User identity authentication method in card-free power taking process

Abstract

The invention relates to the technical field of identity recognition, in particular to a user identity authentication method in a card-free power taking process, which comprises the following steps of: S1, capturing time-space trajectory data of a user stepping in in real time through cooperative detection of a door magnetic sensor and a millimeter wave radar, and triggering a non-inductive identity authentication instruction when detecting that a preset check-in behavior mode is met; s2, executing multi-modal biological characteristic dynamic acquisition; and S3, inputting the multi-modal biological characteristics into cascade verification, when the comprehensive confidence exceeds a dynamic threshold, sending a wireless power taking instruction to a room power distribution system, and synchronously activating a progressive starting program of the lighting equipment. Compared with a traditional single-mode verification method, the overall identity recognition accuracy can be greatly improved, and the adaptability to critical illumination and high-noise environments is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of identity recognition, and in particular to a user identity authentication method in a cardless power taking process. BACKGROUND

[0002] With the popularity of smart hotels, smart apartments and other scenarios, traditional physical door cards or face recognition authentication power taking methods gradually expose many problems. On the one hand, physical door cards have security risks such as easy loss, easy copying, cross-use, etc. On the other hand, identity authentication methods based on single visual or voiceprint features are limited by complex environmental factors (such as light changes, noise interference, multiple users in parallel, etc.), and have great limitations in accuracy, response speed and user experience.

[0003] Existing cardless power taking technologies mostly rely on single modal identity verification, such as face recognition, single-point fingerprint recognition or voiceprint verification. Such methods are often affected by environmental light changes, background noise, occluded objects or insufficient sensor accuracy in actual application, resulting in increased verification failure rate. In addition, some schemes use fixed confidence threshold judgment, which does not have adaptive ability to dynamic changes in the environment and device energy consumption state, and is prone to misjudgment or omission in low power or extreme environmental conditions, affecting system safety and stability.

[0004] In terms of power supply control, existing technologies generally use fixed instruction issuance and single switch control, lack a mechanism for deep linkage with identity verification results, and do not fully consider lighting comfort and energy saving optimization needs. When the user enters the room, the traditional system often directly lights up all the lighting areas, which can easily cause energy waste, and the sudden change in brightness can also easily cause visual discomfort, affecting user experience. SUMMARY

[0005] The present application provides a user identity authentication method in a cardless power taking process, which integrates multi-modal biological features, has dynamic environmental adaptability, and realizes intelligent power supply and progressive lighting collaborative control optimization after verification, to improve the overall safety, reliability and user experience of the system.

[0006] A user identity authentication method in a cardless power taking process, comprising the following steps:

[0007] S1: Through the cooperative detection of a door magnetic sensor and a millimeter wave radar, real-time capture the spatiotemporal trajectory data of the user entering, and when detecting a preset check-in behavior mode, trigger a non-inductive identity verification instruction;

[0008] S2: Perform multi-modal biological feature dynamic acquisition, including:

[0009] Obtain three-dimensional gait features based on a wide-angle TOF camera at the top of the porch;

[0010] Acquire specific orientation voiceprint features through embedded microphone array;

[0011] Extract palm vein topology code using door handle capacitor sensor;

[0012] S3: Input multi-modal biological features for cascade verification, when the comprehensive confidence exceeds the dynamic threshold, send wireless power taking instruction to the room power distribution system, and synchronously activate the progressive start program of the lighting device.

[0013] Optionally, the S1 specifically comprises:

[0014] S11, spatiotemporal data fusion: spatiotemporally align the door leaf opening angle sequence collected by the door magnetic sensor with the human body point cloud trajectory obtained by the millimeter wave radar, and establish a three-dimensional coordinate system with the threshold as the reference;

[0015] S12, behavior pattern analysis: when the preset check-in behavior pattern is met simultaneously, determine it as an effective check-in behavior:

[0016] S13, verification trigger logic: calculate the motion feature matching degree M(t) through a sliding time window, and generate a non-contact identity verification instruction when the motion feature matching degree M(t) of the last 3 windows is greater than 85%.

