Identity recognition access control method and system based on pet facial images

Abstract

The present application relates to the technical field of image recognition, and particularly relates to a pet face image-based identity recognition access control method and system, which collects pet face images in a target area, combines a motion triggering mechanism with image screening logic, extracts image data with time sequence effectiveness, compares the effective image with a registered identity library, outputs an access control instruction according to a similarity score, simultaneously constructs an image time sequence, evaluates recognition continuity, and improves recognition accuracy; after abnormal detection, the system automatically executes a sensitivity correction process, dynamically adjusts interval threshold values and recognition parameters. The system includes image collection, screening, verification, approval and correction modules, and has the technical features of accurate recognition, response closed loop and autonomous learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to a pet face image-based identity recognition access control method and system. BACKGROUND

[0002] With the continuous improvement of people's living standards, the status of pets in family life is increasingly enhanced, and more and more pets are regarded as family members. Pet management and safety control have also attracted widespread attention. In scenarios such as smart homes, community properties, and pet boarding agencies, how to achieve intelligent management of pet access behavior has become a hot direction for technical research and product development.

[0003] Currently, traditional pet access control mainly relies on physical identification devices, such as wearing radio frequency identification (RFID) tags, positioning collars, or inductive keys. These methods have many shortcomings in actual use, such as: pets are prone to losing or removing tags, identification accuracy is greatly affected by interference, accurate individual identity recognition cannot be achieved, and it is difficult to achieve differentiated management in a multi-pet environment. In addition, for unfamiliar animals without identification devices, the system cannot effectively reject their entry, posing a safety hazard.

[0004] In recent years, with the rapid development of computer vision and artificial intelligence technology, face recognition technology has been widely applied in security, finance, transportation, and other fields. Compared with traditional tag identification methods, face recognition has the advantages of non-contact, high precision, and unforgeability. Although face recognition technology is relatively mature, pet face recognition still faces challenges, such as small differences in pet facial features, flexible movements, and large posture changes, which limit its application and promotion in access control.

[0005] Therefore, there is an urgent need to provide a pet face image-based identity recognition access control method and system to achieve efficient identification and safe control of pet individuals, and to make up for the shortcomings of traditional methods in reliability, practicality, and intelligence. SUMMARY

[0006] To solve the above problems, the present application aims to provide a pet face image-based identity recognition access control method, comprising the following steps:

[0007] S1, effective screening: obtaining multiple sets of pet face images in a target area, and screening the face images according to the action changes performed by the pets, and extracting relevant effective face images; the screening includes judging the face orientation in the image, image clarity, face occlusion, etc., and combining image timestamp information to preliminarily filter out interference frames;

[0008] S2, identity verification output: encode the feature vector of each valid face image extracted, and match it with the registered target pet face features in the database, calculate the similarity score; if the score is higher than the preset confidence threshold, trigger the door opening instruction to release; if it is lower than the threshold, trigger the alarm instruction and record the current image and time information;

[0009] S3, identity rationality review: continuously track and collect the target pet image confirmed by comparison, form an image time sequence as the identity tracking basis; statistically analyze the image frame interval in the image time sequence, if the image interval is continuous and falls within the preset reasonable time interval, confirm the time sequence continuity and stability of the identity comparison, and improve the reliability of the recognition result; otherwise, mark the image and subsequent frames as abnormal and submit them for correction;

[0010] S4, sensitivity correction: obtain pet face images with different time intervals, construct a standardized test sequence, and calculate the average interval and fluctuation amplitude of consecutive images in the sequence; determine whether the current sensitivity setting will lead to misjudgment or omission by simulating the identification process; if the test recognition rate is lower than the threshold, automatically generate a correction coefficient and use it to adjust the time interval parameters of the next round of image acquisition and screening, thereby dynamically optimizing the response sensitivity of the system.

[0011] In a preferred technical solution, in step S1, the action change of the pet is perceived to define a judgment period, all face images in the period are extracted as judgment images, and are sorted in chronological order; the time interval between adjacent images is defined as an interval reference value, which is compared with the multiple interval threshold values preset in the system: if the interval reference value is less than any interval threshold value, the image is identified as a valid image; if it is greater than all interval threshold values, the image is not considered as valid data for comparison.

[0012] In a preferred technical solution, the interval threshold values are multiple preset image acquisition interval intervals, each interval sets a matching threshold value as the judgment standard for whether the image enters the comparison process.

[0013] In a preferred technical solution, in step S2, the feature points of the valid image and the standard image in the target pet identity library are matched, and a confidence score is output;

[0014] Compare the score with the confidence threshold preset by the system: if the confidence score is greater than or equal to the confidence threshold S ≥ S t , it is considered that the identity comparison is passed; otherwise, it is marked as identity inconsistency and an alarm is triggered; the alarm content includes the current image data, the comparison score and the alarm triggering time.

