Intelligent door lock with face recognition function and recognition method thereof

Abstract

The invention discloses an intelligent door lock with a face recognition function and a recognition method thereof, and relates to the technical field of intelligent door locks, and the recognition method comprises the following steps: when a trigger event is detected, based on trigger record information associated with the trigger event and historical recognition record information stored in the door lock, calculating a verification session entropy representing access uncertainty; based on the verification session entropy, selecting a corresponding wake-up strategy to execute delayed wake-up and advanced wake-up on the face recognition function, and outputting initial face imaging acquisition parameters; and acquiring continuous face images according to the initial face imaging acquisition parameters, extracting definition evidence, brightness evidence and shielding evidence to form an imaging quality evidence sequence, and calculating an imaging quality evidence drift amount according to the imaging quality evidence sequence. According to the invention, through the calculation step of verifying the session entropy, the face verification process is improved from single recognition result judgment to quantitative analysis of the overall uncertainty of the access session.

Description

Technical Field

[0001] This invention relates to the field of smart door lock technology, and in particular to a smart door lock with facial recognition function and its recognition method. Background Technology

[0002] With the rapid development of smart home and smart security technologies, biometric-based access control has gradually become an important research direction. Among them, contactless identity verification technology, represented by facial recognition, is widely used in the field of smart locks due to its convenience and natural interaction characteristics. Existing technologies typically combine sensor triggering, image acquisition, and feature comparison processes to achieve automated identification and control of access behavior, which not only improves user experience but also has significant implications for the security management of residential and office spaces.

[0003] In practical applications, existing smart door lock facial recognition solutions mostly rely on single recognition results or fixed thresholds as the basis for decision-making. They lack a systematic characterization of the overall stability of the access control process and cannot effectively reflect the continuous impact of complex triggering behaviors and changes in imaging conditions on recognition reliability. In scenarios with fluctuating imaging quality or unstable access behavior, the recognition control logic is prone to becoming rigid, affecting the stability and rationality of access control verification. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent door lock with facial recognition function and its recognition method to solve the problem of difficulty in effectively controlling the stability of access behavior and the evolution of facial imaging quality.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a smart door lock recognition method with facial recognition function, comprising: when a trigger event is detected, calculating a verification session entropy value representing the uncertainty of access control based on trigger record information associated with the trigger event and historical recognition record information stored in the door lock; based on the verification session entropy value, selecting a corresponding wake-up strategy to perform delayed wake-up and early wake-up on the facial recognition function, and outputting initial facial imaging acquisition parameters; according to the initial facial imaging acquisition parameters, acquiring continuous facial images and extracting clarity evidence, brightness evidence, and occlusion evidence to form an imaging quality evidence sequence, and calculating the imaging quality evidence drift based on the imaging quality evidence sequence; based on the imaging quality evidence drift, performing closed-loop correction on the facial imaging acquisition parameters and re-acquiring facial images when continuous deterioration of imaging quality is detected, obtaining a valid facial image; based on the valid facial image, performing facial feature extraction and feature comparison to output a facial verification result, and controlling the door lock to perform unlocking, additional verification, and lock-keeping operations based on the facial verification result, while writing the verification session entropy value and the imaging quality evidence drift into the recognition log.

[0007] As a preferred embodiment of the intelligent door lock recognition method with facial recognition function described in this invention, the triggering event includes one of the following: a sensor triggering event formed by the human body sensor in front of the door lock detecting a human body approaching; a visual triggering event formed by detecting a human figure outline or facial outline in the image captured by the door lock camera; and an interactive triggering event formed by the door lock receiving knocking, button pressing and doorbell operations.

[0008] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific steps for calculating the verification session entropy value, which characterizes the uncertainty of access control, are as follows: After a trigger event is established, the trigger record information associated with the current trigger event and the historical identification record information stored in the door lock are read and organized into a session statistics set; Based on the session statistics dataset, the number of triggers, the duration of triggers, and the proportion of recognition results in the current session are calculated to form access control status distribution data. Based on access control status distribution data, the dispersion between the proportions of different statuses is evaluated, the stability of current access control behavior is characterized, and status dispersion evaluation information is generated. Based on the discrete state evaluation information, the uncertainty of this access control access is uniformly quantified, and the corresponding verification session entropy value is obtained.

[0009] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific steps for outputting the initial facial imaging acquisition parameters are as follows: The face imaging acquisition intensity level corresponding to the current access control is obtained based on the magnitude of the verification session entropy value; based on the face imaging acquisition intensity level, the range of values ​​for the required exposure time, supplementary light intensity, and acquisition frame interval for face imaging is obtained; The initial face imaging acquisition parameters for the current access control are formed by combining the values ​​of exposure time, supplementary light intensity, and acquisition frame interval.

[0010] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific steps for forming the imaging quality evidence sequence are as follows: Based on the initial face imaging acquisition parameters, face images are continuously acquired according to the exposure time, illumination intensity and acquisition frame interval, and arranged in the order of acquisition time to form a time-series face image set; Based on a temporal face image set, sharpness evidence, brightness evidence, and occlusion evidence are extracted from each frame of the same face region to form corresponding frame-level quality evidence entries. Based on frame-level quality evidence items, the sharpness evidence, brightness evidence, and occlusion evidence of each frame are sequentially stitched together in chronological order to form an imaging quality evidence sequence.

[0011] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific process of calculating the image quality evidence drift based on the image quality evidence sequence is as follows. Based on the sequence of imaging quality evidence, the imaging quality evidence corresponding to adjacent frames is compared frame by frame in chronological order to obtain information on the direction and magnitude of quality changes. Based on the information of the direction and magnitude of quality change, cumulative statistics are performed on cases where the direction of change remains consistent across multiple consecutive frames to filter out segments with continuous imaging quality changes. Based on the image quality change segments, the cumulative magnitude of quality change over time is uniformly quantified to form a drift assessment quantity. The drift assessment value is used as the drift value of the imaging quality evidence corresponding to the current access control access.

