The embodiment realizes intention confidence driven parameter customization configuration by dynamically allocating behavior intention level to pressure and action analysis coefficient, breaks through the misjudgment bottleneck caused by traditional fixed threshold, and improves the environmental adaptability of abnormal detection. The mechanical mapping relationship between pressure distribution and joint motion is established by time axis synchronous calibration and space domain feature projection, solving the feature fragmentation problem caused by asynchronous multi-source data. Based on the three-layer rule verification mechanism of the abnormal behavior feature library, the abnormal behavior information containing risk level code and confidence score is generated, which realizes hierarchical response while ensuring the accuracy of judgment: low-risk events trigger data review, high-risk states instantly activate physical protection, achieving the optimized balance of safety and efficiency.

[0056] In one embodiment, the door lock handle pressure information is segmented according to the pressure analysis coefficient to obtain segmented pressure information, including: Step 1: decompose the frequency component of the door lock handle pressure information to obtain high-frequency pressure component and low-frequency pressure component The door lock handle pressure information is signal separated. A preset waveform analysis element is used to implement hierarchical decomposition operation on the original pressure data stream: the primary decomposition extracts the high-frequency transient component in the pressure signal, capturing the pressure rapid fluctuation characteristics; the secondary decomposition separates the medium-frequency slowly changing component, representing the regular operation pressure change mode; the final decomposition retains the low-frequency trend component, reflecting the pressure macro evolution law. The high-frequency pressure component generates a dynamic detail feature set, recording the pressure mutation amplitude and occurrence frequency characteristics; the low-frequency pressure component outputs the baseline evolution trajectory, quantifying the pressure steady-state drift characteristics. The decomposition process implements signal energy standardization processing, eliminating the amplitude deviation caused by individual differences of the sensing unit, ensuring the comparability of the component data. The output component is attached with a time synchronization identifier, which is time-aligned with the behavior intention analysis module.

[0057] The high-frequency pressure component is input into the dynamic detection unit to perform mutation feature analysis. The sensitivity parameter in the pressure analysis coefficient drives the detection mechanism to adaptively adjust: high sensitivity coefficient enables fine detection mode to capture micro-amplitude pressure fluctuation characteristics; low sensitivity coefficient switches to wide-amplitude detection mode to respond only to significant pressure mutation events. The detection process implements three-order operation: calculate the absolute value sequence of the change rate of the high-frequency component, mark the position of the change rate extreme point; suppress signal noise interference; select effective mutation points according to the dynamic detection mechanism.

[0058] The candidate pressure point set is packaged into a spatiotemporal feature sequence: the timestamp records the mutation occurrence time, the spatial coordinates map the pressure sensing array distribution characteristics, and the mutation intensity quantifies the pressure change rate level. Before output, pseudo-signal filtering is performed: adjacent mutation points with a time interval below a critical value are merged into a single point, eliminating abnormal markers caused by sensing noise.

[0059] The low-frequency pressure component is input into the trend analysis unit to perform baseline evolution feature extraction. The identification process implements baseline trajectory modeling operation: construct a polynomial fitting curve of pressure baseline drift, calculate the curvature change extreme point of the curve. The turning point determination adopts a gradient threshold mechanism: when the baseline curvature change rate exceeds a dynamic critical value, it is marked as a trend turning point. The output trend pressure point set is packaged into a spatiotemporal coordinate sequence: the timestamp records the turning time, the baseline offset quantifies the pressure steady-state change amplitude, and the trend type identifier distinguishes the rising / falling / platform turning mode. The identification process is attached with noise suppression processing: when the curvature change duration is below a threshold, it is considered as disturbance noise and automatically filtered, ensuring that the turning point reflects the true operation intention evolution.

[0060] The candidate pressure point set and the trend pressure point set input conflict resolution engine performs spatio-temporal collaborative optimization. The resolution process implements a time window conflict detection mechanism: a sliding time window is established to scan adjacent point sets, and when the time interval between the mutation point and the turning point is below a critical value, a conflict resolution is triggered. The resolution strategy is based on the priority of physical meaning: the mutation point gives priority to the pressure change event, and the turning point focuses on maintaining the continuity of operation. After the resolution, an effective pressure point sequence is generated: the timestamp sequence is strictly arranged in time sequence, the spatial coordinates map the hot area of the pressure sensor array, and the point type identification indicates the mutation / turning attribute. Before output, perform redundant point removal: merge adjacent points of the same type into a single point if the distance is too close, to eliminate decision interference caused by feature overlap.