[0017] Optionally, the preset check-in behavior pattern comprises:

[0018] The door leaf opening angle reaches the expected duration within the preset interval;

[0019] The millimeter wave radar detects that the human body centroid moves along the central axis of the doorway and maintains a speed;

[0020] The Hausdorff distance between the foot trajectory point set and the preset standard check-in path trajectory is less than the preset distance.

[0021] Optionally, the calculation of the motion feature matching degree M(t) comprises: in each sliding window, extracting a trajectory segment within the current time window from the detected foot trajectory point set, performing shape comparison between the trajectory segment and the preset standard check-in path trajectory, quantifying the trajectory deviation degree by calculating the maximum and minimum distance, i.e. the Hausdorff distance, between the two, normalizing the actual Hausdorff distance obtained by calculation and the set maximum allowed distance to generate a matching degree score, reflecting the consistency degree of the user gait trajectory within the current window and the standard path.

[0022] Optionally, in the motion feature matching degree, when the trajectories completely coincide, the matching degree is 100%, when the trajectories deviate to the maximum tolerance limit, the matching degree decreases to 0%, and if the trajectory deviation exceeds the maximum limit, the matching degree is directly determined as 0%.

[0023] Optionally, the S2 specifically comprises:

[0024] S21, three-dimensional gait feature acquisition: through two groups of wide-angle TOF cameras symmetrically arranged on the top of the door, three-dimensional point cloud data of foot movement is synchronously captured, environmental light interference is eliminated by using a noise reduction algorithm, a step vector, a stride cycle vector and an ankle swing angle vector are extracted by performing skeleton node tracking;

[0025] S22, directional voiceprint collection: a six-element microphone array is arranged on both sides of the door frame, the user's head direction is locked through beamforming technology, and the voiceprint features of a preset frequency band are collected, and the environmental echo is eliminated by combining a dynamic time warping algorithm;

[0026] S23, palm vein feature extraction: based on the capacitive sensing matrix built-in the door handle, the capacitive distribution map of the subcutaneous blood vessels of the palm is obtained when the user holds it; the skin capacitance noise is separated by using a frequency domain decomposition method to generate a palm vein topology structure code.

[0027] Optionally, in the S21,

[0028] For a continuous foot movement trajectory, the position of each foot landing event is detected, the spatial coordinates of the current landing point and the spatial coordinates of the next landing point are recorded, and based on the position change of the two landing points, the corresponding step vector is calculated;

[0029] The time interval between the same foot landing twice in succession is measured, the time interval is combined with the corresponding step vector to generate a stride cycle vector, and the coordination of the walking rhythm and the stride is reflected;

[0030] The instantaneous speed vector of the ankle joint in the movement process is extracted, the reference vector in the walking direction is taken as the reference, the included angle change between the ankle joint speed vector and the reference vector is calculated, and the ankle swing angle vector is obtained.

[0031] Optionally, the S2 further comprises a space-time synchronization mechanism, and a timestamp is added to the multi-modal biological feature data.

[0032] Optionally, the S3 specifically comprises:

[0033] S31, multi-modal feature fusion verification: the three-dimensional gait feature vector, the voiceprint feature and the palm vein topology code are input into cascade verification, decision fusion is performed in three stages, and the comprehensive confidence C is output:

[0034] Stage one: the gait features are screened through a lightweight residual network, and the initial user identity confidence P1 is output;

[0035] Phase two: adopt bidirectional LSTM network to model the time sequence of voiceprint features, dynamically adjust the voiceprint weight by combining P1 value, and output enhanced confidence P2;

[0036] Phase three: analyze the palm vein topological structure through the graph convolution network, perform decision-level fusion, and output the comprehensive confidence C;

[0037] S32, dynamic threshold calculation: based on real-time environmental parameters to calculate the dynamic identity verification threshold;

[0038] S33, intelligent power supply control: when the output comprehensive confidence exceeds the dynamic identity verification threshold, execute wireless power taking instruction and progressive lighting activation;

[0039] S34, abnormal blocking mechanism:

[0040] When the comprehensive confidence is lower than the dynamic identity verification threshold for two consecutive times, trigger the security protection protocol, including freezing the biological feature database access permission, temporarily locking and activating the hardware isolation relay in the power distribution circuit.