[0012] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific process for obtaining a valid facial image is as follows: The image quality evidence drift is compared with the historical drift records in the current access control session to determine whether the current image quality is in a state of continuous deterioration. When it is determined that the image quality continues to deteriorate, the exposure time, illumination intensity and acquisition frame interval in the current face imaging acquisition parameters are adjusted according to the change characteristics corresponding to the drift amount of the image quality evidence, and the corrected face imaging acquisition parameters are generated. Based on the corrected face imaging acquisition parameters, face image acquisition is re-executed, and a new temporal face image set is formed according to the acquisition time sequence; For a new temporal set of face images, a new sequence of imaging quality evidence is generated, the corresponding imaging quality evidence drift is calculated, and the corresponding face image is taken as a valid face image when no continuous deterioration of imaging quality is detected.

[0013] As a preferred embodiment of the smart door lock recognition method with facial recognition function described in this invention, the specific process for outputting the facial verification result is as follows: Perform face region localization and standardization processing on valid face images to form standardized face image input; Based on the input of a normalized face image, feature vectors representing facial identity features are extracted, and these feature vectors are used as facial feature descriptions corresponding to the current access control access. The facial feature description is compared with the facial feature template stored in the door lock one by one to obtain the corresponding identity matching degree information; Based on the identity matching information, the face verification status corresponding to the current access control access is output as the face verification result.

[0014] As a preferred embodiment of the intelligent door lock recognition method with facial recognition function described in this invention, the step of simultaneously writing the verification session entropy value and the image quality evidence drift amount into the recognition log is as follows: The facial verification status corresponding to the facial verification result is read; when the verification status is successful, the door lock is controlled to perform an unlocking operation to allow current access; when the identity matching degree of the current facial verification is in an intermediate state between success and failure, the door lock is controlled to enter an additional verification process to request further identity confirmation; when the verification status is unsuccessful, the door lock is controlled to remain locked, denying current access; and the verification session entropy value and image quality evidence drift amount associated with the current access are written into the recognition log.

[0015] Secondly, the present invention provides a smart door lock with facial recognition function, including the aforementioned smart door lock recognition method with facial recognition function, characterized in that it includes: an entropy calculation module, an acquisition control module, an imaging analysis module, a closed-loop adjustment module, and a verification control module; the entropy calculation module is used to calculate a verification session entropy value representing the uncertainty of access control based on the trigger record information associated with the trigger event and the historical recognition record information stored in the door lock when a trigger event is detected; the acquisition control module is used to select a corresponding wake-up strategy to perform delayed wake-up and early wake-up for the facial recognition function based on the verification session entropy value, and output initial facial imaging acquisition parameters; the imaging analysis module is used to... Based on the initial face imaging acquisition parameters, continuous face images are acquired, and clarity evidence, brightness evidence, and occlusion evidence are extracted to form an image quality evidence sequence. The image quality evidence drift is calculated based on the image quality evidence sequence. The closed-loop adjustment module, based on the image quality evidence drift, performs closed-loop correction on the face imaging acquisition parameters and re-acquires face images when continuous deterioration of image quality is detected, to obtain valid face images. The verification control module is used to perform face feature extraction and feature comparison based on the valid face images, output face verification results, and control the door lock to perform unlocking, additional verification, and lock maintenance operations based on the face verification results. At the same time, the verification session entropy value and the image quality evidence drift are written into the recognition log.

[0016] The beneficial effects of this invention are as follows: by verifying the calculation steps of session entropy, the face verification process is upgraded from judging a single recognition result to a quantitative analysis of the overall uncertainty of the access control session; based on the trigger behavior and historical recognition distribution, the access stability is uniformly modeled, and the risk status of the current access control is characterized in the form of entropy value, providing a consistent and interpretable decision basis for face imaging wake-up strategy and collection parameter configuration, thereby improving the stability and control rationality of the face verification process under complex access behaviors. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a smart door lock recognition method with facial recognition functionality.

[0019] Figure 2 This is a schematic diagram of a smart door lock with facial recognition functionality.

[0020] Figure 3 A flowchart for processing image quality evidence.

[0021] Figure 4 This is a flowchart of the closed-loop adjustment for face imaging.

[0022] Figure 5 This is a 3D structural diagram of a smart door lock with facial recognition functionality.

[0023] Figure 6 This is a side-view 3D structural diagram of a smart door lock with facial recognition functionality.

[0024] In the image: 1 Main body shell; 2 Face recognition module; 3 Cat's eye lens module; 4 Mechanical key unlocking hole; 5 Control circuit board; 6 Keypad password area; 7 Electric motor module; 8 Fingerprint recognition module. Detailed Implementation

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0028] Reference Figures 1-6 As one embodiment of the present invention, this embodiment provides a smart door lock with facial recognition function and its recognition method, including the following steps: S1: When a trigger event is detected, calculate the verification session entropy value, which represents the uncertainty of access control, based on the trigger record information associated with the trigger event and the historical identification record information stored in the door lock.

[0029] S1.1: Triggering events include one of the following: sensor triggering events triggered by the human body sensor in front of the door lock detecting a human body approaching; visual triggering events triggered by the human body outline or face outline detected in the image captured by the door lock camera; and interactive triggering events triggered by the door lock receiving knocking, button pressing, and doorbell operations.

[0030] Specifically, when the door lock is in standby mode, the human body sensor in front of the door continuously monitors the area in front of the door and generates a sensor trigger signal when a human body is detected approaching; at the same time, the door lock camera analyzes the captured images in real time and generates a visual trigger signal when a human silhouette or face silhouette is detected; when the door lock receives interactive operation input from knocking, button pressing or doorbell, it generates an interactive trigger signal.

[0031] S1.2: After the trigger event is established, read the trigger record information associated with the current trigger event and the historical identification record information stored in the door lock, and organize them into a session statistics data set.

[0032] Specifically, based on the event identifier and occurrence time of the triggered event, trigger record information matching the triggered event is retrieved from the door lock's trigger record storage area; the access control session time range is obtained based on the occurrence time of the triggered event, and historical identification record information that falls within the access control session time range and is associated with the triggered event type is retrieved from the door lock's identification log storage area; the trigger record information and historical identification record information are time-aligned according to the access control session time range, and the fields are unified and organized in a consistent manner to form a session statistics data set containing the trigger record information field set and the historical identification record information field set.

[0033] It should be noted that the trigger record information includes the trigger event type, the time the trigger event occurred, the duration of the trigger event, and the number of times the trigger event occurred.

[0034] Historical identification records include the time of identification, the circumstances of the identification result, and the label indicating the reason for identification failure.

