Hotel Internet of Things control system based on intelligent hardware integration

Abstract

The invention discloses a hotel internet-of-things control system based on intelligent hardware integration, and belongs to the technical field of hotel intellectualization. Comprising a biometric authentication module, a spatial decision module, a behavior prediction module and an execution control module. The biological authentication module is used for dynamically binding a user identity through three-level biological characteristics which are pre-collected by a mobile terminal and are combined with field non-inductive authentication; the space decision module is used for constructing a real-time environment map based on a multi-sensor fusion SLAM technology and tracking a user moving track; the behavior prediction module is used for predicting a user behavior intention through a space-time sequence model according to the gait dynamics data and the moving track; the execution control module is used for automatically triggering a guest room equipment preparation instruction according to a prediction result to realize full automation of a check-in process; through mobile terminal pre-acquisition and on-site three-level biological characteristic dynamic verification, non-inductive accurate authentication of'walking, namely authentication 'is realized.

Description

Technical Field

[0001] This invention relates to the field of hotel intelligent technology, and more specifically, to a hotel IoT control system based on integrated intelligent hardware. Background Technology

[0002] With the development of IoT technology, the hotel industry is undergoing an intelligent transformation. However, existing systems have significant shortcomings:

[0003] First, identity verification is inefficient. Traditional hotel check-in requires front desk registration, which is not only inefficient, but also requires active cooperation from users using traditional room cards or single biometric identification (such as facial recognition), which takes a long time on average and requires a high degree of cooperation from both guests and staff. It cannot achieve seamless access and is prone to causing queues and congestion during peak hours. In addition, the facial recognition anti-counterfeiting mechanisms in some unmanned hotels are inadequate, affecting the security of hotel property and the personal safety of guests.

[0004] Second, existing behavior prediction systems have low accuracy in complex hotel environments, cannot start the equipment in advance enough, and lack collaborative decision-making mechanisms based on multimodal data such as biometrics, spatial trajectories, and behavioral intentions.

[0005] In addition, the lack of intelligent resource scheduling leads to energy waste. For example, air conditioning and lighting equipment often waste energy due to misjudgment. Actual measurements show that hotel rooms have serious daily waste of ineffective energy. Summary of the Invention

[0006] To address the aforementioned shortcomings, the present invention aims to provide a hotel IoT control system based on integrated smart hardware; to solve the problems of low authentication efficiency, weak environmental perception, and behavior prediction; and to pioneer a system that combines mobile pre-collection with on-site three-level biometric dynamic verification to achieve seamless and accurate authentication through "walking authentication"; to integrate UWB, vision, and point cloud semantic segmentation to construct a high-precision semantic map, and to combine trajectory entropy decision-making to trigger devices in advance, thereby reducing energy consumption; and to innovatively introduce a spatiotemporal graph convolutional network to improve the accuracy of behavior prediction and solve the failure problem of traditional solutions in complex scenarios.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The hotel IoT control system based on smart hardware integration includes a biometric authentication module, a spatial decision-making module, a behavior prediction module, and an execution control module.

[0009] The biometric authentication module is used to dynamically bind user identity through a three-level biometric feature combination of mobile pre-collection and on-site contactless authentication.

[0010] The spatial decision-making module is used to build a real-time environmental map based on multi-sensor fusion SLAM technology and track the user's movement trajectory;

[0011] The behavior prediction module is used to predict user behavioral intentions based on gait dynamics data and movement trajectory using a spatiotemporal sequence model.

[0012] The execution control module is used to automatically trigger guest room equipment preparation instructions based on the prediction results, so as to realize the full automation of the check-in process;

[0013] The three-level non-contact biometric feature includes a mobile pre-collection unit and an on-site verification unit. The mobile pre-collection unit is used to collect features in stages through the hotel APP.

[0014] The on-site verification unit is used to collect biometric features in real time during the user's seamless passage and perform dynamic cross-verification with the pre-collected features on the mobile terminal to complete dual authentication of identity binding and liveness detection.

[0015] Preferably, the mobile pre-collection unit includes:

[0016] Visible light face detection: uses the phone's front-facing camera for liveness detection; Ambient soundprint: records a 5-second voice command and extracts the room's reverberant fingerprint;

[0017] Gait dynamics: using the mobile phone's IMU sensor to collect the frequency domain characteristics of angular velocity or acceleration during walking.