[0041] Optionally, the comprehensive confidence C is calculated as:

[0042] C=βP2+(1-β)V, wherein β is a fusion weight factor, P2 is the identity enhanced confidence after gait+voiceprint fusion, and V represents the palm vein feature matching degree.

[0043] The beneficial effects of the present application are:

[0044] The present application realizes accurate discrimination of the step-in behavior based on the cooperative detection of door magnetic sensors and millimeter wave radars, synchronously collects the three-dimensional gait features, directional voiceprint features and palm vein topological codes of the user, and sequentially completes gait screening, voiceprint dynamic weighting and palm vein graph structure analysis through a three-level fusion mechanism of residual network, bidirectional LSTM and graph convolution network, and outputs the comprehensive confidence which can dynamically adapt to environmental changes. Compared with the traditional single-mode verification method, the overall identity recognition accuracy of the present application can be greatly improved, and the adaptability to critical light and high noise environment is improved.

[0045] The present application dynamically adjusts the identity verification confidence threshold according to the real-time light change rate, environmental noise level and device power state, effectively avoids the misjudgment or omission problem caused by the environmental drastic fluctuation or low device power, adopts the dynamic threshold calculation based on environmental factors, shortens the verification response time while keeping the power taking safety and reliability, improves the user's non-sensing power taking experience, and especially in high dynamic environment, compared with the fixed threshold scheme, the verification error rejection rate is reduced.

[0046] After the invention is verified, the power taking instruction encrypted by the dynamic key generated based on the Lorenz chaotic system is adopted, and the ZigBee-Mesh network transmission is combined, so that the wireless power supply link is safe and reliable, meanwhile, the lighting system adopts a gradual brightness adjustment strategy, follows the standard human eye adaptation curve, avoids the discomfort caused by instantaneous strong light stimulation, dynamically tracks the user movement track through the UWB positioning, adjusts the lighting hotspot area in real time, and ensures that the continuous illuminance within 1.5 meters around is greater than 300 lux. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and other drawings can also be obtained by those skilled in the art without creative effort.

[0048] Fig. 1 The method flowchart of the embodiment of the present application is shown in the figure.

[0049] Fig. 2 The cascade verification schematic diagram of the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0050] The present application will be described in detail below in combination with the drawings and specific embodiments. It should be noted that in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement some known technologies; and the drawings are only used to describe the embodiments more specifically, and are not intended to specifically limit the present application.

[0051] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiments can include specific features, structures or characteristics, but not necessarily every embodiment includes the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in combination with an embodiment, it should be within the knowledge of those skilled in the related art to realize this feature, structure or characteristic in combination with other embodiments (whether or not explicitly described).

[0052] Generally, the terms can be understood at least partly from the use in the context. For example, depending at least partly on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in the singular sense or can be used to describe a combination of features, structures or characteristics in the plural sense. In addition, the term "based on" can be understood as not necessarily conveying a set of exclusive factors, but can instead allow the presence of other factors not necessarily explicitly described, at least partly depending on the context.

[0053] As Figs. 1-2 shown, a user identity authentication method in a cardless power taking process, comprising the following steps:

[0054] S1: Through the cooperative detection of door magnetic sensor and millimeter wave radar, real-time capture the spatiotemporal trajectory data of user stepping in, when detecting the preset entry behavior mode, trigger the non-inductive identity verification instruction;

[0055] S2: Perform multi-modal biometric dynamic acquisition, including:

[0056] Based on the wide-angle TOF camera on the top of the porch to obtain three-dimensional gait features;

[0057] Through the embedded microphone array to collect specific orientation voiceprint features;

[0058] Use the door handle capacitor sensor to extract the palm vein topology code;

[0059] S3: Input the multi-modal biometric features into the cascade verification, when the comprehensive confidence exceeds the dynamic threshold, send the wireless power taking instruction to the room power distribution system, and synchronously activate the progressive start program of the lighting device.

[0060] S1 specifically includes:

[0061] S11, spatiotemporal data fusion: align the door magnetic sensor collected door leaf opening angle sequence θ(t) and the human body point cloud trajectory obtained by the millimeter wave radar in space-time, establish a three-dimensional coordinate system C with the threshold as the reference; wherein θ(t) is the door leaf opening angle of the door magnetic sensor at time t, represents the human body point cloud set obtained by the millimeter wave radar at time t, C represents the three-dimensional rectangular coordinate system established with the threshold center as the origin.