[0015] In a preferred technical solution, the identity comparison result is combined with the access control hardware state for review: if the access control has been opened but no target pet is detected to pass through, or an unauthorized pet is detected to enter in the alarm state, the system automatically marks the comparison image as abnormal and generates a review prompt uploaded to the background server.

[0016] In a preferred technical solution, in the S3 step, the average interval and fluctuation variance of the sequence are calculated by analyzing the time difference between consecutive image frames; if the average interval is within the permitted interval and the fluctuation variance is below the stability threshold, it is determined that the comparison sequence is reasonable; otherwise, it is marked as an unreasonable sequence, and the image and pet information are submitted to the S4 step for processing.

[0017] In a preferred technical solution, the time interval between consecutive images is determined by bidirectional logic, i.e., if the interval reference value falls within the permitted interval, the image is considered to be a reasonable image; otherwise, it is considered to be a to-be-corrected object and enters the to-be-compared sequence as potential abnormal data.

[0018] The present application also provides a pet face image-based identity recognition access control system for implementing the method, comprising:

[0019] An image acquisition module for acquiring multiple sets of pet face images in a target area in real time and attaching time stamps;

[0020] An image screening module for analyzing the dynamic characteristics of the pets in the images and extracting high-quality effective images;

[0021] An identity verification module for comparing the screened images with a target pet identity library and outputting a matching score and an open door instruction;

[0022] An identity approval module for generating an image time sequence, analyzing the image interval, and evaluating the logical continuity of the recognition process;

[0023] A sensitivity correction module for constructing a test process, generating sensitivity correction parameters, and automatically updating the set values of the screening algorithm.

[0024] The present application also provides a pet face image-based identity recognition access control terminal, comprising a processor, a memory, and a bus connecting the elements; the memory stores a computer program, which is executed by the processor to enable the processor to implement all the steps of the method.

[0025] Advantages

[0026] The pet face image-based identity recognition access control method and system provided by the present application have the following technical improvements and effects compared with the existing recognition methods relying on physical tags (such as RFID, collar) or single image comparison:

[0027] 1. Introduce a motion change perception mechanism in the image screening stage, limit the image acquisition period by judging the pet's behaviors such as approaching, staying, turning head, etc., thereby reducing the number of invalid images, improving the quality of image input, and distinguishing from the misidentification risk caused by continuous acquisition without screening in the prior art.

[0028] 2. Adopt an image time interval judgment method, sort the collected images by time and set an effective interval threshold, only keep the images that appear continuously and have reasonable intervals as recognition input, avoiding the misjudgment problem caused by single frame image abnormality. This judgment method is not embodied in the existing label recognition or static image recognition scheme.

[0029] 3. Propose an identity rationality approval step, analyze the sequence time structure of the recognized successful images to further verify whether the recognition process is continuous and consistent, effectively eliminate accidental image mismatch, and solve the defect in the prior art that only relies on one-time comparison and ignores sequence logic. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0031] In order to deepen the understanding of the present application, the present application will be further described in combination with the embodiments below. The present embodiments are only used to explain the present application and do not constitute a limitation on the protection scope of the present application.

[0032] Embodiment one

[0033] According to the figure, the present embodiment provides a pet face image-based identity recognition access control method, including the following steps: Figure 1 S1, effective screening: acquire multiple groups of pet face images in the target area, and according to the action change performed by the pet, screen the face images and extract relevant effective face images; the screening includes judging the face orientation in the image, image clarity, face occlusion situation and other indicators, and combining image timestamp information to preliminarily filter interference frames;

[0034] Among them, the action change of the pet is perceived to define a to-be-judged period, all face images in the period are extracted as to-be-judged images, and are sorted in time sequence; the time interval between adjacent images is defined as an interval reference value

[0035] Compare the value with multiple interval threshold values preset in the system If is less than any interval threshold value , the image is identified as an effective image; if If the distance is greater than all interval thresholds, the image is not considered as valid data for comparison, but is marked as a to-be-corrected image, which is used as a reference for subsequent analysis and parameter adjustment.

