An emotion tree interaction system for public emotion visualization

Abstract

The application relates to the technical field of computers, and specifically discloses an emotion tree interaction system for public emotion visualization, which aims to solve the technical problem that existing public space devices lack emotion perception and bidirectional interaction capability. The system comprises a multi-modal signal acquisition module, a signal preprocessing and feature extraction module, a PAD emotion modeling and mapping module, a central data statistics and group emotion aggregation module, an emotion tree visualization output module, and an interaction control and feedback closed loop unit. The multi-modal signal acquisition module collects user emotion signals through a pressure sensor, a camera and a physiological signal sensor; the PAD emotion modeling module performs weighted fusion and time smoothing processing on the multi-modal signals to obtain comprehensive pleasure, arousal and dominance parameters; the emotion tree visualization module maps the tree crown hue, saturation, particle motion and mechanical posture change according to the PAD parameters; and the interaction control module realizes user proximity detection, embrace judgment and real-time feedback. The single-person interaction tree and the central emotion tree form a hierarchical linkage to generate a public emotion portrait. Through the above technical scheme, the system realizes non-invasive group emotion perception and intuitive visual expression, and enhances the emotional interaction experience in public space.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to an interactive emotion tree system for visualizing public emotions. Background Technology

[0002] With the rapid development of technologies such as 5G and AIoT, intelligent interactive devices in public spaces are becoming increasingly widespread, giving rise to a trend of deep integration between smart cities and human-computer interaction technologies. Against this backdrop, emotion recognition and expression, as a core dimension of human-computer interaction, are gradually extending from laboratory settings to open public spaces. Emotion recognition technology aims to achieve intelligent responses by analyzing human emotional states. Its core lies in acquiring individual psychological state information through multimodal perception and transforming it into data forms that machines can understand and process.

[0003] In the specific field of public emotion interaction, existing technologies exhibit significant limitations. While emotion recognition technology has made remarkable progress in laboratory settings, its application in public spaces faces numerous challenges. Traditional emotion recognition solutions primarily rely on single-modal methods such as facial expression recognition, speech tone analysis, or physiological signal detection. These solutions demonstrate significant inadequacy in handling group, non-contact, and low-psychological-stress public emotion collection scenarios. Specifically, facial expression recognition requires close user cooperation and raises privacy concerns; speech analysis suffers from extremely low signal-to-noise ratios in noisy public environments; and physiological signal collection requires wearing specialized equipment and raises hygiene concerns. These limitations make it difficult for existing technologies to achieve large-scale, natural-state emotion information collection in public spaces, and they fail to meet the application requirements for real-time perception and expression of group emotions.

[0004] Limited by the aforementioned technological bottlenecks, existing public art installations generally exhibit a one-way output characteristic, failing to form an effective two-way emotional interaction channel. More critically, public spaces, as an important "third space" in modern urban life, bear diverse functions such as emotional expression, self-exploration, and social interaction for citizens, yet they have long lacked the technological means to capture and present the emotional state of the group. Users' emotional experiences in this space are consistently neglected, making it difficult for public spaces to fulfill their due functions of emotional connection and social cohesion. The direct consequence is that while public spaces accommodate a large number of individuals, they cannot perceive and respond to the emotional states of these individuals, cannot aggregate discrete emotional information into a meaningful collective emotional landscape, and cannot convey emotional feedback to users through intuitive visual and mechanical changes. This technological predicament severely restricts the development of the emotional interaction capabilities of public spaces and urgently requires a breakthrough through technological innovation. Summary of the Invention

[0005] This application provides an interactive emotion tree system for visualizing public emotions, aiming to solve the technical problem that existing public space installations lack emotion perception and two-way interactive capabilities.

[0006] The first aspect of this application provides an interactive emotion tree system for public emotion visualization, including a multimodal signal acquisition module, a signal preprocessing and feature extraction module, a PAD emotion modeling and mapping module, a central data statistics and group emotion aggregation module, an emotion tree visualization output module, and an interactive control and feedback closed-loop unit. The multimodal signal acquisition module includes a pressure sensing unit, an expression capture unit, and a physiological signal acquisition unit, which are used to automatically acquire the user's multimodal emotional signals when the user approaches the device. The pressure sensing unit is deployed on the trunk surface of the single-person interactive tree and uses a tactile pressure sensor to collect the force signal, duration signal and rhythm change signal when the user hugs the trunk. The facial expression capture unit is automatically activated by the camera when the distance between the user and the device is within 0.5 to 1.5 meters, and is used to capture facial expression images of the user. The physiological signal acquisition unit integrates a PPG heart rate waveform sensor and an EDG skin conductance sensor to collect the user's heart rate variability signal and physiological signals related to emotional arousal. The signal preprocessing and feature extraction module performs unified time source synchronization and noise reduction filtering and baseline correction on various signals acquired by the multimodal signal acquisition module, and extracts statistical and temporal features such as stress intensity features, facial expression pleasure features, and heart rate variability features of various signals. The PAD emotion modeling and mapping module calculates single-modal PAD parameters for stress signals, facial expression signals, and physiological signals, respectively. It forms a comprehensive PAD value through weighted fitting and performs time-smoothing update on the PAD results to obtain smoothed comprehensive pleasure parameters, comprehensive arousal parameters, and comprehensive dominance parameters. The central data statistics and group emotion aggregation module stores the PAD parameters, interaction duration and activation number of each interaction, counts the anonymous emotion distribution and time-period change trend of all users, forms a public emotion profile and outputs a visualized group emotion heat map or time-series curve. The emotion tree visualization output module includes a canopy visual generation unit and a mechanical device activity unit. It maps the canopy hue according to the comprehensive pleasure parameter, the canopy saturation according to the comprehensive arousal parameter, the particle velocity and particle density according to the emotional activity parameter, the node BPM according to the physiological arousal rhythm parameter, and controls the curvature, proportion and movement amplitude of the bionic branches according to the PAD parameter to form mechanical posture changes that match the emotional state. The interactive control and feedback closed-loop unit triggers a hug confirmation mechanism when the user approaches the device. Within a range of 0.5 to 1.5 meters, it automatically enters a receiving state. It judges the intensity of emotions based on the strength, duration, and rhythm of the hug and activates corresponding feedback. The individual interaction tree and the central emotion tree are linked. Individual interaction triggers local changes, and the central emotion tree aggregates all user data to generate an overall public emotion expression.

