A desk lamp adaptive adjustment system based on user behavior analysis

Abstract

This invention relates to the field of intelligent lighting control technology, specifically to an adaptive adjustment system for a desk lamp based on user behavior analysis. The system comprises a multi-source sensing device consisting of an image acquisition unit, an ambient light sensor, an infrared proximity sensor, and a human activity detector deployed around the desk lamp and the user's work environment to collect raw sensing data. Multimodal fusion processing is performed on the raw sensing data to generate a fused behavior sensing vector. This vector is then input into a behavior state decoder equipped with an improved Hidden Markov Model. The state transition probabilities are adjusted based on the temporal continuity constraints of the user's task, outputting a discrete label sequence of user behavior states. This label sequence is then matched with environmental parameters in a lighting demand knowledge base. Based on the matching results, a lighting adjustment target containing target color temperature and target illuminance values ​​is generated, achieving adaptive intelligent adjustment of the desk lamp's illumination.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting control technology, and in particular to an adaptive adjustment system for a desk lamp based on user behavior analysis. Background Technology

[0002] Traditional desk lamps rely heavily on fixed brightness settings or a single ambient light detection method for light adjustment. At the sensing level, they use only a single type of sensor to collect ambient light parameters, failing to deploy diverse and heterogeneous sensing devices within the lamp itself and its surrounding environment. This prevents the simultaneous acquisition of image information, ambient light parameters, human proximity signals, and human activity information. The limited data types and lack of a mechanism for simultaneously acquiring multi-dimensional raw sensing data make it difficult to comprehensively capture the user's actual behavioral state and environmental changes during lamp use.

[0003] Single-sensor data cannot achieve integrated processing of multimodal information; it can only perform simple brightness adjustment based on ambient light values ​​and lacks the ability to fuse multi-source sensor data and construct behavioral representation vectors. Conventional control architectures do not incorporate Hidden Markov Models to construct behavioral decoding modules, and cannot adapt and adjust model state transition probabilities in conjunction with user task temporal continuity constraints, making it difficult to output standardized discrete label sequences of user behavioral states.

[0004] The existing desk lamp control mode lacks a fixed knowledge base architecture for lighting needs, making it impossible to associate and match user behavior status tags with environmental parameters. This makes it difficult to match ideal lighting modes to different behavioral scenarios, and also makes it impossible to automatically generate precise lighting adjustment configurations that include color temperature and illuminance parameters. Consequently, it is difficult to adapt to the personalized lighting needs of different user behavior scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a desk lamp adaptive adjustment system based on user behavior analysis.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a desk lamp adaptive adjustment system based on user behavior analysis, comprising: A multi-source sensing module is used to deploy multi-source sensing devices on a desk lamp and around the user's work environment. The multi-source sensing devices include an image acquisition unit, an ambient light sensor, an infrared proximity sensor, and a human activity detector, which are used to collect raw sensing data. The data fusion module is used to perform multimodal fusion processing on the original sensing data to generate a fused behavior sensing vector; The behavior decoding module is used to input the fused behavior perception vector into the behavior state decoder based on the improved Hidden Markov Model. The improved Hidden Markov Model introduces a temporal continuity constraint matrix on its standard state transition probability matrix. The temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element by element to obtain the modified state transition probability matrix. The input fused behavior perception vector sequence is decoded based on the modified state transition probability matrix, and the discrete label sequence corresponding to the user behavior state is output. The demand matching module is used to input the discrete label sequence of the user's behavior state and the environmental parameters into a preset lighting demand knowledge base for retrieval and matching. The lighting demand knowledge base stores the ideal lighting modes corresponding to various behavior states under different environmental parameters. The target generation module is used to generate illumination adjustment targets containing target color temperature values ​​and target illuminance values ​​based on the search and matching results.

[0007] As a further aspect of the present invention, the original sensing data is subjected to multimodal fusion processing to generate a fused behavior sensing vector, including: The fused behavior perception vector includes user posture features, facial orientation features, eye state features, desktop illuminance features, and ambient light distribution features. The system processes the data from the image acquisition device, determines the user's facial position and orientation through facial key point detection, evaluates eye opening and gaze point changes through eye analysis algorithms, and outputs facial orientation features and eye state features. The data from the human activity detector is processed and combined with the skeletal key point information to determine whether the user's upper body posture is leaning forward, leaning back, or standing upright, and the user's posture features are output. The system processes the data from the ambient light sensor to obtain the intensity and color temperature of the ambient light, calculates the average illuminance on the desktop, and outputs the ambient light distribution characteristics and desktop illuminance characteristics. A unified timestamp is established for each time point, and the facial orientation features, eye state features, user posture features, ambient light distribution features, and desktop illuminance features are aligned to the unified timestamp. All aligned features are concatenated into a multidimensional vector in a predetermined order, and the multidimensional vector is standardized to obtain the fused behavior perception vector.

[0008] As a further aspect of the present invention, the improved Hidden Markov Model adjusts its state transition probabilities based on the temporal continuity constraints of the user task, including: Define a set of hidden states, each corresponding to a different typical user behavior state, including focused reading, writing at a desk, taking intermittent breaks, leaving the seat, and looking up into the distance. Define the observation state as the quantized value of key features extracted from the fused behavior perception vector; Based on the standard state transition probability matrix of the improved Hidden Markov Model, a temporal continuity constraint matrix is ​​introduced. The temporal continuity constraint matrix is ​​set according to the prior knowledge of the task scenario, and is used to increase the transition probability between temporally continuous and task-related behavioral states, while reducing the transition probability between unrelated or anti-logical states. During the decoding process, the temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element by element to obtain the corrected state transition probability matrix. Using the Viterbi algorithm, based on the corrected state transition probability matrix, the observed state probability distribution, and the initial state probability, the input fused behavior perception vector sequence is decoded, and the hidden state sequence, i.e., the discrete label sequence of the user behavior state, is output.

[0009] As a further aspect of the present invention, the retrieval and matching process of the illumination demand knowledge base includes: The lighting demand knowledge base is organized in tabular form, with its row index being different user behavior status labels and its column index being different quantification ranges of environmental parameters. Each cell stores a tuple of lighting patterns consisting of ideal color temperature and ideal illuminance values; The latest label in the discrete label sequence of the user's behavior state is used as the index of the current behavior state; The current environmental parameters, including ambient light intensity and color temperature, are quantized into the corresponding environmental parameter range and used as column indexes. Based on the current behavior state index and column index, locate the corresponding cell in the table of the lighting demand knowledge base; The illumination pattern tuple stored in the corresponding cell is read as the basis for generating the illumination adjustment target.

[0010] As a further aspect of the present invention, the system further includes: The adjustment calculation module calculates the initial color temperature adjustment amount and illuminance adjustment amount based on the difference between the light adjustment target and the current light parameters of the desk lamp. The smoothing control module is used to input the initial color temperature adjustment amount and illuminance adjustment amount to the light change smoothing controller. The light change smoothing controller performs rate planning for the adjustment process based on the user's sensitivity threshold to light changes, and generates a smooth color temperature control command sequence and illuminance control command sequence. The execution drive module is used to send the smooth color temperature control command sequence and illuminance control command sequence to the execution drive circuit of the desk lamp, so as to drive the light-emitting component to realize the adaptive adjustment of the light parameters. The feedback optimization module is used to continuously collect feedback data through the multi-source sensing device during the adjustment process, verify the adjustment effect and user comfort, and dynamically update and optimize the mapping relationship in the light demand knowledge base based on the verification results.

