An intelligent cosmetic mirror personalized adjustment method based on user habit learning

Abstract

The present application relates to user behavior learning and machine learning control technical field, especially to a kind of intelligent makeup mirror individualized adjustment method based on user habit learning, comprising the following steps: obtaining historical makeup session data, extracting action frequency characteristics and parameter preference, build double-track habit model;Current session obtains makeup stage, operation frequency, and the initial stage is compared with frequency benchmark to determine session mode;According to the next makeup stage predicted by makeup stage and session mode, the target control parameter is calculated by combining ambient light and parameter preference, and the control parameter is updated according to the preset slope gradient;According to the slope adjustment of manual adjustment signal and update direction, update double-track habit model after session end.The present application, pre-constructed double-track habit model containing normal mode and time pressure mode, operation frequency is obtained in the initial stage of makeup session compared with frequency benchmark, frequency is greater than benchmark to determine time pressure mode, and the next makeup stage is predicted and target control parameter is generated accordingly.

Description

Technical Field

[0001] This invention relates to the field of user behavior learning and machine learning control technology, and in particular to a personalized adjustment method for an intelligent makeup mirror based on user habit learning. Background Technology

[0002] With the development of smart homes and smart mirror devices, smart makeup mirrors are increasingly being used in fixed home grooming scenarios. Existing smart makeup mirrors typically combine mirror components, lighting components, image acquisition components, ambient light detection components, and user operation components. They collect data on the user's state in front of the mirror, ambient light conditions, and user actions, and control the mirror's lighting parameters and auxiliary adjustment parameters based on the collected results. Lighting parameters may include brightness and color temperature, while auxiliary adjustment parameters may include magnification. Some smart makeup mirrors can also record the user's historical usage data and form parameter preferences based on the user's operation records in different usage scenarios, thus enabling personalized adjustments during subsequent use according to historical preferences.

[0003] Most existing smart makeup mirrors adopt an adjustment method that responds instantly after detecting the current state. Because they cannot identify changes in the user's operating rhythm in the early stages of the session, the adjustment timing is delayed and cannot match the user's actual makeup rhythm. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a personalized adjustment method for smart makeup mirrors based on user habit learning. It aims to improve the problem that most existing smart makeup mirrors adopt an adjustment method that responds instantly after current state detection, which easily leads to delayed adjustment timing and inability to match the user's actual makeup rhythm.

[0005] This invention provides the following technical solution: a personalized adjustment method for an intelligent makeup mirror based on user habit learning, comprising the following steps: S1. Obtain historical makeup session data, extract action frequency features and parameter preferences under usage scenarios, and construct a dual-track habit model including normal mode and time pressure mode, wherein the normal mode records frequency benchmarks; S2. In the current makeup session, obtain the current makeup stage and the current operation frequency. When the current operation frequency is greater than the frequency benchmark of the normal mode at the beginning of the session, determine the current session mode as the time pressure mode; otherwise, determine it as the normal mode. S3. Predict the next makeup stage based on the current makeup stage and the current conversation mode, and calculate the target control parameters based on the current ambient light data and the parameter preferences corresponding to the next makeup stage. Perform a gradual update from the current control parameters to the target control parameters according to a preset slope. S4. During the gradual update process, when a user manual adjustment signal is received, if the manual adjustment signal is consistent with the current gradual update direction, the slope of the gradual update is increased; if it is opposite to the current gradual update direction, the slope of the gradual update is decreased or the gradual update is stopped. S5. When the current makeup session ends, update the dual-track habit model based on the phase sequence of the current makeup session, the current operation frequency, and the manual adjustment signal.

[0006] By adopting the above technical solution, a dual-track habit model containing normal mode and time pressure mode is pre-constructed. At the initial stage of the current makeup session, the current operation frequency is obtained and compared with the frequency benchmark of normal mode. Then, when the operation frequency is greater than the frequency benchmark, the current session mode is directly determined as time pressure mode, and the next makeup stage is predicted and target control parameters are generated accordingly. This improves the problem that most existing smart makeup mirrors adopt the adjustment method of immediate response after current state detection. Since they cannot identify changes in the user's operation rhythm in the early stage of the session, the adjustment timing is delayed and cannot match the user's actual makeup rhythm.

[0007] Furthermore, in S1, the step of obtaining historical makeup session data includes: Retrieve the start time, end time, phase sequence, and operation behavior records of multiple historical makeup sessions; The duration of each historical makeup session is determined based on the start time and the end time, and the usage scenarios are divided based on the duration and the preset time period. The usage scenarios, the phase sequences, and the operation behavior records are correlated and combined to form historical makeup session data.

