A method and device for synchronous collection of pre-blood-drawing data based on face recognition

Abstract

The application discloses a blood sampling pre-data synchronous acquisition method and device based on face recognition, and relates to the technical field of image processing.The method comprises the following steps: marking the face appearance event start time based on the face appearance event marker set; sampling the left and right blood pressure points and the height points; limiting the sampling time within the face appearance event segment coverage period; encapsulating the measurement value, the device sampling time marker and the receiving time marker; generating a three-clock original comparison record set; performing comparison verification and difference registration on the three-clock original comparison record set; solidifying the point position time offset record and the point position transmission time delay record; aligning the left and right blood pressure and the height entries after unifying the time caliber; and generating a blood sampling pre-data result set.The application realizes the dynamic correlation of face features in multiple points, ensures the time continuity of physiological parameters and identity information, solves the matching failure problem caused by shielding interference, and improves the data integrity.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for synchronous acquisition of pre-blood collection data based on face recognition. Background Technology

[0002] With the rapid development of computer vision and image processing technologies, the application of facial recognition in the healthcare field has evolved from basic identity verification to fully automated data acquisition systems. Existing technologies, through multi-point video acquisition and feature extraction strategies, have achieved preliminary automation of pre-blood collection identity verification and physiological parameter acquisition, with data processing gradually transitioning from independent single-device operation to multi-device collaborative acquisition. In pre-blood collection health monitoring scenarios, the system typically relies on a registration entrance camera for identity registration, combined with blood pressure monitors and height measurement devices to collect physiological indicators. Its data processing strategy focuses on local feature extraction and simple correlation.

[0003] Existing technologies have shortcomings in multi-point data synchronization mechanisms: First, there is a lack of dynamic facial feature matching strategies across points. The registration entrance and the blood pressure and height measurement point cameras operate independently, lacking a real-time association algorithm based on facial features. When patients move or lighting changes, differences in perspective and occlusion interference can cause facial features to fail to match effectively, resulting in a disconnect between physiological parameters and identity information time anchors, making it difficult to accurately stitch together key data. Second, the time offset calibration mechanism between devices is imperfect. Each device uses an independent clock source and does not implement dynamic calibration based on facial features, resulting in a systematic offset between registration and measurement times. This necessitates manual alignment, reducing efficiency and making it difficult to meet the real-time and consistency requirements of high-precision medical care. Summary of the Invention

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

[0005] Therefore, this invention provides a method for synchronous acquisition of pre-blood collection data based on face recognition to solve the problems of multi-point face feature matching failure and device time offset.

[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 method for synchronous data acquisition before blood collection based on facial recognition, comprising: performing facial region locking and effective facial frame filtering on the real-time video stream at the registration entrance, extracting facial features, and writing them into the registration camera identifier, registration time stamp, and session effective time window boundary to generate a collection session ticket set; based on the collection session ticket set, acquiring real-time video streams of left and right blood pressure points and height points, extracting facial features at the points, and performing cross-point facial re-identification matching with the facial features in the collection session ticket set, while simultaneously fixing continuous hit segments within the session effective time window to generate a re-identification session locking link set; and processing the re-identification session locking link set... The system performs backtracking and mapping of the hit frame number to the corresponding video time stamp. Stable and continuous frame segments are selected to solidify the face appearance event segments, generating a face appearance event marker set. Based on the start time stamp of the face appearance event in the face appearance event marker set, left and right blood pressure and height points are sampled, and the sampling is limited to the time period covered by the face appearance event segments. The measured values, device sampling time stamps, and receiving time stamps are encapsulated to generate a three-clock original comparison record set. The three-clock original comparison record set is checked and the difference is registered. The point time offset record and point transmission delay record are solidified. After unifying the time caliber, the left and right blood pressure and height entries are aligned to generate a pre-blood collection data result set.

[0007] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps for generating the collection session ticket set are as follows: The system collects real-time video streams from the registration entrance and stores the registration camera identifier, registration time stamp, and continuous replayable frame sequence number to generate a video frame index set for the registration entrance. Based on the video frame index set of the registration entry, candidate face regions are detected frame by frame, and cross-frame association is performed to lock the target face region and generate a face region locking record set. Perform effective face frame filtering on the face region locking record set, write quality labels and keyframe pointers, and aggregate the effective face frame range records to generate an effective face frame filtering result set; The effective face frame screening result set is subjected to face key point alignment and normalization cropping, and face features are extracted. At the same time, the registration camera identifier, registration time marker and session effective time window boundary are bound to generate a collection session ticket set.

[0008] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps of collecting real-time video streams of left and right blood pressure and height points according to the collection session ticket set and extracting facial features at the points are as follows: Based on the collection session ticket set, collect real-time video streams of left and right blood pressure points and height points, and simultaneously extract frame segments within the effective time window of the session to generate a set of effective frame indexes for the point sessions. Based on the valid frame index set of point-based sessions, the target face region is locked at the left and right blood pressure points and height points, and occluded interference frames are removed to generate a set of valid face frame filtering results. The effective face frames of the selected locations are aligned and cropped, and the face features of the locations are extracted to generate a set of face feature records.

[0009] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps for generating the re-identification session locking link set are as follows: For the facial features in the location facial feature record set and the facial features in the collection session ticket set, perform cross-location facial re-identification matching, and summarize them into a hit record sequence according to the frame number; The hit records of adjacent frame numbers in the hit record sequence are spliced ​​into a continuous hit segment, and splicing is performed on the short-term occlusions and the occlusion splicing mark is written to generate a re-identification session locking link set.

[0010] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps of performing backtracking positioning on the hit frame sequence number in the re-identification session locking link set and mapping it to the corresponding video time stamp are as follows: Extract the boundary frame sequence number of the continuously hit segments in the re-identification session locked link set, locate the corresponding original frame sequence number range and frame time marker in the point session valid frame index set, and generate the hit frame sequence number backtracking location set. The frame time stamps of the hit frame sequence number back-location set are sorted and abnormal jumps are removed. The time boundaries of consecutive hit segments are fixed and the point identifiers are bound to generate a video time stamp mapping set.

[0011] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps for generating a face appearance event marker set are as follows: Within the original frame sequence number range corresponding to the video time stamp mapping set, read the face region locking record set and face key points, perform displacement jitter removal and deflection mutation removal, and generate a set of face appearance event segments; The set of face appearance event fragments is combined with face features and location markers, and the original frame sequence range, keyframe pointers and time boundaries are encapsulated to generate a set of face appearance event markers.