[0018] Preferably, the field verification unit includes:

[0019] Visible light face acquisition subunit: An RGB-D camera array deployed at the hotel entrance calculates a 3D face model through multi-view geometric constraints and performs a non-contact 1:N fast comparison with the face features pre-stored on the mobile device;

[0020] Environmental voiceprint authentication subunit: It adopts a distributed microphone array and extracts user voiceprint features through blind source separation to achieve joint verification of voiceprint-spatial reverberation features in natural conversation.

[0021] Gait dynamics subunit: By fusing ground-embedded pressure sensors with millimeter-wave radar, gait cycle features are extracted, pre-stored gait templates are matched in real time, and gait frequency anomaly attacks are detected.

[0022] Preferably, the spatial decision-making module includes:

[0023] Real-time trajectory tracking unit: It integrates UWB positioning and visual odometry data, and uses Kalman filtering to optimize the user's position coordinates. Its state equation is expressed as:

[0024] x t =F t x t-1 +B t ut +w t w t ~N(0,Q) t )

[0025] In the formula: x t F is the state vector at time t, i.e., the user's position coordinates; t B is the state transition matrix; t To control the input matrix; u t w is the control vector. t The noise is process noise, with zero mean and covariance Q. t The distribution is Gaussian; N represents the Gaussian distribution, also known as the normal distribution; 0 is the mean vector of the distribution, i.e., the zero vector; Q t Let be the covariance matrix of the distribution;

[0026] State equation x t =F t x t-1 +B t u t +w t Indicates location prediction; noise term w t ~N(0,Q) t To quantify the uncertainty of the sensor, Kalman filtering dynamically adjusts to suppress noise and outputs optimized coordinates.

[0027] Environmental semantic mapping unit: used to identify key regions through point cloud segmentation. Key regions include elevators, corridors, and guest rooms. Its classification decision function is:

[0028] L i =arg max c∈C P(c∣φ(Voxel i ))

[0029] In the formula: L i This represents the semantic classification result of the i-th spatial unit; C = {elevator, corridor, guest room} is the set of region categories; Voxel i Let represent the i-th voxel unit; φ is the feature extraction function based on the PointNet++C={elevator, corridor, guest room} architecture; P(c|·) is the conditional probability of belonging to class c given voxel features;

[0030] Resource scheduling decision unit: Triggers device pre-start based on user trajectory entropy value; its entropy value calculation function is:

[0031] H t =-∑ r∈R P(r)log2P(r)

[0032] In the formula: H tThe Shannon entropy represents the user's movement trajectory; R = {elevator, corridor, guest room} is the set of reachable areas; P(r) represents the statistical probability of the user appearing in area r;

[0033] When H t When the threshold T is less than 90%, it indicates that the user is definitely heading towards the target room, and the path certainty is greater than 90%. At this time, the room equipment is pre-started. Based on the measured data of the hotel scenario, the recommended interval threshold T∈[0.5,1.2] is determined. This design is the first to apply the information entropy theory to the pre-scheduling decision of hotel equipment, so that the system can accurately start the equipment in advance when the user is 20-50 meters away from the room, which is a distance that traditional methods cannot determine. The measured data reduces energy waste.

[0034] Preferably, the behavior prediction module includes:

[0035] Gait and Intent Association Unit: Used to construct a mapping model between gait features and behavioral intentions;

[0036] Spatiotemporal graph prediction unit: Predicts moving targets using graph convolutional networks;

[0037] Real-time behavior prediction unit: used to generate pre-activation instructions for guest room equipment based on the dynamic parameters of the user's movement trajectory.

[0038] Preferably, the real-time behavior prediction unit is configured as follows:

[0039] (1) Real-time acquisition of the dynamic parameters of the user's movement trajectory, including:

[0040] The predicted arrival time t at the target room, in seconds;

[0041] Instantaneous movement speed v, in meters per second;

[0042] Real-time acceleration a, in meters per second²;

[0043] (2) Perform hierarchical decision-making based on dynamic parameters:

[0044] When the conditions t < 120 and a > 0.3 are met, an emergency rest mode command is generated. The emergency rest mode command includes: rapid cooling program of air conditioner, closing operation of electric curtains and start of bathtub water filling system.

[0045] When the conditions 120≤t<300 and v<0.5 are met, a leisure preparation mode instruction is generated. The leisure preparation mode instruction includes: starting the background music system and adjusting the indoor lighting to a preset low illumination mode.