[0007] In one optional implementation, the pressure sensing unit includes a plurality of tactile pressure sensor arrays distributed at different height positions and circumferential angles on the tree trunk, for collecting spatially distributed pressure signals and overall pressure signals when a user hugs the tree trunk; The output signals of the multiple tactile pressure sensor arrays are amplified by an amplifier circuit and then enter an analog-to-digital converter for digital processing. The digital signal processor performs sliding window feature extraction on the digitized pressure signal to obtain pressure time-series features such as peak pressure, pressure variance, pressure rise rate and pressure duration. The facial expression capture unit uses a near-infrared supplementary light camera. When the user enters the preset distance range, it automatically switches from sleep mode to working mode, performs face detection and key point localization on the captured facial images, extracts the geometric features and optical flow features of the facial key points, and outputs facial expression dimension parameters such as pleasure, sadness, surprise, anger and fear through a pre-trained facial expression classification network. The PPG heart rate waveform sensor of the physiological signal acquisition unit acquires the user's pulse wave signal through photoplethysmography, the EDG skin conductance response sensor acquires the user's skin conductance change signal, and the heart rate variability analysis module performs R-peak detection and interval calculation on the pulse wave signal to obtain heart rate variability index and autonomic nerve tension assessment value. In one optional implementation, the signal preprocessing and feature extraction module includes a time synchronization unit, a noise reduction and filtering unit, a baseline correction unit, and a feature extraction unit. The time synchronization unit uses a unified high-precision clock source to give pressure signals, facial signals, and physiological signals a unified timestamp, ensuring that all types of signals are strictly aligned in the time dimension. The noise reduction and filtering unit adopts an adaptive filtering algorithm, which uses a high-pass filter to remove low-frequency drift in the pressure signal, normalizes the illumination changes in the facial expression image, and uses a notch filter to remove power frequency interference in the physiological signal. The baseline correction unit uses the initial state when the user first comes into contact with the device as the baseline reference, calculates the relative changes of various signals relative to the baseline, and eliminates the absolute value deviation caused by individual differences. The feature extraction unit performs statistical feature extraction and time-series feature extraction on the signal after denoising filtering and baseline correction. The statistical features include mean, variance, peak value, kurtosis and skewness, and the time-series features include autocorrelation function, cross-correlation function and Fourier spectrum features. In one optional implementation, the PAD emotion modeling and mapping module includes a single-modal parameter calculation unit, a multimodal weighted fusion unit, and a time-smoothing update unit; The single-modal parameter calculation unit establishes mapping models between pressure signals and PAD parameters, facial expression signals and PAD parameters, and physiological signals and PAD parameters, respectively. The mapping models are implemented using regression neural networks or lookup tables. The multimodal weighted fusion unit performs weighted fitting based on the confidence and contribution of each modal signal to obtain a comprehensive PAD value. When the quality of a certain modal signal is poor, its weight is reduced, and when the feature of a certain modal signal is significant, its weight is increased. The time-smoothing update unit uses an exponential moving average algorithm to smooth the comprehensive PAD value over time. The smoothing coefficient is adaptively adjusted according to the user's interaction frequency. The smoothed PAD value is used to drive the visualization output of the emotion tree. In one optional implementation, the central data statistics and group emotion aggregation module includes an emotion state recording unit, a group emotion summarization unit, and a public emotion profile generation unit. The emotional state recording unit stores the user identifier, interaction timestamp, PAD parameters, interaction duration, activation count, and signal quality marker for each interaction. The record of each interaction is stored in encrypted form to protect user privacy. The group emotion aggregation unit performs aggregation operations on all anonymous emotion records within a specified time period to obtain a histogram of pleasure distribution, a histogram of arousal distribution, a histogram of dominance distribution, as well as time-series curves of pleasure, arousal, and dominance. The public emotion profile generation unit generates a visualized group emotion heatmap based on the distribution histogram and time-series curve. The color intensity represents the level of pleasure, the position of the heatmap patches represents the distribution of users in public spaces, and the time axis represents the trend of emotion changes. In one optional implementation, the tree canopy visual generation unit of the emotion tree visualization output module dynamically generates the visual representation of the tree canopy based on the pleasure level parameter, arousal level parameter and emotional activity level parameter. The hue mapping algorithm maps the pleasure level parameter to the hue angle of the color space. When the pleasure level is high, it is mapped to a warm hue, and when the pleasure level is low, it is mapped to a cool hue. The saturation mapping algorithm maps the wakefulness parameter to the saturation value in the color space. When the wakefulness is high, the saturation increases, and when the wakefulness is low, the saturation decreases. The particle system controller controls the emission rate and particle velocity of the particle emitters on the canopy surface according to the emotional activity level parameter. When the emotional activity level is high, the particle movement is violent, and when the emotional activity level is low, the particle movement is slow. The node BPM controller controls the flashing rhythm of the LED nodes on the canopy surface according to physiological arousal rhythm parameters, so that it forms a visual rhythmic resonance with the user's average heart rate; In one alternative implementation, the mechanical device of the mood tree visualization output module includes a tree branch drive motor, a reducer, a linkage mechanism, and a bionic branch. The controller calculates the target tree curvature, target tree ratio and target motion amplitude based on PAD parameters, and generates motor drive signals through PID control algorithm; The drive motor drives the linkage mechanism to move through the reducer. The linkage mechanism drives the bionic branches to bend and extend. The bionic branches are made of PLA bionic material, which is lightweight and highly malleable. The activity status of the mechanical device and the visual appearance of the tree canopy form a multi-channel emotional feedback, allowing users to experience the expression of group emotions at the visual, tactile, and psychological levels. In one optional implementation, the interactive control and feedback closed-loop unit includes a proximity detection subunit, an embrace judgment subunit, and a feedback activation subunit; The proximity detection subunit uses a combination of infrared and ultrasonic ranging sensors for detection. When the distance between the user and the device is within 0.5 to 1.5 meters, it is triggered to enter the preparation state. The hugging judgment subunit determines whether the user is hugging based on the output signal of the pressure sensing unit. When the pressure signal exceeds a preset threshold and the duration exceeds a preset duration, the hugging is determined to be successful. The feedback activation subunit activates the corresponding emotion tree visualization output based on the strength, duration, and rhythmic changes of the hug, forming a real-time two-way emotional interaction experience. In one optional implementation, multiple individual interaction trees and at least one central emotion tree are connected via a network to form a distributed emotion perception network. The multiple individual interaction trees are respectively set in different locations in the public space. Each individual interaction tree independently collects multimodal emotion signals from nearby users and uploads the preprocessed feature data to the central server. The central server runs the central data statistics and group emotion aggregation module, which aggregates the feature data uploaded by all individual interaction trees to generate a public emotion profile, and then sends the visualization parameters corresponding to the public emotion profile to all individual interaction trees and the central emotion tree. Each individual interactive tree drives its own canopy vision and mechanical device to form local emotional expression based on locally collected signals. The central emotional tree drives the overall canopy vision and mechanical device to form group emotional expression based on the public emotional profile. A hierarchical emotional expression structure is formed between the individual interactive trees and the central emotional tree. In one optional implementation, the system further includes a privacy protection module for anonymizing the collected signals before storing and transmitting them. The privacy protection module removes all personally identifiable information during the signal preprocessing stage, retaining only the emotion feature vector, timestamp, and signal quality marker. The emotion feature vector is processed by dimensionality reduction and quantization encoding to compress high-dimensional emotion features into low-dimensional anonymous feature representation. The stored and transmitted data are all anonymized to ensure that an individual's identity cannot be deduced from the data. Compared with the prior art, the advantages and positive effects of the present invention are as follows: By adopting a technical solution that combines multimodal signal acquisition with PAD emotion modeling, non-intrusive emotion perception and visualization of a large number of users in public spaces are realized, solving the technical problem of the lack of group-oriented emotion acquisition methods in existing technologies; By integrating data from a tactile pressure sensor, camera, PPG heart rate waveform sensor, and EDG skin conductance sensor, the psychological threshold for user participation is lowered, making emotional expression a natural behavioral feedback rather than a deliberate data collection behavior. By employing multimodal weighted fusion and time-smooth update processing of stress signals, facial expression signals, and physiological signals, the robustness and reliability of emotion recognition results are improved, and the volatility and uncertainty of single-modal recognition results are avoided. Through the coordinated control of the canopy visual generation unit and the mechanical device activity unit, abstract emotional data is transformed into intuitive color changes, particle movements and mechanical shape changes, enabling users to intuitively feel the expression of their own emotions and the emotions of the group. Through the hierarchical linkage design between the individual interactive tree and the central emotion tree, multi-scale emotion visualization, from individual emotion expression to group emotion aggregation, is realized, enriching the emotional interaction dimension in public spaces. The privacy protection module anonymizes all collected data, retaining only the emotional feature vectors and not storing personally identifiable information, thus protecting users' privacy rights while realizing the emotion perception function. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the overall technical solution architecture of the emotion tree interactive system for public emotion visualization proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the PAD emotion modeling and mapping module in this invention; Figure 3 This is a flowchart illustrating the logic flow of the multimodal signal acquisition and preprocessing stage according to an embodiment of this application. Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the single-person interaction tree and the central emotion tree according to an embodiment of this application; Figure 5 This is a schematic diagram of the framework of the emotion tree visualization output module according to an embodiment of this application. Detailed Implementation