[0011] As a further aspect of the present invention, the illumination change smoothing controller performs rate planning for the adjustment process based on the user's sensitivity threshold to illumination changes, including: Obtain a pre-calibrated user illumination change sensitivity threshold, wherein the sensitivity threshold defines the maximum allowable range of color temperature and illuminance change per unit time; Divide the initial color temperature adjustment amount and illuminance adjustment amount by the corresponding sensitivity threshold to obtain the number of adjustment time steps required under unconstrained conditions; Obtain the adjustment time step value corresponding to color temperature adjustment and the adjustment time step value corresponding to illuminance adjustment respectively, and take the maximum value of the two step values ​​as the common time step number for the final adjustment; The initial color temperature adjustment and illuminance adjustment are evenly distributed to each step of the common time step, and the color temperature adjustment increment and illuminance adjustment increment of each step are calculated. In chronological order, color temperature control instructions and illuminance control instructions for each step are generated sequentially and combined to form the smooth color temperature control instruction sequence and illuminance control instruction sequence.

[0012] As a further aspect of the present invention, during the adjustment process, feedback data is continuously collected through the multi-source sensing device to verify the adjustment effect and user comfort, including: After the desk lamp begins executing the smooth color temperature control command sequence and illuminance control command sequence, it continuously collects a new round of raw sensory data. Process the new round of raw sensory data to obtain the adjusted fused behavior perception vector; Extract the actual desktop illuminance features and the desk lamp illuminance features separated from the ambient light from the fused behavior perception vector; The actual desktop illuminance characteristics are compared with the target illuminance value to calculate the illuminance achievement index; By analyzing the adjusted facial orientation features and eye state features, combined with a preset comfort model, the user's visual comfort index is evaluated. By combining the aforementioned illuminance achievement index and user visual comfort index, a verification result of the adjustment effect is generated.

[0013] As a further aspect of the present invention, the mapping relationships in the illumination demand knowledge base are dynamically updated and optimized based on the verification results, including: When the results of the adjustment effect verification show that the illuminance achievement is high and the user visual comfort index is good, it is confirmed that the illuminance pattern tuple retrieved from the illuminance demand knowledge base is effective and its confidence weight is increased. When the adjustment effect verification results show that the illuminance achievement is low or the user visual comfort index is poor, it is determined that the currently retrieved illuminance mode tuple does not match the user's actual needs, and the optimization process is initiated. In the optimization process, a corrected lighting mode is calculated in reverse based on the actual user behavior and environmental parameters, combined with the adjusted and measured user visual comfort index. The modified lighting pattern is weighted and fused with the original lighting pattern tuple to update the content of the corresponding cell in the lighting requirement knowledge base, so as to adapt to the personalized preferences of specific users.

[0014] As a further aspect of the present invention, the modified lighting pattern is weighted and fused with the original lighting pattern tuple to update the content of the corresponding cell in the lighting requirement knowledge base, in order to adapt to the personalized preferences of a specific user, including the following steps: Calculate the first confidence weight of the original illumination pattern tuple; Calculate the second confidence weight of the corrected illumination pattern; Based on the first confidence weight and the second confidence weight, calculate the weighted fusion coefficients corresponding to the ideal color temperature value and ideal illuminance value in the original illumination mode tuple and the corrected color temperature value and corrected illuminance value in the corrected illumination mode, respectively. Using the weighted fusion coefficient, the ideal color temperature value in the original lighting mode tuple and the corrected color temperature value in the corrected lighting mode are weighted and averaged to calculate the updated color temperature value. Using the weighted fusion coefficient, the ideal illuminance value in the original illumination mode tuple and the corrected illuminance value in the corrected illumination mode are weighted and averaged to calculate the updated illuminance value. The updated color temperature value and the updated illuminance value are combined to form an updated illumination mode tuple; The updated lighting pattern tuple replaces the previously stored original lighting pattern tuple in the corresponding cell of the lighting requirement knowledge base, thus completing a personalized update of the mapping relationship.

[0015] As a further aspect of the present invention, in the process of driving the light-emitting component to achieve adaptive adjustment of illumination parameters, dynamic compensation for ambient light disturbances is also included: The system monitors the data from the ambient light sensor in real time to detect any sudden changes in ambient light. When a sudden change in ambient light is detected, the currently executing sequence of smooth control commands is immediately interrupted; Based on the changed ambient lighting parameters and combined with the latest user behavior status, new lighting adjustment targets are retrieved and generated again from the lighting demand knowledge base; Based on the new lighting adjustment targets, a new lighting adjustment process is recalculated and initiated to quickly counteract the impact of ambient light disturbances on the user's visual experience.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Multiple sensing devices are deployed on the desk lamp itself and in the user's working environment to build a multi-dimensional information acquisition architecture, simultaneously acquiring raw sensing data related to images, ambient light, infrared sensing, and human activity. Multimodal fusion processing is performed on these heterogeneous raw sensing data to construct a unified fused behavioral perception vector. This broadens the dimensions of information acquisition related to user behavior and environmental status, expands the sources of information for behavioral feature representation, overcomes the limitations of a single-sensor acquisition mode with its limited information dimensions, and improves the data infrastructure for behavioral analysis and recognition.

[0017] A behavior state decoder is built based on an improved Hidden Markov Model. The internal state transition probabilities of the model are adjusted according to the temporal continuity constraints of the user task to complete the logical parsing of continuous behavior features, outputting a regularized discrete label sequence of user behavior states. Hierarchical parsing of behavior states is completed based on the model's own temporal constraint rules, establishing a mapping relationship between continuous behavior states and discrete labels, achieving a standardized abstract expression of user behavior states over continuous time.

[0018] The discrete tag sequence of user behavior states and environmental parameters are jointly accessed into a pre-defined lighting requirement knowledge base for retrieval and matching. Based on the ideal lighting patterns associated with various behavior states and environmental parameters stored in the knowledge base, the system automatically associates scenes with lighting parameters. According to the matching results, a lighting adjustment target containing target color temperature and target illuminance values ​​is generated. This establishes a logical correspondence between user behavior states, environmental conditions, and lighting parameters, forming a complete and coherent autonomous control chain that adapts to the autonomous adjustment of lighting parameters under different usage scenarios. Attached Figure Description

[0019] Figure 1 This is a timing diagram of a desk lamp adaptive adjustment system based on user behavior analysis as described in this invention. Figure 2 A flowchart illustrating the workflow for generating fused behavior perception vectors by performing multimodal fusion processing on raw sensing data; Figure 3A flowchart illustrating the process of adjusting state transition probabilities for an improved Hidden Markov Model. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] See Figure 1 This invention provides a desk lamp adaptive adjustment system based on user behavior analysis, the overall implementation scheme of which is as follows: A multi-source sensing module deploys multi-source sensing devices on the desk lamp and around the user's work environment. These devices include an image acquisition unit, an ambient light sensor, an infrared proximity sensor, and a human activity detector. The module's function is to collect raw sensing data. A data fusion module receives the raw sensing data and performs multimodal fusion processing to generate a fused behavior sensing vector. A behavior decoding module inputs the fused behavior sensing vector into a behavior state decoder based on an improved Hidden Markov Model (HMM). This improved HMM adjusts its state transition probabilities according to the temporal continuity constraints of the user's task. The decoder outputs a discrete label sequence corresponding to the user's behavior state. A demand matching module inputs the discrete label sequence of the user's behavior state along with environmental parameters into a preset lighting demand knowledge base for retrieval and matching. This knowledge base stores ideal lighting patterns corresponding to various behavior states under different environmental parameters. A target generation module generates a lighting adjustment target containing the target color temperature and target illuminance values ​​based on the retrieval and matching results.