[0061] The effective pressure point sequence drives the pressure data stream segmentation processing. The division process implements three-order operation: cutting the original pressure data stream with adjacent effective points as the start and end boundaries, adding time window expansion compensation to the head and tail edge segments, and marking the segment attribute characteristics based on point type. The output segmented pressure information is constructed as a structured data set: the time segment records the start and end timestamps and the duration, the spatial thermal map matrix stores the pressure distribution characteristics, and the attribute label labels the mutation dominant type / trend dominant type / mixed type segment category. The division process implements segment feature enhancement: extract the pressure mean and variance statistics within the segment, calculate the pressure gradient direction change spectrum, generate the pressure center displacement trajectory vector, and form a multi-dimensional feature description system.

[0062] The embodiment separates high-frequency transient and low-frequency trend pressure components through multi-scale frequency decomposition technology, realizes the synchronous capture of differentiated threats such as violent pressure and exploratory lock picking, and breaks through the detection blind area of traditional single frequency band analysis. Based on the dynamic sensitivity adjustment mechanism of the pressure analysis coefficient, the fine detection and wide detection modes are adaptively switched according to the risk scenario, which reduces the false alarm rate of environmental interference while maintaining high threat recognition rate. The spatio-temporal collaborative conflict resolution strategy is adopted to preferentially retain high-risk mutation points representing violent door breaking, ensuring that the security response is not delayed due to trend analysis, and strengthening the priority of high-risk event handling.

[0063] In one embodiment, the door lock is switched according to the abnormal behavior information Hardware state grading, get hierarchical control instruction, including: The abnormal behavior information input preset data set executes confidence quantization analysis. The abnormal behavior data set constructs a three-dimensional judgment matrix, maps the abnormal type code to the risk level weight factor, associates the risk confidence value with the dynamic threshold adjustment curve, and binds the spatial distribution verification rule with the influence range vector. The judgment process implements a dual-track mechanism of static rule matching and dynamic threshold correction: in the static rule matching stage, the feature deviation is calculated by searching the data set standard risk template through the abnormal type code; in the dynamic threshold correction stage, the confidence compensation coefficient is generated according to the historical abnormal occurrence frequency and the handling record, and the judgment threshold boundary is adjusted in real time.

[0064] The output decision result is a composite data structure, including binary high / low confidence markers, quantized confidence scores in the 0-100 interval, decision failure timestamps, and conflict feature index tables. When the confidence score exceeds the critical threshold and the conflict feature index is empty, it is marked as a high-confidence decision result; when the confidence score falls into the ambiguous interval, the data set incremental learning mechanism is triggered to update the weight factor allocation strategy of the abnormal type code.

[0065] High-confidence decision results trigger the pre-activation protocol of the actuator. The pre-activation protocol implements four-stage control of mechanism type analysis, energy preloading, mechanical pre-positioning, and safety interlock release: mechanism type analysis matches the actuator type according to the abnormal type code, electromagnetic lock, motor, or mechanical brake; energy preloading sends a pulse width modulation signal to the target mechanism power supply module, boosting the drive voltage to the standby threshold; mechanical pre-positioning controls the motor to output a pre-torque, eliminating the gear transmission gap; safety interlock release cuts off the electromagnetic holding circuit of the physical safety lock.

[0066] The generation of the drive preparation instruction encapsulates a four-dimensional control vector: the target mechanism address code identifies the operation object, the preloaded energy level sets the voltage parameter, the mechanical pre-positioning parameter defines the torque value, and the safety interlock state identifies the physical lock status. The instruction is attached with a time-sensitive marker, and a state rollback mechanism is triggered if it is not executed within a certain time, ensuring that a false activation operation can be reversed to a safe baseline state.

[0067] The drive preparation instruction inputs the motor control bus to execute standby state configuration. Standby control implements instruction analysis and energy adaptation operations, extracts the preloaded energy level parameter in the instruction and converts it into a motor coil drive current reference value, and calibrates the drive circuit output level according to the reference value. The phase pre-calibration process generates a rotor position compensation pulse based on the mechanical pre-positioning parameter of the drive preparation instruction to eliminate permanent magnet positioning errors. The dynamic response optimization phase binds the target mechanism address code with the preset response curve to control the motor to maintain a no-load micro-vibration state and continuously monitor the back electromotive force waveform. The health monitoring unit samples coil impedance change data in real time, and automatically switches to a safe standby mode when abnormal fluctuations exceed the limit. The output motor standby signal includes the drive current reference value, phase compensation identifier, vibration frequency setting parameter, and 800 millisecond effective action countdown. When the countdown ends without receiving an execution signal, the power is automatically cut off to generate a fault log, blocking the risk of accidental motor action.