[0036] S2, identity verification output: encode the feature vector of each extracted valid face image, and match it with the registered target pet face features in the database to calculate the similarity score; specifically, extract the image vector and the registered image vector Then, calculate the cosine similarity as the credibility S:

[0037]

[0038] Wherein, represents the dot product, represents the vector length;

[0039] Compare the credibility with the system preset credibility threshold : if , it is considered that the identity comparison is passed, triggering the door opening instruction to release, and the comparison image, comparison time and score information are recorded in the system log;

[0040] If , it is marked as identity inconsistency and triggers an alarm; the alarm content includes the current image data, comparison score and alarm trigger time; at the same time, the state of the access control hardware is reviewed and judged:

[0041] The system obtains the door lock execution state M (open = 1, close = 0) and the access detection result E (with passing behavior = 1, without passing behavior = 0), if M ≠ E occurs, that is, the recognition and access control response are inconsistent, the current comparison image is marked as abnormal, and a review prompt information is generated and uploaded to the background server for subsequent manual intervention or parameter correction.

[0042] S3, identity rationality review: continuously track and collect the target pet image confirmed by comparison to form an image time sequence as the identity tracking basis; the image time sequence is represented as , each image frame is attached with an accurate time stamp, the system counts the time interval of adjacent image frames, and calculates the average value and variance of all frame intervals in the sequence:

[0043]

[0044]

[0045] The system will Compare with the permitted interval, if it is within the preset time interval and the variance Below the stability threshold set by the system , then the image sequence is judged to have reasonable continuity, and the temporal structure of this round of identity comparison can be confirmed to be stable;

[0046] Otherwise, the current image and subsequent continuous images are marked as abnormal objects and passed to step S4 as content to be corrected for further processing; this method ensures that identity recognition is not only based on single-frame image judgment, but also can perform secondary verification of the recognition results based on time continuity.

[0047] S4. Sensitivity Calibration: Obtain pet facial images at different time intervals, construct a standardized image sequence for testing, simulate the recognition process, and perform statistical analysis on the recognition results of the images in the sequence. By comparing the recognition results with the database labels, the following two indicators are calculated:

[0048] Missed judgment rate ( ): refers to the proportion of registered pets that are misclassified as non-targets;

[0049] False positive rate ( ): refers to the proportion of unregistered pets that are mistakenly identified as target pets;

[0050] If any of the above indicators is higher than the system's allowable threshold, the system will calculate the correction coefficient :

[0051]

[0052] in, 、 is a weight factor used to balance the impact of the two types of errors; the system is based on the correction coefficient Automatically adjust key parameters in the image screening process, including interval threshold sets and confidence thresholds , so that the sensitivity configuration of the next round of image acquisition and recognition is more adapted to the actual recognition performance, and ultimately achieves a dynamic balance between the sensitivity and stability of the recognition system; the updated parameters take effect immediately and are used for all subsequent image acquisition and judgment processes.

[0053] Example 2: Identity recognition access control method based on pet facial images

[0054] This embodiment is implemented on an access control terminal, which includes a visible light camera, an infrared fill light, an ultrasonic distance sensor, an embedded processor, a relay-type door lock driver module, and a local identity database.

[0055] The terminal is fixed at a distance of 40 cm from the ground, and the power supply is 12 volts DC. The system software runs according to the following flow, and each step is executed in a loop unless otherwise specified.

[0056] S1 Validity screening:

[0057] S11 When the distance sensor detects that the target body is between 15-70 cm and stays for more than two seconds, the camera enters the continuous shooting mode, collecting facial images at a frequency of six frames per second, and adding a millisecond-level timestamp to each frame.

[0058] S12 The system combines the pet's orientation, clarity, and occlusion detection results to perform screening on each frame: if the proportion of front-facing orientation is high, there is no strong light overflow in the key facial area, and the outline is complete, it is determined as a clear image; otherwise, the image is discarded.

[0059] S13 The system records the time interval between the two adjacent frames, and compares it with three preset thresholds: 0.5 seconds, 1 second, and 2 seconds. If the time interval is not greater than any threshold, the current frame is identified as a continuous image and retained; otherwise, the frame is marked as a to-be-corrected image but does not participate in identification temporarily.

[0060] S2 Identity verification output:

[0061] S21 For continuous images, call the local face feature extraction model to generate vector data, and match it with the registered vector in the identity database to calculate the similarity score.

[0062] S22 The system sets the confidence threshold to 85%. When the score is higher than the threshold, the processor outputs an open door command through the relay, and records "pass" in the log, the timestamp, and the image path; when the score is lower than the threshold, the system writes to the alarm buffer, triggers the red light to flash, and uploads the image and time to the background.

[0063] S3 Identity rationality review:

[0064] S31 After identification, the system continues to collect images within a 3-second observation period, maintaining the same frame rate as S1.