[0023] In one embodiment of the present invention, see [reference] Figure 2The fused behavior perception vector includes user posture features, facial orientation features, eye state features, desktop illuminance features, and ambient light distribution features. It processes data from the image acquisition unit, determines the user's facial position and orientation through facial keypoint detection, and evaluates eye opening and gaze point changes through eye analysis algorithms, outputting facial orientation features and eye state features. It processes data from the human activity detector, combining skeletal keypoint information to determine whether the user's upper body posture is forward-leaning, backward-leaning, or upright, outputting user posture features. It processes data from the ambient light sensor, acquiring the intensity and color temperature of ambient light, and calculates the average desktop illuminance, outputting ambient light distribution features and desktop illuminance features. A unified timestamp is established for each time point, aligning the facial orientation features, eye state features, user posture features, ambient light distribution features, and desktop illuminance features to this unified timestamp. All aligned features are concatenated into a multi-dimensional vector in a predetermined order, and this multi-dimensional vector is standardized to obtain the fused behavior perception vector.

[0024] In practical implementation, the multi-source sensing module, comprising an image acquisition unit, an ambient light sensor, an infrared proximity sensor, and a human activity detector, works collaboratively to collect multimodal signals constituting the raw sensing data. When processing data from the image acquisition unit, based on computer vision algorithms, the facial key point detection algorithm first locates the face region and outputs a series of two-dimensional coordinates of facial key points. Based on these coordinates, a facial orientation feature vector representing the projection of the facial normal direction into three-dimensional space is calculated. The eye analysis algorithm calculates the ratio of distances between eyelid key points to determine the degree of eye opening, representing the level of alertness, and obtains the gaze point change trajectory by tracking the sequence of changes in the relative position of the pupils, collectively constituting the eye state features. In practical implementation, the video stream or depth map data output by the human activity detector is input into the pose estimation algorithm. The algorithm extracts key point information from the human skeleton, calculates the angle between the line connecting the spinal key points and the vertical direction, and compares it with a preset threshold to determine whether the user's upper body posture is forward, backward, or upright. This determination result is encoded into discrete numerical labels, which are output as user pose features. In some embodiments, the ambient light sensor collects the raw intensity and spectral data of ambient light, and calculates the absolute light intensity value in lux and the color temperature value in Kelvin through a built-in algorithm or an external microcontroller. These two values ​​together constitute the ambient light distribution characteristics. At the same time, by analyzing the desktop area image captured by the image acquisition device or combining the readings of a photoelectric sensor with a specific orientation, the average illuminance value of the desktop area is calculated as the desktop illuminance characteristic.

[0025] In practical implementation, establishing a unified timestamp for each time point is fundamental to achieving multimodal data alignment. The data fusion module maintains a global clock, and every frame of data collected by all sensing devices is stamped with the same timestamp based on that current global clock when it enters the module. In practice, facial orientation features, eye state features, user posture features, ambient light distribution features, and desktop illuminance features aligned to the unified timestamp are concatenated in a predetermined order. For example, a sequence containing [user posture feature encoding, facial orientation yaw angle, facial orientation pitch angle, eye opening / closing, gaze point x-coordinate, gaze point y-coordinate, ambient light intensity, ambient light color temperature, average desktop illuminance] is concatenated into a nine-dimensional multi-dimensional vector. This concatenation order is predefined and fixed to ensure the consistency of the feature vector structure. In practice, the concatenated multi-dimensional vector undergoes standardization to eliminate the influence of differences in feature units and numerical ranges. The standardization process employs a method based on historical data statistics. For a feature dataset containing n samples, the mean and standard deviation for each feature dimension are calculated, and the newly generated multidimensional vectors are then standardized using z-scores. This process can be expressed by the following formula: in: This indicates that the input multidimensional vector is at the th... The original values ​​in each feature dimension Represents the pre-calculated first... The average of each feature dimension across all historical samples Represents the pre-calculated first... The standard deviation of each feature dimension across all historical samples Then it means that the standardized multidimensional vector is at the th... The results are obtained in each feature dimension. The nine-dimensional vector after this standardization process is the final fused behavior perception vector. In some embodiments, the human activity detector and the image acquisition device can be designed as a single hardware unit, such as a depth camera, which simultaneously provides RGB images, depth information, and a skeleton keypoint data stream calculated in real time based on this information. In this case, the data from the human activity detector and the image acquisition device have a synchronized timestamp basis at the hardware level. Optionally, for acquiring desktop illuminance features, in addition to image analysis, a miniature illuminance sensor facing the desktop can be integrated into the lamp head to directly read the desktop illuminance value as the desktop illuminance feature. It can be understood that regardless of the specific sensor combination and data acquisition method used, the ultimate goal is to generate a standardized, multi-dimensional, and time-aligned fused behavior perception vector for use by the subsequent behavior decoding module. During the standardization process of the multi-dimensional vector in the data fusion module, the mean and standard deviation of each feature dimension need to be calculated in advance based on historical sample data. When a certain feature dimension takes the same value in all historical samples, the standard deviation of that dimension is zero. In this boundary case, a zero denominator in the standardization formula will cause an aberration in the calculation, thus causing the fused behavior perception vector to fail. To address this boundary case, the system performs a check step after calculating the standard deviation of each dimension: if the standard deviation of a certain feature dimension is zero, it is determined that this dimension has no variability in the current sample set and does not provide effective information for distinguishing subsequent behavioral states. In this case, no division operation is performed, and the standardized result corresponding to that dimension is directly assigned to zero. Simultaneously, during the system initialization phase, calibration data covering various user behavioral states and lighting environments is collected to ensure that each feature dimension has sufficient sample variability, reducing the possibility of zero standard deviation from the data source.

[0026] In one embodiment of the present invention, see [reference] Figure 3A set of hidden states is defined, each corresponding to a different typical user behavior state, including focused reading, writing at a desk, taking a break, leaving the seat, and looking into the distance. Observational states are defined as quantized values ​​of key features extracted from the fused behavior perception vector. Based on the standard state transition probability matrix of the improved Hidden Markov Model, a temporal continuity constraint matrix is ​​introduced. This matrix is ​​set according to prior knowledge of the task scenario and is used to increase the transition probability between temporally continuous and task-related behavior states, while reducing the transition probability between unrelated or logically inverse states. During decoding, the temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element-wise to obtain the modified state transition probability matrix. Using the Viterbi algorithm, based on the modified state transition probability matrix, the observation state probability distribution, and the initial state probability, the input fused behavior perception vector sequence is decoded, outputting a sequence of hidden states, i.e., a discrete label sequence of user behavior states.