[0046] When the conditions t≥300 and a<0.1 are met, a normal check-in mode instruction is generated, which includes: starting the basic lighting system and starting the fresh air system;

[0047] (3) The generated instruction pattern is pushed to the user's mobile terminal in real time through the hotel APP data interface, and the following information is displayed synchronously on the APP interface:

[0048] The currently active mode types include Emergency Rest Mode Instruction, Leisure Preparation Mode Instruction, and Regular Check-in Mode Instruction;

[0049] List of triggered device operations;

[0050] The countdown to the completion of guest room preparation is underway;

[0051] (4) The generated device commands are sent to the guest room execution terminal in real time via the MQTT protocol;

[0052] (5) When the user's movement parameters cross the decision threshold, the mode switch and instruction update are completed within 200ms;

[0053] This display allows users to understand the system status in real time and intervene manually. By realizing bidirectional visualization of predictive commands for the first time in a hotel IoT control system, it not only executes equipment operations but also provides real-time feedback on the system's decision-making logic through the APP, thereby improving user experience and system transparency.

[0054] Preferably, the three-level biometric authentication adopts a cascaded decision-making mechanism, and the authentication result is determined by the following formula:

[0055]

[0056] in:

[0057] Auth result This is the final output of Level 3 biometric authentication; s i The confidence score represents the i-th level biometric feature (i=1: face, i=2: voiceprint, i=3: gait); ∏ represents the product operator; α represents the multiplication operator. i Represents the dynamic decision threshold for the i-th level of authentication; I(·) is the indicator function:

[0058]

[0059] Hierarchical independent validation: Confidence scores s are calculated for each level of biometric feature (face / voiceprint / gait). i Dynamic threshold α i The system adjusts in real time according to the environment, such as lowering the face recognition threshold when there is insufficient light; the indicator function I outputs a binary result:

[0060] And through cascading decision-making: the final output is Auth only when all three levels of results are 1 (i.e., ∏ = 1). result=1, any failure at any level (resulting in 0) immediately causes the product to become zero, i.e., authentication fails.

[0061] Preferably, the gait and intention association unit is configured as follows:

[0062] The input is a gait feature vector containing the three-dimensional distribution of gait frequency, stride length, and plantar pressure;

[0063] Temporal features are extracted using an attention-enhanced bidirectional LSTM network, dynamically focusing on key frames for steering or speed changes;

[0064] Output the probability distribution of user behavior intent: [walking to the elevator, going to the guest room, staying in the public area];

[0065] The spatiotemporal graph prediction unit is configured as follows:

[0066] Construct a spatiotemporal graph: Node = real-time user coordinates (x, y, t), spatial edge = Euclidean distance between users + direction angle, time edge = dynamic sliding historical trajectory;

[0067] Computed via an adaptive spatiotemporal graph convolutional network:

[0068]

[0069] In the formula: H (l+1) It is the node feature matrix of the l-th layer; Θ (l) It is the trainable weight matrix of the l-th layer;

[0070] H (l) Θ (l) It is a matrix multiplication that multiplies the node feature matrix H. (l) With weight matrix Θ (l) Multiply them to obtain the transformed feature matrix; To enhance the adjacency matrix: This is the result of adding the identity matrix I to the original adjacency matrix A, i.e. Where: A is the adjacency matrix, representing the connection relationship between nodes in the graph, and I is the identity matrix, representing the connection between each node and itself; For the enhancement degree matrix: This is the enhancement adjacency matrix. The degree matrix; The degree matrix is ​​raised to the power of -1 / 2, that is, for Each diagonal element is raised to the power of -1 / 2.

[0071] Output the probability distribution of the moving target area: [elevator, corridor, guest room].

[0072] Compared with the prior art, the beneficial effects of the present invention are:

[0073] 1. This hotel IoT control system, based on integrated smart hardware, achieves seamless and efficient identity verification through a three-level biometric authentication dynamic circuit breaker mechanism, solving the problem of low efficiency in traditional authentication.

[0074] By combining mobile pre-collection (face / voiceprint / gait) with on-site contactless verification, the user passage speed is greatly improved. Compared with the traditional solution, the authentication time is reduced from 30 seconds to 1.2 seconds, which improves the security of face anti-counterfeiting; eliminates queuing congestion during peak hours, and solves the efficiency bottleneck of face anti-counterfeiting recognition at hotel entrances.