[0062] S12, behavior pattern analysis: when the following conditions are met at the same time, it is determined as valid entry behavior:

[0063] S121, the door leaf opening angle is within the set interval for a duration of 60°≤θ(t)≤90° and the duration is ≥0.5 seconds;

[0064] S122, the human body centroid moves along the door porch central axis direction, and the speed satisfies: v center (t)≤1.2m / s;

[0065] S123, the foot trajectory point set and the preset entry path Hausdorff distance satisfies:

[0066] Wherein, v center(t) represents the instantaneous velocity of the human body's center of mass along the direction of the doorway's central axis, is the current detected foot trajectory point set, is the preset standard entry path trajectory, is the set is the Hausdorff distance between .

[0067] S13, verification trigger logic: adopt sliding time window mechanism (window length 2 seconds, step length 0.1 seconds), calculate motion feature matching degree M(t) in each sliding window, when the following conditions are met for three consecutive sliding windows:

[0068] M(t)>85%, generate non-contact identity verification instruction, wherein M(t) represents the motion feature matching degree calculated in the current time window. The calculation method of motion feature matching degree M(t) in each sliding window (length 2 seconds, step length 0.1 seconds) is as follows:

[0069] 1. Select the trajectory subset in the current time window from the foot trajectory point set

[0070] 2. Calculate the Hausdorff distance d between the standard entry path : H :

[0071] 3. Map the actual Hausdorff distance to the matching degree score M(t), according to the following proportion:

[0072]

[0073] wherein, represents the Hausdorff distance between the detected trajectory in the current window and the standard trajectory, d max represents the maximum allowed Hausdorff distance (set to 15 cm), M(t) is the motion feature matching degree (percentage), when d H =0, M(t)=100% (completely consistent), when d H =d max , M(t)=0% (just over the boundary), if d H >d max , forcibly set M(t)=0%.

[0074] The foot trajectory point set is obtained as follows:

[0075] From the human point cloud set output by the millimeter wave radar in real time In the ,points in the low height area close to the ground (the height threshold is set to h ≤ 0.25m);

[0076] Perform density threshold-based clustering on the extracted low-height point set, cluster the locally densely distributed points into the foot region, and extract the center of gravity point f(t) of the foot region at each time t, and continuously record it on the time axis to form the foot trajectory point set: Among them, f(t k ) indicates that at time t k The detected centroid coordinates of the foot area, where n is the number of samples in the current time window.

[0077] Standard occupancy path trajectory The presets include:

[0078] With the centerline of the porch as a reference, define an idealized straight line path in the three-dimensional coordinate system C. The standard path can be obtained by averaging historical guest data, or by initial engineering design, as a trajectory from a certain position outside the door (0.5 meters outside the door) along the central axis straight into the room. The standard path trajectory point set is expressed as: Among them, r m Represents the position coordinates of the mth discrete sampling point on the standard trajectory (the standard trajectory is evenly sampled, with one point every 10-20 cm).

[0079] S2 specifically includes:

[0080] S21, 3D gait feature acquisition:

[0081] S211, foot 3D point cloud collection: The two sets of TOF cameras on the top scan synchronously to collect the foot 3D point cloud set at each time t Among them, (x i (t),y i (t),z i (t)) is the three-dimensional coordinate of the i-th point cloud point, N t is the number of point clouds detected at time t;

[0082] S212, adaptive noise reduction processing: for ambient light interference, Point cloud intensity value I i (t) Perform threshold filtering Among them, I i (t) is the light intensity echo value at the i-th point, I thresh is the set light intensity noise threshold, light intensity noise threshold I thresh According to the echo intensity distribution of the TOF camera when there is no user and only ambient background light, the mean background light intensity μ over a period of time is calculated. bg and standard deviation σ bg, the threshold is set to: I thresh =μ bg +2σ bg , that is, leaving a certain margin based on the environmental background fluctuation range and filtering out more than 99% of the background noise points.