[0027] In the specific implementation, the behavior state decoder based on the improved Hidden Markov Model in the behavior decoding module receives the fused behavior perception vector sequence from the data fusion module. In this implementation, a set of five hidden states is defined, corresponding to five typical user behavior states: focused reading, writing at a desk, taking a break, leaving the seat, and looking into the distance. In this implementation, an observation state is defined, which is the quantified value of key features extracted from the fused behavior perception vector according to preset rules. For example, user posture feature encoding, eye opening / closing value, and the sine of the facial tilt angle extracted from the fused behavior perception vector are used as components of the three-dimensional observation vector. In this implementation, the standard state transition probability matrix of the improved Hidden Markov Model is a 5x5 matrix, where each element represents the probability of transitioning from one hidden state to another. The standard state transition probability matrix can be obtained through statistical training on historical user behavior sequence data.

[0028] In practical implementation, the temporal continuity constraint matrix introduced on top of the standard state transition probability matrix is ​​also a 5x5 matrix. The temporal continuity constraint matrix is ​​set based on prior knowledge of the task scenario. Each element value in the matrix is ​​manually set or generated through rules based on an understanding of the behavioral logic in typical learning or office tasks. In practical implementation, the temporal continuity constraint matrix is ​​used to increase the transition probability between temporally continuous and task-related behavioral states. For example, the probability of transitioning from a "focused reading" state to a "looking up and gazing into the distance" state is set to a relatively high value to reflect the logic that the user might look up to rest after reading. In practical implementation, the temporal continuity constraint matrix also reduces the transition probability between unrelated or counter-logical states. For example, the probability of directly transitioning from a "focused reading" state to a "leaning from the seat" state is set to an extremely low value or zero, because users usually have transitional behaviors such as "looking up" or "leaning back" before leaving their seats. In some embodiments, the temporal continuity constraint matrix can be designed as a sparse matrix, where non-zero values ​​are only found at positions corresponding to strongly logically related state transitions, and zero values ​​are found at all other positions.

[0029] In practice, during each step of the decoding process, the temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element-wise. This operation can be represented by the following formula: in: This represents the standard state transition probability matrix. Represents the temporal continuity constraint matrix. Indicates from hidden state Transition to hidden state The standard probability, Represents hidden state Transition to hidden state The timing constraint coefficients, Indicates the modified hidden state Transition to hidden state The state transition probabilities. Element-wise multiplication can be understood as directly multiplying the values ​​at corresponding positions in two matrices to obtain the corrected state transition probability matrix. In some embodiments, to ensure the corrected state transition probability matrix... Each row of the matrix still represents a probability distribution. After element-wise multiplication, all elements in each row are normalized so that the sum of all elements in that row equals 1. This formula performs element-wise multiplication of the standard state transition probability matrix with the temporal continuity constraint matrix to obtain the modified state transition probability matrix. After element-wise multiplication, the sum of the elements in each row of the modified matrix may no longer equal one, failing to satisfy the normalization constraint of the probability distribution. Therefore, after the element-wise multiplication, a normalization operation is performed on each row of the modified matrix by dividing the value of each element in that row by the sum of all elements in that row, so that the modified matrix satisfies the transition matrix defined by the probability distribution.

[0030] In practical implementation, the corrected state transition probability matrix is ​​obtained. Then, combined with the observed state probability distribution and initial state probability vector The Viterbi algorithm is used to decode the input fused behavior-aware vector sequence. This can be understood as the probability distribution of the observed states. This describes the probability distribution of the observed state, representing the likelihood of observing a specific feature vector given a certain hidden state. The training data can be fitted using a Gaussian mixture model or an equal probability model. In practice, the Viterbi algorithm uses dynamic programming to find an optimal hidden state path that maximizes the probability of generating the observed fused behavior-aware vector sequence. The hidden state sequence output by the Viterbi algorithm is the discrete label sequence of the user's behavior state; for example, a possible output sequence might be ["Focused Reading", "Focused Reading", "Looking Up into the Distance", "Focused Reading"]. Optionally, in the backtracking path of the Viterbi algorithm, in addition to the optimal path, the top N paths with the highest probability values ​​can be retained to provide a certain degree of ambiguity tolerance for subsequent processing.

[0031] In one embodiment of the present invention, the illumination demand knowledge base is organized in tabular form, with row indices representing different user behavior state labels and column indices representing different quantization ranges of environmental parameters. Each cell stores a illumination pattern tuple consisting of an ideal color temperature value and an ideal illuminance value. The latest label in the discrete label sequence of user behavior states is used as the current behavior state index. The current environmental parameters, including ambient light intensity and color temperature, are quantized to the corresponding environmental parameter range and used as column indices. Based on the current behavior state index and the column index, the corresponding cell is located in the table of the illumination demand knowledge base. The illumination pattern tuple stored in the corresponding cell is read as the basis for generating illumination adjustment targets.

[0032] In practical implementation, the demand matching module receives discrete tag sequences of user behavior states from the behavior decoding module and current environmental parameters obtained from the ambient light sensor. These environmental parameters include ambient light intensity and ambient light color temperature. In practice, the lighting demand knowledge base is organized in tabular form in the system's non-volatile memory. The row indices of the lighting demand knowledge base table represent different user behavior state tags, and the column indices represent different quantization ranges of the environmental parameters. A typical lighting demand knowledge base is shown in Table 1.

[0033] Table 1: Knowledge Base Table of Light Requirements In specific implementations, each cell stores a tuple of lighting patterns consisting of ideal color temperature and ideal illuminance values. For example, the tuple (5500K, 500lx) represents a target color temperature of 5500 Kelvin and a target illuminance of 500 lux. In specific implementations, the latest label in the discrete label sequence of user behavior states is used as the current behavior state index. For example, if the discrete label sequence of user behavior states is [...], "Looking into the distance", "Focusing on reading"], then the latest "Focusing on reading" is taken as the current behavior state index. In some embodiments, if the system detects a rapid switch in the discrete label sequence of user behavior states within a very short time, the demand matching module can use a sliding window voting mechanism to select the behavior state with the highest frequency among the most recent N labels as the current behavior state index.

[0034] Table 1 divides the ambient illuminance range into four intervals: 0 to 100 lux, 101 to 300 lux, 301 to 500 lux, and 501 to 750 lux. The widths of each interval are 100 lux, 199 lux, 199 lux, and 249 lux, respectively, and are not equal. In practice, the system maintains a list of interval boundary values ​​corresponding to the column definitions in Table 1. The measured ambient illuminance values ​​are compared segment by segment with the lower and upper limits of each interval. When a measured value is above the lower limit of an interval but below its upper limit, the corresponding column number is used as the column index. When the measured ambient illuminance value is less than the lower limit of the first interval in Table 1, the system directly assigns the measured value to the first interval, 0 to 100 lux, using this as the column index to ensure that no measured value cannot be located due to being below the lower limit. Table 1 covers two environmental parameter dimensions: ambient illuminance and ambient color temperature. The ambient color temperature dimension is divided and mapped into independent intervals in the same way as ambient illuminance, generating a color temperature column index. The illuminance column index and the color temperature column index are combined into a composite column index key value, which is used to locate cells in the multidimensional table of the lighting demand knowledge base. Each interval uniformly follows the left-closed, right-open rule, that is, the interval includes the lower limit value but does not include the upper limit value. The critical value of 100 lux belongs to the second interval, i.e., the interval from 101 to 300 lux; the critical value of 300 lux belongs to the third interval, i.e., the interval from 301 to 500 lux; and the critical value of 500 lux belongs to the fourth interval, i.e., the interval from 501 to 750 lux.