[0012] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the sampling of left and right blood pressure points and height points is performed based on the start time marker of the face appearance event in the face appearance event marker set, and the steps are as follows. The start and end time markers of face occurrence events in the face occurrence event marker set are organized according to the location identifier and written into the original frame sequence range and keyframe pointer to generate a sampling arrangement list; Based on the sampling arrangement list, sampling is performed at the sampling entrances of left and right blood pressure points and height points, and the measured values ​​and equipment sampling time stamps are transmitted back to generate a point sampling transmission record set.

[0013] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps for generating the three-clock original comparison record set are as follows: Write the receiving time stamp into the sampling and feedback record set of the point, and combine it with the start time stamp and end time stamp of the face appearance event to establish the correspondence of the coverage period and generate a three-clock entry record set; The three-clock entry record set is merged according to the left and right blood pressure points and height points, and the point identifier, original frame number range and key frame pointer are written to generate the three-clock original comparison record set.

[0014] As a preferred embodiment of the pre-blood collection data synchronization method based on face recognition described in this invention, the steps for generating the pre-blood collection data result set are as follows: The correspondence between the location markers and the start time markers of the face appearance event in the original three-clock comparison record set is verified, and the verification status marker is written to generate a three-clock comparison verification item set; Based on the three-clock comparison verification item set, point time offset comparison verification and difference registration are performed on the start time marker of the face appearance event and the device sampling time marker to generate a point time offset record set; Based on the point time offset record set, perform point transmission delay comparison verification and difference registration on the device sampling time mark and receiving time mark, and generate a three-clock difference registration result set; Based on the three-clock difference registration result set, the device sampling time stamp is converted to a unified time caliber, and the left and right blood pressure and height entries are aligned to generate a pre-blood collection data result set.

[0015] Secondly, the present invention provides a pre-blood collection data synchronization acquisition device based on face recognition, comprising, The cabinet serves as the overall load-bearing structure for the pre-blood collection data synchronization acquisition device, providing a stable installation environment and protection for each functional module, and ensuring the overall coordinated operation of the equipment. The left-side blood pressure acquisition module is used to acquire left-side blood pressure measurements and record the device sampling time markers. It also acquires real-time video streams, extracts facial features at the locations, and performs cross-location re-identification matching with facial features at the registration entrance. The right-side blood pressure acquisition module is used to collect right-side blood pressure measurements, device sampling time markers, and facial features at the sampling points, enabling simultaneous acquisition of blood pressure data from both sides and identity binding. The height acquisition module is used to collect the patient's height measurement value and record the device sampling time marker. At the same time, it collects real-time video stream and extracts facial features to participate in cross-point face re-identification and matching, ensuring the temporal continuity and correlation between height data and identity information and blood pressure data. Locking casters are used for moving and locking the cabinet.

[0016] The beneficial effects of this invention are as follows: by cross-point face re-identification matching and continuous hit segment splicing, the dynamic association of facial features at multiple points is realized, ensuring the temporal continuity of physiological parameters and identity information, solving the matching failure problem caused by occlusion interference, and improving data integrity; by three-clock verification and difference registration, the time reference between devices is automatically calibrated without manual intervention, ensuring real-time consistency and clinical accuracy of data before blood collection. 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 method for synchronously collecting pre-blood collection data based on facial recognition.

[0019] Figure 2 This is a schematic diagram of a face recognition-based pre-blood collection data synchronization device.

[0020] Figure 3 This is a curve comparing the occlusion ratio and the coverage ratio.

[0021] Figure 4 A bar chart comparing the number of valid entries.

[0022] Figure 2 In the middle: 1. Cabinet; 2. Left blood pressure acquisition module; 3. Right blood pressure acquisition module; 4. Height acquisition module; 5. Locking casters. Detailed Implementation

[0023] 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.

[0024] 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.

[0025] 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.

[0026] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for synchronously collecting blood collection data before collection based on face recognition, including the following steps: S1: The real-time video stream at the registration entrance is used to perform face region locking and valid face frame filtering, extract face features, and write them into the registration camera identifier, registration time marker and session valid time window boundary to generate a collection session ticket set; S1.1: Collect the real-time video stream at the registration entrance, and fix the registration camera identifier, registration time stamp and continuous replayable frame sequence number to generate a video frame index set at the registration entrance; Furthermore, the real-time video stream of the registration entrance is acquired and kept continuously readable. The video frames of the registration entrance are decoded sequentially in time order to obtain continuous image frames. A registration camera identifier is written to each frame, and a registration time stamp is fixed at the frame acquisition time. At the same time, a continuous playable frame sequence number is assigned to each frame, and the sequence number is filled in and rearranged for lost or out-of-order frames to ensure that the continuous playable frame sequence number is monotonically increasing. The continuous playable frame sequence number is bound to the position pointer of the corresponding frame to support playback by sequence number. Then, the registration camera identifier, registration time stamp, and continuous playable frame sequence number are aggregated at the frame granularity and encapsulated and organized according to a fixed field structure to generate a registration entrance video frame index set.

[0027] S1.2: Based on the registration entry video frame index set, detect the face candidate region frame by frame, perform cross-frame association, lock the target face region, and generate a face region locking record set; Furthermore, the corresponding image frames are located sequentially according to the continuous replayable frame sequence number of the registered entry video frame index set, and face candidate region detection is performed frame by frame. The face candidate regions of each frame are solidified into candidate region records bound with frame sequence numbers and written with candidate region position indication information. Cross-frame association is performed on the candidate region records of adjacent frames. Cross-frame association is constrained by the spatial proximity relationship and scale consistency relationship of the candidate region position indication information. The candidate region records that meet the constraints are concatenated into candidate region trajectories. Multiple face candidate regions appearing in the same frame are prioritized according to the continuity of the candidate region trajectories. The candidate region trajectory with the highest priority is selected as the target face region locking trajectory. The target face region position indication information is solidified for each frame along the target face region locking trajectory and written into the continuous replayable frame sequence number range and the registration time mark range. Trajectory filling is performed on short-term missing frames in the target face region locking trajectory and filling mark is written to generate a face region locking record set.