[0009] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] In the accompanying drawings, the size of constituent elements, the thickness of layers, or areas may sometimes be exaggerated for clarity. Therefore, any implementation of this disclosure is not necessarily limited to the dimensions shown in the drawings, and the shapes and sizes of the components in the drawings do not reflect true proportions. Furthermore, the drawings schematically illustrate ideal examples, and any implementation of this disclosure is not limited to the shapes or values ​​shown in the drawings.

[0011] Reference Figure 1 , Figure 1 This is a schematic diagram of the overall technical architecture of an interactive emotion tree system for public emotion visualization proposed in one embodiment of this application. As shown in the figure, the system includes: a multimodal signal acquisition module, a signal preprocessing and feature extraction module, a PAD emotion modeling and mapping module, a central data statistics and group emotion aggregation module, an emotion tree visualization output module, and an interactive control and feedback closed-loop unit.

[0012] In this embodiment, the core design concept of the emotion tree interactive system for public emotion visualization is to collect emotional signals from a massive number of users in public spaces through non-invasive multimodal perception methods. After professional emotion modeling and mapping processing, abstract emotional states are transformed into intuitive visual expressions. In specific implementation, the system achieves comprehensive perception and accurate capture of users' emotional states through a diversified sensor network composed of pressure sensing units, facial expression capture units, and physiological signal acquisition units.

[0013] The pressure sensing unit in the multimodal signal acquisition module is deployed on the trunk surface of the single-person interactive tree. It uses a tactile pressure sensor array to collect the force, duration, and rhythm changes when the user hugs the trunk. In this embodiment, the pressure sensing unit is specifically arranged as follows: multiple tactile pressure sensor arrays are evenly distributed at different heights and circumferential angles on the trunk. The acquisition frequency of each sensor point is set to 100 Hz, and the sampling accuracy is 16-bit analog-to-digital conversion resolution. When the user hugs the trunk with both arms, the sensors at different locations are simultaneously subjected to pressure, generating a voltage signal proportional to the pressure magnitude. This voltage signal is amplified by an amplification circuit and then digitized by an analog-to-digital converter. The digital signal processor performs sliding window feature extraction on the digitized pressure signal, with a sliding window length of 500 milliseconds and a step size of 100 milliseconds. Through sliding window analysis, the system can extract pressure temporal features such as peak pressure, pressure variance, pressure rise rate, and pressure duration. These feature data will serve as important input parameters for subsequent PAD emotion modeling.