[0008] Furthermore, in S1, the step of extracting action frequency features and parameter preferences in the usage context includes: Historical makeup session data were categorized according to usage context, and the number of operations and duration of each makeup stage were statistically analyzed under the same usage context. The action frequency characteristics are calculated based on the number of operations and the duration of the phase, and the statistical mean of the action frequency characteristics is determined as the frequency benchmark in the normal mode. Extract the deviation between the control parameters manually adjusted by the user and the preset reference parameters corresponding to the current ambient light data, and determine the deviation as the parameter preference.

[0009] Furthermore, in S1, the step of constructing a dual-track habit model including a normal pattern and a time-pressure pattern includes: Based on the action frequency characteristics in historical makeup conversation data, historical makeup conversations are divided into normal conversations and time-pressured conversations. The transition frequency between each makeup stage is statistically analyzed for both the normal session and the time-pressure session, and the stage transition probability is calculated for the normal mode and the time-pressure mode. The stage transition probability, the parameter preference, and the frequency benchmark under the normal mode are bound and stored to construct a dual-track habit model.

[0010] Furthermore, in S2, the step of obtaining the current makeup stage and the current operation frequency includes: Obtain the distance between the current facial focus area and the face mirror to identify the current makeup stage; When a face is detected in front of a mirror and there is no effective makeup application, the system enters a hovering state, during which the current operation frequency is paused and statistics are not collected. When a valid makeup action is detected to be restored, the hovering state is exited, the actual duration and hovering duration within the current time window are obtained, and the hovering duration is subtracted from the actual duration to obtain the effective duration. The current operation frequency is calculated based on the number of valid operations within the current time window and the effective duration.

[0011] Further, in S2, the step of determining the current session mode as the time pressure mode when the current operation frequency is greater than the frequency reference of the normal mode during the initial stage of the session includes: Calculate the ratio of the current operating frequency to the frequency reference of the normal mode. If the ratio is greater than a preset ratio threshold, generate a candidate time pressure marker. The difference between the current operating frequency and the operating frequency at the previous moment is calculated to obtain the operating acceleration; In the initial stage of the current makeup session, when the candidate time pressure marker exists and the operation acceleration is greater than zero, the current session mode is determined to be the time pressure mode.

[0012] Furthermore, in S3, the step of predicting the next makeup stage based on the current makeup stage and the current conversation mode includes: Read the historical transition probabilities from the current makeup stage to each candidate next makeup stage under the current session mode determined in S2; Calculate the degree of match between the current makeup stage duration and the historical average duration of this stage under the current session mode; The prediction confidence of each candidate next makeup stage is calculated based on the historical transition probability and the matching degree. The candidate next makeup stage with the highest confidence and greater than the preset confidence threshold is taken as the predicted next makeup stage.

[0013] Furthermore, in S3, the step of gradually changing the current control parameter to the target control parameter according to a preset slope includes: Based on the current ambient light data, a preset baseline control parameter is determined. The preset baseline control parameter is then superimposed with the parameter preference corresponding to the next makeup stage obtained in S3 to calculate the target control parameter. Calculate the parameter difference between the target control parameter and the current control parameter; When the current session mode is the time pressure mode, the preset fading time is shortened proportionally according to the ratio of the current operation frequency to the frequency reference of the normal mode, and the preset slope is calculated based on the parameter difference and the shortened preset fading time. Within the shortened preset gradient duration, the current control parameters are gradually adjusted to the target control parameters according to the preset slope.

[0014] Further, in S4, the step of increasing the slope of the gradual update if the manual adjustment signal is consistent with the current gradual update direction includes: The manual adjustment signal is converted into a parameter change vector, and the current gradient update direction is converted into a gradient vector; Calculate the vector dot product of the parameter change vector and the gradient vector; When the vector dot product is greater than zero, it is determined that the manual adjustment signal is consistent with the current gradual update direction, and the slope of the gradual update is increased according to the preset gain coefficient.

[0015] Furthermore, in S5, the step of updating the dual-track habit model based on the current makeup session's phase sequence, the current operation frequency, and the manual adjustment signal includes: Update the phase transition probability in the corresponding session mode based on the phase sequence of this makeup session; Update the action frequency characteristics in the corresponding session mode or the frequency reference in the normal mode based on the current operation frequency; Update the parameter preferences for the corresponding session mode and makeup stage based on the adjustment amount corresponding to the manual adjustment signal; After applying exponentially decaying weights to the historical data, the updated stage transition probabilities, action frequency characteristics, and parameter preference increments are written into the dual-track habit model.