[0028] S1.3: Perform effective face frame filtering on the face region locking record set, write quality labels and keyframe pointers, and simultaneously aggregate the effective face frame range records to generate an effective face frame filtering result set; Furthermore, the target face region is located along the target face region locking trajectory in the face region locking record set, and the range of consecutive replayable frame numbers corresponding to the target face region is located. The target face region image blocks are captured frame by frame, and the image blocks are subjected to clarity judgment, occlusion judgment, and pose deviation judgment to distinguish between usable and unusable frames. Frames that meet the judgment conditions are registered as valid face frames, and frames that do not meet the judgment conditions are registered as invalid face frames. For valid face frames, quality labels are written according to quality to indicate the degree of preference. At the same time, representative frames are selected from the valid face frames and written into keyframe pointers to support subsequent fast location and playback. Connectivity segment merging is performed on the valid face frames according to the consecutive replayable frame numbers. The consecutive replayable frame numbers of adjacent valid face frames are concatenated to form a valid face frame range record, and the start and end boundaries of discontinuous segments are registered respectively. Finally, the quality labels, keyframe pointers, and valid face frame range records are aggregated and encapsulated according to a fixed field structure to generate a set of valid face frame filtering results.

[0029] It should be noted that the sharpness determination is as follows: the edge gradient response and texture detail response of the target face region image block are checked for consistency. When the edge contour is continuous and the proportion of high-frequency details remains stable, it is registered as passing the sharpness test. When the edge response collapses or the details disappear in patches, it is registered as failing the sharpness test.

[0030] Occlusion determination: Perform skin color distribution connectivity check and facial key point visibility check on the target face region image block. If the skin color connectivity is broken or the facial key points are missing, reaching the occlusion indication state, it is registered as occlusion failure. If the skin color connectivity is complete and the facial key point visibility is complete, it is registered as occlusion success.

[0031] Pose deviation determination: Perform symmetry and projection deformation checks on the geometric relationship of facial key points in the target face region image block. If the symmetry of the key points is broken and abnormal lateral or pitch deformation occurs, it is registered as a failed pose deviation. If the geometric relationship of the key points is stable and the projection deformation is within a stable morphological cluster, it is registered as a passed pose deviation.

[0032] S1.4: Perform facial key point alignment and normalization cropping on the effective face frame screening result set, extract facial features, and bind the registration camera identifier, registration time marker and session effective time window boundary to generate a collection session ticket set.

[0033] Furthermore, keyframes are located based on the keyframe pointers in the effective face frame screening result set. A sequence of effective face frames covering the keyframes is extracted based on the effective face frame range record. Facial key points are located frame by frame in the effective face frame sequence, and abnormal frames are removed to ensure stable and usable facial key points. The screened effective face frames undergo rotation and scale correction based on facial key points and are then aligned. Normalized cropping is performed on the aligned effective face frames to obtain uniformly sized face cropped image blocks. Grayscale normalization and brightness consistency are then performed on the face cropped image blocks to reduce the impact of lighting differences. Facial features are then extracted from the face cropped image blocks, and these features are aggregated by frame number to form a session-wide face feature record. Based on the start and end registration time markers of each connected segment recorded in the effective face frame range record, a session-wide effective time window segment set is generated. The registered camera identifier, registration time markers, session-wide effective time window boundaries, and session-wide effective time window segment set are bound and encapsulated according to a fixed field structure to generate a collection session ticket set.

[0034] It should be noted that facial key points are a set of feature coordinate points in a cropped face image block used to stably locate fixed structural positions such as the corners of the eyes, the tip of the nose, and the corners of the mouth, and are used to support rotation correction, scale correction, and facial key point alignment.

[0035] Figure 2As shown, cabinet 1 serves as the overall supporting structure for the pre-blood collection data synchronization acquisition device, providing a stable installation environment and protection for each functional module, while also organizing the wiring layout to ensure the overall coordinated operation of the equipment. The left-side blood pressure acquisition module 2 is equipped with a blood pressure acquisition component. On one hand, it acquires blood pressure measurements on the left side and records the device's sampling time marker; on the other hand, it acquires real-time video streams through an identity recognition node, extracting facial features at the location for cross-location re-identification matching with facial features at the registration entrance, ensuring accurate association between blood pressure data and patient identity. The right-side blood pressure acquisition module 3 is structurally and functionally symmetrical to the left-side blood pressure acquisition module 3, also equipped with a blood pressure acquisition component and a... The system includes a recognition node that collects right-side blood pressure measurements, device sampling time markers, and facial features at the location, enabling simultaneous collection and identity binding of blood pressure data from both sides, thus improving the comprehensiveness and accuracy of blood pressure measurement data. A height acquisition module 4, located between the left and right blood pressure collection points, is equipped with a height acquisition component and an identity recognition node. The height acquisition component obtains the patient's height measurement and records the device sampling time marker. The identity recognition node collects real-time video streams and extracts facial features, participating in cross-location facial re-identification matching to ensure the temporal continuity and correlation between height data, identity information, and blood pressure data. A locking caster wheel 5 is installed at the bottom of the cabinet, providing both movement and locking functions.

[0036] S2: Based on the collection session ticket set, collect real-time video streams of left and right blood pressure points and height points, extract facial features at the points, and perform cross-point facial re-identification matching with facial features in the collection session ticket set. At the same time, solidify the continuous hit segments within the effective time window of the session to generate a re-identification session lock link set. S2.1: Based on the collection session ticket set, collect real-time video streams of the left and right blood pressure points and height points, and simultaneously extract frame segments within the effective time window of the session to generate a set of effective frame indexes for the point sessions. Furthermore, the session ticket set is written into the point acquisition entry of the left and right blood pressure points and height points, and the effective time window boundary of the session is fixed. Real-time video stream acquisition is started for the left and right blood pressure points and height points respectively, and the real-time video stream of the point points is decoded frame by frame to generate point point video frames. A point point identifier is written into each frame, and a frame time marker is fixed with the frame acquisition time. At the same time, a continuous replayable frame sequence number is assigned and bound to the point point video frame position pointer to support playback by sequence number. Point point video frames whose frame time markers fall within the boundary of the effective time window of the session are registered as valid frames of the session, and point point video frames whose frame time markers are earlier or later than the boundary of the effective time window of the session are registered as invalid frames of the session. Frame segment aggregation is performed on the valid frames of the session according to the continuous replayable frame sequence number, and the frame sequence number range and frame time marker range of the valid frames of the session are fixed. The boundaries of the discontinuous frame segments that appear in the valid frames of the session are registered and frame segment sequence number markers are written, generating a point point session valid frame index set.