[0035] S1.3: Based on the session statistics data set, calculate the proportion of trigger occurrences, trigger duration, and recognition results in the current session to form access control status distribution data.

[0036] Specifically, the system reads the trigger occurrence count and trigger duration records from the trigger record information field set in the session statistics set. Using the access control session time range as the statistical boundary, it performs a summary count on the trigger occurrence count record and a duration accumulation on the trigger duration record to obtain the total number of trigger occurrences and the total trigger duration corresponding to the current access control session.

[0037] The system reads the records of recognition results occurrence from the historical recognition record information field set in the session statistics data set, and performs classification and merging of the records of recognition results occurrence according to the recognition result category to count the number of occurrences. Based on the total number of trigger occurrences, the total duration of triggers, and the number of occurrences of each recognition result category, the system calculates the percentage of trigger occurrences, the percentage of trigger duration, and the percentage of recognition result occurrences, and organizes them according to the field names to form access control status distribution data.

[0038] It should be noted that the recognition results include at least three categories: passed, failed, and uncertain. Pass indicates that the face comparison result meets the unlocking conditions, fail indicates that the face comparison result is below the minimum pass threshold, and uncertain indicates that the identity cannot be directly determined due to insufficient imaging quality, failure of liveness detection, or timeout. The minimum pass threshold is obtained statistically based on the similarity distribution of historical successfully unlocked samples, for example, 0.75 to 0.85 (calculated as cosine similarity), or the mean minus 1 to 2 times the standard deviation is taken as the pass threshold.

[0039] S1.4: Based on the access control access status distribution data, assess the degree of dispersion between the proportions of different statuses, characterize the stability changes of the current access control behavior, and generate status dispersion assessment information.

[0040] Specifically, the percentage of trigger occurrences, the percentage of trigger durations, and the percentage of recognition results in the access control status distribution data are read and organized into a unified status percentage set for the same session. A percentage difference calculation is performed on the status percentage set, calculating the difference between the maximum and minimum percentages and the deviation of each percentage from the average percentage of the status percentage set, to obtain a dispersion measure. This dispersion measure is then correlated with the access control session time range corresponding to the access control status distribution data to form status dispersion assessment information.

[0041] Let the set of state proportions be { }, corresponding to the percentage of trigger occurrences, the percentage of trigger duration, and the percentage of recognition results occurrence, respectively, and the average percentage of the state percentage set, the formula is: ; In the formula, This represents the average proportion of the set of states. This indicates the percentage of times the event occurred. This indicates the percentage of the duration that the event was triggered. This indicates the percentage of occurrences of the recognition result.

[0042] The range of the set of state proportions is given by the formula: ; In the formula, The range represents the proportion of states in the set. This represents the maximum value operation. This represents the minimum value operation.

[0043] The average deviation of the state proportion set is given by the formula: ; In the formula, This represents the average deviation of the set of states.

[0044] The discreteness characteristic is expressed by the following formula: ; In the formula, Represents the measure of dispersion. This represents the stabilizing factor.

[0045] in, The stability factor was obtained through statistical analysis of the distribution of discreteness values ​​in historical access control sessions. The smallest positive number that does not affect the interval determination interval was selected as the stabilization factor. For example, a lower bound of one order of magnitude of the smallest non-zero value of the historical discreteness value could be used. An example value is... The formulas for the average proportion of the state proportion set, the range of the state proportion set, the average deviation of the state proportion set, and the discreteness characterization quantity are all dimensionless.

[0046] S1.5: Based on the discrete state evaluation information, the uncertainty of this access control access is uniformly quantified, and the corresponding verification session entropy value is obtained.

[0047] Specifically, the discreteness representation quantity in the discrete state evaluation information and the access control session time range associated with the discrete state evaluation information are read, and the discreteness representation quantity is used as the input quantity for the uncertainty quantification of this access control access; interval mapping processing is performed on the discreteness representation quantity, and the discreteness representation quantity is compared with the statistical range of the discreteness representation quantity of the corresponding access control session time range in the historical identification record information, and the relative position of the uncertainty of this access control access in the historical discrete fluctuation is obtained based on the comparison relationship; an uncertainty quantification value is generated based on the relative position, and the uncertainty quantification value is associated and organized with the access control session time range to form a verification session entropy value.

[0048] A superior approach, compared to the usual method of judging access control based on similarity scores, confidence thresholds, or a fixed number of failures in a single identification process, expands the single identification judgment to a behavioral stability analysis of the entire access control session by reading the discrete representation quantity in the discrete evaluation information of the state and combining it with the corresponding access control session time range. It compares the current discrete representation quantity with the discrete fluctuation range of similar access control sessions in historical identification records to obtain the relative position of access control uncertainty in the historical behavior distribution, and further forms a verification session entropy value. This uniformly quantifies the session-level uncertainty of access control, reflects the risk changes under complex access behaviors, and improves the rationality and robustness of access control.

[0049] The mapping function obtains the relative position, and the formula is: ; In the formula, This represents the relative position of the current access control uncertainty within the historical discrete fluctuation statistical range. Indicates relative position value, This represents the discreteness representation of the current access control session. Indicates the current access control session. This represents the minimum value of the discreteness representation obtained statistically from historical identification record information. This represents the maximum value of the discreteness representation obtained statistically from historical identification record information.

[0050] The formula for verifying session entropy is: ; In the formula, This represents the verification session entropy value corresponding to the current access control session. This indicates the time range of the access control session corresponding to the verification session entropy value. Represents the natural logarithm operation.

[0051] in The normalized proportion has a range of values. The formulas for obtaining relative position and verifying session entropy in the mapping function are all dimensionless quantities, and logarithmic operations only apply to dimensionless quantities.

[0052] Both the access control session time range and the access control session refer to the same time boundary concept.

[0053] S2: Based on the verification session entropy value, select the corresponding wake-up strategy to perform delayed wake-up and early wake-up on the face recognition function, and output the initial face imaging acquisition parameters.

[0054] S2.1: Obtain the face imaging acquisition intensity level corresponding to the current access control based on the magnitude of the verification session entropy value.

[0055] Specifically, the verification session entropy value is read and matched one by one with the face image acquisition intensity level judgment conditions stored in the door lock. The face image acquisition intensity level corresponding to the current access is obtained by comparing the verification session entropy value in the interval position corresponding to each judgment condition.

[0056] It should be noted that the judgment criteria refer to the rules for dividing the verification session entropy value range to distinguish different access control risk levels.