[0016] The present invention has the following beneficial effects: 1. In this invention, a dual-track habit model containing normal mode and time pressure mode is pre-constructed, and the current operation frequency is obtained at the initial stage of the current makeup session and compared with the frequency benchmark of normal mode. Then, when the operation frequency is greater than the frequency benchmark, the current session mode is directly determined as time pressure mode, and the next makeup stage is predicted and target control parameters are generated accordingly. This improves the problem that most existing smart makeup mirrors adopt the adjustment method of immediate response after current state detection. Since they cannot identify changes in the user's operation rhythm in the early stage of the session, the adjustment timing is delayed and cannot match the user's actual makeup rhythm.

[0017] 2. In this invention, by receiving user manual adjustment signals during the automatic gradual update of mirror parameters, and increasing or decreasing the slope of the gradual update based on the consistency determination result between the manual adjustment signal and the current gradual update direction, the user's real-time intervention is integrated into the parameter change trajectory without interrupting the automatic adjustment process. This improves the problem that most existing smart makeup mirrors adopt a processing method in which user manual adjustment and automatic adjustment are independent or simply superimposed. Since the directional intention of user intervention cannot be distinguished, a conflict of control occurs between automatic control and user operation.

[0018] 3. In this invention, by identifying situations where the face is in front of the mirror but there is no effective makeup action during the acquisition of the current operation frequency and entering a hovering state, the current operation frequency is paused during the hovering state. After exiting the hovering state, the hovering time is deducted from the actual duration to calculate the effective operation frequency. This improves upon the fact that most existing smart makeup mirrors directly judge the stage and perform operation statistics based on time or the state of the face. Because non-makeup pause behaviors such as picking up makeup products are equated with effective makeup time, the problem of misjudging the operation rhythm and incorrect stage switching occurs. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall process of a personalized adjustment method for an intelligent makeup mirror based on user habit learning, as described in this invention. Figure 2 This is a flowchart of the historical makeup conversation data processing and dual-track habit model construction of the present invention; Figure 3 This is a flowchart illustrating the current session mode determination and next makeup stage prediction process of the present invention. Figure 4 This is a flowchart illustrating the collaborative processing of target control parameter calculation, gradual update, and manual adjustment in this invention. Figure 5 This is a flowchart of the present invention for updating the dual-track habit model based on the current makeup session. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: In the first embodiment of the present invention, a personalized adjustment method for a smart makeup mirror based on user habit learning is provided, such as... Figure 1 As shown, the process includes the following steps: S1. Obtain historical makeup session data, extract action frequency features and parameter preferences in the usage context, and construct a dual-track habit model including normal mode and time pressure mode. The normal mode records frequency benchmarks. Furthermore, in S1, the steps for obtaining historical makeup session data include: Retrieve the start time, end time, phase sequence, and operation behavior records of multiple historical makeup sessions; The duration of each historical makeup session is determined based on the start and end times, and the usage scenarios are divided based on the duration and preset time periods. The usage context, stage sequence, and operational behavior records are correlated and combined to form historical makeup session data.

[0022] Specifically, for each user makeup session, the start and end timestamps of the session are recorded. Simultaneously, facial images and adjustment events on the mirror control panel are continuously collected at preset time intervals. Based on the shift sequence of the user's gaze area in the facial image and the triggering order of the adjustment events, a session is divided into several ordered makeup stages, forming a stage sequence. The type of adjustment event, the occurrence time, and the duration of the facial focus area within each stage are recorded as operational behaviors. The start and end timestamps of each session are read, and the time difference between them is calculated to obtain the duration. The start timestamp of each session is matched with a preset time period division rule. This rule divides the day into weekday morning, weekday daytime, weekday evening, rest daytime, and rest evening periods according to the hour interval of the device's local clock. Matching determines the usage context of each session. The stage sequences and operational behavior records of multiple sessions belonging to the same usage context are associated and stored according to session numbers, forming structured historical makeup session data.

[0023] Furthermore, in S1, the steps for extracting action frequency features and parameter preferences in the usage context include: Historical makeup session data were categorized according to usage context, and the number of operations and duration of each makeup stage were statistically analyzed under the same usage context. The action frequency characteristics are calculated based on the number of operations and the duration of each phase, and the statistical mean of the action frequency characteristics is determined as the frequency benchmark in the normal mode. Extract the deviation between the control parameters manually adjusted by the user and the preset reference parameters corresponding to the current ambient light data, and determine the deviation as the parameter preference.