[0035] In specific implementations, the corresponding cell is located in the table of the lighting demand knowledge base based on the current behavior state index and column index. Specifically, the current behavior state index determines the row coordinates, and the calculated column index determines the column coordinates; the row and column coordinates together uniquely identify a cell in the table. The lighting mode tuple stored in the corresponding cell is read; for example, based on the current behavior state index "Focused Reading" and the column index range "101-300", the tuple (5200K, 450lx) is located and read from Table 1. The read lighting mode tuple is directly passed to the target generation module as the basis for generating lighting adjustment targets. In some embodiments, if the located cell is empty or uninitialized, the system can use a predefined default lighting mode tuple, or interpolate from tuples in adjacent rows and columns to obtain a new lighting mode tuple. Optionally, the lighting demand knowledge base table can be stored in a multi-dimensional data structure, not limited to a two-dimensional table format. For example, it can be organized using a tree structure or key-value pairs. However, its core function remains key-value lookup based on behavior status labels and environmental parameter ranges. It can be understood that the specific values ​​in the lighting demand knowledge base table are preset at the system's factory and dynamically updated and optimized subsequently through the feedback optimization module.

[0036] In one embodiment of the present invention, the adjustment calculation module calculates the preliminary color temperature adjustment amount and illuminance adjustment amount based on the difference between the light adjustment target and the current light parameters of the desk lamp. The smoothing control module inputs the preliminary color temperature adjustment amount and illuminance adjustment amount to the light change smoothing controller. The light change smoothing controller performs rate planning on the adjustment process based on the user's sensitivity threshold to light changes, generating a smooth color temperature control command sequence and illuminance control command sequence. The specific rate planning includes obtaining a pre-calibrated user light change sensitivity threshold, which defines the maximum allowable amplitude of color temperature and illuminance changes per unit time. The preliminary color temperature adjustment amount and illuminance adjustment amount are divided by the corresponding sensitivity threshold to obtain the number of adjustment time steps required under unconstrained conditions. The adjustment time step values ​​corresponding to color temperature adjustment and illuminance adjustment are obtained respectively, and the maximum value of the two step values ​​is taken as the common time step for final adjustment. The preliminary color temperature adjustment amount and illuminance adjustment amount are evenly distributed to each step of the common time step, and the color temperature adjustment increment and illuminance adjustment increment of each step are calculated. Following a chronological order, color temperature control commands and illuminance control commands for each step are generated sequentially, forming a smooth sequence of color temperature control commands and illuminance control commands. The execution drive module sends these smooth sequence of color temperature control commands and illuminance control commands to the lamp's execution drive circuit, driving the light-emitting components to adaptively adjust the illumination parameters. The process of driving the light-emitting components to adaptively adjust the illumination parameters also includes dynamic compensation for ambient light disturbances. Data from the ambient light sensor is monitored in real time to detect sudden changes in ambient light. When a sudden change in ambient light is detected, the currently executing smooth control command sequence is immediately interrupted. Based on the changed ambient light parameters and the latest user behavior, a new illumination adjustment target is retrieved from the illumination demand knowledge base and generated. Based on the new illumination adjustment target, a new illumination adjustment process is recalculated and initiated. The feedback optimization module continuously collects feedback data through multi-source sensing devices during the adjustment process, verifies the adjustment effect and user comfort, and dynamically updates and optimizes the mapping relationship in the illumination demand knowledge base based on the verification results.

[0037] In practice, after the target generation module outputs the lighting adjustment target containing the target color temperature and target illuminance values, the adjustment calculation module begins operation. The adjustment calculation module receives the lighting adjustment target and reads the current color temperature and illuminance values ​​of the lamp's light-emitting components from the lamp's driver controller as the lamp's current lighting parameters. Specifically, the adjustment calculation module calculates the initial color temperature adjustment by subtracting the current color temperature from the target color temperature, and the initial illuminance adjustment by subtracting the current illuminance from the target illuminance. These two differences represent the absolute adjustment range required.

[0038] In practical implementation, the smoothing control module receives the initial color temperature and illuminance adjustment values ​​output by the adjustment calculation module. The light change smoothing controller stores pre-calibrated user light change sensitivity thresholds, which typically include color temperature and illuminance sensitivity thresholds. The color temperature sensitivity threshold defines the maximum Kelvin change in color temperature per unit time, and the illuminance sensitivity threshold defines the maximum lux change in illuminance per unit time. The light change smoothing controller performs rate planning for the adjustment process based on the user's sensitivity thresholds to light changes. In practical implementation, the first step of rate planning is to divide the absolute value of the initial color temperature adjustment by the color temperature sensitivity threshold to obtain the theoretical number of time steps required to complete the color temperature adjustment under unconstrained conditions, and to divide the absolute value of the initial illuminance adjustment by the illuminance sensitivity threshold to obtain the theoretical number of time steps required to complete the illuminance adjustment under unconstrained conditions. In practice, the maximum value between the theoretical time step value corresponding to color temperature adjustment and the theoretical time step value corresponding to illuminance adjustment is taken, and this maximum value is used as the common time step number for the final adjustment. This calculation can be expressed as: in: This indicates the initial color temperature adjustment amount. Indicates the threshold of sensitivity to color temperature changes. This indicates the initial illuminance adjustment amount. Indicates the sensitivity threshold to changes in illuminance, symbol This indicates the rounding up operation. This represents the calculated number of common time steps. In practice, the initial color temperature adjustment is evenly distributed across the common time steps. In each step, the color temperature adjustment increment is calculated, and the initial illuminance adjustment is evenly distributed across the common time steps. In each step, the illuminance adjustment increment is calculated. Color temperature control commands and illuminance control commands are generated sequentially according to time sequence. The color temperature control command value for each step is the current value plus the color temperature adjustment increment, and the illuminance control command value is the current value plus the illuminance adjustment increment. These commands combine to form a smooth sequence of color temperature control commands and illuminance control commands. It can be understood that by evenly distributing and executing these commands over multiple time steps, the rate of change of color temperature and illuminance is limited within the sensitivity threshold, achieving a smooth transition of illumination. During the rate planning process in the smooth control module, when the illumination adjustment target is completely consistent with the current illumination parameters of the desk lamp, the calculated initial color temperature adjustment and initial illuminance adjustment values ​​are both zero. Dividing these values ​​by the corresponding sensitivity threshold yields a theoretical number of zero time steps, and the common time step count after taking the maximum value is also zero. If the common time step count is zero and subsequent processes still attempt to evenly distribute and generate commands according to this number of steps, a division-by-zero exception or an empty command sequence error will occur. To address this scenario, a decision step is added after calculating the common time step: if the common time step is zero, it is determined that no lighting adjustment is needed at present. The smoothing control module does not generate any control command sequence and sends a no-operation signal to the execution drive module, keeping the desk lamp's current lighting parameters unchanged. Only when the common time step is greater than zero will the subsequent adjustment increment allocation calculation and command sequence generation steps continue to be executed, ensuring stable system operation when the adjustment amount is zero.