[0068] Hardware cooperative control performs state synchronization verification and energy level dynamic matching operations. State synchronization verification compares the effective countdown of the standby signal with the time-sensitive marker of the preparation instruction, and if the time deviation is greater than the threshold, a clock compensation protocol is started to reset the time reference. The energy level dynamic matching process integrates the drive current reference value and the preloaded energy level parameter to perform power balancing calculations and generate a drive composite parameter set that matches the motor torque output characteristics and energy supply curve.

[0069] The physical linkage packaging stage integrates mechanical pre-positioning parameters and phase compensation identifiers, constructs a mechanical and electrical combined control vector, and injects a temperature compensation coefficient. Finally, a hierarchical control instruction package is generated as an executable binary code stream: a high-confidence emergency response instruction fuses 24-volt reinforced preloading parameters and high-frequency vibration signals to activate 0.1-second motor locked-rotor protection; a medium-confidence preparation instruction couples 12-volt standard preloading and medium-frequency monitoring signals to maintain a 500-millisecond reversible standby window; and a low-confidence monitoring instruction binds health monitoring fault codes to trigger a multi-sensor deep self-checking protocol. Before the instruction output, triple safety verification is implemented: mechanical and electrical parameter compatibility inspection detects interface protocol conflicts, power transient impact test verifies voltage fluctuation tolerance, and safety interlock loop diagnosis confirms the integrity of the physical lock state.

[0070] The embodiment realizes intention confidence-driven parameter customization configuration by dynamically allocating stress and motion analysis coefficients according to behavior intention levels, breaks through the misjudgment bottleneck caused by traditional fixed thresholds, and improves the environmental adaptability of abnormal detection. The hierarchical joint trajectory clustering technology is used to analyze hand motion information, and the operation mode classification label is generated based on space grid division and motion feature fusion, which eliminates the interference of joint coordinate drift in complex environments on intention judgment and enhances the biological rationality of gesture recognition. The stress segmentation and multi-modal spatio-temporal fusion processing of motion intention are implemented, the mechanical mapping relationship between stress distribution and joint motion is established through time axis synchronous calibration and space domain feature projection, and the feature fragmentation problem caused by asynchronous multi-source data is solved.

[0071] In one embodiment, when the determination result is a low-confidence determination result, further comprising: The low-confidence determination result triggers a master control unit hibernation state migration protocol. The hibernation mode switching performs multi-level power consumption control operations: the main processor switches to a low-frequency clock running mode, and the floating point operation unit and cache function unit are closed; the unnecessary external device module power supply circuit is cut off through the bus isolation circuit, and the oblique downward-looking camera and millimeter wave sensor enter the physical power-off state; the dynamic allocation of memory areas performs data compression storage protocol, and the behavior intention level history record and abnormal feature key data set are reserved; the security coprocessor is maintained in an active state, and the basic risk monitoring function is taken over. The output peripheral hibernation instruction set constructs a three-dimensional control structure: the power supply control vector identifies the power-on / off state of each peripheral module, the clock management parameter defines the running frequency ratio of the master control unit, and the wake-up trigger configures the preset stress mutation threshold and the voiceprint feature matching condition. The instruction set embeds an event-driven tag mechanism, which only interrupts the hibernation state when the preset wake-up condition is detected, eliminating the invalid power consumption generated by periodic polling.

[0072] The sampling frequency adjustment implements a self-adaptive control strategy of the sensing module. According to the clock management parameter of the instruction set, the working mode of the pressure sensing array is adjusted: the sampling rate of the distributed pressure sensor is reduced to the reference low frequency value, and the regional polling collection protocol is enabled; the horizontal view camera is switched to the moving detection trigger collection mode, the oblique downward view camera is closed to the light supplement unit and the exposure interval period is extended; the hand joint tracking algorithm is switched to the key point simplified topology, and the number of joint feature points is compressed to the core skeletal points.

[0073] The adjustment process continuously calculates the energy consumption accuracy balance parameter, quantizes the corresponding relationship between the feature loss rate caused by the sampling rate reduction and the power consumption reduction ratio, and automatically increases the sampling rate of the specified sensor when the balance parameter is out of limit. The output energy saving monitoring signal encapsulates four data entities: the sensing state matrix records the real-time sampling parameter configuration, the accuracy compensation factor quantizes the feature loss influence value, the wake-up ready flag identifies the critical condition set that can trigger full function recovery, and the power consumption record vector accumulates the historical data of the running energy consumption of each module.