[0014] The facial expression capture unit in the multimodal signal acquisition module is automatically activated when the distance between the user and the device is within 0.5 to 1.5 meters, and is used to capture facial expression images of the user. In this embodiment, the facial expression capture unit uses a near-infrared supplementary lighting camera, which is positioned at the top of the single-person interaction tree, facing the direction in which the user is likely to stand. The camera's built-in human detection module continuously monitors whether a user appears in the field of view. When the infrared ranging sensor and the ultrasonic ranging sensor fuse to detect and determine that the user has entered the preset distance range of 0.5 to 1.5 meters, the camera automatically switches from sleep mode to working mode. The camera's working parameters are set as follows: image acquisition resolution of 640×480 pixels, frame rate of 30 frames per second, wavelength of near-infrared supplementary lighting of 850 nanometers, and supplementary lighting intensity automatically adjusted according to ambient light. The acquired facial image is first processed by a face detection algorithm to locate the face region, and then the geometric coordinates of 68 key points on the face are extracted by a key point localization algorithm. Subsequently, the geometric features and optical flow features of the facial key points are extracted. These feature data are input into a pre-trained facial expression classification network, which is trained based on a convolutional neural network architecture and can output five facial expression dimensions: joy, sadness, surprise, anger, and fear. Each parameter has a value ranging from 0 to 1 as a floating-point number.

[0015] The physiological signal acquisition unit in the multimodal signal acquisition module integrates a PPG heart rate waveform sensor and an EDG skin conductance response sensor to collect physiological signals related to the user's heart rate variability and emotional arousal. In this embodiment, the PPG heart rate waveform sensor uses the photoplethysmography principle to acquire pulse wave signals. This sensor is integrated into a dedicated wearing area on the surface of the single-person interactive tree trunk. When the user places their finger in this area, the sensor emits green light waves (wavelength 525 nm) to irradiate the skin surface, and the pulse wave signal is obtained by detecting the periodic changes in the intensity of the reflected light. The sampling rate of the PPG sensor is set to 100 Hz, and the sampling accuracy is 12 bits. The EDG skin conductance response sensor uses a two-electrode method to measure changes in skin conductivity. The electrode material is silver-silver chloride electrodes, which are attached to the fingertips of the user's non-dominant index and middle fingers. The sampling rate of the EDG sensor is set to 10 Hz, and the sampling accuracy is 16 bits. The heart rate variability analysis module performs R-peak detection and interval calculation on the acquired pulse wave signal. It uses classic heart rate variability time-domain analysis and frequency-domain analysis methods to calculate heart rate variability indices, including SDNN (standard deviation of normal RR interval), RMSSD (root mean square difference of adjacent RR intervals), LF / HF (low-frequency to high-frequency power ratio), etc. At the same time, it outputs autonomic nerve tension assessment values ​​to characterize the balance between the sympathetic and parasympathetic nervous systems.

[0016] In this embodiment, when a user approaches the device, the proximity detection subunit in the interactive control and feedback closed-loop unit uses a fusion detection method combining infrared and ultrasonic ranging sensors to determine the distance between the user and the device. When the system detects that the user has entered a preset distance range of 0.5 to 1.5 meters, it automatically triggers the multimodal signal acquisition module to enter the working state, and simultaneously prompts the user to interact through voice prompts or visual guidance. When the distance between the user and the device is within the effective interaction range, the facial expression capture unit automatically switches from sleep mode to working state and begins to capture the user's facial expression images.

[0017] Reference Figure 3 , Figure 3 This is a logical flowchart of the multimodal signal acquisition and preprocessing stage according to an embodiment of this application. The signal preprocessing and feature extraction module performs unified time source synchronization, noise reduction filtering, and baseline correction on various signals acquired by the multimodal signal acquisition module, and extracts statistical and temporal features such as stress intensity features, facial expression pleasure features, and heart rate variability features from various signals.

[0018] In this embodiment, the signal preprocessing and feature extraction module includes a time synchronization unit, a noise reduction and filtering unit, a baseline correction unit, and a feature extraction unit. The time synchronization unit uses a unified high-precision clock source with an accuracy better than 1 millisecond to assign a unified timestamp to pressure signals, facial expression signals, and physiological signals, ensuring strict alignment of all signals in the time dimension. Specifically, each sensor module adds a timestamp label while collecting data. The timestamps are uniformly allocated by the central processing unit after synchronization with a standard time server via the Network Time Protocol (NTP), ensuring that the time deviation between multimodal signals is controlled within 2 milliseconds.

[0019] The denoising and filtering unit employs an adaptive filtering algorithm, using different filtering strategies for different types of signals. For low-frequency drift components in pressure signals, a high-pass filter with a cutoff frequency of 0.5 Hz is used for removal. This high-pass filter is a second-order Butterworth filter design, and the phase delay of the filter is compensated for through pre-filtering. For illumination variations in facial expression images, a normalization method is used. First, the average brightness value of the entire image is calculated, and then the gray value of each pixel is divided by the average brightness value to normalize the image, making the brightness distribution more uniform. For power frequency interference in physiological signals, a notch filter is used for removal. The center frequency of the notch filter is set to 50 Hz or 60 Hz (adaptively selected according to the local power grid frequency), the notch width is ±2 Hz, and the notch depth is greater than 40 dB.

[0020] The baseline correction unit uses the user's initial state upon first contact with the device as a baseline reference, calculating the relative changes of various signals relative to the baseline to eliminate absolute value deviations caused by individual differences. Specifically, when the user first contacts the device, the system automatically records the data collected from various signals within the first 10 seconds, calculating the average value of each signal channel within this time window as the baseline value. In subsequent data processing, all real-time acquired signal values ​​are subtracted from the corresponding baseline value to obtain the relative changes for further analysis. This baseline correction method effectively eliminates physiological differences between different users, making emotion recognition results more objective and accurate.