[0065] S32 The system calculates the average time interval and the maximum interval difference of all images in this period: the average interval should be between 0.5 seconds and 2 seconds, and the maximum difference should not exceed 0.2 seconds. If both conditions are met, this identification sequence is marked as "continuous and stable"; if either condition is not met, this identification sequence is marked as "interval abnormality", and all images enter the correction list.

[0066] S33 If an animal is detected crossing the threshold when the door is closed and an alarm occurs, the system records the current image as "identification - behavior inconsistency", and displays a prompt message in the background.

[0067] S4 Sensitivity correction:

[0068] S41, the system starts an offline correction task at a predetermined time every day, reads all abnormal images and associated time intervals in the previous day's correction list.

[0069] S42, the system replays these images to perform an offline identification with the same process as S2, and counts the number of missed and false positives: missed indicates that a registered pet is rejected, and false positive indicates that an unregistered pet is released.

[0070] S43, if the number of missed is higher than ten or the number of false positives is higher than five, the system adjusts the parameters in the following order:

[0071] a. Reduce the upper limit of the interval threshold by 0.2 seconds;

[0072] b. Reduce the credibility threshold by five percentage points, but not lower than 80%;

[0073] c. If the false positive is still over-standard, increase the frame rate from six frames per second to eight frames per second.

[0074] S44, after the parameter adjustment is completed, the system automatically restarts the camera service and archives the "parameter update" label in the log; if the statistical value is already in the safe range, keep the original parameters and clear the correction list.

[0075] Example three

[0076] The present embodiment provides a pet face image-based identity recognition access control system for implementing the access control method, comprising the following modules:

[0077] An image acquisition module for acquiring multiple sets of pet face images in a target area in real time and attaching a timestamp. This module is based on the trigger signal of an external distance sensor and starts continuous image acquisition after detecting that a pet has entered a pre-set identification area. Each frame of image is automatically attached with time information to ensure that the image has complete time sequence information when it enters the subsequent processing link. The image acquisition module supports stable operation under different lighting conditions and completes day and night image acquisition tasks in cooperation with the light supplement device.

[0078] An image screening module for analyzing the dynamic features of the pet in the image and extracting high-quality valid images. The module receives the original image stream provided by the image acquisition module, and performs validity screening on the image by detecting whether the face orientation in the image deviates from the normal view angle, whether the image has motion blur or occlusion area, and combining the time interval information between adjacent image frames. The time interval of adjacent images is defined as an interval reference value, which is compared with a plurality of interval threshold values preset in the system: if the interval reference value falls within any threshold value range, the image is identified as a valid image; if it exceeds all threshold values, it is temporarily excluded from the identification process and only retained for sensitivity correction processing.

[0079] An identity verification module for comparing the screening image with the target pet identity library and outputting a matching score and a door opening instruction. The module generates a face vector for each valid image frame through a feature extraction algorithm, and performs similarity matching with the target pet image vector registered in the identity database. The similarity calculation uses the cosine similarity formula between vectors to output a credibility score. The score is compared with the threshold value set by the system: if the score is not lower than the threshold value, a "successful identity matching" signal is output and the access control is triggered; if the score is lower than the threshold value, a "matching failed" signal is output and an alarm is triggered, and the current image and related information are recorded for review. The module also interacts with the access control hardware, and performs bidirectional verification combined with the door lock state and the access behavior. If the identification result is inconsistent with the physical state of the access control, the system records the anomaly and issues a background prompt.

[0080] An identity approval module for generating an image time sequence, analyzing the image interval, and evaluating the logical continuity of the identification process. After successful identification, the module continues to track the pet face image for several frames, constructs an image sequence with a time stamp, extracts the time interval between image frames, and calculates the average interval and fluctuation amplitude of the time sequence. By judging whether the average time interval is within the permitted range and whether the time interval variance is below the stability threshold, it is determined whether the identification sequence is continuous and reliable. If the result is a non-continuous sequence or the interval fluctuation is too large, the sequence is marked as an abnormal sequence and submitted to the sensitivity correction module for processing.

[0081] A sensitivity correction module for constructing a test process, generating sensitivity correction parameters, and automatically updating the setting values of the screening algorithm. The module receives all abnormal image sequences and performs offline identification simulation on them, respectively calculates the missed judgment rate and the misjudgment rate in the identification, and automatically generates correction coefficients based on the current identification performance. The system automatically adjusts the time interval threshold value of the screening stage and the credibility threshold value of the identity comparison stage according to the coefficients. The updated parameters are directly applied to the next round of image acquisition and screening process, realizing the dynamic optimization of the system identification accuracy and sensitivity.