[0074] The motor torque lock implements the pre-protective braking control. The sensing state matrix and the accuracy compensation factor of the energy saving monitoring signal are analyzed to generate the torque lock grading parameter: the static holding torque is enabled in the low frequency sampling state, and the dynamic micro-vibration lock is set in the medium frequency sampling mode. The lock operation implements three-stage protection: the magnetic pole phase pre-calibration eliminates the positioning dead zone, the coil constant current drive suppresses the temperature drift effect, and the overload fuse sets the torque protection threshold. The output lock state signal contains the rotor position verification code, the coil current reference value, the mechanical braking state identifier and the thermal protection threshold parameter, and the heat dissipation unit is automatically triggered when the abnormal temperature rise exceeds the limit.

[0075] The hardware sleep control implements a hierarchical strategy fusion: the wake-up trigger configuration of the sleep instruction set and the mechanical braking identifier of the lock state signal are compared, and the braking protection protocol is preferentially executed when there is a conflict; the energy level matching engine integrates the power supply control vector and the power consumption record vector to generate the limit energy saving parameter threshold; the safety monitoring unit continuously checks the relationship between the temperature compensation coefficient and the thermal protection threshold. The final hierarchical control instruction outputs a binary code stream: the deep sleep instruction (D1 level) completely closes the non-safety unit, and maintains the lock torque > 5 N·m; the energy saving standby instruction (D2 level) retains 10% of the basic sensor, and the lock torque ≤ 2 N·m; the pre-wake-up instruction (D3 level) enables the single sensor high frequency monitoring. Before the instruction is executed, three-stage verification is completed: power transient recovery test, braking redundancy verification, and clock drift compensation diagnosis, to ensure that the sleep state can be restored to full function operation within 0.5 seconds.

[0076] The embodiment triggers the multi-level sleep control of the master control unit through the low-confidence determination result, realizes the cooperative energy saving of the processor running at a reduced frequency and the unnecessary physical power-off of the peripheral device, breaks through the continuous power consumption defect of the traditional standby mode, and significantly prolongs the endurance period of the door lock. Based on the peripheral sleep instruction set, the dynamic adjustment of the sensing sampling rate is implemented, the region polling collection and the mobile detection triggering mechanism are adopted, the average power consumption is compressed to below the reference value under the premise of maintaining the basic security monitoring capability, and the energy waste problem caused by the fixed sampling frequency is solved. Through the torque lock grading control strategy, the static retention and dynamic micro-vibration locking mode are automatically switched according to the energy-saving monitoring signal, the electromagnetic loss in the motor standby state is eliminated, and at the same time, the brake response capability can be restored within 0.5 seconds in the event of a sudden threat.

[0077] In one embodiment, the alarm protocol matching is performed based on the hierarchical control instruction and the user behavior information, and alarm execution information is obtained, including: The hierarchical control instruction is input into the hardware state decoding unit to perform a physical layer analysis operation. The instruction structure is disassembled to separate the motor control field and the safety policy field, the motor control field is mapped to the drive register address space to analyze the rotor positioning identifier, the coil current reference value and the brake state flag. The safety policy field is converted into a standard safety level classification through a preset encoding table, and a four-dimensional safety level code is output: risk response level (R1-R4), protocol trigger priority (P0-P3), resource occupation matrix mark sensors and communication modules to be enabled, and time limit constraint parameter defines the longest execution time and countdown rule of the protocol.

[0078] The motor state identifier encapsulates the motion state parameters (rotor phase angle / idle vibration frequency / locked-rotor protection threshold), energy control parameters, and mechanical state identifier (gear meshing depth / brake pad clearance value). The decoding process implements an error checking mechanism, and when the register mapping conflicts, a redundant backup register reading is triggered, and when the security code is invalid, the historical safety level cache data is enabled to ensure the physical reliability of the output identifier and code.

[0079] Based on the motor state identifier, the user behavior information is dynamically segmented. When the motor is in a high vibration frequency state, a short window dense segmentation strategy is adopted, the window length is adapted to the high frequency vibration characteristics to strengthen the mobile direction mutation detection; when the motor is in a low power standby state, a long window sparse segmentation mode is enabled, the window length is extended to three times the reference value and the acceleration feature weight is weakened.