[0075] 2. By using semantic map fusion and trajectory entropy decision-making, precise equipment pre-scheduling is achieved, solving the problem of energy waste caused by weak environmental perception. By constructing high-precision maps of elevators, corridors and guest rooms, the equipment start-up distance is advanced to 20-50 meters, the average daily energy consumption of guest rooms is reduced by 37%, and the equipment false start rate is reduced from 40% to 1.2%. This overcomes the defects of traditional positioning technology errors and lack of semantic understanding, and solves the problem of energy waste caused by prediction errors.

[0076] 3. By using a cross-modal behavior prediction model, accurate intent recognition in complex scenarios is achieved, solving the problem of behavior prediction lag. The integration of three-dimensional plantar pressure distribution features and attention BiLSTM network improves the accuracy of turning intent recognition. Furthermore, a two-way visual control closed loop enables real-time user-system interaction, solving the problem of opaque command execution. Device commands are displayed synchronously through the APP, improving user experience satisfaction.

[0077] This invention constructs a fully intelligent closed loop of "perception-decision-execution-feedback", compressing the hotel check-in process from 30 minutes to 120 seconds of seamless access, reducing operating costs, and defining the next generation of intelligent hotel industry standards. Attached Figure Description

[0078] Figure 1 This is the overall flowchart of the hotel IoT control system based on integrated smart hardware;

[0079] Figure 2 This is an overall diagram of the spatial decision-making module of the hotel IoT control system based on integrated smart hardware.

[0080] Figure 3 This is an overall diagram of the field verification unit for the hotel IoT control system based on integrated smart hardware;

[0081] Figure 4 This is an overall diagram of the behavior prediction module of the hotel IoT control system based on integrated smart hardware. Detailed Implementation

[0082] Please see Figure 1 - Figure 4The hotel IoT control system based on integrated smart hardware includes a biometric authentication module, a spatial decision-making module, a behavior prediction module, and an execution control module.

[0083] The biometric authentication module is used to dynamically bind user identity through a three-level biometric feature combination of mobile pre-collection and on-site contactless authentication.

[0084] The spatial decision-making module is used to build a real-time environmental map based on multi-sensor fusion SLAM technology and track the user's movement trajectory;

[0085] The behavior prediction module is used to predict user behavioral intentions based on gait dynamics data and movement trajectory using a spatiotemporal sequence model.

[0086] The execution control module is used to automatically trigger guest room equipment preparation instructions based on the prediction results, so as to realize the full automation of the check-in process;

[0087] The three-level non-contact biometric feature includes a mobile pre-collection unit and an on-site verification unit. The mobile pre-collection unit is used to collect features in stages through the hotel APP.

[0088] The on-site verification unit is used to collect biometric features in real time during the user's seamless passage and perform dynamic cross-verification with the pre-collected features on the mobile terminal to complete dual authentication of identity binding and liveness detection.

[0089] In a further embodiment, the mobile terminal pre-collection unit includes:

[0090] Visible light face detection: Uses the phone's front-facing camera for liveness detection;

[0091] Ambient soundprint: Record a 5-second voice command and extract the room's reverberant fingerprint;

[0092] Gait dynamics: using the mobile phone's IMU sensor to collect the frequency domain characteristics of angular velocity or acceleration during walking.

[0093] In a further embodiment, the field verification unit includes:

[0094] Visible light face acquisition subunit: An RGB-D camera array deployed at the hotel entrance calculates a 3D face model through multi-view geometric constraints and performs a non-contact 1:N fast comparison with the face features pre-stored on the mobile device;

[0095] Environmental voiceprint authentication subunit: It adopts a distributed microphone array and extracts user voiceprint features through blind source separation to achieve joint verification of voiceprint-spatial reverberation features in natural conversation.

[0096] Gait dynamics subunit: By fusing ground-embedded pressure sensors with millimeter-wave radar, gait cycle features are extracted, pre-stored gait templates are matched in real time, and gait frequency anomaly attacks are detected.

[0097] In a further embodiment, the spatial decision-making module includes:

[0098] Real-time trajectory tracking unit: It integrates UWB positioning and visual odometry data, and uses Kalman filtering to optimize the user's position coordinates. Its state equation is expressed as:

[0099] x t =F t x t-1 +B t u t +w t w t ~N(0,Q) t )

[0100] In the formula: x t F is the state vector at time t, i.e., the user's position coordinates; t B is the state transition matrix; t To control the input matrix; u t w is the control vector. t The noise is process noise, with zero mean and covariance Q. t The distribution is Gaussian; N represents the Gaussian distribution, also known as the normal distribution; 0 is the mean vector of the distribution, i.e., the zero vector; Q t Let be the covariance matrix of the distribution;

[0101] State equation x t =F t x t-1 +B t u t +w t Indicates location prediction; noise term w t ~N(0,Q) t To quantify the uncertainty of the sensor, Kalman filtering dynamically adjusts to suppress noise and outputs optimized coordinates.