[0021] The feature extraction unit performs statistical and temporal feature extraction on the denoised, filtered, and baseline-corrected signal. Statistical features include mean, variance, peak value, kurtosis, and skewness, calculated using a sliding window method with a window length of 5 seconds and a step size of 1 second. Temporal features include autocorrelation function, cross-correlation function, and Fourier spectrum features. The autocorrelation function analyzes the signal's periodicity, the cross-correlation function analyzes the correlation between different signal channels, and the Fourier spectrum features extract the energy distribution of the signal at different frequency components using Fast Fourier Transform (FFT). These feature data constitute a high-dimensional feature vector for subsequent PAD sentiment modeling.

[0022] Reference Figure 2 , Figure 2 This is a schematic diagram of the core principle framework of the PAD emotion modeling and mapping module in this invention. The PAD emotion modeling and mapping module calculates single-modal PAD parameters for stress signals, facial expression signals, and physiological signals respectively, forms a comprehensive PAD value through weighted fitting, and performs time-smoothing update processing on the PAD results to obtain smoothed comprehensive pleasure parameters, comprehensive arousal parameters, and comprehensive dominance parameters.

[0023] In this embodiment, the PAD emotion modeling and mapping module includes a single-modal parameter calculation unit, a multimodal weighted fusion unit, and a time-smoothing update unit. The single-modal parameter calculation unit establishes mapping models between stress signals and PAD parameters, facial expression signals and PAD parameters, and physiological signals and PAD parameters, respectively. The mapping models are implemented using a regressive neural network, which is a three-layer fully connected neural network structure. The first hidden layer contains 128 neurons, the second hidden layer contains 64 neurons, and the output layer contains 3 neurons corresponding to the three dimensions of pleasure, arousal, and dominance. The network uses the ReLU activation function, and the output layer uses the Sigmoid activation function to map the output values ​​to the 0-1 range. The network's training data comes from a pre-collected multimodal emotion signal dataset with emotion labels, and training uses the mean squared error loss function and the Adam optimizer.

[0024] Specifically, for the mapping model between stress signals and PAD parameters, the input features include stress time-series features such as peak pressure, pressure variance, pressure rise rate, and pressure duration, and the output is the corresponding three-dimensional PAD parameters. For the mapping model between facial expression signals and PAD parameters, the input features include facial expression dimension parameters such as pleasure, sadness, surprise, anger, and fear, and the output is the corresponding three-dimensional PAD parameters. For the mapping model between physiological signals and PAD parameters, the input features include heart rate variability indicators (SDNN, RMSSD, LF / HF, etc.) and autonomic nervous system tone assessment values, and the output is the corresponding three-dimensional PAD parameters.

[0025] The multimodal weighted fusion unit performs a weighted fitting based on the confidence and contribution of each modality signal to obtain a comprehensive PAD value. The confidence assessment mechanism is as follows: For stress signals, when the variance of the stress signal is within a reasonable range (between 0.1 and 0.8), the confidence is high; when the variance is too small (less than 0.1), it indicates that the user may not have performed an effective hug; when the variance is too large (greater than 0.8), it may indicate abnormal interference, and the confidence is correspondingly reduced. For facial expression signals, when the face detection confidence is greater than 0.8 and the facial key point integrity is greater than 0.9, the confidence is high; for physiological signals, when the signal-to-noise ratio of the PPG signal is greater than 10 dB and the heart rate is within the normal range (50 to 120 beats per minute), the confidence is high. The weight calculation formula is: the weight of a certain modality is equal to the confidence of that modality divided by the sum of the confidence of all modalities. The final comprehensive PAD value is calculated by weighted averaging of the PAD values ​​of each modality and their corresponding weights.

[0026] The time-smoothing update unit uses an exponential moving average algorithm to smooth the overall PAD value over time, with the smoothing coefficient adaptively adjusted based on the user's interaction frequency. When the user interaction frequency is high (the interval between two adjacent interactions is less than 30 seconds), the smoothing coefficient is set to 0.3 to make the system response more sensitive; when the user interaction frequency is low, the smoothing coefficient is set to 0.8 to make the changes in emotional state more gradual. The smoothed PAD value is used to drive the visualization output of the emotion tree, ensuring that the visual presentation does not have abrupt jumps.

[0027] In this embodiment, through the processing of the PAD emotion modeling and mapping module, the system can transform the user's multimodal emotion signals into a unified three-dimensional PAD representation. The pleasure parameter reflects the user's emotional valence, with positive values ​​indicating positive emotions and negative values ​​indicating negative emotions; the arousal parameter reflects the user's emotional activation level, with high arousal indicating excitement or agitation and low arousal indicating calmness or depression; the dominance parameter reflects the user's degree of control over emotional expression, with high dominance indicating strong initiative in emotional expression and low dominance indicating passive expression. These three dimensions provide a complete description of the emotional state for the subsequent emotion tree visualization output.

[0028] Reference Figure 4 , Figure 4 This is a schematic diagram illustrating the multi-level interaction relationship and data flow between individual user interaction trees and a central emotion tree according to an embodiment of this application. Multiple individual user interaction trees and at least one central emotion tree are connected via a network to form a distributed emotion perception network. The multiple individual user interaction trees are respectively located in different positions within a public space. Each individual user interaction tree independently collects multimodal emotion signals from nearby users and uploads the preprocessed feature data to a central server.

[0029] In this embodiment, the distributed emotion perception network adopts a star topology. Each individual interaction tree acts as an independent perception node, establishing a communication connection with the central server via a wired network (Ethernet) or a wireless network (Wi-Fi, 5G). The central server runs a central data statistics and group emotion aggregation module, aggregating the feature data uploaded by all individual interaction trees to generate a public emotion profile, and then distributing the visualization parameters corresponding to the public emotion profile to all individual interaction trees and the central emotion tree.