[0080] Rotor position marker is added with trajectory curvature extreme point capture algorithm in pre-engagement state, and the torso deflection gradient analysis weight is inhibited in brake state marker activation state. The segmentation process intercepts the behavior data stream in time window units, extracts trajectory curvature extreme points, acceleration change rate and torso deflection gradient in the window to generate time-stamped behavior segment units. The output behavior segment set encapsulates the time sequence data chain structure: segment ID identifies the space-time position, time window records the start and end timestamps, feature vector stores the core motion parameters, and state association marker binds the motor vibration frequency and brake state.

[0081] The motor state mutation event triggers the backtracking segmentation mechanism, and the vibration frequency jump threshold is added to the backtracking segmentation interval of the current time window, which blocks the risk of truncation of key behavior characteristics. The segmentation result continuously calibrates the mechanical vibration interference, and the trajectory smoothing algorithm is enabled to eliminate data distortion when the rotor phase angle is abnormal.

[0082] The security level code input alarm protocol library performs hierarchical indexing operation. The protocol library constructs a three-dimensional search architecture, maps the risk response level to the protocol trigger condition matrix, binds the protocol trigger priority to the response action sequence, and associates the resource occupation matrix with the hardware resource configuration strategy.

[0083] The indexing process implements dynamic path selection: filtering the basic protocol channel set according to the risk response level, sorting the protocols in the channel to generate an execution sequence, and enabling the corresponding sensor and communication module configuration parameters according to the resource occupation matrix.

[0084] The target protocol type output is a composite instruction structure, which includes the protocol core action set definition sound and light alarm, cloud upload and mechanical anti-lock operation instruction chain, time and space constraint parameters limit the upper limit of each action execution time and start interval, and abnormal handling plan encapsulates the degradation response strategy when protocol execution fails. The indexing process integrates conflict resolution mechanism, and generates the optimal execution path tree according to the risk level and resource occupation when multiple protocol channels are activated at the same time, to ensure that the protocol selection and hardware state are strictly matched.

[0085] The behavior segment set and the target protocol type input trigger the matching engine to perform two-dimensional alignment operation. The space-time phase alignment establishes a mapping relationship between the behavior segment timestamp sequence and the protocol action time axis, and the phase deviation is super-limited to enable the segment time stretching compensation algorithm to calibrate the feature timing.

[0086] Feature rule matching extracts trajectory curvature extreme value, acceleration change rate and torso deflection gradient of behavior feature vector, compares the preset abnormal feature template of protocol library to calculate the feature deviation degree and generate matching confidence score. Dynamic strategy adjustment responds according to confidence score classification: high confidence matching directly activates protocol core action set, medium confidence matching adds multi-modal data review process, and low confidence matching enables abnormal handling plan.

[0087] The alarm execution information is packaged as a five-element control instruction body: a protocol execution sequence defines an action execution order chain, a resource occupation configuration specifies a camera view angle and a communication channel, a time and space constraint parameter controls an action execution duration and delay, a dynamic strategy identifier records an execution decision logic, and a fuse mechanism presets a backup channel switching strategy when the cloud does not respond. Three verifications are performed before output: a protocol action and a motor state compatibility test detects physical conflicts, a resource conflict diagnosis checks sensor occupation conflicts, and a time and cost feasibility analysis verifies the executability of the countdown rule.

[0088] The embodiment solves the feature truncation problem of the traditional fixed segmentation mode in the fast operation scene by the hardware state driven behavior dynamic segmentation strategy, adjusts the behavior analysis time window size in real time according to the motor vibration frequency and the brake state, significantly improves the behavior capture integrity. Based on the three-dimensional search architecture of the alarm protocol library, the cooperative indexing mechanism of the risk response level, the protocol priority and the hardware resource is realized, which breaks through the limitation that the static rule library cannot adapt to the variable risk scene, and enhances the accuracy and timeliness of the protocol matching. Through the time and space stretching compensation algorithm of the behavior segment and the protocol action, the time sequence phase deviation caused by the hardware delay is dynamically calibrated, the mis-matching risk caused by the time dislocation is eliminated, and the reliability of the alarm triggering is ensured.

[0089] In one embodiment, it also includes: The alarm execution information performs protocol conversion processing and implements a hierarchical structured packaging operation. The action instruction chain of the protocol execution sequence is extracted and mapped to the international security communication standard field system, including alarm type identification, resource occupation configuration and action execution parameter. The constraint parameter is converted into a timestamp sequence and a geographic coordinate coding format, and the fuse mechanism in the dynamic strategy identifier is compressed into a binary decision tree structure using Huffman coding.

