Face recognition access control living body detection and anti-counterfeiting method in complex environment

Abstract

The application provides a face recognition access control living body detection and anti-fake method in a complex environment, and relates to the technical field of face recognition. The method comprises the following steps: in response to an access control authentication request, a directional airflow is blown along the surface of a transparent eye protection piece transversely, and visible light and thermal imaging image sequences are acquired during this period; an eye area is determined, and a condensation disappearance boundary is determined in the image area of the transparent eye protection piece; when the condensation disappearance boundary moves to the eye area, local face features are extracted from a target clear area; the edge extension exhalation heat trace, source end position and tail end moving direction of a breathing protection piece edge are determined according to the thermal imaging image; if the source end position remains at the edge of the breathing protection piece and the tail end moving direction is the same as the airflow, the local face features are taken as living body identity features, matched with pre-stored eye area features, and an identification result is obtained. The application utilizes the same airflow to associate condensation disappearance and exhalation heat trace response, and improves the reliability of access control identification in a protective wearing state.

Description

Technical Field

[0001] This application relates to the field of facial recognition technology, and in particular to a method for liveness detection and anti-counterfeiting of facial recognition access control systems in complex environments. Background Technology

[0002] When personnel wearing goggles and respirators undergo access control authentication, the visible light image primarily preserves the area around the eyes, while the nose, mouth, and cheek areas are obscured by the respirator. The warm, moist air exhaled by the personnel escapes from the edges of the respirator. Some of this warm, moist air reaches the surface of the transparent goggle and condenses, obscuring the eye contours and periocular texture. The remaining warm, moist air diffuses along the outer edge of the respirator, forming a continuous thermal area in the thermal imaging image. Therefore, access control authentication requires obtaining identity data from the gradually receding periocular area while the personnel remain protected, while simultaneously confirming that the thermal changes on the outer edge of the respirator originate from the personnel's real-time breathing.

[0003] When wearing transparent eye protection and respiratory protection to pass through access control, condensation on the surface of the transparent eye protection can obscure the eye contours and periocular texture, while the respiratory protection can obscure the mouth and nose area. This reduces the facial area available for identification in visible light images, and thermal changes related to respiration in thermal imaging are unreliable as a sole indicator of liveness.

[0004] Current access control facial recognition systems typically use visible light facial images for identity comparison, while liveness detection can combine infrared imaging, motion coordination, or temperature changes to determine the status of a real person.

[0005] When protective gear is worn continuously and condensation occurs on the transparent surface, the visible area around the eyes is incomplete, and the mouth and nose area cannot be used as a basis for routine facial comparison, which can easily lead to a decrease in the reliability of access control identity authentication and liveness detection results. Summary of the Invention

[0006] This application provides a method for liveness detection and anti-counterfeiting of facial recognition access control in complex environments, which solves the problem in the prior art that the reliability of facial recognition liveness detection of personnel wearing protective gear is insufficient because transparent eye protection parts condense and obscure the area around the eyes and respiratory protection parts cover the mouth and nose.

[0007] To address the aforementioned technical problems, in a first aspect, this application provides a method for liveness detection and anti-counterfeiting in complex environments using facial recognition access control, for authenticating the identity of individuals wearing transparent eye and respiratory protective gear, including: In response to access control authentication requests, a directional airflow is blown laterally along the surface of the transparent eye protection component; During the application of directional airflow, multiple frames of visible light images and multiple frames of thermal images of the person to be identified are acquired in chronological order. The periocular area covered by the transparent eye protection device is determined based on multiple frames of visible light images, and the condensation area, the clear target area, and the condensation fading boundary between the condensation area and the clear target area are determined within the image area of ​​the transparent eye protection device. When the condensation disappearance boundary moves to the peri-eye area, local facial features are determined based on the clear target area within the peri-eye area. Based on multi-frame thermal imaging images, the exhaled thermal trace extending outward from the edge of the respiratory protective device, as well as the source position and tail direction of the exhaled thermal trace, were determined. If the source end position remains at the edge of the respiratory protective device, and the direction of movement of the tail end is consistent with the direction of the directional airflow, then the local facial features will be identified as liveness identification features; The liveness identification features are matched with pre-stored periorbital facial features to generate access control recognition results.