[0024] Specifically, the system reads all historical makeup session data belonging to the same usage scenario, iterates through each makeup stage in each session, accumulates the total number of adjustment events within that stage as the number of operations, and records the time span from the entry time to the exit time of that stage as the stage duration. For a specific makeup stage in a given usage scenario, the number of operations in that stage is divided by the stage duration to obtain the action frequency of that stage in that session. The arithmetic mean of all action frequencies of the same stage in the same usage scenario is calculated, and this mean is taken as the action frequency characteristic of that stage in the normal mode, and this mean is used as the frequency benchmark for the normal mode. The system extracts the control parameter value after the user's last manual adjustment in each session, and reads the preset benchmark parameter corresponding to the current ambient light data from the ambient light-benchmark parameter mapping table stored in the device. The difference between the control parameter value and the preset benchmark parameter is calculated, and this difference is stored as the parameter preference for that makeup stage in the given usage scenario.

[0025] like Figure 2 As shown, further, in S1, the steps for constructing a dual-track habit model including a normal pattern and a time-pressure pattern include: Based on the action frequency characteristics in historical makeup conversation data, historical makeup conversations are divided into normal conversations and time-pressured conversations. The transition frequency between each makeup stage was counted for normal and time-pressured sessions, and the stage transition probability was calculated for normal and time-pressured modes. The stage transition probability, parameter preference, and frequency benchmark under normal mode are bound and stored to form a dual-track habit model.

[0026] Specifically, for all historical makeup sessions under the same usage scenario, sessions where the frequency of actions in the makeup stage exceeds a preset proportion threshold and is greater than the normal mode frequency benchmark are marked as time-pressure sessions, and the remaining sessions are marked as normal sessions. In this scheme, the preset proportion threshold is preferably 0.6. For both the normal session set and the time-pressure session set, the number of times each makeup stage transitions to other makeup stages is counted. The number of transitions is divided by the total number of occurrences of that stage to obtain the stage transition probability between each stage under normal mode and time-pressure mode. For stage pairs with zero transition occurrences, Laplace smoothing is applied, and the smoothing factor can be 0.1. The stage transition probability, parameter preferences of each stage, and frequency benchmark under normal mode are bound into one data record, and the stage transition probability and parameter preferences of each stage under time-pressure mode are bound into another data record. The two data records together constitute the dual-track habit model under this usage scenario and are written into the storage component.

[0027] S2. In the current makeup session, obtain the current makeup stage and the current operation frequency. If the current operation frequency is greater than the normal mode frequency benchmark at the beginning of the session, determine the current session mode as the time pressure mode; otherwise, determine it as the normal mode. Furthermore, in S2, the steps for obtaining the current makeup stage and the current operation frequency include: Obtain the distance between the current facial focus area and the face mirror to identify the current makeup stage; When the face is detected to be in front of the mirror and there is no effective makeup application, the system enters a hovering state, during which the current operation frequency is paused and statistics are not collected. When a valid makeup action is detected to be restored, the hover state is exited. The actual duration and hover duration within the current time window are obtained. The hover duration is subtracted from the actual duration to obtain the effective duration. The current operation frequency is calculated based on the number of valid operations and the effective duration within the current time window.

[0028] Specifically, the system acquires user facial images through an image acquisition component, extracts facial key points to determine the user's facial orientation and gaze area, and combines the distance between the face and the mirror obtained by a distance sensor. The current gaze area is then matched with facial region templates for each preset makeup stage, and the makeup stage corresponding to the template with the highest matching degree is selected as the current makeup stage. The system continuously acquires sequences of user hand images, extracts hand key point trajectories, and calculates the relative position change rate between the hand and face and the variance of the facial tilt angle. If, within N consecutive frames, the relative position change rate between the hand and face is lower than a preset displacement threshold and the variance of the facial tilt angle is lower than a preset variance threshold, it is determined that there is no effective makeup action. In this scheme, N is preferably 10, the preset displacement threshold is preferably 0.05 times the mirror width per frame, and the preset variance threshold is... The threshold is preferably 0.01 square radians. When the user's face remains within the detection area in front of the mirror for multiple consecutive sampling periods and is determined to have no effective makeup action, the user is determined to enter a hovering state. During the hovering state, the statistical process of the current operation frequency is suspended, and the accumulated number of operations and the timing value are maintained. When the relative position change rate of the hand and face is detected to exceed the preset displacement threshold or the variance of the facial tilt angle exceeds the preset variance threshold, the user is determined to resume effective makeup action and exit the hovering state. The statistical process resumes, the actual clock elapsed within the current sliding time window is read, and the duration of this hovering is subtracted from it to obtain the effective duration. The accumulated number of effective adjustment operations within the sliding time window is divided by the effective duration to obtain the current operation frequency.