[0083] S213, skeleton node tracking extracts gait features, performs skeleton node tracking on the point cloud sequence, and extracts the step vector S step (t), stride period vector T stride (t), ankle joint swing angle vector A ankle (t);

[0084] Step vector S step (t) is the spatial displacement vector of two consecutive landing points, defined as:

[0085] S step (t) = p foot (t next )-p foot (t), where p foot (t) is the current foot center of mass position, t next For the next landing moment;

[0086] Stride period vector T stride (t) represents the time interval between two consecutive landing events of the same foot, and is combined with the step length to form the stride rhythm vector: T stride (t)=(t next -t)×S step (t);

[0087] Ankle joint swing angle vector A ankle (t) represents the swing angle of the ankle joint motion calculated based on the tangent vector of the foot trajectory:

[0088]

[0089] Among them, v ankle (t) is the instantaneous velocity vector of the ankle joint, v ref is the walking direction reference vector.

[0090] S22, directional voiceprint collection:

[0091] S221, head azimuth locking: a six-element microphone array is deployed on both sides of the door frame. The six-element microphone array outputs the user's head direction angle θ based on beamforming. head (t), determined by the direction of maximum sound energy:

[0092] Among them, w i (θ) is the weighting coefficient of the i-th microphone for direction θ, S i(t) is the signal received by the i-th microphone.

[0093] S222, Voiceprint feature collection and processing: Extract the energy feature vector F in the frequency band 1.2-4kHz voice (t), apply the dynamic time warping (DTW) algorithm to calculate the similarity between the current voiceprint sequence and the template voiceprint:

[0094] D DTW (S input ,S template ), where S input is the current user’s voiceprint sequence, S template To pre-register the guest’s voiceprint sequence, D DTW Represents the cumulative shortest distance of dynamic time warping.

[0095] S23, palm vein feature extraction: The door handle's built-in 128×96 capacitance sensor matrix records the capacitance distribution map E(x, y, t) when the user grips the door handle. E is the capacitance per unit area of ​​each pixel. Fast Fourier transform (FFT) is used to convert the capacitance signal into the frequency domain to separate the deep vein features:

[0096] E f (u,v)=FFT(E(x,y,t)), select low-frequency components to reconstruct the palm vein map V palm , to suppress high-frequency skin noise, where E(x,y,t) represents the capacitance value collected at coordinate (x,y) and time t. Represents the processed palm vein topology coding.

[0097] S3 specifically includes:

[0098] S31, multimodal feature fusion verification, cascade verification is performed by inputting 3D gait feature vectors, voiceprint features, and palm vein topology encoding, and decision fusion is performed in three stages:

[0099] Stage 1: Gait screening (lightweight residual network):

[0100] 1. The extracted three-dimensional gait feature vector group: Input lightweight residual network for classification;

[0101] 2. After multiple layers of residual block calculations, Sigmoid normalization is finally performed at the output layer to obtain the preliminary identity confidence: in, Represents the Sigmoid activation function, W gait is the residual network output layer weight matrix, b gait is the output layer bias.

[0102] Phase two: voiceprint timing modeling (bidirectional LSTM):

[0103] 1. The voiceprint feature F voice (t) is input into the bidirectional LSTM network to obtain the voiceprint matching degree S, and the hidden state of the LSTM output at time t is h t , and the final voiceprint matching degree is: S = CosineSimilarity(h t , h template ), wherein h t is the LSTM encoding feature of the current voiceprint timing, h template is the encoding feature of the pre-registered voiceprint template, and CosineSimilarity represents the calculation of the cosine similarity of the included angle of two vectors, with a range of [0, 1].

[0104] 2. Dynamically adjust the voiceprint fusion weight to calculate the enhanced confidence P2:

[0105] Voiceprint credibility coefficient α setting rule:

[0106] When P1>0.7, α=0.3;

[0107] When P1≤0.7, α linearly increases, and the maximum does not exceed 0.6;

[0108] The enhanced confidence calculation formula is: P2=P1+α(1-P1)S.

[0109] Phase three: palm vein topology fusion (graph convolution network):

[0110] 1. The palm vein topology encoding is modeled as a vein node graph: G=(V, E), wherein V is a node set, each node represents a feature point on the palm vein graph, and E is an edge set, representing the connection relationship of the vein structure;

[0111] 2. Input the graph convolution network to obtain the vein matching degree V: Wherein W vein is the graph convolution output weight matrix, and b vein is the output bias;

[0112] 3. The final comprehensive confidence C is calculated as:

[0113] C=βP2+(1-β)V, wherein β is dynamically adjusted according to the voiceprint signal quality and environmental parameters, and is set in the range of 0.6-0.8.