[0037] S2.2: Based on the valid frame index set of point-based sessions, the target face region is locked at the left and right blood pressure points and height points, and occluded interference frames are removed to generate a set of valid face frame filtering results. Furthermore, based on the point-based session valid frame index set, the effective frame number range of the left and right blood pressure points and height points is located, and the corresponding point-based video frames are extracted frame by frame. Face candidate regions are detected in the point-based video frames, and cross-frame association is performed on the face candidate regions of adjacent frames to form target face region locking trajectories. The target face region locking trajectories are encapsulated and organized according to the point identifier, frame number range, and target face region position indication information to generate a point-based face region locking record set and solidify the target face regions of the left and right blood pressure points and height points. At the same time, occlusion judgment and multi-target interference judgment are performed frame by frame on the frame segments covered by the point-based face region locking record set. Frames with occlusion ratio exceeding the limit and frames with multi-target overlap are registered as occlusion interference frames and removed. The retained frames are registered as valid face frames and are aggregated, encapsulated, and written into the point identifier and frame number range according to the frame number to generate a point-based valid face frame filtering result set.

[0038] It should be noted that the occlusion determination writes the localization results of the facial key points corresponding to the target face region into the key point visibility status, and defines the occlusion ratio by the proportion of the number of key point localization failures to the total number of fixed key points. The occlusion determination threshold value is registered according to the video frame resolution of the point and the pixel scale of the target face region (e.g., 0.35 to 0.55 in the example). The occlusion determination threshold value is adaptively adjusted according to the resolution level, the pixel scale level of the target face region, and the detection stability value, and is fixed with the frame number.

[0039] S2.3: Perform effective face frame alignment and cropping on the effective face frame screening result set of the point location, extract the face features of the point location, and generate a face feature record set of the point location; Furthermore, the effective face frames are traversed according to the result set of effective face frames at each location. Facial key points are located frame by frame and recorded. Abnormal frames are removed. The remaining frames are rotated and scaled according to the facial key points to complete the alignment of facial key points. Normalization cropping is performed on the effective face frames that have completed the alignment of facial key points to obtain point-based face cropped image blocks and brightness uniformity is performed. Point-based face features are extracted from the point-based face cropped image blocks and aggregated according to the frame sequence number to form point-based face feature records within the point session. The point-based face feature records within the point session are bound and encapsulated with point identifiers, frame sequence ranges, and frame time stamp ranges to generate a point-based face feature record set.

[0040] S2.4: Perform cross-location face re-identification matching on the face features in the location face feature record set and the collection session ticket set, and summarize them into a hit record sequence according to the frame number; Furthermore, for each facial feature in the location facial feature record set, the corresponding frame time stamp range is located according to the location identifier and frame number. The location facial features are matched with the facial features in the collected session ticket set one by one, and cross-location facial re-identification matching is performed. The similarity value of each matching pair is calculated and written into the similarity label. Multiple matching results generated by the same frame number are sorted according to the similarity label and the highest-ranked matching result is retained. At the same time, the coverage relationship of the session effective time window boundary of the matching results is verified and out-of-bounds results are removed. The retained matching results are written into the hit record entries in ascending order of frame number and the location identifier, frame number, frame time stamp range and similarity label are fixed. The hit record entries of all frame numbers are gathered to form a hit record sequence.

[0041] The formula for calculating the similarity value is: ; in, Indicates the similarity value. This represents the location-based facial feature vector in the location-based facial feature record set. This represents the facial feature vectors collected from the session ticket set. Represents the vector dot product. Represents the facial feature vector at a specific location. The 2-norm, This represents the facial feature vector in the collected session ticket set. The 2-norm, To represent numerically stable terms, avoid having a denominator of zero (e.g., take Example 10). 12 ).

[0042] S2.5: Concatenate the hit records of adjacent frame numbers in the hit record sequence into a continuous hit segment, and perform concatenation on short-term occlusions and write occlusion concatenation markers to generate a re-identification session lock link set.

[0043] Furthermore, the hit record entries are traversed sequentially according to the ascending frame number of the hit record sequence. Adjacent hit record entries with continuously ascending frame numbers are aggregated into the same continuous hit segment based on the point identifier. The start frame number, end frame number, and corresponding frame time marker range of the continuous hit segment are fixed and registered as segment boundary records. The frame number interval between adjacent continuous hit segments is detected and short-term occlusion interruptions are determined. The frame number interval is registered as the number of interrupted frames and is checked for consistency with the allowable gap condition. The allowable gap condition is determined by the point video frame rate marker corresponding to the point identifier and the occlusion splicing rule. The criteria are jointly provided and registered in the form of frames (e.g., Example 0, Example 3, Example 6, Example 10). When a short-term occlusion interruption meets the allowable gap condition, the consecutive hit segments on both sides of the interruption are spliced ​​together, and an occlusion splicing mark is written at the splicing position. When a short-term occlusion interruption does not meet the allowable gap condition, the boundary record of the consecutive hit segments remains unchanged. For each consecutive hit segment, the similarity label statistics are aggregated and written into the segment quality label. The consecutive hit segments, occlusion splicing mark, and segment quality label, together with the point identifier and the effective time window boundary of the session, are encapsulated and organized to generate a re-identification session locking link set.

[0044] It should be noted that, for the pre-examination process of outpatient blood collection, examinees first swipe their ID card or scan their examination form at the registration entrance to obtain an identity verification certificate. During the same registration action, the real-time video stream at the registration entrance is collected to perform face region locking and effective face frame filtering. After extracting facial features, the identity verification certificate, registration time stamp, and session effective time window boundary are written into the collection session ticket set. Examinees then enter the height measurement point and left and right blood pressure measurement points respectively. The corresponding real-time video stream is collected at each point and the facial features at the point are extracted. Cross-point face re-identification matching is performed with the facial features in the collection session ticket set. At the same time, the identity verification certificate is used as the session merging key to solidify the continuous hit segments within the session effective time window, generating a re-identification session locking link set for unified binding and traceability verification of subsequent multi-point measurement items.

[0045] Even after identity verification via ID card or examination form, height and left / right blood pressure measurements may still result in misbinding or inconsistent timeframes between measurement entries and the same subject's session due to queuing intervals, movement across locations, inconsistencies in local clocks of different devices, and fluctuations in data upload link latency. To address this, within the effective time window of the session locked by the re-identification session link set, the measurement output of each location is synchronously recorded with the measurement value, device sampling time stamp, and receiving time stamp to form a three-clock comparison entry. Based on the three-clock comparison entry, the location time offset and location transmission latency are registered and converted to a unified timeframe. Then, the merging relationship of multi-location measurement entries is constrained by identity verification credentials and consecutive hit segments, thereby enabling subsequent steps to stably bind height and left / right blood pressure entries to the same session and have a verifiable chain of consistent evidence.