[0102] Environmental semantic mapping unit: used to identify key regions through point cloud segmentation. Key regions include elevators, corridors, and guest rooms. Its classification decision function is:

[0103] L i =arg max c∈C P(c∣φ(Voxel i ))

[0104] In the formula: L iThis represents the semantic classification result of the i-th spatial unit; C = {elevator, corridor, guest room} is the set of region categories; Voxel i Let represent the i-th voxel unit; φ is the feature extraction function based on the PointNet++C={elevator, corridor, guest room} architecture; P(c|·) is the conditional probability of belonging to class c given voxel features;

[0105] The process is as follows:

[0106] Input: 3D point cloud data of the hotel environment, generated by LiDAR or depth camera;

[0107] Processing flow:

[0108] graph LR

[0109] A [Original point cloud] --> B {Voxelization process}

[0110] B-->C[Voxel Unit V_oxel_i]

[0111] C --> D [PointNet++ Feature Extraction φ]

[0112] D-->E[Class probability P(c|φ)]

[0113] E-->F[Decision L_i = argmax P]

[0114] F-->G [Semantic Map]

[0115] Output: Each voxel is labeled as a class C = {elevator, corridor, guest room}; finally, a semantically labeled environment map is generated for path planning and resource scheduling.

[0116] Resource scheduling decision unit: Triggers device pre-start based on user trajectory entropy value; its entropy value calculation function is:

[0117] H t =-∑ r∈R P(r)log2P(r)

[0118] In the formula: H t The Shannon entropy represents the user's movement trajectory; R = {elevator, corridor, guest room} is the set of reachable areas; P(r) represents the statistical probability of the user appearing in area r;

[0119] When H tWhen the threshold T is less than 90%, it indicates that the user is definitely heading towards the target room, and the path certainty is greater than 90%. At this time, the room equipment is pre-started. Based on the measured data of the hotel scenario, the recommended interval threshold T∈[0.5,1.2] is determined. This design is the first to apply the information entropy theory to the pre-scheduling decision of hotel equipment, so that the system can accurately start the equipment in advance when the user is 20-50 meters away from the room, which is a distance that traditional methods cannot determine. The measured data reduces energy waste.

[0120] In a further embodiment, the behavior prediction module includes:

[0121] Gait and Intent Association Unit: Used to construct a mapping model between gait features and behavioral intentions;

[0122] Spatiotemporal graph prediction unit: Predicts moving targets using graph convolutional networks;

[0123] Real-time behavior prediction unit: used to generate pre-activation instructions for guest room equipment based on the dynamic parameters of the user's movement trajectory.

[0124] In a further embodiment, the real-time behavior prediction unit is configured as follows:

[0125] (1) Real-time acquisition of the dynamic parameters of the user's movement trajectory, including:

[0126] The predicted arrival time t at the target room, in seconds;

[0127] Instantaneous movement speed v, in meters per second;

[0128] Real-time acceleration a, in meters per second²;

[0129] (2) Perform hierarchical decision-making based on dynamic parameters:

[0130] When the conditions t < 120 and a > 0.3 are met, an emergency rest mode command is generated. The emergency rest mode command includes: rapid cooling program of air conditioner, closing operation of electric curtains and start of bathtub water filling system.

[0131] When the conditions 120≤t<300 and v<0.5 are met, a leisure preparation mode instruction is generated. The leisure preparation mode instruction includes: starting the background music system and adjusting the indoor lighting to a preset low illumination mode.

[0132] When the conditions t≥300 and a<0.1 are met, a normal check-in mode instruction is generated, which includes: starting the basic lighting system and starting the fresh air system;

[0133] (3) The generated instruction pattern is pushed to the user's mobile terminal in real time through the hotel APP data interface, and the following information is displayed synchronously on the APP interface:

[0134] The currently active mode types include Emergency Rest Mode Instruction, Leisure Preparation Mode Instruction, and Regular Check-in Mode Instruction;

[0135] List of triggered device operations;

[0136] The countdown to the completion of guest room preparation is underway;

[0137] (4) The generated device commands are sent to the guest room execution terminal in real time via the MQTT protocol;

[0138] (5) When the user's movement parameters cross the decision threshold, the mode switch and instruction update are completed within 200ms;

[0139] This display allows users to understand the system status in real time and intervene manually. By realizing bidirectional visualization of predictive commands for the first time in a hotel IoT control system, it not only executes equipment operations but also provides real-time feedback on the system's decision-making logic through the APP, thereby improving user experience and system transparency.