[0030] The specific data flow process is as follows: First, each individual interaction tree, through its local multimodal signal acquisition module and signal preprocessing and feature extraction module, completes the acquisition, preprocessing, and feature extraction of user emotional signals, outputting an emotional feature vector. Each individual interaction tree packages the locally acquired emotional feature vector with a timestamp and location identifier, and uploads it to the central server in JSON format. The upload frequency is dynamically adjusted according to the actual interaction situation, uploading every 500 milliseconds when there is user interaction, and remaining in standby mode when there is no user interaction. After receiving the data uploaded by each individual interaction tree, the central server performs deduplication, verification, and format unification processing, and then stores it in a time-series database.

[0031] The central data statistics and group emotion aggregation module includes an emotion state recording unit, a group emotion summarization unit, and a public emotion profile generation unit. The emotion state recording unit stores the user identifier, interaction timestamp, PAD parameters, interaction duration, activation count, and signal quality marker for each interaction. Each interaction record is stored in encrypted form to protect user privacy; the encryption algorithm uses AES-256 symmetric encryption, and the key is uniformly managed by the central server security management module. The group emotion summarization unit performs aggregation operations on all anonymous emotion records within a specified time period to obtain a histogram of pleasure distribution, an arousal distribution histogram, a dominance distribution histogram, and time-series curves for pleasure, arousal, and dominance. The time window for aggregation operations can be set to minutes, hours, or days as needed. The public emotion profile generation unit generates a visualized group emotion heatmap based on the distribution histogram and time-series curves, using color intensity to represent the level of pleasure, the position of heatmap patches to represent the distribution of users in the public space, and a time axis to represent the trend of emotion changes.

[0032] Each individual interactive tree, driven by locally collected signals, uses its own canopy visuals and mechanical devices to express local emotions. The central emotion tree, based on a shared emotional profile, drives the overall canopy visuals and mechanical devices to express collective emotions. A hierarchical structure of emotional expression is formed between the individual interactive trees and the central emotion tree: the individual interactive trees respond to the emotional state of individual users, providing immediate one-to-one emotional feedback; the central emotion tree aggregates data from all individual interactive trees, presenting the overall emotional atmosphere of the public space. The visual representations of both are kept consistent through a unified PAD parameter mapping algorithm, allowing users to intuitively perceive the connection between personal and collective emotions.

[0033] Reference Figure 5 , Figure 5This is a schematic diagram of the framework of the emotion tree visualization output module according to an embodiment of this application. The emotion tree visualization output module includes a canopy visual generation unit and a mechanical device activity unit. It maps the canopy hue according to the comprehensive pleasure parameter, maps the canopy saturation according to the comprehensive arousal parameter, maps particle velocity and particle density according to the emotional activity level parameter, maps the node BPM according to the physiological arousal rhythm parameter, and controls the curvature, proportion, and bionic branch movement amplitude according to the PAD parameter to form mechanical posture changes that match the emotional state.

[0034] In this embodiment, the tree canopy visual generation unit of the emotion tree visualization output module dynamically generates the visual representation of the tree canopy based on the pleasure parameter, arousal parameter, and emotional activity level parameter. The hue mapping algorithm maps the pleasure parameter to the hue angle in the HSV color space. The pleasure parameter ranges from -1 to 1, and the mapped hue angle ranges from 0 degrees to 300 degrees. Specifically, when the pleasure parameter is 1 (highest positive emotion), the hue angle is mapped to 0 degrees (warm red); when the pleasure parameter is 0 (neutral emotion), the hue angle is mapped to 120 degrees (green); and when the pleasure parameter is -1 (highest negative emotion), the hue angle is mapped to 240 degrees (cool blue). This mapping method ensures an intuitive visual association between emotional valence and color temperature. The saturation mapping algorithm maps the arousal parameter to the saturation value in the HSV color space. The arousal parameter ranges from 0 to 1, and the mapped saturation value ranges from 30% to 100%. High arousal corresponds to high saturation, making colors more vivid and intense; low arousal corresponds to low saturation, making colors more soft and elegant.

[0035] The particle system controller adjusts the emission rate and particle velocity of the particle emitters on the canopy surface based on an emotional activity level parameter. The emotional activity level parameter is calculated by combining arousal and dominance parameters, using the formula: emotional activity level equals 0.7 times arousal plus 0.3 times dominance. When the emotional activity level parameter is high (greater than 0.7), the particle emission rate is set to 200 particles per second, the initial particle velocity is set to 5 meters per second, and the particle color uses a highly saturated warm tone. When the emotional activity level parameter is low (less than 0.3), the particle emission rate is set to 20 particles per second, the initial particle velocity is set to 0.5 meters per second, and the particle color uses a low-saturation cool tone. The particle system's physics simulation uses a GPU-accelerated particle rendering engine, with each particle's lifespan set to 2 seconds. During movement, particles are affected by gravity and air resistance, creating natural falling and scattering effects.

[0036] The node BPM controller controls the blinking rhythm of the LED nodes on the tree canopy surface based on physiological arousal rhythm parameters, creating a visual rhythmic resonance with the user's average heart rate. The physiological arousal rhythm parameters are calculated from heart rate data collected by a PPG sensor, reflecting the user's heart rhythm in real time. The LED node array is distributed across the branches and trunks of the tree canopy, comprising 500 independently controllable LED nodes, each with a diameter of 5 mm and a spacing of 2 cm. The node BPM controller calculates the target blinking frequency based on the physiological arousal rhythm parameters and converts this frequency into a control signal sent to the LED driver circuit. When the user's average heart rate is 75 beats per minute, the LED nodes blink at a frequency of 75 times per minute, with a blinking pattern of full brightness for 100 milliseconds followed by full darkness for 700 milliseconds. This visual feedback of rhythmic resonance allows the user to visually perceive a representation of their own physiological state.