[0046] Figure 3 This paper presents a comparison of the coverage percentage as a function of occlusion ratio under different occlusion ratios, used to measure the session continuity and data integrity of cross-location face re-identification matching under occlusion interference. The horizontal axis represents the occlusion ratio, and the vertical axis represents the coverage percentage. Multiple curves correspond to the stitching strategies for the allowable gap conditions registered in the occlusion stitching rule entries (e.g., example frames 0, 3, 6, and 10). The coverage percentage for each curve is obtained by statistically analyzing the hit record sequence after aggregating consecutive hit segments and writing occlusion stitching tags. The red dashed rectangle in the figure marks the difference window in the high occlusion interval, and a magnified view is provided below. In the magnified view, multiple curves are aligned at the same occlusion ratio value points, and the position of the largest difference is marked at the curve inflection point, highlighting the difference in coverage percentage caused by different allowable gap condition strategies under high occlusion conditions.

[0047] S3: Perform backtracking location on the hit frame sequence number in the re-identification session lock link set, map it to the corresponding video time stamp, select stable continuous frame segments to solidify face appearance event segments, and generate a face appearance event stamp set; S3.1: Extract the boundary frame sequence number of the continuously hit segments in the re-identification session locked link set, locate the corresponding original frame sequence number range and frame time marker in the point session valid frame index set, and generate the hit frame sequence number backtracking location set; Furthermore, for each link in the re-identification session locking link set, the point identifier and the start frame number and end frame number of the consecutive hit segment are extracted, forming a pair of consecutive hit segment boundary frame numbers. Based on the point identifier, the corresponding point frame segment is located in the point session valid frame index set, and the frame records with the same sequence number within the frame segment are retrieved according to the consecutive hit segment boundary frame number pair. The frame records corresponding to the start frame number and the end frame number of the consecutive hit segment are solidified into segment boundary positioning records. The original frame number range covered by the consecutive hit segment is obtained by expanding along the segment boundary positioning records, and the frame time stamp sequence within the original frame number range is aggregated. The original frame number range, the frame time stamp sequence, and the point identifier are bound, encapsulated, and written into the consecutive hit segment boundary frame number pair to generate a hit frame number backtracking positioning set.

[0048] S3.2: Sort and remove abnormal jumps from the frame time stamps of the hit frame sequence number backtracking location set, solidify the time boundaries of consecutive hit segments, bind the point identifiers, and generate a video time stamp mapping set; Furthermore, the point identifiers and frame time stamp sequences are extracted one by one from the hit frame sequence number backtracking location set, and sorted according to time sequence. The sorted frame time stamp sequences are compared point by point with adjacent time differences, and abnormal jumps are identified. When an abnormal jump meets the jump characteristics, the corresponding frame time stamp is removed, and a jump removal mark is written to the removal position. The continuity check is performed on the removed frame time stamp sequence, and the continuity interruption position is divided into multiple continuous time periods. For each continuous time period, the earliest frame time stamp is taken and solidified as the continuous hit segment start time stamp, and the latest frame time stamp is taken and solidified as the continuous hit segment end time stamp. The continuous hit segment start time stamp and continuous hit segment end time stamp are bound and encapsulated with point identifiers, continuous hit segment boundary frame sequence number pairs, and original frame sequence number range to generate a video time stamp mapping set.

[0049] It should be noted that the abnormal jump removal involves verifying the adjacent time differences of the sorted frame time stamp sequence one by one and identifying sudden increases or regression time points that are significantly inconsistent with the surrounding time progression patterns. The corresponding frame time stamps are then removed and written into the jump removal flag to ensure that continuous time periods can be stably fixed.

[0050] A jump feature refers to a time anomaly in which the time difference between two adjacent frames in a frame time stamp sequence does not match the increment of the sequence number of consecutive replayable frames and there is a sudden increase or decrease, and the sudden increase or decrease cannot return to a stable increasing rhythm within a small number of adjacent frames (e.g., the time difference between adjacent frames suddenly increases to 5 seconds or time reversal occurs).

[0051] S3.3: Within the range of the original frame sequence number corresponding to the video time stamp mapping set, read the face region locking record set and face key points, perform displacement jitter removal and deflection mutation removal, and generate a set of face appearance event segments; Furthermore, the system locates each point identifier and the original frame number range according to the video time stamp mapping set, and traverses the frame numbers within the original frame number range. Based on the frame number, it extracts the target face region location indication information from the face region locking record set corresponding to the same point and simultaneously extracts the face key point positioning results. It performs displacement continuity verification on the target face region location indication information of adjacent frames and marks displacement jitter frames. It performs attitude continuity verification on the face key point positioning results of adjacent frames and marks deflection mutation frames. It removes displacement jitter frames and deflection mutation frames and retains stable frames. For the retained stable frames, it performs connected segment merging according to the frame number continuity and fixes the start frame number and end frame number of each connected segment. At the same time, it writes the start time mark and end time mark of the continuous hit segment corresponding to the connected segment into the connected segment time boundary, and selects representative stable frames to write into the key frame pointer. It encapsulates and organizes the connected segments along with the point identifiers to generate a face appearance event segment set.

[0052] S3.4: Combine the set of face appearance event fragments according to face features and location identifiers, and encapsulate the original frame sequence range, keyframe pointer and time boundary to generate a set of face appearance event markers.

[0053] Furthermore, for each face occurrence event segment set, the location identifier, start frame number, end frame number, start time marker and end time marker of consecutive hit segments, and keyframe pointer are extracted. Segments are then grouped according to the location identifier. For each location identifier group, representative stable frames are located according to the keyframe pointer, and face features are extracted. Face occurrence event segments with similar face features are grouped into the same event cluster and an event cluster number is written. Adjacent face occurrence event segments within the same event cluster are spliced ​​at frame number intervals and a segment splicing mark is written to obtain the event segment frame number range. The event segment frame number range is backfilled with the original frame number range in the video time stamp mapping set, and the corresponding time boundaries are converged. The original frame number range, keyframe pointer, time boundaries, location identifier, and face features are bound, encapsulated, and organized to generate a face occurrence event marker set.