[0029] like Figure 3 As shown, further, in S2, when the current operating frequency is greater than the frequency reference of the normal mode at the initial stage of the session, the step of determining the current session mode as the time pressure mode includes: Calculate the ratio of the current operating frequency to the frequency reference of the normal mode. If the ratio is greater than the preset ratio threshold, generate a candidate time pressure marker. Calculate the difference between the current operating frequency and the operating frequency at the previous moment to obtain the operating acceleration; In the initial stage of the current makeup session, if there is a candidate time pressure marker and the operation acceleration is greater than zero, the current session mode is determined to be the time pressure mode.

[0030] Specifically, the current operating frequency and the normal mode frequency reference calculated in S2 are read, and the ratio between the two is calculated. ;in This is the current operating frequency, measured in times per second. The frequency reference for normal mode is measured in times per second. If the ratio... If the ratio exceeds a preset threshold, a candidate time pressure marker is generated. In this scheme, the preset threshold is preferably 1.3; the operation frequency of reading the previous sliding time window is also considered. Calculate operational acceleration ;in The time interval between the center moments of the two sliding time windows, in seconds. The unit is times per square second; within the first few sliding time windows after the start of the makeup session, if candidate time pressure markers exist simultaneously and the operation acceleration... If the value is greater than zero, the current session mode will be set to time pressure mode; otherwise, it will remain in normal mode.

[0031] S3. Predict the next makeup stage based on the current makeup stage and the current conversation mode, and calculate the target control parameters based on the current ambient light data and the parameter preferences corresponding to the next makeup stage. Perform a gradual update from the current control parameters to the target control parameters according to the preset slope. Furthermore, in S3, predicting the steps for the next makeup stage based on the current makeup stage and the current conversation pattern includes: Read the historical transition probabilities from the current makeup stage to each candidate next makeup stage under the current session mode determined in S2; Calculate the degree of match between the current makeup stage duration and the historical average duration of this stage under the current session mode; The prediction confidence of each candidate next makeup stage is calculated based on the historical transition probability and matching degree. The candidate next makeup stage with the highest confidence and greater than the preset confidence threshold is selected as the predicted next makeup stage.

[0032] Specifically, based on the current conversation mode determined in S2, the system reads the current makeup stage from the dual-track habit model. Move on to the next makeup stage for each candidate Historical transition probability Read the current makeup stage The duration of And read the historical average duration of this phase under this pattern from the dual-track habitual model. Calculate the duration matching degree ,when Greater than At that time, The limit is no less than 0.1; for each candidate, the next makeup stage... Calculate its prediction confidence level ;in and These are the weighting coefficients in this scheme. The preferred value is 0.6. The preferred value is 0.4; the stage with the highest prediction confidence among all candidate next makeup stages and greater than the preset confidence threshold is taken as the predicted next makeup stage. In this scheme, the preset confidence threshold is preferably 0.5.

[0033] like Figure 4 As shown, further, in S3, the step of gradually changing the current control parameter to the target control parameter according to a preset slope includes: Based on the current ambient light data, the preset baseline control parameters are determined. The preset baseline control parameters are then superimposed with the parameter preferences corresponding to the next makeup stage obtained in S3 to calculate the target control parameters. Calculate the parameter difference between the target control parameter and the current control parameter; When the current session mode is time pressure mode, the preset gradient duration is shortened proportionally according to the ratio of the current operation frequency to the frequency benchmark of the normal mode, and the preset slope is calculated based on the parameter difference and the shortened preset gradient duration. Within the shortened preset gradient duration, the current control parameters are gradually adjusted to the target control parameters according to the preset slope.

[0034] Specifically, the system reads the current ambient light data and retrieves the corresponding preset reference control parameters from a pre-defined ambient light-reference parameter mapping table. This mapping table divides the ambient light intensity into several intervals, each corresponding to a set of reference brightness and reference color temperature values. It then reads the next makeup stage predicted in S3, obtains the parameter preferences for that stage in this usage scenario from the dual-track habit model, and adds the preset reference control parameters to the parameter preferences item by item to obtain the target control parameters. Finally, it calculates the difference between the target control parameters and the current control parameters. If the current session mode is a time-pressure mode, it reads the ratio of the current operation frequency to the normal mode frequency reference. The preset default gradient duration shortened to And not less than the preset minimum gradient duration In this scheme, the default gradient duration is... The preferred duration is 3 seconds, with the shortest gradient duration. The preferred value is 0.8 seconds. The preset slope is calculated based on the parameter difference and the shortened transition time, which is the parameter change per unit time. Within the shortened transition time, the current control parameter is gradually adjusted in equal steps according to the preset slope until the target control parameter is reached.