[0140] In this embodiment, the three-level biometric authentication adopts a cascaded decision-making mechanism, and the authentication result is determined by the following formula:

[0141]

[0142] in:

[0143] Auth result This is the final output of Level 3 biometric authentication; s i The confidence score represents the i-th level biometric feature (i=1: face, i=2: voiceprint, i=3: gait); ∏ represents the product operator; α represents the multiplication operator. i Represents the dynamic decision threshold for the i-th level of authentication; I(·) is the indicator function:

[0144]

[0145] Hierarchical independent validation: Confidence scores s are calculated for each level of biometric feature (face / voiceprint / gait). i Dynamic threshold α i The system adjusts in real time according to the environment, such as lowering the face recognition threshold when there is insufficient light; the indicator function I outputs a binary result:

[0146] And through cascading decision-making: the final output is Auth only when all three levels of results are 1 (i.e., ∏ = 1). result =1, failure at any level (resulting in 0) immediately causes the product to return to zero, i.e., authentication fails;

[0147] In a hotel entrance authentication scenario:

[0148] Normal users:

[0149] Face s1 = 0.92 ( ＞ α1=0.8)→I1=1;

[0150] Voiceprint = 0.88 ＞ α² = 0.75) → I² = 1;

[0151] Gait = 0.95 ＞ α3=0.7)→I3=1;

[0152] Product = 1 → Auth result =1 (Door lock unlocks automatically)

[0153] Attacker (forged face):

[0154] Face = 0.45 ＜ α1) → I1 = 0;

[0155] Product = 0 → Auth result =0 (System alarm triggered and defense protocol activated)

[0156] Failure at any level terminates the subsequent authentication process, saving time and improving authentication efficiency; attackers must simultaneously forge three levels of biometric features (3D face + voiceprint + gait dynamics) to enhance security and improve anti-forgery capabilities.

[0157] In a further embodiment, the gait and intention association unit is configured as follows:

[0158] The input is a gait feature vector containing the three-dimensional distribution of gait frequency, stride length, and plantar pressure;

[0159] Temporal features are extracted using an attention-enhanced bidirectional LSTM network, dynamically focusing on key frames for steering or speed changes;

[0160] Output the probability distribution of user behavior intent: [walking to the elevator, going to the guest room, staying in the public area];

[0161] The spatiotemporal graph prediction unit is configured as follows:

[0162] Construct a spatiotemporal graph: Node = real-time user coordinates (x, y, t), spatial edge = Euclidean distance between users + direction angle, time edge = dynamic sliding historical trajectory;

[0163] Computed via an adaptive spatiotemporal graph convolutional network:

[0164]

[0165] In the formula: H (l+1)This is the node feature matrix of the l-th layer; its dimension is N×F, where N is the number of nodes in the graph and F is the feature dimension of each node. In the hotel scenario, nodes represent users, and features include location coordinates (x, y), timestamp (t), velocity (v), acceleration (a), etc.

[0166] Θ (l) It is the trainable weight matrix of the l-th layer, with dimensions F×F', where F is the input feature dimension and F' is the output feature dimension. This matrix is ​​used to perform linear transformation on the node features (similar to a fully connected layer) and learn a representation suitable for the task.

[0167] H (l) Θ (l) It is a matrix multiplication that multiplies the node feature matrix H. (l) With weight matrix Θ (l) Multiplying them together yields the transformed feature matrix, which has dimensions N×F'.

[0168] To enhance the adjacency matrix: This is the result of adding the identity matrix I to the original adjacency matrix A, i.e. Where: A is the adjacency matrix, which represents the connection relationship between nodes in the graph (for example, if user i and user j are within 3 meters and the angle is within ±60°, then A_{ij} = 1, otherwise it is 0);

[0169] I is the identity matrix, representing the connection between each node and itself. This is done to ensure that nodes take their own features into account during convolution.

[0170] For the enhancement degree matrix: This is the enhancement adjacency matrix. The degree matrix is ​​a diagonal matrix, consisting of the degree (number of connections) of each node plus self-loops (so each node has at least a degree of 1). The degree matrix is ​​used to normalize the adjacency matrix.