[0037] In this embodiment, the mechanical device of the emotion tree visualization output module includes a branch drive motor, a reducer, a linkage mechanism, and a bionic branch. The controller calculates the target branch curvature, target branch proportion, and target motion amplitude based on PAD parameters. The target branch curvature is related to the pleasure level parameter: the higher the pleasure level, the greater the branch curvature, exhibiting an upward stretching posture; the lower the pleasure level, the smaller the branch curvature, exhibiting a downward convergent posture. The target branch proportion is related to the arousal level parameter: at high arousal, the branch proportion is set to 1.2 times the baseline length, exhibiting an expanding posture; at low arousal, the branch proportion is set to 0.8 times the baseline length, exhibiting a contracting posture. The target motion amplitude is related to the dominance level parameter: at high dominance, the motion amplitude is set to ±30 degrees, exhibiting active posture changes; at low dominance, the motion amplitude is set to ±5 degrees, exhibiting passive slight swaying.

[0038] The drive motor is a brushless DC motor with a rated power of 50 watts and a rated speed of 3000 rpm. The motor drive signal is generated by a PID control algorithm, with the proportional coefficient of the PID controller set to Kp = 10, the integral coefficient to Ki = 0.5, and the derivative coefficient to Kd = 2. The motor drives the linkage mechanism through a reducer with a reduction ratio of 30:1, converting the high-speed rotation of the motor into a low-speed, high-torque output. The linkage mechanism adopts a four-bar design, including a driving link, a driven link, a connecting link, and a rocker arm. The rotation of the driving link drives the driven link to reciprocate. The linkage mechanism drives the biomimetic branches to bend and extend. The biomimetic branches are made of PLA biomimetic material and formed using 3D printing technology, resulting in a lightweight and highly malleable structure. The activity of the mechanical device and the visual expression of the tree crown create multi-channel emotional feedback, allowing users to experience the expression of group emotions on visual, tactile, and psychological levels.

[0039] In one embodiment, this application also provides a specific implementation of an interactive control and feedback closed-loop unit in an emotion tree interactive system for public emotion visualization. The interactive control and feedback closed-loop unit includes a proximity detection subunit, an embrace judgment subunit, and a feedback activation subunit.

[0040] The proximity detection subunit employs a fusion detection method combining an infrared ranging sensor and an ultrasonic ranging sensor. The infrared ranging sensor has a ranging range of 0.1 to 5 meters, a ranging accuracy of ±2 centimeters, and a response time of 50 milliseconds; the ultrasonic ranging sensor has a ranging range of 0.2 to 8 meters, a ranging accuracy of ±1 centimeter, and a response time of 100 milliseconds. The fusion detection algorithm uses a weighted average method, with the infrared ranging sensor having a weight of 0.4 and the ultrasonic ranging sensor having a weight of 0.6. When the fused distance value is within the range of 0.5 to 1.5 meters, the system is triggered to enter a ready state, and an interactive prompt signal is sent to the user.

[0041] The hugging judgment subunit determines whether the user has initiated a hugging action based on the output signal of the pressure sensing unit. The judgment algorithm sets two threshold conditions: first, the peak pressure signal exceeds a preset pressure threshold of 10 Newtons per square centimeter; second, the pressure duration exceeds a preset duration of 500 milliseconds. When both conditions are met simultaneously, the system determines that the hugging action is valid and sends a hugging confirmation signal to the feedback activation subunit.

[0042] The feedback activation subunit activates the corresponding emotion tree visualization output based on the strength, duration, and rhythmic variations of the hug. The strength is mapped to the offset of the tree crown hue; the greater the strength, the warmer the hue. The duration is mapped to the continuous emission time of the particle system; the longer the duration, the more persistent the particle effect. The rhythmic variations are mapped to the movement frequency of the mechanical device; the faster the rhythm, the more intense the movement. Through this multi-dimensional feedback mechanism, the system creates a real-time, two-way emotional interaction experience, allowing users to intuitively feel their emotional state being recognized and responded to.

[0043] In one embodiment, this application also provides a specific implementation of a privacy protection module in an interactive emotion tree system for public emotion visualization. The system further includes a privacy protection module for anonymizing the collected signals before storage and transmission.

[0044] The privacy protection module removes all personally identifiable information during the signal preprocessing stage, retaining only the emotion feature vector, timestamp, and signal quality marker. The specific anonymization process is as follows: First, face images detected by the face detection algorithm are not stored; only expression feature parameters are extracted, and the original image data is immediately discarded. Then, stress signals and physiological signals are de-identified, removing any metadata related to the user's identity. Finally, only the extracted emotion feature vector (128 dimensions) is retained; this feature vector cannot be used to infer the original physiological signals or facial image.

[0045] The emotional feature vectors undergo dimensionality reduction and quantization encoding to compress high-dimensional emotional features into low-dimensional anonymous feature representations. Dimensionality reduction employs Principal Component Analysis (PCA) to reduce the 128-dimensional feature vector to 32 dimensions while retaining over 95% of the information variance. Quantization encoding converts the 32-dimensional floating-point feature vector into an 8-bit integer vector with 256 quantization levels using uniform quantization. The feature representations after dimensionality reduction and quantization can no longer be reconstructed from the original signal or identify the individual.

[0046] All stored and transmitted data is anonymized to ensure that individual identities cannot be deduced from the data. The data stored on the central server only contains anonymized emotion feature vectors, timestamps, signal quality markers, and location identifiers, and does not contain any information that can directly identify the user. Data transmission uses the TLS 1.3 encryption protocol to ensure that data is not eavesdropped on or tampered with during transmission. Through this multi-layered privacy protection mechanism, the system fully protects the user's privacy rights while realizing emotion perception functionality.