[0035] S4. During the gradual update process, when a user manual adjustment signal is received, if the manual adjustment signal is consistent with the current gradual update direction, the slope of the gradual update is increased; if it is opposite to the current gradual update direction, the slope of the gradual update is decreased or the gradual update is stopped. like Figure 4 As shown, further, in S4, if the manual adjustment signal is consistent with the current gradient update direction, the step of increasing the gradient update slope includes: Convert the manual adjustment signal into a parameter change vector, and convert the current gradient update direction into a gradient vector; Calculate the dot product of the parameter change vector and the gradual change vector; When the vector dot product is greater than zero, it is determined that the manual adjustment signal is consistent with the current gradual update direction, and the slope of the gradual update is increased according to the preset gain coefficient.

[0036] Specifically, the user-manually adjusted signal is decomposed into adjustment components in each adjustable parameter dimension. In this scheme, the multi-dimensional parameter change vector consists of the brightness change. and color temperature change Composition, that is The current gradient update direction is used to construct a gradient vector with the same parameter dimensions. ; Calculate the dot product of two vectors ,when When the value is greater than zero, it indicates that the angle between the user-manually adjusted direction and the current automatic gradient direction is less than 90 degrees, and the two directions are determined to be consistent; the current gradient change slope is multiplied by a preset gain coefficient. The increased slope is obtained. In this scheme, the gain coefficient is preset. The preferred value is 1.5, if If the value is less than zero, the gradually changed slope will be multiplied by a preset attenuation coefficient. In this scheme, a preset attenuation coefficient is used. The preferred value is 0.5, or the slope can be set to zero to stop the gradual update.

[0037] S5. When the current makeup session ends, update the dual-track habit model based on the current makeup session's phase sequence, current operation frequency, and manual adjustment signals. like Figure 5 As shown, further, in S5, the steps for updating the dual-track habit model based on the current makeup session's phase sequence, current operation frequency, and manual adjustment signals include: Update the phase transition probability in the corresponding session mode based on the phase sequence of this makeup session; Update the action frequency characteristics in the corresponding session mode or the frequency benchmark in the normal mode based on the current operation frequency; Update the parameter preferences for the corresponding session mode and makeup stage based on the adjustment amount corresponding to the manual adjustment signal; After applying exponentially decaying weights to historical data, the updated stage transition probabilities, action frequency characteristics, and parameter preference increments are written into the dual-track habit model.

[0038] Specifically, the process reads the complete sequence of stages in this makeup session, determines the actual transition paths between each stage, and increments the count of each actual transition path by 1 in the stage transition frequency counter for the corresponding usage scenario and session mode, recalculating the stage transition probability for that mode. It also reads the current operation frequency within each sliding time window of this session. If the session mode is normal, it performs an exponentially weighted moving average of the mean operation frequency of each window with the historical frequency benchmark, updating the formula as follows: ;in To update the coefficients, The frequency reference before the update, The frequency baseline for normal mode is updated using the average operation frequency of this session. If the session mode is time-pressure mode, the action frequency characteristics under this mode are updated using the same update formula. The final control parameter value corresponding to the manual adjustment signal is read, and its deviation from the preset baseline parameter corresponding to the current ambient light data is calculated. This deviation is then subjected to an exponentially weighted moving average with the historical parameter preferences for the corresponding usage scenario and makeup stage. The update formula is as follows: ;in For the parameter preferences before the update, The parameter preferences are updated based on the deviation calculated for this session; when weighting and fusing the aforementioned historical data, an exponentially decaying weight is applied to the historical session data. ;in The preferred attenuation coefficient is 0.05. To establish the interval between the historical session number and the current session number, incremental writing is performed based on the weighted historical data and the current session data. The updated stage transition probability, action frequency characteristics, frequency benchmark, and parameter preferences are then written into the corresponding storage location in the dual-track habit model.