[0114] S32, dynamic identity verification threshold calculation:

[0115] S321, set the basic security threshold: T base =0.92;

[0116] S322, calculate the environmental interference factor γ:

[0117] According to the illumination change rate ΔL and noise intensity N, set:

[0118] When ΔL>50lux / s and N>65dB, γ=0.15;

[0119] Under normal conditions, γ decreases proportionally, with a minimum of 0.05;

[0120] S323, calculate the power compensation coefficient ΔE, and for every 10% decrease in power, ΔE increases by 0.02;

[0121] S324, final dynamic identity authentication threshold T is calculated: T = T base ×(1+γΔE).

[0122] S33, intelligent power supply control, if the comprehensive confidence C satisfies: C>T, then the intelligent power supply is triggered;

[0123] S331, generates wireless power-off instructions and uses the Lorenz system to generate chaotic keys:

[0124]

[0125] After the initial value is set, the integral generates a 1024-bit random sequence, which is used as a dynamic key.

[0126] S332, start the progressive lighting program and control the brightness according to the following process:

[0127] Light up the background light of the user's zone to 20% of the target brightness within 50ms;

[0128] After a delay of 100ms, the adaptive dimming algorithm is activated, causing the light brightness to rise linearly to the set value within 0.5 seconds;

[0129] The lighting area is dynamically adjusted based on UWB positioning data to ensure that the illumination within 1.5 meters around the user is greater than or equal to 300 lux.

[0130] S34, abnormal blocking mechanism: When the comprehensive confidence is lower than the dynamic threshold twice in a row, that is: C <T(连续两次检测),则触发安全防护协议,具体措施包括:

[0131] Freeze access to biometric databases;

[0132] Send a temporary locking command to the door locking system;

[0133] Activate the hardware isolation relay in the power distribution circuit to physically block the power control link.

[0134] The present application encompasses any alternatives, modifications, equivalent methods and solutions made to the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0135] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for user identity authentication in a cardless electricity withdrawal process, characterized in that: The following steps are involved: S1: Through the coordinated detection of the door magnetic sensor and millimeter-wave radar, the user's temporal and spatial trajectory data is captured in real time. When the preset check-in behavior pattern is detected, the non-sensing identity verification command is triggered; S2: Perform dynamic multimodal biometric collection, including: Acquire three-dimensional gait features based on a wide-angle TOF camera on the top of the porch; Collect specific directional voiceprint features through the embedded microphone array; Extracting palm vein topology codes using door handle capacitance sensors; S3: Multimodal biometric input is cascaded for verification. When the comprehensive confidence exceeds the dynamic threshold, a wireless power-off instruction is sent to the room power distribution system to synchronously activate the progressive startup procedure of the lighting equipment.

2. The method for user identity authentication in a cardless electricity withdrawal process according to claim 1, characterized in that: Said S1 specifically includes: S11, spatiotemporal data fusion: The door opening angle sequence collected by the door magnetic sensor is spatiotemporally aligned with the human body point cloud trajectory obtained by the millimeter wave radar to establish a three-dimensional coordinate system based on the door threshold; S12, behavior pattern analysis: When the preset check-in behavior patterns are met at the same time, it is determined to be a valid check-in behavior: S13, verification trigger logic: calculate the motion feature matching degree M(t) through the sliding time window, and generate a non-sensing identity authentication instruction when the motion feature matching degree M(t) of three consecutive windows is greater than 85%.

3. The method for user identity authentication in a cardless electricity withdrawal process according to claim 2, characterized in that: The preset check-in behavior modes include: The door opening angle remains within the preset range for the expected duration; The millimeter-wave radar detects that the center of mass of the human body moves along the central axis of the porch and maintains the speed; The Hausdorff distance between the foot trajectory point set and the preset standard check-in path trajectory is less than the preset distance.