[0039] In practical implementation, the execution drive module receives a smooth sequence of color temperature control commands and an illuminance control command sequence. The execution drive module includes a microcontroller and a drive circuit. The microcontroller sequentially sends the color temperature control command value and illuminance control command value for each step to the execution drive circuit of the desk lamp via a pulse width modulation signal or a digital bus, according to the time intervals of the command sequence. In practical implementation, the execution drive circuit adjusts the current or voltage applied to the light-emitting component based on the received command values, thereby changing the driving current of the light-emitting diode or the brightness ratio of each color LED in the mixed light source. This drives the light-emitting component to achieve adaptive adjustment of the illumination parameters, gradually bringing the actual color temperature and illuminance of the desk lamp closer to the illumination adjustment target.

[0040] In specific implementations, during the adaptive adjustment of illumination parameters by driving the light-emitting components, the system simultaneously performs dynamic compensation for ambient light disturbances. The dynamic compensation mechanism monitors the data stream from the ambient light sensor in real time, detecting sudden changes in ambient light intensity and color temperature by analyzing real-time readings. In some embodiments, the criterion for detecting a sudden change is that the change in ambient light intensity within a shorter time window exceeds a preset threshold. In specific implementations, when a sudden change in ambient light is detected, the illumination change smoothing controller immediately interrupts the currently executing smoothing control command sequence, regardless of whether the sequence has been completed. In specific implementations, based on the altered ambient light parameters and the latest user behavior status output in real time by the behavior decoding module, the demand matching module retrieves and generates a new illumination adjustment target from the illumination demand knowledge base. In specific implementations, based on the new illumination adjustment target, the adjustment calculation module and the smoothing control module recalculate the initial adjustment amount and plan a new smoothing control command sequence, subsequently initiating a new illumination adjustment process by the execution driving module. Understandably, this process aims to quickly offset the impact of sudden changes in ambient light (such as the sudden switching on or off of indoor main lights, or changes in natural light caused by changes in cloud cover outside the window) on the user's visual experience. In some embodiments, to cope with frequent minor fluctuations in ambient light, the system can set a dead zone threshold, triggering a complete recalculation and adjustment process only when the amplitude of the ambient light change exceeds this dead zone threshold, in order to avoid unnecessary frequent adjustments.

[0041] Table 2: Thresholds for User Sensitivity to Light Changes In practical implementation, referring to Table 2, three preset user illumination change sensitivity threshold configurations are shown. The system can determine the applicable configuration mode based on user selection or through initial learning. In practical implementation, the illumination change smoothing controller reads the corresponding color temperature change sensitivity threshold and illuminance change sensitivity threshold from Table 1 for rate planning calculation. Optionally, the sensitivity threshold can be dynamically adaptive; the system fine-tunes the color temperature change sensitivity threshold and illuminance change sensitivity threshold based on user comfort data from historical adjustment processes. It can be understood that smooth control of the adjustment process, combined with dynamic compensation for sudden changes in ambient light, jointly ensures the comfort and stability of the illumination adjustment process.

[0042] In one embodiment of the present invention, after the desk lamp begins executing a smooth color temperature control command sequence and an illuminance control command sequence, a new round of raw perception data is continuously collected. The new round of raw perception data is processed to obtain an adjusted fused behavior perception vector. The actual desktop illuminance features and the desk lamp illuminance features separated from the ambient light are extracted from the adjusted fused behavior perception vector. The actual desktop illuminance features are compared with the target illuminance value to calculate the illuminance achievement index. By analyzing the adjusted facial orientation features and eye state features, combined with a preset comfort model, the user's visual comfort index is evaluated. The illuminance achievement index and the user's visual comfort index are combined to generate an adjustment effect verification result. When the adjustment effect verification result shows a high illuminance achievement and a good user visual comfort index, the current lighting pattern tuple retrieved from the lighting demand knowledge base is confirmed to be effective, and its confidence weight is increased. When the adjustment effect verification result shows a low illuminance achievement or a poor user visual comfort index, it is determined that the currently retrieved lighting pattern tuple does not match the user's actual needs, and an optimization process is initiated. In the optimization process, a corrected lighting mode is calculated by reverse engineering based on actual user behavior and environmental parameters, combined with the adjusted and measured user visual comfort index. The corrected lighting mode is then weighted and fused with the original lighting mode tuple, updating the corresponding cell in the lighting requirement knowledge base. The specific weighted fusion process includes calculating the first confidence weight of the original lighting mode tuple; calculating the second confidence weight of the corrected lighting mode; and, based on the first and second confidence weights, calculating the weighted fusion coefficients corresponding to the ideal color temperature and ideal illuminance values ​​in the original lighting mode tuple and the corrected color temperature and corrected illuminance values ​​in the corrected lighting mode, respectively. Using these weighted fusion coefficients, a weighted average is calculated between the ideal color temperature values ​​in the original lighting mode tuple and the corrected color temperature values ​​in the corrected lighting mode to obtain the updated color temperature value. Finally, using the weighted fusion coefficients, a weighted average is calculated between the ideal illuminance values ​​in the original lighting mode tuple and the corrected illuminance values ​​in the corrected lighting mode to obtain the updated illuminance value. The updated color temperature value is combined with the updated illuminance value to form an updated lighting pattern tuple. The updated lighting pattern tuple replaces the previously stored lighting pattern tuple in the corresponding cell of the lighting requirements knowledge base.

[0043] In implementation, the feedback optimization module initiates its workflow after the desk lamp begins executing a smooth sequence of color temperature and illuminance control commands, continuously collecting a new round of raw sensory data. This new round of raw sensory data is acquired by the multi-source sensing module within a time window after the adjustment action is executed. In implementation, the data fusion module processes this new round of raw sensory data to obtain an adjusted fused behavior perception vector. This vector includes user posture features, facial orientation features, eye state features, desktop illuminance features, and ambient light distribution features at the adjusted time point. In implementation, the actual desktop illuminance features are extracted from the adjusted fused behavior perception vector. These features are direct measurements or estimates of the current illuminance in the desktop area. In implementation, by subtracting the known light contribution from the current desk lamp source from the adjusted ambient light distribution features, the desk lamp illumination features can be separated from the ambient light. These features include the illuminance and color temperature components contributed by the desk lamp. In practical implementation, the actual desktop illuminance characteristics are compared with the target illuminance value. The ratio of the absolute difference between the two to the target illuminance value is calculated to obtain an illuminance achievement index between 0 and 1, or in percentage form. The calculation method is, for example: (1 - |actual illuminance - target illuminance| / target illuminance). In practical implementation, the user's visual comfort index is evaluated by analyzing the adjusted facial orientation and eye state characteristics, combined with a preset comfort model. The preset comfort model can be a rule-based function whose inputs include whether the face frequently turns to avoid the light source, whether the eye opening is consistently too low (e.g., squinting), or whether blinking is frequent. Its output is a scalar value representing the degree of comfort, i.e., the user visual comfort index. In practical implementation, the illuminance achievement index and the user visual comfort index are combined to generate a verification result of the adjustment effect. The verification result can be a binary judgment ("good" or "poor") or a comprehensive score.