[0047] Thus, this application, in its emotion tree interactive system for public emotion visualization, achieves non-invasive emotion perception and visualization of a massive number of users in public spaces through a technical solution combining multimodal signal acquisition and PAD emotion modeling. By fusing data acquisition from tactile pressure sensors, cameras, PPG heart rate waveform sensors, and EDG skin conductance sensors, the psychological barrier to user participation is lowered, making emotion expression a natural behavioral feedback rather than a deliberate data collection behavior. The use of multimodal weighted fusion and time-smooth update processing of pressure signals, facial expression signals, and physiological signals improves the robustness and reliability of emotion recognition results. Through the coordinated control of the canopy visual generation unit and the mechanical device activity unit, abstract emotion data is transformed into intuitive color changes, particle movements, and mechanical shape changes. The hierarchical linkage design between the individual interactive tree and the central emotion tree enables multi-scale emotion visualization, from individual emotion expression to group emotion aggregation. A privacy protection module anonymizes all collected data, retaining only emotion feature vectors and not storing personally identifiable information, thus protecting user privacy while achieving emotion perception functionality.

[0048] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0049] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0054] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0055] The above provides a detailed description of the emotion tree interactive system for visualizing public emotions provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An interactive emotion tree system for visualizing public emotions, characterized in that, include: The multimodal signal acquisition module is used to acquire the user's multimodal emotional signals; The signal preprocessing and feature extraction module, connected to the multimodal signal acquisition module, is used to perform time synchronization, noise reduction filtering, and feature extraction on the acquired signals; the PAD emotion modeling and mapping module, connected to the signal preprocessing module, is used to calculate and weightedly fuse PAD parameters based on the extracted features to obtain a comprehensive PAD value; the central data statistics and group emotion aggregation module, connected to the PAD emotion modeling module, is used to aggregate PAD parameters from multiple users to form a public emotion profile. The emotion tree visualization output module is connected to the PAD emotion modeling module and the central data statistics module, respectively, and is used to generate the visual representation of the tree crown and drive the mechanical device action according to the PAD parameter mapping; the interactive control and feedback closed loop unit is connected to the multimodal signal acquisition module and the emotion tree visualization output module, respectively, and is used to detect the user's approach, judge the hugging action and trigger the corresponding visual feedback.

2. The emotion tree interaction system according to claim 1, characterized in that, The multimodal signal acquisition module includes: a pressure sensing unit deployed on the tree trunk surface for acquiring pressure signals when a user hugs the tree trunk; an expression capture unit for automatically activating and acquiring facial expression images when the distance between the user and the device is 0.5 to 1.5 meters; and a physiological signal acquisition unit for acquiring heart rate variability signals and skin conductance response signals.

3. The emotion tree interactive system according to claim 2, characterized in that, The pressure sensing unit includes multiple tactile pressure sensor arrays distributed at different heights and circumferential angles on the tree trunk to collect spatially distributed pressure signals. The output signals of the sensor arrays are amplified by an amplifier circuit and then digitized by an analog-to-digital converter. The digital signal processor performs sliding window feature extraction on the digitized pressure signals to obtain peak pressure, pressure variance, pressure rise rate, and pressure duration.

4. The emotion tree interaction system according to claim 2, characterized in that, The physiological signal acquisition unit includes a PPG heart rate waveform sensor and an EDG skin conductance response sensor, which are used to acquire the user's pulse wave signal and skin conductance change signal, respectively. The heart rate variability analysis module performs R-peak detection and interval calculation on the pulse wave signal to obtain heart rate variability index.

5. The emotion tree interactive system according to claim 1, characterized in that, The signal preprocessing and feature extraction module includes: a time synchronization unit that uses a unified clock source to assign a unified timestamp to each signal; a noise reduction and filtering unit that uses an adaptive filtering algorithm to perform high-pass filtering on pressure signals and notch filtering on physiological signals; a baseline correction unit that uses the initial state when the user first comes into contact with the device as a baseline reference; and a feature extraction unit that extracts statistical features including mean, variance, peak value, kurtosis, and skewness, as well as time-series features including autocorrelation function and Fourier spectrum features.

6. The emotion tree interaction system according to claim 1, characterized in that, The PAD emotion modeling and mapping module includes: a single-modal parameter calculation unit, which establishes mapping models between stress signals, facial expression signals, physiological signals and PAD parameters respectively; a multimodal weighted fusion unit, which performs weighted fitting based on the confidence level of each modality signal to obtain a comprehensive PAD value; and a time-smoothing update unit, which uses an exponential moving average algorithm to perform time-smoothing processing on the comprehensive PAD value.

7. The emotion tree interaction system according to claim 1, characterized in that, The emotion tree visualization output module includes a canopy visual generation unit and a mechanical device activity unit; the canopy visual generation unit maps the canopy hue according to the pleasure parameter, maps the saturation according to the arousal parameter, controls the particle speed and particle density according to the emotional activity parameter, and controls the LED node flashing rhythm according to the physiological arousal rhythm parameter; the mechanical device activity unit controls the tree branch curvature and the movement amplitude of the bionic branches according to the PAD parameter.

8. The emotion tree interaction system according to claim 7, characterized in that, The mechanical device's moving unit includes a tree branch drive motor, a reducer, a linkage mechanism, and a bionic branch; the controller calculates the curvature and motion amplitude of the target tree branch based on PAD parameters, generates a motor drive signal through a PID control algorithm, and drives the motor to drive the linkage mechanism through the reducer to move the bionic branch in bending and extending motion.

9. The emotion tree interactive system according to claim 1, characterized in that, Multiple individual interaction trees and at least one central emotion tree are connected via a network to form a distributed emotion perception network. Each individual interaction tree independently collects multimodal emotion signals from nearby users and uploads feature data to the central server. The central server runs the central data statistics and group emotion aggregation module, which aggregates the feature data uploaded by all individual interaction trees to generate a public emotion profile and distributes visualization parameters to all individual interaction trees and the central emotion tree. Individual interaction trees drive local emotion expression based on local signals, while the central emotion tree drives overall emotion expression based on the public emotion profile, forming a hierarchical emotion expression structure.

10. The emotion tree interaction system according to claim 1, characterized in that, It also includes a privacy protection module for anonymizing the collected signals; the privacy protection module removes all personally identifiable information during the signal preprocessing stage, retaining only the emotion feature vector, timestamp, and signal quality marker; The emotion feature vector is processed by dimensionality reduction and quantization encoding to compress high-dimensional emotion features into low-dimensional anonymous feature representations.

Images