4. The method for user identity authentication in a cardless electricity withdrawal process according to claim 3, characterized in that: The calculation of the motion feature matching degree M(t) includes extracting the trajectory segments within the current time window from the detected foot trajectory points in each sliding window, performing shape comparison between the trajectory segments and the preset standard entry path trajectory, and quantifying the degree of trajectory deviation by calculating the maximum and minimum distance between the two, namely the Hausdorff distance. The calculated actual Hausdorff distance is normalized with the set maximum allowable distance to generate a matching score that reflects the degree of consistency between the user's gait trajectory in the current window and the standard path.

5. The method for user identity authentication in a cardless electricity withdrawal process according to claim 4, characterized in that: In the motion feature matching degree, the matching degree is 100% when the trajectories completely overlap, and the matching degree drops to 0% when the trajectories deviate to the maximum tolerance limit; if the trajectory deviation exceeds the maximum limit, the matching degree is directly determined to be 0%.

6. The method for user identity authentication in a cardless electricity withdrawal process according to claim 1, characterized in that: The S2 specifically includes: S21, 3D gait feature acquisition: Two sets of wide-angle TOF cameras symmetrically arranged on the top of the porch synchronously capture 3D point cloud data of foot movement. A noise reduction algorithm is used to eliminate ambient light interference. Skeleton node tracking is performed to extract the step length vector, stride period vector, and ankle joint swing angle vector. S22, directional voiceprint collection: A six-element microphone array deployed on both sides of the door frame uses beamforming technology to lock the user's head position, collect voiceprint characteristics in a preset frequency band, and combine it with a dynamic time warping algorithm to eliminate environmental echoes; S23, palm vein feature extraction: Based on the capacitive sensing matrix built into the door handle, a capacitance distribution map of the subcutaneous blood vessels in the palm is obtained when the user holds the handle. The frequency domain decomposition method is used to separate the epidermal capacitance noise and generate the palm vein topology structure code.

7. The method for user identity authentication in a cardless electricity withdrawal process according to claim 6, characterized in that: In said S21: For the continuous foot motion trajectory, the position of each foot landing event is detected, the spatial coordinates of the current footfall point and the spatial coordinates of the next footfall point are recorded, and the corresponding step length vector is calculated based on the position change of the two footfall points; The time interval between two consecutive foot landings is measured and combined with the corresponding stride length vector to generate a stride period vector, which reflects the coordination of walking rhythm and stride length. The instantaneous velocity vector of the ankle joint during movement is extracted, and the angle change between the ankle joint velocity vector and the reference vector is calculated with the reference vector of the walking direction as the benchmark to obtain the ankle joint swing angle vector.

8. The method for user identity authentication in a cardless electricity withdrawal process according to claim 6, characterized in that: The S2 also includes a spatiotemporal synchronization mechanism, including adding a timestamp to the multimodal biometric data.

9. The method for user identity authentication in a cardless electricity withdrawal process according to claim 1, characterized in that: The S3 specifically includes: S31, multimodal feature fusion verification: The three-dimensional gait feature vector, voiceprint feature, and palm vein topology encoding are input into the cascade verification, and the decision fusion is performed in three stages to output the comprehensive confidence C: Stage 1: Filter gait features through a lightweight residual network and output preliminary user identity confidence P1; Phase 2: Use a bidirectional LSTM network to perform time series modeling on the voiceprint features, dynamically adjust the voiceprint weight based on the P1 value, and output the enhanced confidence P2; Phase 3: Analyze the palm vein topology through graph convolutional networks, perform decision-level fusion, and output the comprehensive confidence C; S32, dynamic threshold calculation: calculating a dynamic authentication threshold based on real-time environment parameters; S33, intelligent power supply control: When the output comprehensive confidence exceeds the dynamic authentication threshold, the wireless power supply command and progressive lighting activation are executed; S34, abnormal blocking mechanism: When the comprehensive confidence falls below the dynamic authentication threshold twice in a row, the security protection protocol is triggered, including freezing access to the biometric database, temporarily locking, and activating the hardware isolation relay in the power distribution circuit.

10. The method for user identity authentication in a cardless electricity withdrawal process according to claim 9, characterized in that: The comprehensive confidence C is calculated as: C = βP2 + (1-β)V, where β is the fusion weight factor, P2 is the identity enhancement confidence after gait + voiceprint fusion, and V represents the palm vein feature matching degree.

Images