[0057] The verification session entropy value can be divided into intervals according to the quantiles of historical samples. For example, P25, P50, and P75 can be used as thresholds to divide the data into low, medium, and high imaging intensity levels. The interval thresholds are obtained by the door lock from the statistics of historical access data.

[0058] When the verification session entropy value is in the low-risk range, the face recognition function is controlled to enter the delayed wake-up mode, and face collection is started only after continuous triggering is detected or the distance meets the judgment conditions; when the verification session entropy value is in the medium-high risk range, the face recognition function is controlled to enter the early wake-up mode.

[0059] S2.2: Based on the intensity level of face imaging acquisition, obtain the range of values ​​for the required exposure time, supplementary light intensity, and acquisition frame interval for face imaging.

[0060] Specifically, the system reads the facial imaging intensity level corresponding to the current access control access and uses it as a parameter index. Historically successful unlocking sessions are grouped according to the verification session entropy value range. Within each group, the distribution of exposure time, fill light intensity, and acquisition frame interval is statistically analyzed. After removing outliers, the upper and lower quantile boundaries of each parameter distribution are extracted as parameter variation ranges, forming a mapping relationship between the verification session entropy value range and the imaging parameter value range. In the current access control access, the facial imaging intensity level is used as an index condition, and the corresponding exposure time, fill light intensity, and acquisition frame interval value ranges are obtained from the mapping relationship.

[0061] The upper and lower quantile boundaries were obtained using the statistical quantile method. The lower quantile was the 25th percentile (P25) of the historical successfully unlocked samples, and the upper quantile was the 75th percentile (P75). Before the statistics were compiled, outliers (samples exceeding Q1−1.5×IQR or Q3+1.5×IQR) were removed using the interquartile range method.

[0062] S2.3: Based on the range of values ​​for exposure time, supplementary light intensity, and acquisition frame interval, the initial face imaging acquisition parameters for the current access control are combined to form the parameters.

[0063] Specifically, based on the range of exposure time, fill light intensity, and acquisition frame interval, parameter values ​​that match the current access control face imaging acquisition intensity level are selected, and then integrated according to the parameter combination relationship of exposure time, fill light intensity, and acquisition frame interval to generate the initial face imaging acquisition parameters for the current access control.

[0064] The parameter values ​​can be selected from the parameter values ​​that appear most frequently in the historical successful unlock sessions within the corresponding interval.

[0065] S3: Based on the initial face imaging acquisition parameters, acquire continuous face images and extract sharpness evidence, brightness evidence and occlusion evidence to form an imaging quality evidence sequence, and calculate the imaging quality evidence drift based on the imaging quality evidence sequence.

[0066] S3.1: Based on the initial face imaging acquisition parameters, face images are continuously acquired according to the exposure time, illumination intensity and acquisition frame interval, and arranged in the order of acquisition time to form a time-series face image set.

[0067] Specifically, the initial face imaging acquisition parameters corresponding to the current access control are read, and the exposure duration is written to the camera exposure control item, the fill light intensity is written to the fill light control item, and the acquisition frame interval is written to the acquisition rhythm control item. After the exposure duration, fill light intensity, and acquisition frame interval take effect, the face image acquisition process is started. The acquisition time is recorded each time acquisition occurs, and the acquisition time is bound and stored with the corresponding face image. Face image acquisition is repeated according to the acquisition frame interval until the acquisition termination condition of the current access control is reached. The bound and stored face images are sorted according to the acquisition time, and the sorted face images are arranged sequentially by time index to form a time-series face image set.

[0068] It should be noted that the acquisition termination condition is that during continuous acquisition, when the acquired face image meets the preset requirements for face region integrity and basic imaging quality, or when the continuous acquisition reaches the maximum acquisition time or maximum number of acquisition frames allowed by the current access control, the face image acquisition will be terminated.

[0069] The preset requirements for face region integrity and basic imaging quality are obtained by statistical analysis of face images from successful unlocking sessions in the history of door locks; the face region occupies the image area in the face region integrity requirement.

[0070] For example, the ratio can be set to 40%–80%; the sharpness evaluation value in the basic imaging quality requirements can be set to the lower quantile value that is greater than the average sharpness value of historical successful samples, and the lower quantile value is clearly defined as the 25th percentile; the brightness evaluation value can be set to the middle interval of the brightness distribution of historical successful samples, and the middle interval is clearly defined as [Q1, Q3], where Q1 is the first quartile and Q3 is the third quartile.

[0071] S3.2: Based on the temporal face image set, extract sharpness evidence, brightness evidence and occlusion evidence for each frame of the same face region and form corresponding frame-level quality evidence entries.

[0072] Specifically, face region localization is performed on the first frame of the temporal face image set to obtain the face region boundary, and the face region boundary is used as the tracking and updating cropping reference. The same face region is cropped for each frame of the temporal face image set. Within the cropped face region, sharpness evidence representing edge sharpness, brightness evidence representing exposure level and brightness stability, and occlusion evidence representing the proportion of usable face texture area are calculated respectively. The acquisition time, sharpness evidence, brightness evidence, and occlusion evidence corresponding to each frame image are combined to form a frame-level quality evidence entry that corresponds one-to-one with the frame index.

[0073] Among them, clarity evidence is calculated using gradient or Laplacian variance, brightness evidence is calculated using the mean and fluctuation of grayscale in the face region, and occlusion evidence is calculated using the percentage of visible face pixels or key points.

[0074] S3.3: Based on frame-level quality evidence items, the sharpness evidence, brightness evidence, and occlusion evidence of each frame are sequentially spliced ​​in chronological order to form an imaging quality evidence sequence.

[0075] Specifically, the frame-level quality evidence entries corresponding to each frame of the time-series face image set are read, and the frame-level quality evidence entries are sorted by time based on the acquisition time in the frame-level quality evidence entries to obtain the order of the frame-level quality evidence entries; the sharpness evidence, brightness evidence, and occlusion evidence corresponding to each frame are extracted in the sorting order and written into the same sequence position to form sequence elements; all sequence elements are arranged continuously in time sorting order, and the sequence elements are associated with the corresponding acquisition time and saved to form an imaging quality evidence sequence.

[0076] S3.4: Based on the imaging quality evidence sequence, compare the imaging quality evidence corresponding to adjacent frames frame by frame in chronological order to obtain information on the direction and magnitude of quality changes.