[0054] S4: Based on the start time marker of the face appearance event in the face appearance event marker set, sample the left and right blood pressure points and height points, and limit them to the time period covered by the face appearance event segment. Encapsulate the measurement value, device sampling time marker and receiving time marker to generate a three-clock original comparison record set. S4.1: Organize the start and end time markers of face occurrence events in the face occurrence event marker set according to the location identifier, and write them into the original frame sequence range and keyframe pointer to generate a sampling arrangement list; Furthermore, for each face appearance event marker set, the point identifier, face appearance event start time marker, face appearance event end time marker, original frame sequence number range, and keyframe pointer are extracted and grouped according to the point identifier. The face appearance event start time marker and face appearance event end time marker under the same point identifier are sorted according to time sequence and time overlap entries are identified. The time overlap entries are prioritized according to the coverage length of the original frame sequence number range and the density of keyframe pointers, and the highest priority entries are retained. At the same time, the sampling task sequence number is written to the retained entries, and the sampling task sequence number is bound to the point identifier to form a point sampling trigger entry. The allowed sampling start time is written as the face appearance event start time marker and the allowed sampling end time is written as the face appearance event end time marker for the point sampling trigger entry. The original frame sequence number range and keyframe pointer are fixed as playback positioning pointers. The point sampling trigger entries are encapsulated and organized according to a fixed field structure to generate a sampling arrangement list.

[0055] S4.2: Based on the sampling arrangement list, sampling is performed at the sampling entrances of left and right blood pressure points and height points, and the measured values ​​and device sampling time stamps are transmitted back to generate a point sampling transmission record set; Furthermore, sampling trigger entries for left and right blood pressure and height points are sequentially issued according to the sampling task sequence number in the sampling arrangement list. Point identifiers and sampling task sequence numbers are fixed at the sampling entry points for left and right blood pressure and height points, and the allowed start and end times for sampling are written in. Reaching the allowed start time triggers sampling, while the allowed end time serves as the sampling cutoff condition. After triggering the sampling start condition, point measurement sampling is performed and measured values ​​are generated. Simultaneously, a device sampling time stamp is fixed on the device side at the point, and the device sampling time stamp is bound to the measured value for transmission. The transmitted entries undergo consistency checks for the sampling task sequence number and point identifier, and inconsistent entries are eliminated. Transmitted entries that pass the checks are written with the point identifier, sampling task sequence number, measured value, and device sampling time stamp, and then aggregated and packaged according to the sampling task sequence number to generate a point sampling transmission record set.

[0056] S4.3: Write the receiving time stamp into the sampling and transmission record set of the point, and combine it with the start time stamp and end time stamp of the face appearance event to establish the correspondence of the coverage period and generate a three-clock entry record set; Furthermore, for each point sampling return record set, a return entry is received and a receiving time stamp is fixed at the receiving entry point. The receiving time stamp is bound to the point identifier, sampling task number, measurement value and device sampling time stamp and written into the return entry to form a return entry with a receiving time stamp. According to the sampling task number, the corresponding face appearance event start time stamp and face appearance event end time stamp are located in the sampling arrangement list and checked for consistency with the point identifier. The face appearance event start time stamp and face appearance event end time stamp are written into the return entry with a receiving time stamp and the coverage period correspondence is registered. The coverage period correspondence requires that the device sampling time stamp falls within the time period defined by the face appearance event start time stamp and face appearance event end time stamp, and that the receiving time stamp and device sampling time stamp have the same sampling task number binding relationship. Return entries that do not meet the coverage period correspondence are marked with an out-of-bounds flag and removed. Return entries that meet the coverage period correspondence are encapsulated and organized according to a fixed field structure to generate a three-clock entry record set.

[0057] S4.4: Merge the three-clock entry record set according to the left and right blood pressure points and height points, and write the point identifier, original frame number range and keyframe pointer to generate the three-clock original comparison record set.

[0058] Furthermore, the three-clock entry record set is grouped according to the location identifier, and the left and right blood pressure location entries and height location entries are distinguished. The left and right blood pressure location entries and height location entries under the same sampling task number are merged and arranged in ascending order of sampling task number and written into the merged sequence number mark; according to the sampling task number, the corresponding original frame sequence number range and keyframe pointer are located in the sampling arrangement list and checked for consistency with the location identifier. The original frame sequence number range and keyframe pointer are written into the merged left and right blood pressure location entries and height location entries to form playback positioning pointer entries; the fields of the playback positioning pointer entries are checked for completeness and entries with missing original frame sequence number range or missing keyframe pointer are removed. The verified playback positioning pointer entries are encapsulated and organized according to a fixed field structure to generate the three-clock original comparison record set.

[0059] S5: Perform comparison verification and difference registration on the original three-clock comparison record set, solidify the point time offset record and point transmission delay record, and align the left and right blood pressure and height entries after unifying the time caliber to generate the pre-blood collection data result set.

[0060] S5.1: Verify the correspondence between the location markers and the start time markers of the face appearance event in the original three-clock comparison record set, and write the verification status marker to generate a three-clock comparison verification item set; Furthermore, for each record in the original three-clock comparison record set, the location identifier, the start time marker of the face appearance event, the device sampling time marker, the reception time marker, the original frame sequence number range, and the keyframe pointer are extracted, and verification groups are established according to the location identifier. The sampling task sequence number entry corresponding to the location identifier is located in the sampling arrangement list according to the location identifier, and the start time marker and end time marker of the face appearance event are extracted. A consistency check of the location identifier is performed, and inconsistent entries are written to the verification failure status flag. For entries with consistent location identifiers, a face appearance event start time marker coverage check is performed. The coverage check requires that the start time marker of the face appearance event in the original three-clock comparison record set falls within the time period defined by the start and end time markers of the face appearance event in the sampling arrangement list. Entries that fail the coverage check are written to the out-of-bounds status flag, and entries that pass the coverage check are written to the verification pass status flag, generating a three-clock comparison verification entry set.