[0090] The core parameters of the alarm level are separated to generate an independent alarm level identifier, including a three-dimensional vector: a risk level code defines high, medium and low risk levels, a response time limit range is set for execution delay, and a resource priority weight marks the network bandwidth and storage resource preemption coefficient. The standard alarm data packet forms a four-layer packaging structure: the protocol header encapsulates the version number and the checksum field, the metadata layer stores the resource and space coding, the payload layer carries the compressed action instruction chain, and the verification layer adds the digital signature and the timestamp hash value. The conversion process implements real-time compatibility verification, detects non-standard protocol fields, and enables historical template interpolation filling, ensuring the compatibility of heterogeneous security platforms.

[0091] Multi-level queue allocation strategy is executed according to the alarm level identifier. The risk level code and the response time level cooperate to trigger the queue classification mechanism: when the high risk level code L1 level and the response time T0 level are triggered, the real-time preemption queue is allocated, and the dedicated transmission channel is enabled; when the medium risk L2 level and the response time T1 level are triggered, the high priority queue is allocated, and 80% of the shared bandwidth resources are guaranteed; when the low risk L3 level and the response time T2 level are triggered, the regular queue is allocated, and the remaining bandwidth resources are used. When the resource priority weight exceeds the threshold, the queue level is automatically promoted, and the queue expansion mechanism is triggered by continuous data packets of the same level.

[0092] The alarm queue to be transmitted is constructed into a composite control structure: the real-time preemption queue adopts a ring buffer structure, the high priority queue is constructed into a double linked list structure, and the regular queue adopts a first-in-first-out structure. The queue output includes queue type identifier, data packet index sequence, bandwidth and storage resource allocation ratio, invalid countdown parameter and expansion state flag. The allocation process performs load balancing monitoring, and when the queue delay exceeds the limit, the data packet is migrated to the low load queue to block the transmission congestion risk.

[0093] The input transmission path detection unit of the alarm queue to be transmitted performs network topology decision analysis. The real-time state parameters of the available communication link are scanned, including wireless channel delay, data packet loss rate and bandwidth occupancy rate. Based on the resource priority weight calculation of the alarm level identifier, the path transmission time cost and energy cost are calculated, a weighted decision matrix is constructed to implement dynamic path selection.

[0094] The output optimal transmission path is packaged into a four-tuple structure: the main path protocol identifier transmits the channel type, the standby path sequence records the emergency communication link, the estimated delay parameter marks the end-to-end transmission time, and the energy consumption indicator quantifies the unit data packet transmission power consumption.

[0095] The redundant backup node selection follows the geographical dispersion principle, and the backup node level is dynamically configured according to the data packet invalid countdown parameter: the core backup node is deployed in the city edge computing center, the secondary backup node is located in the cross-regional cloud service center, and the emergency backup node is integrated in the local gateway storage unit. The link quality pre-check is implemented in the node selection process, and when the signal strength is lower than the threshold, the standby node level is automatically switched.

[0096] The standard alarm data packet is distributed to the cloud through the optimal transmission path. The hierarchical storage strategy is implemented: the protocol header and metadata layer are stored in the low-delay solid-state array, and the payload layer is stored in the cold-hot mixed storage pool after being processed by the encryption algorithm.

[0097] The generation storage location three-dimensional coordinate identification data center physical topology is encapsulated as an intelligent pointer structure: logical address mapping virtual storage location, physical coordinate record data center longitude and latitude, access key binding equipment biological characteristics, and storage hierarchical label hot storage area level. The distribution process integrates a fault-tolerant control mechanism, switches to an alternative path when transmission is interrupted, and triggers local cache retransmission when data block verification fails. The cloud storage location is additionally provided with a self-destruction protection protocol, and data block erasing instructions are activated when unauthorized access attempts exceed the limit.

[0098] The redundant backup node and the cloud storage location perform bidirectional synchronization verification. A lightweight Merkle tree is constructed to compare core data hash values, timestamp sequence verification is used to check the consistency of the last write version, and random block sampling is used to implement bit-level data integrity verification.

[0099] The output backup consistency result is defined as a five-state classification code: a completely consistent state record verification pass, metadata offset marks header information difference, payload data loss identifies content missing, version conflict marks timeline misalignment, and node failure indicates communication interruption.