[0077] Specifically, the order of each sequence element in the imaging quality evidence sequence is obtained based on the acquisition time associated with the imaging quality evidence sequence; two adjacent sequence elements are used as a comparison group to extract the sharpness evidence, brightness evidence, and occlusion evidence corresponding to the preceding sequence element and the sharpness evidence, brightness evidence, and occlusion evidence corresponding to the following sequence element.

[0078] Difference calculations are performed on the sharpness evidence, the brightness evidence, and the occlusion evidence. The sign of the difference is used as the direction of quality change, and the absolute value of the difference is used as the magnitude of quality change. The direction and magnitude of quality change corresponding to the sharpness evidence, the brightness evidence, and the occlusion evidence are written into the quality change information set in chronological order to form the information on the direction and magnitude of quality change.

[0079] S3.5: Based on the information of quality change direction and change magnitude, cumulative statistics are performed on cases where the change direction remains consistent across multiple consecutive frames to filter out segments with continuous imaging quality changes.

[0080] Specifically, the direction and magnitude of quality changes are read in chronological order of acquisition, and consistency judgment is performed on the direction of quality changes corresponding to sharpness evidence, brightness evidence, and occlusion evidence at adjacent time positions. When at least one type of quality change direction remains consistent across multiple consecutive frames, cumulative statistics are performed on the number of consecutive frames and the magnitude of quality changes. Accumulation ends when the direction of quality change is reversed or interrupted. The continuity judgment is performed on the continuous change segments obtained from the cumulative statistics, segments with insufficient continuity length are removed, and segments with continuity length meeting the trend requirements are retained. The retained segments are regarded as continuous imaging quality change segments.

[0081] It should be noted that when the number of frames changes continuously in the same direction... Cumulative change range And duration The segment was identified as a valid change segment, among which Obtained from the statistical quantile values ​​of historically successfully unlocked samples.

[0082] S3.6: Based on the image quality change segments, the cumulative situation of the quality change amplitude in the time dimension is uniformly quantified to form a drift evaluation quantity; the drift evaluation quantity is used as the image quality evidence drift quantity corresponding to the current access control access.

[0083] Specifically, the starting acquisition time, ending acquisition time, direction of quality change, and cumulative quality change amplitude are obtained for the corresponding image quality change segments. Within the same image quality change segment, time span normalization processing is performed on the quality change amplitudes of sharpness evidence, brightness evidence, and occlusion evidence. The cumulative quality change amplitude is converted using the time difference between the starting acquisition time and the ending acquisition time as the normalization benchmark to obtain a characterization of the change intensity per unit time.

[0084] The intensity of change per unit time for clarity evidence, brightness evidence, and occlusion evidence is combined and summarized to form a drift assessment quantity, which is then used as the drift quantity of the imaging quality evidence corresponding to the current access control session.

[0085] A superior approach, compared to existing technologies that only judge the imaging quality of a single frame or discrete frames, is to uniformly quantify the evolution trend of imaging quality over time, forming the imaging quality evidence drift amount for closed-loop correction and access control decisions.

[0086] The start and end times of the image quality change segment are respectively (Start time of data collection) and (Time when data collection ended), and > .

[0087] The intensity of change per unit time is characterized by the following formula: ; In the formula, This characterizes the intensity of change in clarity evidence per unit time. Evidence of clarity This indicates the cumulative magnitude of quality change in the sharpness evidence within a segment of image quality variation. The intensity of the change in brightness per unit time is characterized by its representation. Indicates evidence of brightness. This indicates the cumulative magnitude of quality change in brightness evidence within a segment of image quality variation. The intensity of change per unit time representing the obstruction of evidence. This indicates that the evidence is being obscured. This indicates the cumulative magnitude of quality change in the occluded evidence within the segment showing the change in image quality.

[0088] The cumulative quality change amplitude is obtained by accumulating the change amplitude of corresponding quality evidence between adjacent frames within the imaging quality change segment.

[0089] , and Both are representations of the intensity of change per unit time obtained by normalizing the change amplitude of the corresponding imaging quality evidence over the same time span. The numerator is a dimensionless evaluation quantity, and the denominator is a time quantity, with consistent dimensions.

[0090] S4: Based on the image quality evidence drift, when continuous deterioration of image quality is detected, closed-loop correction is performed on the face image acquisition parameters and the face image is reacquired to obtain a valid face image.

[0091] S4.1: Compare the image quality evidence drift with the historical drift records in the current access control session to determine whether the current image quality is in a state of continuous deterioration.

[0092] Specifically, the system reads the image quality evidence drift corresponding to the current access control session and retrieves historical drift records from the door lock that are adjacent to or related to the same type of scenario in the time range of the current access control session. The system compares the current image quality evidence drift with the drift trend in the historical drift records item by item to determine whether the current image quality evidence drift shows an increasing or deteriorating characteristic in the same direction in the continuous time dimension. When the current image quality evidence drift deviates continuously from the historical drift records and does not fall back, it is determined that the current image quality is in a state of continuous deterioration.

[0093] It should be noted that when the drift of imaging quality evidence in multiple consecutive time intervals shows the same direction of change relative to their respective historical drift record baselines and the magnitude of change does not show a reverse decline, it is considered to be an increase or deterioration in the same direction. The same direction of change means that the current drift of imaging quality evidence in consecutive time intervals is positively deviated from the previous time interval and the corresponding historical average. A reverse decline means that there is no situation in the consecutive time interval where the drift decreases and returns to the historical fluctuation range.

[0094] Similar scenarios refer to historical access scenarios that are consistent with the current access control access, based on the lighting level obtained from the range of supplementary light intensity and ambient brightness, the time period divided by daytime, nighttime or high-frequency and low-frequency usage periods, and the installation environment with the same door lock location as the reference.

[0095] S4.2: When it is determined that the image quality continues to deteriorate, based on the change characteristics corresponding to the drift amount of the image quality evidence, targeted adjustments are made to the exposure time, supplementary light intensity and acquisition frame interval in the current face imaging acquisition parameters to generate corrected face imaging acquisition parameters.

[0096] Specifically, when it is determined that the image quality is continuously deteriorating, the change characteristics reflected in the image quality evidence drift are read, and the source of image quality deterioration is distinguished; when the image quality evidence drift shows that the sharpness is continuously decreasing, the acquisition frame interval in the face imaging acquisition parameters is shortened or the exposure time is adjusted to improve the imaging stability.