[0008] Optionally, after confirming that the source end position remains at the edge of the respiratory protective device and that the tail end movement direction is consistent with the directional airflow direction, the initial airflow direction is recorded, and the exhaled heat trace is verified a second time by randomly switching to the verification airflow.

[0009] In response to the source end position remaining at the edge of the respiratory protective device and the tail end moving in the same direction as the directional airflow, the direction of the directional airflow is recorded as the initial airflow direction; Switch the directional airflow to a verification airflow with a direction opposite to the initial airflow direction; During the application of the verification airflow, multiple frames of verification thermal imaging images of the person to be identified are acquired in chronological order. Based on multiple frames of verification thermal imaging images, the verification exhalation thermal trace extending outward from the edge of the respiratory protective device, as well as the location of the verification source end and the direction of movement of the verification tail end of the verification exhalation thermal trace, are determined. If the source end position and the verification source end position are located at the same edge of the respiratory protective device, the direction of the tail end movement is consistent with the initial airflow direction, and the verification exhalation heat trace responds effectively to the direction switching of the verification airflow, then the local facial features are identified as liveness identification features.

[0010] Optionally, the directional airflow is switched to a verification airflow with a direction opposite to the initial airflow direction, including: The moment when the source end position is kept at the edge of the respiratory protective device and the tail end movement direction is consistent with the initial airflow direction is determined as the verification start moment; Randomly select one acquisition time from at least two predetermined thermal imaging acquisition times during the period after the verification start time and before the directional airflow has switched, and determine the selected acquisition time as the verification switching time; When the verification switching time is reached, the directional airflow is switched to the verification airflow, and the thermal imaging images collected from the time of verification switching are determined as the verification thermal imaging images.

[0011] Optionally, determining the verification exhalation thermal trace extending outward from the edge of the respiratory protective device, as well as the verification source position and verification tail movement direction of the verification exhalation thermal trace, based on multiple frames of verification thermal imaging images, includes: Identify the continuous thermal region connected to the edge of the respiratory protective gear from each frame of the verification thermal imaging image; The location where the continuous hot zone connects to the edge of the respiratory protective device is determined as the verification source location; The location furthest from the source end of the continuous hot zone is defined as the tail end of the verification. The direction of movement of the verification tail end is determined based on the position of the verification tail end at different acquisition times, and the moment when the direction of movement of the verification tail end first aligns with the direction of the verification airflow is determined. The verification response duration is determined based on the verification switching time and the thermal trace turning time. If the verification response time does not exceed the preset response time limit, the verification of the exhaled heat trace is deemed to be effective in verifying the direction switching response of the airflow.

[0012] Optionally, determining the source location and verifying that the source location remains at the same edge of the respiratory protective device includes: The edge reference line of the respiratory protective device is determined based on multiple frames of thermal imaging images and verification thermal imaging images; Map the source position and the verification source position to the edge reference line respectively to obtain two edge mapping positions; Determine the mapping error of the control points during image registration, and determine the allowable positional deviation along the edge reference line based on the mapping error; if the distance between the two edge mapping positions along the edge reference line is not greater than the allowable positional deviation, then determine that the source end position and verify that the source end position are kept at the same edge position of the respiratory protective device.

[0013] Optionally, the periocular area covered by the transparent eye protection device and the condensation fading boundary between the condensation area and the target clear area are determined based on multiple frames of visible light images, including: The outline and key points of the transparent eye protection device are extracted from multiple visible light images. The perianal region is determined based on the key points of the eye located within the outline. The condensation fading process is tracked in adjacent image blocks within the outline of the transparent eye protection device. Image blocks that reveal perianal texture or extract key points of the eye are identified as target clear regions. Image blocks that are adjacent to the target clear regions but still obscure perianal texture are identified as condensation regions. The boundary line between the image blocks of the target clear regions and the condensation regions is identified as the condensation fading boundary.