[0100] The fault recovery strategy implements a hierarchical reconstruction mechanism: local cache reconstruction is enabled when there is metadata offset, data is recovered from an emergency node when there is payload loss, the latest timestamp is used to cover the strategy when there is version conflict, and a secondary node switching operation is triggered when there is node failure. The recovery process generates a fault feature vector and updates the historical fault library of the transmission path to optimize the decision weight of subsequent path detection.

[0101] The present embodiment realizes lossless adaptation of alarm execution information and international security communication standards through the hierarchical encapsulation mechanism of protocol format conversion, solves the compatibility problem of heterogeneous systems, and significantly improves the reliability of cross-platform data interaction. Based on the dynamic queue allocation strategy of alarm level identification, real-time preemption / high priority / regular three-level transmission channels are automatically configured according to the risk level and time limit, breaking through the resource contention bottleneck of traditional single queue and ensuring zero-delay transmission of critical alarms. Through the weighted decision model of transmission path detection, the optimal path is dynamically selected by combining link state and energy cost, effectively avoiding transmission interruption caused by network congestion and improving the success rate of cloud distribution. A geographical dispersion deployment architecture of redundant backup nodes is constructed, combined with a five-state fault diagnosis mechanism of data synchronization verification, to realize second-level switching and data reconstruction when the node fails, and to ensure the high availability of alarm information storage.

[0102] Referring to Figure 2 The application also provides a door lock opening intention recognition system based on behavior analysis, which is applied to the door lock opening intention recognition method based on behavior analysis. The acquisition module is used to acquire user behavior information, hand movement information and door lock handle pressure information collected by the door lock, and perform identification calculation to obtain a behavior intention level. The analysis module is used for operating intention analysis on the door lock handle pressure information and the hand action information according to the behavior intention level, and obtaining abnormal behavior information; The correlation module is used for performing hardware state grading switching on the door lock according to the abnormal behavior information, and obtaining grading control instructions; The processing module is used for performing alarm protocol matching based on the grading control instructions and the user behavior information, and obtaining alarm execution information.

[0103] The application provides a door lock opening intention recognition system based on behavior analysis, which breaks through the limitation of single dimension judgment, significantly reduces the intention misjudgment rate, and improves the accuracy of the opening intention recognition by ternary collaborative analysis on user behavior information, hand action information and door lock handle pressure information, establishes a behavior intention level, and directly converts a safety threat determination result into a physical layer control instruction to realize millisecond level risk blocking response and solve the safety protection hysteresis problem. The application links user behavior data based on the grading control instructions, establishes a multi-dimensional matching mechanism of the alarm protocol, automatically adapts to the field risk level through real-time mapping of abnormal behavior characteristics and hardware response state, and improves the resource utilization efficiency of the security system.

[0104] It should be noted that, for the convenience and brevity of description, the specific working processes of the system and each module described above can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0105] The above only describes the preferred embodiments of the application, and does not limit the patent scope of the application, and any equivalent structure or equivalent process transformation based on the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the application.

Claims

1. A method for recognizing door lock unlocking intent based on behavior analysis, characterized in that, include: The system acquires user behavior information, hand gesture information, and door lock handle pressure information collected by the door lock, performs identification calculations, and obtains the behavior intent level. Based on the behavioral intent level, the operation intent is analyzed by the door lock handle pressure information and the hand movement information to obtain abnormal behavior information; Based on the abnormal behavior information, the door lock is subjected to hardware state hierarchical switching to obtain hierarchical control commands; Alarm protocol matching is performed based on the hierarchical control instructions and the user behavior information to obtain alarm execution information.

2. The door lock unlocking intent recognition method based on behavior analysis according to claim 1, characterized in that, The process of acquiring user behavior information, hand gesture information, and door lock handle pressure information collected by the door lock, and performing identification calculations to obtain the behavior intent level, includes: The user behavior information is obtained by extracting the behavior trajectory from the continuous image frame sequence captured by the horizontal view camera of the door lock; Hand feature points are detected in a continuous sequence of image frames captured by a downward-facing camera of the door lock to obtain the hand movement information; The door lock handle pressure information is obtained by identifying the handle pressure of the raw pressure data collected by the door lock's sensors. Based on the door lock handle pressure information, modal intent recognition is performed on the user behavior information and the hand movement information to obtain the behavior intent level.