[0097] When the intensity of the change in brightness evidence per unit time remains unfavorable over multiple consecutive acquisition periods, and continues to increase without decreasing relative to historical drift records, it is determined that the drift amount of the imaging quality evidence indicates a continuous deterioration in brightness-related evidence, and adjustments are made to the exposure time or supplementary light intensity in the face imaging acquisition parameters.

[0098] When the intensity of the change in unit time corresponding to the occlusion evidence increases in the same direction in continuous image quality change segments, and the change amplitude is consistently higher than that of the sharpness evidence and brightness evidence or higher than the occlusion change level of similar historical access control sessions, it is determined that the change in occlusion-related evidence has intensified. The acquisition frame interval is adjusted to improve the probability of obtaining effective frames. The adjusted exposure time, supplementary light intensity and acquisition frame interval are combined to form the corrected face imaging acquisition parameters.

[0099] It should be noted that the change characteristics refer to the intensity and direction of change per unit time of the sharpness evidence, brightness evidence, and occlusion evidence, respectively, during the formation of the image quality evidence drift.

[0100] After determining that the image quality continues to deteriorate, the imaging parameters are finely adjusted in a closed loop according to a preset step size. The exposure time is adjusted by a step size Δe, the fill light intensity by a step size Δl, and the acquisition frame interval by a step size Δf. Each parameter is always limited to the range of values ​​for the exposure time, fill light intensity, and acquisition frame interval. The preset step size is set by the adjustment range of the imaging parameters in historical successful unlock sessions. For example, the exposure time is 5%–10% of the adjustable range, the fill light intensity is 5%–15%, and the acquisition frame interval is 1–2 frames. It is limited to the safe upper and lower limits that do not cause imaging instability or overexposure.

[0101] S4.3: Re-execute face image acquisition based on the corrected face imaging acquisition parameters, and form a new temporal face image set according to the acquisition time sequence.

[0102] Specifically, the exposure time, fill light intensity, and acquisition frame interval in the corrected face imaging acquisition parameters are written into the corresponding imaging control items to apply the corrected face imaging acquisition parameters; the face image acquisition process is restarted, and the corresponding acquisition time is recorded and associated with the acquired face image during each face image acquisition; face image acquisition is continuously performed according to the acquisition frame interval in the corrected face imaging acquisition parameters, and acquisition ends after the termination condition for re-acquisition is met; the re-acquisitioned face images are sorted according to the acquisition time, and the sorted face images are arranged sequentially to form a new temporal face image set.

[0103] S4.4: Regenerate the imaging quality evidence sequence for the new temporal face image set, calculate the corresponding imaging quality evidence drift, and take the corresponding face image as a valid face image when no continuous deterioration of imaging quality is detected.

[0104] Specifically, for the new temporal face image set, clarity evidence, brightness evidence, and occlusion evidence are re-extracted according to the established face region localization method, and a new imaging quality evidence sequence is generated in the order of acquisition time. Based on the new imaging quality evidence sequence, the corresponding imaging quality evidence drift is calculated, and the new imaging quality evidence drift is compared with the historical drift record in the current access control session. When the new imaging quality evidence drift does not show a continuously increasing or accumulating deterioration trend in the same direction, it is determined that the current imaging quality has reached a stable state, and the face images in the new temporal face image set that meet the requirements of face region integrity and basic imaging quality are taken as valid face images.

[0105] Among them, when the drift of new imaging quality evidence does not show a continuous increase in the same direction in the continuous acquisition time series, and the value is within the stable fluctuation range corresponding to the historical drift record or shows a downward trend, it is determined that there is no deterioration trend of continuous increase or accumulation in the same direction.

[0106] S5: Based on the valid face image, perform face feature extraction and feature comparison to output the face verification result, and control the door lock to perform unlocking, additional verification and lock maintenance operations according to the face verification result. At the same time, write the verification session entropy value and the image quality evidence drift amount into the recognition log.

[0107] S5.1: Perform face region localization and standardization processing on valid face images to form standardized face image input.

[0108] Specifically, face region localization is performed on the valid face image to obtain the face region boundary, and the valid face image is cropped according to the face region boundary to remove non-face regions; within the cropped face region, scale normalization and position alignment processing are performed on the face image to keep the size ratio and spatial position of the face in the image consistent, and the brightness and contrast of the face image are basically normalized to form a normalized face image input.

[0109] S5.2: Based on the normalized face image input, extract the feature vector representing the face identity features, and use the feature vector as the face feature description corresponding to the current access control access.

[0110] Specifically, the normalized face image is input into the face feature extraction process to extract texture features from the local texture regions in the normalized face image. The texture features are then combined with the spatial distribution relationship of each local texture region in a unified coordinate system to form a multi-dimensional numerical representation of the differences in face identity. The multi-dimensional numerical representations are combined to form a face feature vector that can distinguish different identities. The face feature vector is associated with the current access control session and saved, and the face feature vector is used as the face feature description corresponding to the current access control session.

[0111] Among them, the local texture region refers to the image sub-region in a normalized face image that has a stable texture structure, can reflect identity differences, and maintains relative consistency under multiple imaging conditions.

[0112] The facial feature extraction process can use a convolutional neural network to encode features in a normalized facial image, outputting an embedding vector with a dimension of 128 or 512, and then normalizing and comparing the feature vectors using cosine similarity to characterize the degree of difference between different identities.

[0113] S5.3: Compare the facial feature description with the facial feature template stored in the door lock one by one to obtain the corresponding identity matching degree information.

[0114] Specifically, the system sequentially reads the stored facial feature templates from the door lock, compares the facial feature description corresponding to the current access control with each facial feature template one by one, calculates the similarity between the facial feature description and the corresponding facial feature template, quantifies the matching relationship between the facial feature description and the corresponding facial feature template, and summarizes the similarity values ​​obtained from each comparison to obtain the corresponding identity matching information.

[0115] It should be noted that the similarity is calculated by normalizing the feature vector similarity between the facial feature description and the facial feature template.

[0116] S5.4: Based on the identity matching degree information, output the face verification status corresponding to the current access control as the face verification result.