[0044] In practical implementation, when the adjustment effect verification results show high illuminance achievement and good user visual comfort index, the system confirms that the currently retrieved lighting pattern tuple from the lighting demand knowledge base is valid. In practical implementation, for valid mapping relationships, the system increases their confidence weight, for example, by incrementing a confidence counter associated with the cell storing the lighting pattern tuple. In practical implementation, when the adjustment effect verification results show low illuminance achievement or poor user visual comfort index, the system determines that the currently retrieved lighting pattern tuple does not match the user's actual needs and immediately initiates the optimization process. In some embodiments, the determination threshold is preset; for example, the optimization process is triggered when the illuminance achievement is below 0.85 or the user visual comfort index is below a certain threshold.

[0045] In practical implementation, during the optimization process, the system, based on the actual user behavior and environmental parameters at the time of optimization, and combined with the adjusted and measured user visual comfort index, reverse-calculates a corrected lighting pattern. The corrected lighting pattern includes a corrected color temperature value and a corrected illuminance value. The logic of the reverse calculation can be as follows: if the user visual comfort index is poor and the measured desktop illuminance is higher than the target illuminance, it is inferred that the target illuminance is too high, and the corrected illuminance value can be adjusted downwards by an amount proportional to the difference in comfort based on the measured desktop illuminance; if the color temperature is considered uncomfortable, the corrected color temperature value can be adjusted by a fixed step towards warmer or cooler. In some embodiments, the reverse calculation can be based on a simple gradient descent idea, so that the corrected lighting pattern is expected to bring a higher user visual comfort index. In practical implementation, the corrected lighting pattern is weighted and fused with the original lighting pattern tuple, updating the content of the corresponding cell in the lighting requirement knowledge base. The weighted fusion process aims to integrate newly acquired user personalized preference information into the existing general knowledge.

[0046] In practice, the first step of the weighted fusion process is to calculate the first confidence weight of the original illumination pattern tuple. This first confidence weight can be based on the number of times the tuple has been successfully validated, or an initial confidence level that decays over time. Next, the second confidence weight of the corrected illumination pattern is calculated. This second confidence weight can be based on the degree of poorness of the user's visual comfort index at the time of this optimization trigger, or a fixed, lower weight representing the confidence level of a single observation. Based on the first and second confidence weights, weighted fusion coefficients are calculated for the ideal color temperature and ideal illuminance values ​​in the original illumination pattern tuple, and the corrected color temperature and corrected illuminance values ​​in the corrected illumination pattern. These weighted fusion coefficients can be the normalized ratio of the first and second confidence weights. Finally, using these weighted fusion coefficients, a weighted average is calculated between the ideal color temperature values ​​in the original illumination pattern tuple and the corrected color temperature values ​​in the corrected illumination pattern to obtain the updated color temperature value. The calculation process can be expressed as follows: in: This represents the ideal color temperature value in the original lighting mode tuple. This indicates the corrected color temperature value in the modified lighting mode. Indicates the weight of the first confidence level. This indicates the weight of the second confidence level. This represents the calculated updated color temperature value. In practice, using the same weighted fusion coefficient, the ideal illuminance value in the original lighting pattern tuple and the corrected illuminance value in the corrected lighting pattern are weighted and averaged to calculate the updated illuminance value, similar to the color temperature update. In practice, the updated color temperature value and the updated illuminance value are combined to form the updated lighting pattern tuple. In practice, the updated lighting pattern tuple replaces the previously stored original lighting pattern tuple in the corresponding cell of the lighting requirement knowledge base, completing a personalized update of the mapping relationship. Optionally, while updating the lighting pattern tuple, the confidence weight associated with that cell is also reset or updated; for example, the sum of the first confidence weight and the second confidence weight is stored as the new confidence weight. The first confidence weight of the original illumination pattern tuple is assigned a preset positive integer value during system initialization and increments after each valid verification. The second confidence weight of the modified illumination pattern is also assigned a preset positive integer value. Since both weights are always positive, the denominator of the weighted fusion formula is always greater than zero, preventing division by zero anomalies. Regarding the calculation of observation probabilities in the Viterbi algorithm, when the calculated probability density of an observation vector in a certain hidden state is extremely low, the system uses logarithmic probability instead of the original probability for calculation, converting multiplication into addition to avoid numerical precision loss due to floating-point underflow.

[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A desk lamp adaptive adjustment system based on user behavior analysis, characterized in that, include: A multi-source sensing module is used to deploy multi-source sensing devices on a desk lamp and around the user's work environment. The multi-source sensing devices include an image acquisition unit, an ambient light sensor, an infrared proximity sensor, and a human activity detector, which are used to collect raw sensing data. The data fusion module is used to perform multimodal fusion processing on the original sensing data to generate a fused behavior sensing vector; The behavior decoding module is used to input the fused behavior perception vector into the behavior state decoder based on the improved Hidden Markov Model. The improved Hidden Markov Model introduces a temporal continuity constraint matrix on its standard state transition probability matrix. The temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element by element to obtain the modified state transition probability matrix. The input fused behavior perception vector sequence is decoded based on the modified state transition probability matrix, and the discrete label sequence corresponding to the user behavior state is output. The demand matching module is used to input the discrete label sequence of the user's behavior state and the environmental parameters into a preset lighting demand knowledge base for retrieval and matching. The lighting demand knowledge base stores the ideal lighting modes corresponding to various behavior states under different environmental parameters. The target generation module is used to generate illumination adjustment targets containing target color temperature values ​​and target illuminance values ​​based on the search and matching results.

2. The adaptive adjustment system for a desk lamp based on user behavior analysis according to claim 1, characterized in that, The original sensing data is subjected to multimodal fusion processing to generate a fused behavior sensing vector, including: The fused behavior perception vector includes user posture features, facial orientation features, eye state features, desktop illuminance features, and ambient light distribution features. The system processes the data from the image acquisition device, determines the user's facial position and orientation through facial key point detection, evaluates eye opening and gaze point changes through eye analysis algorithms, and outputs facial orientation features and eye state features. The data from the human activity detector is processed and combined with the skeletal key point information to determine whether the user's upper body posture is leaning forward, leaning back, or standing upright, and the user's posture features are output. The system processes the data from the ambient light sensor to obtain the intensity and color temperature of the ambient light, calculates the average illuminance on the desktop, and outputs the ambient light distribution characteristics and desktop illuminance characteristics. A unified timestamp is established for each time point, and the facial orientation features, eye state features, user posture features, ambient light distribution features, and desktop illuminance features are aligned to the unified timestamp. All aligned features are concatenated into a multidimensional vector in a predetermined order, and the multidimensional vector is standardized to obtain the fused behavior perception vector.