[0082] The system described in the embodiment can realize the identification, tracking and access control of pet individuals through the cooperative operation of image acquisition, image screening, identity verification, identity approval and sensitivity correction modules, and has closed-loop control capability and parameter self-adaptive capability, and is suitable for the intelligent access control demand in the family, community and pet management scene.

[0083] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for access control based on pet facial image identification, characterized in that: The following steps are involved: S1. Validity screening: Acquire multiple sets of pet facial images within the target area and screen the facial images based on the pet's movements to extract relevant valid facial images. The screening includes determining facial orientation, image clarity, and facial occlusion indicators in the images, and preliminarily filtering out interfering frames based on image timestamp information. By sensing the pet's movement changes, a period of time to be judged is defined, and all facial images within this period are extracted as images to be judged and sorted in chronological order; The time interval between adjacent images is defined as an interval reference value, which is compared with multiple sets of interval thresholds preset in the system: if the interval reference value is less than any interval threshold, the image is considered a valid image; if it is greater than all interval thresholds, the image is temporarily not included in the comparison as valid data; The interval threshold corresponds to a plurality of preset image acquisition intervals, and a matching threshold is set for each interval as a criterion for determining whether the image enters the comparison process; S2. Identity Verification Output: Each extracted valid facial image is encoded as a feature vector and matched with the facial features of the target pet registered in the database to calculate a credibility score. If the score is higher than or equal to the preset credibility threshold, the door is opened and the pet is allowed to pass. If the score is lower than the threshold, an alarm is triggered and the current image and time information are recorded. S3. Identity Verification: Continuously track and capture images of the target pet that have been confirmed through comparison to form an image time series, which serves as the basis for identity tracking. Statistical analysis is performed on the intervals between image frames in the image time series. If the intervals between image frames are continuous and fall within the preset reasonable time interval, the temporal continuity and stability of the identity comparison are confirmed, improving the reliability of the recognition result. Otherwise, the image and subsequent frames are marked as abnormal and submitted for correction. S4. Sensitivity correction: Acquire pet facial images with different time intervals, construct a standardized test sequence, and calculate the average interval and fluctuation amplitude of consecutive images in the sequence; determine whether the current sensitivity setting will lead to misjudgment or missed judgment through a simulated recognition process; if the test recognition rate is lower than the threshold, automatically generate a correction coefficient and use it to adjust the interval threshold for the next round of image acquisition and screening, thereby dynamically optimizing the system response sensitivity.

2. The method for access control based on pet facial images according to claim 1, characterized in that: In step S2, feature points of the valid image are matched with the standard image in the target pet identity database, and a credibility score is output; Compare the score with the system preset credibility threshold: If the credibility score is greater than or equal to the credibility threshold , the identity comparison is considered to be successful; otherwise it is marked as identity mismatch and an alarm is triggered; the alarm content includes the current image data, comparison score and alarm trigger time.

3. The method for access control based on pet facial images according to claim 2, characterized in that: The above identity comparison results are reviewed in combination with the access control hardware status: if the access control is open but the target pet is not detected passing through, or an unauthorized pet is detected entering in the alarm state, the system automatically marks the image as abnormal and generates a review prompt to upload to the background server.

4. The method for access control based on pet facial images according to claim 3, characterized in that: In step S3, the time difference between consecutive image frames is analyzed to calculate the average interval and fluctuation variance of the sequence; if the average interval is within the permitted range and the fluctuation variance is lower than the stability threshold, the sequence is judged to be reasonable; otherwise, it is marked as an unreasonable sequence, and the image and pet information are submitted to step S4 for processing.

5. The method for access control based on pet facial images according to claim 4, characterized in that: The time interval judgment of the continuous images is a bidirectional logic, that is, if the average interval falls within the permitted range, the image is considered to be a reasonable image; otherwise, it is regarded as an object to be corrected and entered into the sequence to be compared as potential abnormal data.

6. An identity recognition access control system based on pet facial images, used to implement the method according to any one of claims 1 to 5, characterized in that: include: An image acquisition module, used to acquire multiple sets of pet facial images in the target area in real time with timestamps; Image screening module, used to analyze the dynamic features of pets in images and extract high-quality valid images; The identity verification module is used to compare the screening image with the target pet identity database and output a matching score and door opening instruction; Identity verification module, which is used to generate image time series, analyze image intervals, and evaluate the logical continuity of the recognition process; The sensitivity correction module is used to build test procedures, generate sensitivity correction parameters and automatically update the set values ​​of the screening algorithm.

7. An identity recognition access control terminal based on pet facial images, characterized by: A bus comprising a processor, a memory and connecting elements; a computer program is stored in the memory, and the computer program is executed by the processor so that the processor implements all the step logics in the method according to any one of claims 1 to 5.

Images