[0117] Specifically, the system reads the identity matching degree information and compares it with the pre-stored identity matching judgment range in the door lock. Based on the range in which the identity matching degree information falls, it determines the face verification status corresponding to the current access control access. When the identity matching degree information falls into the pass judgment range, the face verification status is marked as pass. When the identity matching degree information falls into the uncertain range, the face verification status is marked as uncertain. When the identity matching degree information falls into the fail judgment range, the face verification status is marked as fail. The determined face verification status is then used as the face verification result for the current access control access.

[0118] It should be noted that the distribution statistics of identity matching information corresponding to historically successful unlocking sessions are performed, and the stable high-value interval of the successful sample distribution is extracted as the pass determination interval; the distribution statistics of identity matching information corresponding to historically failed identification sessions are performed, and the stable low-value interval of the failed sample distribution is extracted as the fail determination interval; the transition interval between the pass determination interval and the fail determination interval is determined as the uncertain interval.

[0119] The pass / fail judgment interval can be set as the degree of identity matching being greater than or equal to the 75th percentile of the successful sample distribution, and the fail / fail judgment interval can be set as the degree of identity matching being less than or equal to the 25th percentile of the failed sample distribution. The degree of identity matching between the pass / fail judgment interval and the fail / fail judgment interval is the uncertain interval.

[0120] S5.5: Read the face verification status corresponding to the face verification result. When the verification status indicates that the verification is successful, control the door lock to perform an unlocking operation to allow the current access control.

[0121] Specifically, the system reads the facial verification status corresponding to the current access control access and determines whether the facial verification status indicates success. It then compares the identity matching information with the pre-set pass judgment range in the door lock. When the identity matching information falls into the pass judgment range, the system determines that the facial verification status indicates success and sends an unlocking command to the door lock's actuator to drive the bolt to complete the unlocking action, allowing the current access control access to enter.

[0122] It should be noted that after cleaning and denoising the identity matching information in the successful unlocking sessions of the door lock in the past, the information is summarized according to the same scenario, and its distribution characteristics are statistically analyzed. Stable quantile intervals (e.g., the lower quantile value of the successful sample is used as the lower bound and the upper quantile value is used as the upper bound) are selected as the set pass judgment intervals.

[0123] S5.6: The current facial verification identity matching degree is in an intermediate state between passing and failing. Control the door lock to enter the additional verification process and request further identity confirmation operation.

[0124] Specifically, when the identity matching information is in the middle range between the pass and fail judgment ranges, it is marked as an uncertain state and the unlocking operation is suspended; when the additional verification process triggers the uncertain state, it selects and executes additional identity verification methods, such as fingerprint or password verification, and guides the execution of supplementary verification operations by issuing a further identity confirmation request.

[0125] S5.7: When the verification status representation fails, control the door lock to remain locked and refuse the current access control access; and write the verification session entropy value and imaging quality evidence drift associated with the current access control access into the recognition log.

[0126] Specifically, the system reads the identity matching degree information and compares it with the judgment range. If the matching degree is lower than the upper limit of the failure judgment range, the face verification status is determined to be failed, the unlock command is prohibited, and the door lock is kept locked to refuse the current access control access. At the same time as refusing access, the verification session entropy value and imaging quality evidence drift amount corresponding to the current access control access session are associated and organized with the failed face verification status and written into the recognition log.

[0127] The identification log is indexed by the session identifier and includes at least the access control session time range, trigger event type, verification session entropy value, image quality evidence drift, face verification status, and failure reason label.

[0128] This embodiment also provides a smart door lock with face recognition function, including: an entropy calculation module, an acquisition control module, an imaging analysis module, a closed-loop adjustment module, and a verification control module; the entropy calculation module is used to calculate the verification session entropy value, which characterizes the uncertainty of access control, based on the trigger record information associated with the trigger event and the historical recognition record information stored in the door lock when a trigger event is detected; the acquisition control module is used to select the corresponding wake-up strategy to perform delayed wake-up and early wake-up for the face recognition function based on the verification session entropy value, and output the initial face imaging acquisition parameters; the imaging analysis module is used to acquire continuous face images based on the initial face imaging acquisition parameters. The system extracts facial images and gathers clarity, brightness, and occlusion evidence to form an imaging quality evidence sequence. Based on this sequence, it calculates the imaging quality evidence drift. A closed-loop adjustment module, based on the drift, corrects the facial imaging acquisition parameters and re-acquires facial images when continuous deterioration in imaging quality is detected, obtaining valid facial images. A verification control module, based on the valid facial images, performs facial feature extraction and feature comparison, outputting facial verification results. Based on these results, it controls the door lock to perform unlocking, additional verification, and lock maintenance operations, while simultaneously writing the verification session entropy value and the imaging quality evidence drift into the recognition log.

[0129] In summary, this invention elevates the face verification process from judging a single recognition result to a quantitative analysis of the overall uncertainty of the access control session by: verifying the calculation step of session entropy; and by uniformly modeling access stability based on triggering behavior and historical recognition distribution, using entropy to characterize the current access control risk state, providing a consistent and interpretable decision-making basis for face imaging wake-up strategies and acquisition parameter configuration, thereby improving the stability and control rationality of the face verification process under complex access behaviors.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart door lock recognition method with facial recognition function, characterized in that: include, When a trigger event is detected, the verification session entropy value, which characterizes the uncertainty of access control, is calculated based on the trigger record information associated with the trigger event and the historical identification record information stored in the door lock. Based on the verification session entropy value, the corresponding wake-up strategy is selected to perform delayed wake-up and early wake-up on the face recognition function, and the initial face imaging acquisition parameters are output. Based on the initial face imaging acquisition parameters, continuous face images are acquired and clarity evidence, brightness evidence and occlusion evidence are extracted to form an imaging quality evidence sequence. The imaging quality evidence drift is calculated based on the imaging quality evidence sequence. Based on the image quality evidence drift, when continuous deterioration of image quality is detected, closed-loop correction is performed on the face image acquisition parameters and face images are reacquired to obtain valid face images. Based on valid face images, face feature extraction and feature comparison are performed to output face verification results. Based on the face verification results, the door lock is controlled to perform unlocking, additional verification, and lock maintenance operations. At the same time, the verification session entropy value and the image quality evidence drift amount are written into the recognition log.