[0171] The degree matrix is ​​raised to the power of -1 / 2, that is, for Each diagonal element is raised to the power of -1 / 2. This is to normalize the adjacency matrix so that the feature propagation will not be biased due to different node degrees.

[0172] This is a normalized symmetric adjacency matrix, obtained by left multiplication and right multiplication. We obtained a variant of a normalized Laplacian matrix, which, when multiplied by the node features, can achieve a weighted average of the features of neighboring nodes.

[0173] Output the probability distribution of the moving target area: [Elevator, Corridor, Guest Room]

[0174] First, perform a linear transformation on the node features of the current layer (multiplied by the weight matrix Θ). (l) ).

[0175] Then, the features of neighboring nodes (including themselves) are aggregated using a normalized adjacency matrix.

[0176] Finally, nonlinearity is introduced through the ReLU activation function.

[0177] This operation combines the graph's structural information (adjacency matrix) with node feature information to generate new node representations. In hotel control systems, this is used to predict users' movement target areas, such as elevators, corridors, and guest rooms.

[0178] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A hotel IoT control system based on integrated smart hardware, characterized in that: The system includes: a biometric authentication module, a spatial decision-making module, a behavior prediction module, and an execution control module; The biometric authentication module is used to dynamically bind user identity through a three-level biometric feature combination of mobile pre-collection and on-site contactless authentication. The spatial decision-making module is used to build a real-time environmental map based on multi-sensor fusion SLAM technology and track the user's movement trajectory; The behavior prediction module is used to predict user behavioral intentions based on gait dynamics data and movement trajectory using a spatiotemporal sequence model. The execution control module is used to automatically trigger guest room equipment preparation instructions based on the prediction results, so as to realize the full automation of the check-in process; The three-level non-contact biometric feature includes a mobile pre-collection unit and an on-site verification unit. The mobile pre-collection unit is used to collect features in stages through the hotel APP. The on-site verification unit is used to collect biometric features in real time during the user's seamless passage and perform dynamic cross-verification with the pre-collected features on the mobile terminal to complete dual authentication of identity binding and liveness detection.

2. The hotel IoT control system based on intelligent hardware integration according to claim 1, characterized in that: The mobile terminal pre-collection unit includes: Visible light face detection: Uses the phone's front-facing camera for liveness detection; Ambient soundprint: Record a 5-second voice command and extract the room's reverberant fingerprint; Gait dynamics: using the mobile phone's IMU sensor to collect the frequency domain characteristics of angular velocity or acceleration during walking.

3. The hotel IoT control system based on intelligent hardware integration according to claim 1, characterized in that: The on-site verification unit includes: Visible light face acquisition subunit: An RGB-D camera array deployed at the hotel entrance calculates a 3D face model through multi-view geometric constraints and performs a non-contact 1:N fast comparison with the face features pre-stored on the mobile device; Environmental voiceprint authentication subunit: It adopts a distributed microphone array and extracts user voiceprint features through blind source separation to achieve joint verification of voiceprint-spatial reverberation features in natural conversation. Gait dynamics subunit: By fusing ground-embedded pressure sensors with millimeter-wave radar, gait cycle features are extracted, pre-stored gait templates are matched in real time, and gait frequency anomaly attacks are detected.

4. The hotel IoT control system based on intelligent hardware integration according to claim 1, characterized in that: The spatial decision-making module includes: Real-time trajectory tracking unit: It integrates UWB positioning and visual odometry data, and uses Kalman filtering to optimize the user's position coordinates. Its state equation is expressed as: x t =F t x t-1 +B t u t +w t ,w t ~N(0,Q t ) In the formula: x t F is the state vector at time t, i.e., the user's position coordinates; t B is the state transition matrix; t To control the input matrix; u t w is the control vector. t The noise is process noise, with zero mean and covariance Q. t The distribution is Gaussian; N represents the Gaussian distribution, also known as the normal distribution; 0 is the mean vector of the distribution, i.e., the zero vector; Q t Let be the covariance matrix of the distribution; Environmental semantic mapping unit: used to identify key regions through point cloud segmentation. Key regions include elevators, corridors, and guest rooms. Its classification decision function is: L i =arg max c∈C P(c∣φ(Voxel i )) In the formula: L i This represents the semantic classification result of the i-th spatial unit; C = {elevator, corridor, guest room} is the set of region categories; Voxel i Let represent the i-th voxel unit; φ is the feature extraction function based on the PointNet++C={elevator, corridor, guest room} architecture; P(c|·) is the conditional probability of belonging to category c given voxel features; Resource scheduling decision unit: Triggers device pre-start based on user trajectory entropy value; its entropy value calculation function is: H t =-∑ r∈R P(r)log2P(r) In the formula: H t Let R represent the Shannon entropy of the user's movement trajectory; R = {elevator, corridor, guest room} is the set of reachable areas; P(r) represents the statistical probability of the user appearing in area r.