[0039] Example 2: Users typically have ample time for a complete makeup routine, but they often demand a higher level of refinement. They frequently adjust parameters such as brightness and color temperature manually on top of automatic adjustments to achieve the ideal effect for a specific atmosphere. However, existing smart makeup mirrors, when frequently manually intervened by the user, tend to misinterpret temporary adjustments as long-term preferences or fail to accurately identify the user's directional intent. This leads to interference between automatic and manual adjustments, disrupting the smoothness of the makeup process. To address these issues, the present invention employs a personalized adjustment method for smart makeup mirrors based on user habit learning. The specific implementation process of this method is as follows: First, in daily use, the smart makeup mirror has accumulated historical makeup session data of users in different situations through the method described in Example 1, and constructed a dual-track habit model including normal mode and time pressure mode. The normal mode records the user's operation frequency benchmark and parameter preferences at each stage in relaxed situations such as rest days evenings. When a user stands in front of the mirror to start applying makeup on a weekend evening, the smart makeup mirror uses its camera to obtain the current facial focus area and the distance to the mirror, identifies that the user is currently in the base makeup stage, and begins to count the current operation frequency. Since the user's operation pace is relatively relaxed at this time, the current operation frequency does not exceed the frequency benchmark of the normal mode, and the system determines the current session mode as the normal mode. During this process, if the user picks up a makeup brush or checks their phone, the system detects that the face is still in front of the mirror but there is no effective makeup action, so it automatically enters a hovering state and pauses the operation frequency count. The hovering time is deducted after the user resumes the makeup action, so that the operation frequency count is not affected by non-makeup pauses. After determining the normal mode, the smart makeup mirror predicts the next stage as the eyebrow and eye makeup stage based on the current base makeup stage and the stage transition probability in the normal mode, combined with the matching degree between the duration of the current stage and the historical average duration. At the same time, it reads the current ambient light data to determine the baseline control parameters, superimposes the parameter preferences corresponding to the eyebrow and eye makeup stage, calculates the target brightness and color temperature in advance, and gradually adjusts the current control parameters with the gentle slope corresponding to the normal mode, so that when the user enters the eyebrow and eye makeup stage, the lighting has smoothly transitioned to a lighting state suitable for fine eye makeup. During the gradient update process, the user manually rotated the color temperature knob to achieve a specific makeup effect, hoping the light would be warmer. The smart makeup mirror converts this manual adjustment signal into a parameter change vector, performs a dot product operation with the current gradient direction vector, and automatically increases the gradient update slope after determining that the two directions are consistent, helping the user achieve the desired warm light state more quickly. If the user subsequently adjusts the brightness downwards, the system calculates and determines that the adjustment direction is opposite to the current automatic gradient direction, so it reduces the gradient slope, thereby avoiding a conflict of control between automatic adjustment and user manual operation. Once the user completes the overall check and leaves the mirror, the makeup session ends. The smart makeup mirror updates the stage transition probability in normal mode based on the complete stage sequence experienced in this session, updates the frequency benchmark of normal mode based on the current operation frequency, and updates the parameter preferences for the corresponding scenario and stage based on the deviation between the final control parameter value corresponding to the manual adjustment signal and the current ambient light benchmark parameter. During the update process, an exponential decay weight is applied to earlier historical data, so that the new fine-tuning habits exhibited by the user in this session can be gradually integrated into the dual-track habit model. As a result, in subsequent makeup sessions on rest days, the smart makeup mirror can more accurately predict the user's preference for a warm lighting atmosphere and adjust the lighting in advance to a state that better matches the user's expectations.

[0040] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for personalized adjustment of an intelligent makeup mirror based on user habit learning, characterized in that, Includes the following steps: S1. Obtain historical makeup session data, extract action frequency features and parameter preferences under usage scenarios, and construct a dual-track habit model including normal mode and time pressure mode, wherein the normal mode records frequency benchmarks; S2. In the current makeup session, obtain the current makeup stage and the current operation frequency. When the current operation frequency is greater than the frequency benchmark of the normal mode at the beginning of the session, determine the current session mode as the time pressure mode; otherwise, determine it as the normal mode. S3. Predict the next makeup stage based on the current makeup stage and the current conversation mode, and calculate the target control parameters based on the current ambient light data and the parameter preferences corresponding to the next makeup stage. Perform a gradual update from the current control parameters to the target control parameters according to a preset slope. S4. During the gradual update process, when a user manual adjustment signal is received, if the manual adjustment signal is consistent with the current gradual update direction, the slope of the gradual update is increased; if it is opposite to the current gradual update direction, the slope of the gradual update is decreased or the gradual update is stopped. S5. When the current makeup session ends, update the dual-track habit model based on the phase sequence of the current makeup session, the current operation frequency, and the manual adjustment signal.

2. The personalized adjustment method for a smart makeup mirror based on user habit learning according to claim 1, characterized in that, In S1, the step of obtaining historical makeup session data includes: Retrieve the start time, end time, phase sequence, and operation behavior records of multiple historical makeup sessions; The duration of each historical makeup session is determined based on the start time and the end time, and the usage scenarios are divided based on the duration and the preset time period. The usage scenarios, the phase sequences, and the operation behavior records are correlated and combined to form historical makeup session data.

3. The personalized adjustment method for a smart makeup mirror based on user habit learning according to claim 1, characterized in that, In S1, the step of extracting action frequency features and parameter preferences in the usage context includes: Historical makeup session data were categorized according to usage context, and the number of operations and duration of each makeup stage were statistically analyzed under the same usage context. The action frequency characteristics are calculated based on the number of operations and the duration of the phase, and the statistical mean of the action frequency characteristics is determined as the frequency benchmark in the normal mode. Extract the deviation between the control parameters manually adjusted by the user and the preset reference parameters corresponding to the current ambient light data, and determine the deviation as the parameter preference.