3. The adaptive adjustment system for a desk lamp based on user behavior analysis according to claim 1, characterized in that, The improved Hidden Markov Model adjusts its state transition probabilities based on the temporal continuity constraint of the user task, including: Define a set of hidden states, each corresponding to a different typical user behavior state, including focused reading, writing at a desk, taking intermittent breaks, leaving the seat, and looking up into the distance. Define the observation state as the quantized value of key features extracted from the fused behavior perception vector; Based on the standard state transition probability matrix of the improved Hidden Markov Model, a temporal continuity constraint matrix is ​​introduced. The temporal continuity constraint matrix is ​​set according to the prior knowledge of the task scenario, and is used to increase the transition probability between temporally continuous and task-related behavioral states, while reducing the transition probability between unrelated or anti-logical states. During the decoding process, the temporal continuity constraint matrix and the standard state transition probability matrix are multiplied element by element to obtain the corrected state transition probability matrix. Using the Viterbi algorithm, based on the corrected state transition probability matrix, the observed state probability distribution, and the initial state probability, the input fused behavior perception vector sequence is decoded, and the hidden state sequence, i.e., the discrete label sequence of the user behavior state, is output.

4. The adaptive adjustment system for a desk lamp based on user behavior analysis according to claim 1, characterized in that, The retrieval and matching process of the illumination demand knowledge base includes: The lighting demand knowledge base is organized in tabular form, with its row index being different user behavior status labels and its column index being different quantification ranges of environmental parameters. Each cell stores a tuple of lighting patterns consisting of ideal color temperature and ideal illuminance values; The latest label in the discrete label sequence of the user's behavior state is used as the index of the current behavior state; The current environmental parameters, including ambient light intensity and color temperature, are quantized into the corresponding environmental parameter range and used as column indexes. Based on the current behavior state index and column index, locate the corresponding cell in the table of the lighting demand knowledge base; The illumination pattern tuple stored in the corresponding cell is read as the basis for generating the illumination adjustment target.

5. The adaptive adjustment system for a desk lamp based on user behavior analysis according to claim 1, characterized in that, The system also includes: The adjustment calculation module calculates the initial color temperature adjustment amount and illuminance adjustment amount based on the difference between the light adjustment target and the current light parameters of the desk lamp. The smoothing control module is used to input the initial color temperature adjustment amount and illuminance adjustment amount to the light change smoothing controller. The light change smoothing controller performs rate planning for the adjustment process based on the user's sensitivity threshold to light changes, and generates a smooth color temperature control command sequence and illuminance control command sequence. The execution drive module is used to send the smooth color temperature control command sequence and illuminance control command sequence to the execution drive circuit of the desk lamp, so as to drive the light-emitting component to realize the adaptive adjustment of the light parameters. The feedback optimization module is used to continuously collect feedback data through the multi-source sensing device during the adjustment process, verify the adjustment effect and user comfort, and dynamically update and optimize the mapping relationship in the light demand knowledge base based on the verification results.

6. The adaptive adjustment system for a desk lamp based on user behavior analysis according to claim 5, characterized in that, The illumination change smoothing controller performs rate planning for the adjustment process based on the user's sensitivity threshold to illumination changes, including: Obtain a pre-calibrated user illumination change sensitivity threshold, wherein the sensitivity threshold defines the maximum allowable range of color temperature and illuminance change per unit time; Divide the initial color temperature adjustment amount and illuminance adjustment amount by the corresponding sensitivity threshold to obtain the number of adjustment time steps required under unconstrained conditions; Obtain the adjustment time step value corresponding to color temperature adjustment and the adjustment time step value corresponding to illuminance adjustment respectively, and take the maximum value of the two step values ​​as the common time step number for the final adjustment; The initial color temperature adjustment and illuminance adjustment are evenly distributed to each step of the common time step, and the color temperature adjustment increment and illuminance adjustment increment of each step are calculated. In chronological order, color temperature control instructions and illuminance control instructions for each step are generated sequentially and combined to form the smooth color temperature control instruction sequence and illuminance control instruction sequence.

7. A desk lamp adaptive adjustment system based on user behavior analysis according to claim 5, characterized in that, During the adjustment process, feedback data is continuously collected through the multi-source sensing device to verify the adjustment effect and user comfort, including: After the desk lamp begins executing the smooth color temperature control command sequence and illuminance control command sequence, it continues to collect a new round of raw sensory data. Process the new round of raw sensory data to obtain the adjusted fused behavior perception vector; Extract the actual desktop illuminance features and the desk lamp illuminance features separated from the ambient light from the fused behavior perception vector; The actual desktop illuminance characteristics are compared with the target illuminance value to calculate the illuminance achievement index; By analyzing the adjusted facial orientation features and eye state features, combined with a preset comfort model, the user's visual comfort index is evaluated. By combining the aforementioned illuminance achievement index and user visual comfort index, a verification result of the adjustment effect is generated.

8. A desk lamp adaptive adjustment system based on user behavior analysis according to claim 7, characterized in that, Based on the verification results, the mapping relationships in the illumination demand knowledge base are dynamically updated and optimized, including: When the results of the adjustment effect verification show that the illuminance achievement is high and the user visual comfort index is good, it is confirmed that the illuminance pattern tuple retrieved from the illuminance demand knowledge base is effective and its confidence weight is increased. When the adjustment effect verification results show that the illuminance achievement is low or the user visual comfort index is poor, it is determined that the currently retrieved illuminance mode tuple does not match the user's actual needs, and the optimization process is initiated. In the optimization process, a corrected lighting mode is calculated in reverse based on the actual user behavior and environmental parameters, combined with the adjusted and measured user visual comfort index. The modified lighting pattern is weighted and fused with the original lighting pattern tuple to update the content of the corresponding cell in the lighting requirement knowledge base, so as to adapt to the personalized preferences of specific users.

9. A desk lamp adaptive adjustment system based on user behavior analysis according to claim 8, characterized in that, The modified lighting pattern is weighted and fused with the original lighting pattern tuple to update the content of the corresponding cell in the lighting requirement knowledge base, in order to adapt to the personalized preferences of specific users. This includes the following steps: Calculate the first confidence weight of the original illumination pattern tuple; Calculate the second confidence weight of the corrected illumination pattern; Based on the first confidence weight and the second confidence weight, calculate the weighted fusion coefficients corresponding to the ideal color temperature value and ideal illuminance value in the original illumination mode tuple and the corrected color temperature value and corrected illuminance value in the corrected illumination mode, respectively. Using the weighted fusion coefficient, the ideal color temperature value in the original lighting mode tuple and the corrected color temperature value in the corrected lighting mode are weighted and averaged to calculate the updated color temperature value. Using the weighted fusion coefficient, the ideal illuminance value in the original illumination mode tuple and the corrected illuminance value in the corrected illumination mode are weighted and averaged to calculate the updated illuminance value. The updated color temperature value and the updated illuminance value are combined to form an updated illumination mode tuple; The updated lighting pattern tuple replaces the previously stored original lighting pattern tuple in the corresponding cell of the lighting requirement knowledge base, thus completing a personalized update of the mapping relationship.

10. A desk lamp adaptive adjustment system based on user behavior analysis according to claim 5, characterized in that, The process of adaptively adjusting illumination parameters by driving the light-emitting components also includes dynamic compensation for ambient light disturbances: The system monitors the data from the ambient light sensor in real time to detect any sudden changes in ambient light. When a sudden change in ambient light is detected, the currently executing sequence of smooth control commands is immediately interrupted; Based on the changed ambient lighting parameters and combined with the latest user behavior status, new lighting adjustment targets are retrieved and generated again from the lighting demand knowledge base; Based on the new lighting adjustment targets, a new lighting adjustment process is recalculated and initiated to quickly counteract the impact of ambient light disturbances on the user's visual experience.

Images