5. The hotel IoT control system based on intelligent hardware integration according to claim 1, characterized in that: The behavior prediction module includes: Gait and Intent Association Unit: Used to construct a mapping model between gait features and behavioral intentions; Spatiotemporal graph prediction unit: Predicts moving targets using graph convolutional networks; Real-time behavior prediction unit: used to generate pre-activation instructions for guest room equipment based on the dynamic parameters of the user's movement trajectory.

6. The hotel IoT control system based on intelligent hardware integration according to claim 5, characterized in that: The real-time behavior prediction unit is configured as follows: (1) Real-time acquisition of the dynamic parameters of the user's movement trajectory, including: The predicted arrival time t at the target room, in seconds; Instantaneous movement speed v, in meters per second; Real-time acceleration a, in meters per second²; (2) Perform hierarchical decision-making based on dynamic parameters: When the conditions t < 120 and a > 0.3 are met, an emergency rest mode command is generated. The emergency rest mode command includes: rapid cooling program of air conditioner, closing operation of electric curtains and start of bathtub water filling system. When the conditions 120≤t<300 and v<0.5 are met, a leisure preparation mode instruction is generated. The leisure preparation mode instruction includes: starting the background music system and adjusting the indoor lighting to a preset low illumination mode. When the conditions t≥300 and a<0.1 are met, a normal check-in mode instruction is generated, which includes: starting the basic lighting system and starting the fresh air system; (3) The generated instruction pattern is pushed to the user's mobile terminal in real time through the hotel APP data interface, and the following information is displayed synchronously on the APP interface: The currently active mode types include Emergency Rest Mode Instruction, Leisure Preparation Mode Instruction, and Regular Check-in Mode Instruction; List of triggered device operations; The countdown to the completion of guest room preparation is underway; (4) The generated device commands are sent to the guest room execution terminal in real time via the MQTT protocol; (5) When the user's movement parameters cross the decision threshold, the mode switch and instruction update are completed within 200ms.

7. The hotel IoT control system based on intelligent hardware integration according to claim 1, characterized in that: The three-level biometric authentication adopts a cascaded decision-making mechanism, and the authentication result is determined by the following formula: In the formula: Auth result This is the final output of Level 3 biometric authentication; s i The confidence score represents the i-th level biometric feature (i=1: face, i=2: voiceprint, i=3: gait); ∏ represents the product operator; α represents the multiplication operator. i Represents the dynamic decision threshold for the i-th level of authentication; I(·) is the indicator function:

8. The hotel IoT control system based on intelligent hardware integration according to claim 5, characterized in that: The gait and intent association unit is configured as follows: The input is a gait feature vector containing the three-dimensional distribution of gait frequency, stride length, and plantar pressure; Temporal features are extracted using an attention-enhanced bidirectional LSTM network, dynamically focusing on key frames for steering or speed changes; Output the probability distribution of user behavior intent: [walking to the elevator, going to the guest room, staying in the public area]; The spatiotemporal map prediction unit is configured as follows: Construct a spatiotemporal graph: Node = real-time user coordinates (x, y, t), spatial edge = Euclidean distance between users + direction angle, time edge = dynamic sliding historical trajectory; Computed via an adaptive spatiotemporal graph convolutional network: In the formula: H (l+1) It is the node feature matrix of the l-th layer; Θ (l) H is the trainable weight matrix of the l-th layer; (l) Θ (l) It is a matrix multiplication that multiplies the node feature matrix H. (l) With weight matrix Θ (l) Multiply them to obtain the transformed feature matrix; To enhance the adjacency matrix: This is the result of adding the identity matrix I to the original adjacency matrix A, i.e. Where: A is the adjacency matrix, representing the connection relationships between nodes in the graph. I is the identity matrix, representing the connection between each node and itself; For the enhancement degree matrix: This is the enhancement adjacency matrix. The degree matrix; The degree matrix is ​​raised to the power of -1 / 2, that is, for Each diagonal element is raised to the power of -1 / 2; Output the probability distribution of the moving target area: [elevator, corridor, guest room].

Images