4. The personalized adjustment method for a smart makeup mirror based on user habit learning according to claim 1, characterized in that, In S1, the steps of constructing a dual-track habit model including a normal pattern and a time-pressure pattern include: Based on the action frequency characteristics in historical makeup conversation data, historical makeup conversations are divided into normal conversations and time-pressured conversations. The transition frequency between each makeup stage is statistically analyzed for both the normal session and the time-pressure session, and the stage transition probability is calculated for the normal mode and the time-pressure mode. The stage transition probability, the parameter preference, and the frequency benchmark under the normal mode are bound and stored to construct a dual-track habit model.

5. The personalized adjustment method for an intelligent makeup mirror based on user habit learning according to claim 1, characterized in that, In S2, the steps of obtaining the current makeup stage and the current operation frequency include: Obtain the distance between the current facial focus area and the face mirror to identify the current makeup stage; When a face is detected in front of a mirror and there is no effective makeup application, the system enters a hovering state, during which the current operation frequency is paused and statistics are not collected. When a valid makeup action is detected to be restored, the hovering state is exited, the actual duration and hovering duration within the current time window are obtained, and the hovering duration is subtracted from the actual duration to obtain the effective duration. The current operation frequency is calculated based on the number of valid operations within the current time window and the effective duration.

6. The personalized adjustment method for a smart makeup mirror based on user habit learning according to claim 1, characterized in that, In S2, the step of determining the current session mode as the time pressure mode when the current operating frequency is greater than the frequency reference of the normal mode during the initial stage of the session includes: Calculate the ratio of the current operating frequency to the frequency reference of the normal mode. If the ratio is greater than a preset ratio threshold, generate a candidate time pressure marker. The difference between the current operating frequency and the operating frequency at the previous moment is calculated to obtain the operating acceleration; In the initial stage of the current makeup session, when the candidate time pressure marker exists and the operation acceleration is greater than zero, the current session mode is determined to be the time pressure mode.

7. The personalized adjustment method for an intelligent makeup mirror based on user habit learning according to claim 1, characterized in that, In S3, the step of predicting the next makeup stage based on the current makeup stage and the current conversation mode includes: Read the historical transition probabilities from the current makeup stage to each candidate next makeup stage under the current session mode determined in S2; Calculate the degree of match between the current makeup stage duration and the historical average duration of this stage under the current session mode; The prediction confidence of each candidate next makeup stage is calculated based on the historical transition probability and the matching degree. The candidate next makeup stage with the highest confidence and greater than the preset confidence threshold is taken as the predicted next makeup stage.

8. The personalized adjustment method for a smart makeup mirror based on user habit learning according to claim 1, characterized in that, In S3, the step of gradually changing the current control parameter to the target control parameter according to a preset slope includes: Based on the current ambient light data, a preset baseline control parameter is determined. The preset baseline control parameter is then superimposed with the parameter preference corresponding to the next makeup stage obtained in S3 to calculate the target control parameter. Calculate the parameter difference between the target control parameter and the current control parameter; When the current session mode is the time pressure mode, the preset fading time is shortened proportionally according to the ratio of the current operation frequency to the frequency reference of the normal mode, and the preset slope is calculated based on the parameter difference and the shortened preset fading time. Within the shortened preset gradient duration, the current control parameters are gradually adjusted to the target control parameters according to the preset slope.

9. A personalized adjustment method for an intelligent makeup mirror based on user habit learning according to claim 1, characterized in that, In S4, the step of increasing the slope of the gradual update if the manual adjustment signal is consistent with the current gradual update direction includes: The manual adjustment signal is converted into a parameter change vector, and the current gradient update direction is converted into a gradient vector; Calculate the vector dot product of the parameter change vector and the gradient vector; When the vector dot product is greater than zero, it is determined that the manual adjustment signal is consistent with the current gradual update direction, and the slope of the gradual update is increased according to the preset gain coefficient.

10. A personalized adjustment method for an intelligent makeup mirror based on user habit learning according to claim 1, characterized in that, In S5, the step of updating the dual-track habit model based on the current makeup session's phase sequence, the current operation frequency, and the manual adjustment signal includes: Update the phase transition probability in the corresponding session mode based on the phase sequence of this makeup session; Update the action frequency characteristics in the corresponding session mode or the frequency reference in the normal mode based on the current operation frequency; Update the parameter preferences for the corresponding session mode and makeup stage based on the adjustment amount corresponding to the manual adjustment signal; After applying exponentially decaying weights to the historical data, the updated stage transition probabilities, action frequency characteristics, and parameter preference increments are written into the dual-track habit model.

Images