[0061] Figure 4The number of valid entries in different experimental groups was statistically analyzed using a bar chart to verify the completeness and traceability of the pre-collection data set from the perspective of "entry size / available records". The horizontal axis represents the experimental groups, which were determined by the conditions of the simulation experiment: while keeping the non-variable parameters such as the effective session time window, number of points, and number of sampling events at normal average values, the occlusion intensity, short-term interruption gap strategy, and time consistency condition were set as variables, and each set of variable levels was fixed into an experimental group; the statistical summary results of a batch of session samples corresponding to each experimental group were numbered sequentially to form group 1 to group 8. The vertical axis represents the number of valid entries, with scientific numbers used to indicate the order of magnitude, facilitating comparison of the differences in retention scale between different experimental groups under the same coordinate system. Each bar represents the total number of valid record entries that, under the corresponding experimental group conditions, entered the control verification process after session locking and sampling encapsulation, and were ultimately retained as entries suitable for alignment. A taller bar indicates that more usable entries can be stably formed under those conditions, indirectly reflecting stronger session coverage and fewer lost entries in cross-location face re-identification matching and continuous hit segment splicing prompts. The difference in the number of valid entries across different experimental groups in the figure indicates that variations in occlusion intensity, short-term interruption gap strategies, and time consistency conditions affect the final scale of usable entries. The bar comparison visually presents the quantitative results of "improved data integrity" and provides entry scale support for subsequent control verification and difference registration based on the three-clock original control record set, as well as for unifying the time caliber to align left and right blood pressure and height entries.

[0062] S5.2: Based on the three-clock comparison verification item set, perform point time offset comparison verification and difference registration on the start time marker of the face appearance event and the device sampling time marker, and generate a point time offset record set; Furthermore, the three-clock comparison verification item set is filtered and retained according to the verification pass status flag, and grouped according to the location identifier. The location identifier, face appearance event start time flag, and device sampling time flag of each verification pass item are extracted as comparison verification input pairs. The execution time sequence of the comparison verification input pairs is checked and abnormal comparison verification input pairs are identified. Abnormal comparison verification input pairs are written with an abnormal status flag and removed. The retained comparison verification input pairs are subjected to location time offset comparison verification and difference registration. The difference registration result is bound to the location identifier and written into the offset registration item, and the offset direction flag and offset amount label are fixed. The offset registration items are aggregated according to the location identifier, and the consistency check and deduplication are performed on the duplicate offset registration items with the same sampling task sequence number. The deduplicated offset registration items are encapsulated and organized according to a fixed field structure to generate a location time offset record set.

[0063] It should be noted that the point time offset comparison verification is to establish a one-to-one correspondence between the start time marker of the face appearance event under the same point identifier and the sampling time marker of the device, and to verify the consistency of the time sequence and the consistency of the offset direction of the correspondence in order to confirm that the offset registration entry can be fixed and written into the point time offset record set.

[0064] S5.3: Based on the point time offset record set, perform point transmission delay comparison verification and difference registration on the device sampling time mark and receiving time mark, and generate a three-clock difference registration result set; Furthermore, the point identifiers are located one by one according to the point time offset record set and associated with the same point identifier entries in the three-clock comparison verification entry set. The device sampling time mark and the receiving time mark are extracted and formed into a transmission verification input pair. The execution time sequence of the transmission verification input pair is checked and entries with receiving time marks earlier than device sampling time marks are removed. For the retained entries, point transmission delay comparison verification and difference registration are performed, and the difference registration results are written into the transmission delay registration entry. The transmission delay registration entry is fixed with the point identifier and delay direction mark and written into the delay amount label. The transmission delay registration entry and the point time offset record set are combined according to the point identifier and written into the three-clock difference registration entry and the associated sequence number mark is registered. The three-clock difference registration entries with the same sampling task sequence number are checked for duplicates and deduplicated. The result set of the three-clock difference registration is generated by encapsulating and organizing according to the fixed field structure.

[0065] S5.4: Based on the three-clock difference registration result set, the device sampling time stamp is converted to a unified time caliber, and the left and right blood pressure and height entries are aligned to generate a pre-blood collection data result set.

[0066] Furthermore, based on the three-clock difference registration result set, the point identifier, equipment sampling time marker, corresponding point time offset record, and point transmission delay record are extracted one by one. The point time offset record and point transmission delay record are combined to generate point time conversion parameters and bound to the point identifier. A unified time caliber conversion is performed on each equipment sampling time marker. The converted unified time caliber sampling time marker is written into the corresponding entry, and the correlation between the point identifier and the measured value is retained. Entries with abnormal conversion are marked as conversion failures and removed. The left and right blood pressure point entries and height point entries are sorted by time according to the unified time caliber sampling time marker and aligned and paired according to the sampling task sequence number. Entries with incomplete alignment and pairing are marked as alignment failures and removed. The successfully aligned left and right blood pressure and height entries are encapsulated and organized according to a fixed field structure and written into the unified time caliber sampling time marker and point identifier to generate the pre-blood collection data result set.

[0067] This embodiment also provides a pre-blood collection data synchronization acquisition device based on face recognition, including: The cabinet serves as the overall load-bearing structure for the pre-blood collection data synchronization acquisition device, providing a stable installation environment and protection for each functional module, and ensuring the overall coordinated operation of the equipment. The left-side blood pressure acquisition module is used to acquire left-side blood pressure measurements and record the device sampling time markers. It also acquires real-time video streams, extracts facial features at the locations, and performs cross-location re-identification matching with facial features at the registration entrance. The right-side blood pressure acquisition module is used to collect right-side blood pressure measurements, device sampling time markers, and facial features at the sampling points, enabling simultaneous acquisition of blood pressure data from both sides and identity binding. The height acquisition module is used to collect the patient's height measurement value and record the device sampling time marker. At the same time, it collects real-time video stream and extracts facial features to participate in cross-point face re-identification and matching, ensuring the temporal continuity and correlation between height data and identity information and blood pressure data. Locking casters are used for moving and locking the cabinet.

[0068] In summary, this invention achieves dynamic association of facial features across multiple locations through cross-location face re-identification matching and continuous hit segment splicing, ensuring the temporal continuity of physiological parameters and identity information, solving the matching failure problem caused by occlusion interference, and improving data integrity; through three-clock verification and difference registration, it achieves automated calibration of time base between devices without manual intervention, ensuring real-time consistency and clinical accuracy of data before blood collection.