2. The smart door lock recognition method with facial recognition function as described in claim 1, characterized in that: The triggering events include one of the following: a sensory triggering event triggered by the human body sensor in front of the door lock detecting a human body approaching; a visual triggering event triggered by the human figure or face outline detected in the image captured by the door lock camera; and an interactive triggering event triggered by the door lock receiving knocking, button pressing, and doorbell operations.

3. The smart door lock recognition method with facial recognition function as described in claim 2, characterized in that: The specific steps for calculating the verification session entropy value, which characterizes the uncertainty of access control, are as follows: After a trigger event is established, the trigger record information associated with the current trigger event and the historical identification record information stored in the door lock are read and organized into a session statistics set; Based on the session statistics dataset, the number of triggers, the duration of triggers, and the proportion of recognition results in the current session are calculated to form access control status distribution data. Based on access control status distribution data, the dispersion between the proportions of different statuses is evaluated, the stability of current access control behavior is characterized, and status dispersion evaluation information is generated. Based on the discrete state evaluation information, the uncertainty of this access control access is uniformly quantified, and the corresponding verification session entropy value is obtained.

4. The smart door lock recognition method with facial recognition function as described in claim 3, characterized in that: The specific steps for outputting the initial face image acquisition parameters are as follows: The facial imaging intensity level corresponding to the current access control access is obtained based on the magnitude of the verification session entropy value. Based on the intensity level of face imaging acquisition, obtain the range of values ​​for the required exposure time, supplementary light intensity, and acquisition frame interval for face imaging; The initial face imaging acquisition parameters for the current access control are formed by combining the values ​​of exposure time, supplementary light intensity, and acquisition frame interval.

5. The smart door lock recognition method with facial recognition function as described in claim 4, characterized in that: The specific steps for forming the imaging quality evidence sequence are as follows: Based on the initial face imaging acquisition parameters, face images are continuously acquired according to the exposure time, illumination intensity and acquisition frame interval, and arranged in the order of acquisition time to form a time-series face image set; Based on a temporal face image set, sharpness evidence, brightness evidence, and occlusion evidence are extracted from each frame of the same face region to form corresponding frame-level quality evidence entries. Based on frame-level quality evidence items, the sharpness evidence, brightness evidence, and occlusion evidence of each frame are sequentially stitched together in chronological order to form an imaging quality evidence sequence.

6. The smart door lock recognition method with facial recognition function as described in claim 5, characterized in that: The specific process for calculating the image quality evidence drift based on the image quality evidence sequence is as follows. Based on the sequence of imaging quality evidence, the imaging quality evidence corresponding to adjacent frames is compared frame by frame in chronological order to obtain information on the direction and magnitude of quality changes. Based on the information of the direction and magnitude of quality change, cumulative statistics are performed on cases where the direction of change remains consistent across multiple consecutive frames to filter out segments with continuous imaging quality changes. Based on the image quality change segments, the cumulative magnitude of quality change over time is uniformly quantified to form a drift assessment quantity. The drift assessment value is used as the drift value of the imaging quality evidence corresponding to the current access control access.

7. The smart door lock recognition method with facial recognition function as described in claim 6, characterized in that: The specific process for obtaining a valid facial image is as follows. The image quality evidence drift is compared with the historical drift records in the current access control session to determine whether the current image quality is in a state of continuous deterioration. When it is determined that the image quality continues to deteriorate, the exposure time, illumination intensity and acquisition frame interval in the current face imaging acquisition parameters are adjusted according to the change characteristics corresponding to the drift amount of the image quality evidence, and the corrected face imaging acquisition parameters are generated. Based on the corrected face imaging acquisition parameters, face image acquisition is re-executed, and a new temporal face image set is formed according to the acquisition time sequence; For a new temporal set of face images, a new sequence of imaging quality evidence is generated, the corresponding imaging quality evidence drift is calculated, and the corresponding face image is taken as a valid face image when no continuous deterioration of imaging quality is detected.

8. The smart door lock recognition method with facial recognition function as described in claim 7, characterized in that: The specific process for outputting the face verification result is as follows: Perform face region localization and standardization processing on valid face images to form standardized face image input; Based on the input of a normalized face image, feature vectors representing facial identity features are extracted, and these feature vectors are used as facial feature descriptions corresponding to the current access control access. The facial feature description is compared with the facial feature template stored in the door lock one by one to obtain the corresponding identity matching degree information; Based on the identity matching information, the face verification status corresponding to the current access control access is output as the face verification result.

9. The smart door lock recognition method with facial recognition function as described in claim 8, characterized in that: Simultaneously, the verification session entropy value and the image quality evidence drift amount are written into the recognition log. The specific process is as follows: Read the face verification status corresponding to the face verification result. When the verification status indicates that the verification is successful, control the door lock to perform an unlocking operation to allow the current access control. The current facial verification is in an intermediate state between success and failure. The door lock is then activated to enter an additional verification process, requesting further identity confirmation. When the verification status representation fails, the control door lock remains locked, denying the current access control access; and the verification session entropy value and imaging quality evidence drift associated with the current access control access are written to the recognition log.

10. A smart door lock with facial recognition function, based on the smart door lock recognition method with facial recognition function according to any one of claims 1 to 9, characterized in that: It includes an entropy calculation module, an acquisition and control module, an imaging analysis module, a closed-loop adjustment module, and a verification and control module; The entropy calculation module is used to calculate the verification session entropy value, which represents the uncertainty of access control, based on the trigger record information associated with the trigger event and the historical identification record information stored in the door lock when a trigger event is detected. The acquisition control module is used to select the corresponding wake-up strategy based on the verification session entropy value to perform delayed wake-up and early wake-up on the face recognition function, and output the initial face imaging acquisition parameters. The imaging analysis module is used to acquire continuous face images based on the initial face imaging acquisition parameters and extract clarity evidence, brightness evidence and occlusion evidence to form an imaging quality evidence sequence, and calculate the imaging quality evidence drift based on the imaging quality evidence sequence. The closed-loop adjustment module, based on the image quality evidence drift, performs closed-loop correction on the face imaging acquisition parameters and re-acquires face images when it detects continuous deterioration of image quality, in order to obtain effective face images. The verification control module is used to perform facial feature extraction and feature comparison based on a valid facial image, output facial verification results, and control the door lock to perform unlocking, additional verification, and locking operations according to the facial verification results. At the same time, the verification session entropy value and the image quality evidence drift amount are written into the recognition log.

Images