3. The door lock unlocking intent recognition method based on behavior analysis according to claim 2, characterized in that, Based on the door lock handle pressure information, modal intent recognition is performed on the user behavior information and the hand gesture information to obtain the behavior intent level, including: The user behavior information is segmented into directional trajectories to obtain proximity direction information; The hand gesture information is processed for gesture recognition, and the gesture feature information is output. Gradient analysis is performed on the pressure information of the door lock handle to obtain pressure change information; Based on preset intent recognition rules, the operation intent is calculated using the approach direction information, the action feature information, and the pressure change information to obtain the behavior intent level.

4. The door lock unlocking intent recognition method based on behavior analysis according to claim 1, characterized in that, The step involves analyzing the operation intent of the door lock handle pressure information and the hand movement information based on the behavioral intent level to obtain abnormal behavior information, including: The behavioral intention levels are dynamically assigned to obtain stress analysis coefficients and action analysis coefficients; The pressure information of the door lock handle is segmented according to the pressure analysis coefficient to obtain segmented pressure information. Based on the motion analysis coefficients, the hand motion information is clustered into joint trajectories to obtain motion intent information; The segmented pressure information and the action intention information are fused using multimodal methods to obtain behavioral feature information; The abnormal behavior information is obtained by identifying abnormal intent based on the pre-set abnormal behavior feature database.

5. The door lock unlocking intent recognition method based on behavior analysis according to claim 4, characterized in that, The step of segmenting the door lock handle pressure information according to the pressure analysis coefficient to obtain segmented pressure information includes: The pressure information of the door lock handle is decomposed into frequency components to obtain high-frequency pressure components and low-frequency pressure components. Based on the pressure analysis coefficients, abrupt change points are detected in the high-frequency pressure components to obtain a candidate pressure point set. The trend inflection points of the low-frequency pressure components are identified to obtain a set of trend pressure points. Conflict resolution is performed on the candidate pressure point set and the trend pressure point set to obtain the effective pressure point sequence; The pressure information of the door lock handle is segmented according to the effective pressure point sequence to obtain the segmented pressure information.

6. The door lock unlocking intent recognition method based on behavior analysis according to claim 1, characterized in that, The step of performing hardware state hierarchical switching on the door lock based on the abnormal behavior information to obtain hierarchical control instructions includes: The confidence level of the abnormal behavior information is determined based on a preset abnormal behavior dataset to obtain the determination result; When the determination result is a high confidence determination result, the actuator of the door lock is activated in a preparatory state according to the high confidence determination result, and a drive preparatory command is obtained; Based on the drive preparation command, the motor drive module of the door lock is put into standby control to obtain a motor standby signal; The drive preparation command and the motor standby signal are controlled in a door lock hardware coordination manner to obtain the hierarchical control command.

7. The door lock unlocking intent recognition method based on behavior analysis according to claim 6, characterized in that, Also includes: When the determination result is a low confidence determination result, the main control unit of the door lock is switched to sleep mode according to the low confidence determination result to obtain the peripheral sleep instruction set; The sampling frequency of the door lock is adjusted based on the peripheral sleep instruction set to obtain an energy-saving monitoring signal; Based on the energy-saving monitoring signal, the motor drive module is torque-locked to obtain a lock status signal; By combining the peripheral sleep instruction set with the lock status signal, the door lock hardware sleep control is performed to obtain the hierarchical control instruction.

8. The door lock unlocking intent recognition method based on behavior analysis according to claim 1, characterized in that, The alarm protocol matching based on the hierarchical control instructions and the user behavior information to obtain alarm execution information includes: The door lock is hardware status decoded according to the hierarchical control command to obtain the door lock motor status identifier and security level code; The user behavior information is segmented based on the door lock motor status identifier to obtain a set of behavior fragments. The security level encoding is indexed by a preset alarm protocol library to obtain the target protocol type; The alarm execution information is obtained by matching the set of behavioral fragments with the target protocol type.

9. A door lock unlocking intent recognition system based on behavior analysis, characterized in that, The door lock unlocking intent recognition method based on behavior analysis, applied to any one of claims 1-8, includes: The data acquisition module is used to acquire user behavior information, hand gesture information, and door lock handle pressure information collected by the door lock, and to perform identification calculations to obtain the behavior intent level. The analysis module is used to analyze the operation intent of the door lock handle pressure information and the hand movement information according to the behavior intent level, and obtain abnormal behavior information. The association module is used to perform hardware state hierarchical switching of the door lock based on the abnormal behavior information to obtain hierarchical control instructions. The processing module is used to perform alarm protocol matching based on the hierarchical control instructions and the user behavior information to obtain alarm execution information.

Images