[0069] 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 method for synchronously acquiring pre-blood collection data based on facial recognition, characterized in that: include, The real-time video stream at the registration entrance is used to perform face region locking and valid face frame filtering, extract face features, and write them into the registration camera identifier, registration time marker and session valid time window boundary to generate a collection session ticket set; Based on the collection session ticket set, real-time video streams of left and right blood pressure points and height points are collected, facial features of the points are extracted, and cross-point facial re-identification matching is performed with facial features in the collection session ticket set. At the same time, continuous hit segments within the effective time window of the session are fixed to generate a re-identification session lock link set. Perform backtracking location on the hit frame sequence number in the re-identification session locked link set, map it to the corresponding video time stamp, select stable continuous frame segments to solidify face appearance event segments, and generate a face appearance event stamp set; Based on the face appearance event start time marker of the face appearance event marker set, the left and right blood pressure points and height points are sampled and restricted to the time period covered by the face appearance event segment. The measurement value, face appearance event marker, device sampling time marker and receiving time marker are encapsulated to generate a three-clock original comparison record set. Perform comparison verification and difference registration on the original three-clock reference record set, solidify the point time offset record and point transmission delay record, and align the left and right blood pressure and height entries after unifying the time caliber to generate a pre-blood collection data result set.

2. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 1, characterized in that: The steps for generating the collection session ticket set are as follows: The system collects real-time video streams from the registration entrance and stores the registration camera identifier, registration time stamp, and continuous replayable frame sequence number to generate a video frame index set for the registration entrance. Based on the video frame index set of the registration entry, candidate face regions are detected frame by frame, and cross-frame association is performed to lock the target face region and generate a face region locking record set. Perform effective face frame filtering on the face region locking record set, write quality labels and keyframe pointers, and aggregate the effective face frame range records to generate an effective face frame filtering result set; The effective face frame screening result set is subjected to face key point alignment and normalization cropping, and face features are extracted. At the same time, the registration camera identifier, registration time marker and session effective time window boundary are bound to generate a collection session ticket set.

3. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 2, characterized in that: The steps for collecting real-time video streams of left and right blood pressure and height measurements based on the collected session ticket set and extracting facial features at these measurements are as follows: Based on the collection session ticket set, collect real-time video streams of left and right blood pressure points and height points, and simultaneously extract frame segments within the effective time window of the session to generate a set of effective frame indexes for the point sessions. Based on the valid frame index set of point-based sessions, the target face region is locked at the left and right blood pressure points and height points, and occluded interference frames are removed to generate a set of valid face frame filtering results. The effective face frames of the selected locations are aligned and cropped, and the face features of the locations are extracted to generate a set of face feature records.

4. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 3, characterized in that: The steps for generating the re-identification session lock link set are as follows: For the facial features in the location facial feature record set and the facial features in the collection session ticket set, perform cross-location facial re-identification matching, and summarize them into a hit record sequence according to the frame number; The hit records of adjacent frame numbers in the hit record sequence are spliced ​​into a continuous hit segment, and splicing is performed on the short-term occlusions and the occlusion splicing mark is written to generate a re-identification session locking link set.

5. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 4, characterized in that: The steps for performing backtracking location on the hit frame sequence number in the re-identification session locked link set and mapping it to the corresponding video time stamp are as follows: Extract the boundary frame sequence number of the continuously hit segments in the re-identification session locked link set, locate the corresponding original frame sequence number range and frame time marker in the point session valid frame index set, and generate the hit frame sequence number backtracking location set. The frame time stamps of the hit frame sequence number back-location set are sorted and abnormal jumps are removed. The time boundaries of consecutive hit segments are fixed and the point identifiers are bound to generate a video time stamp mapping set.

6. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 5, characterized in that: The steps for generating the face appearance event marker set are as follows: Within the original frame sequence number range corresponding to the video time stamp mapping set, read the face region locking record set and face key points, perform displacement jitter removal and deflection mutation removal, and generate a set of face appearance event segments; The set of face appearance event fragments is combined with face features and location markers, and the original frame sequence range, keyframe pointers and time boundaries are encapsulated to generate a set of face appearance event markers.

7. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 6, characterized in that: The sampling of left and right blood pressure and height points is performed based on the start time marker of the face appearance event using the face appearance event marker set. The steps are as follows: The start and end time markers of face occurrence events in the face occurrence event marker set are organized according to the location identifier and written into the original frame sequence range and keyframe pointer to generate a sampling arrangement list; Based on the sampling arrangement list, sampling is performed at the sampling entrances of left and right blood pressure points and height points, and the measured values ​​and equipment sampling time stamps are transmitted back to generate a point sampling transmission record set.

8. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 7, characterized in that: The steps for generating the original three-clock comparison record set are as follows: Write the receiving time stamp into the sampling and feedback record set of the point, and combine it with the start time stamp and end time stamp of the face appearance event to establish the correspondence of the coverage period and generate a three-clock entry record set; The three-clock entry record set is merged according to the left and right blood pressure points and height points, and the point identifier, original frame number range and key frame pointer are written to generate the three-clock original comparison record set.

9. The method for synchronous data acquisition before blood collection based on face recognition as described in claim 8, characterized in that: The steps for generating the pre-blood collection data set are as follows: The correspondence between the location markers and the start time markers of the face appearance event in the original three-clock comparison record set is verified, and the verification status marker is written to generate a three-clock comparison verification item set; Based on the three-clock comparison verification item set, point time offset comparison verification and difference registration are performed on the start time marker of the face appearance event and the device sampling time marker to generate a point time offset record set; Based on the point time offset record set, perform point transmission delay comparison verification and difference registration on the device sampling time mark and receiving time mark, and generate a three-clock difference registration result set; Based on the three-clock difference registration result set, the device sampling time stamp is converted to a unified time caliber, and the left and right blood pressure and height entries are aligned to generate a pre-blood collection data result set.

10. A face recognition-based pre-blood collection data synchronization acquisition device, based on the face recognition-based pre-blood collection data synchronization acquisition method according to any one of claims 1 to 9, characterized in that: include, The cabinet serves as the overall load-bearing structure for the pre-blood collection data synchronization acquisition device, providing a stable installation environment and protection for each functional module, and ensuring the overall coordinated operation of the equipment. The left-side blood pressure acquisition module is used to acquire left-side blood pressure measurements and record the device sampling time markers. It also acquires real-time video streams, extracts facial features at the locations, and performs cross-location re-identification matching with facial features at the registration entrance. The right-side blood pressure acquisition module is used to collect right-side blood pressure measurements, device sampling time markers, and facial features at the sampling points, enabling simultaneous acquisition of blood pressure data from both sides and identity binding. The height acquisition module is used to collect the patient's height measurement value and record the device sampling time marker. At the same time, it collects real-time video stream and extracts facial features to participate in cross-point face re-identification and matching, ensuring the temporal continuity and correlation between height data and identity information and blood pressure data. Locking casters are used for moving and locking the cabinet.

Images