[0014] Optionally, the liveness identification features are matched with pre-stored periorbital facial features to generate access control recognition results, including: Based on the location of the person's identity card and liveness feature in the periorbital area carried in the access control authentication request, the pre-stored periorbital facial features of the same person and corresponding location are selected. The feature similarity between the liveness feature and the selected pre-stored periorbital facial features is determined. When the feature similarity reaches a preset matching threshold, an access control recognition result that allows passage is generated; otherwise, an access control recognition result that denies passage is generated.

[0015] Optionally, when the condensation disappearance boundary moves to the periocular region, local facial features are determined based on a clearly defined target area within the periocular region, including: Identify the overlapping area between the clear target area and the peri-eye area. When all key points of the same eye are located in the overlapping area, the area enclosed by all key points of the eye in the overlapping area is identified as the feature extraction area, and local facial features are extracted from the feature extraction area.

[0016] Optionally, based on multiple frames of thermal imaging images, the expiratory thermal trace extending outward from the edge of the respirator, as well as the source location and tail direction of the expiratory thermal trace, are determined, including: Continuous thermal regions connected to the edge of the respirator and with a temperature higher than the adjacent background area are extracted from each frame of thermal imaging images. These continuous thermal regions are identified as exhaled thermal traces. The connection position between the continuous thermal regions and the edge of the respirator is identified as the source end position. The direction of tail movement is determined based on the cumulative position change of the tail position farthest from the source end position over multiple consecutive acquisition times.

[0017] Optionally, in response to an access control authentication request, a directional airflow is blown laterally along the surface of the transparent eye protection, including: The system acquires the initial visible light image at the time of the access control authentication request. When at least one key eye point is extracted from the initial visible light image, the system determines the initial peri-eye region based on the initial visible light image and defines the region from which the key eye point is extracted as the initial clear region. The system determines the distances from the initial clear region to the left and right lateral sides of the initial peri-eye region and defines the lateral side corresponding to the distance with the smallest value among the two distances as the starting blowing side. When the two distance values ​​are equal, either distance cannot be determined, or the initial clear region does not exist, the system defines the pre-registered lateral side as the starting blowing side and blows a directional airflow in the direction from the starting blowing side to the initial peri-eye region.

[0018] The technical solution provided in this application has the following beneficial effects: This application delivers a directional airflow along the surface of a transparent eye protection device, causing the condensation fading boundary to move with the airflow within the image area of ​​the transparent eye protection device and enter the reflected periorbital region, thereby forming a clear target area that reveals periorbital texture or allows for the extraction of key eye points. Simultaneously, the tail end of the exhaled heat trail extending from the edge of the respiratory protective device deflects along the same airflow direction, and can form an effective deflection result after the airflow reverses. Only when the source end position remains at the edge of the respiratory protective device and the tail end movement direction corresponds to the airflow direction, the local facial features generated in the clear target area are identified as liveness identification features. Thus, the condensation fading result, local facial features, and real-time respiratory response are causally correlated through the same airflow process, improving the ability of access control facial recognition results to resist forgery by photos, displayed images, and independent heat sources without the person removing the protective device. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the face recognition access control liveness detection and anti-counterfeiting method provided in an embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] This application applies to scenarios where personnel are still required to wear goggles, transparent face shields, masks, or respirators when entering access control points. After personnel enter the access control buffer zone from the low-temperature work area, condensation is likely to form on the transparent protective surfaces around the eyes, and the mouth and nose areas are blocked by respiratory protective gear. Access control authentication needs to be completed when the area around the eyes is partially visible and the exhaled breath heat traces can be captured by thermal imaging.

[0023] Therefore, this embodiment uses visible light images and thermal imaging images within the same certification period to record local information around the eyes and thermal changes on the outer edge of the respiratory protective device after the condensation has subsided.

[0024] like Figure 1As shown, the face recognition access control liveness detection and anti-counterfeiting method in complex environments is executed sequentially from S101 to S107.

[0025] S101, in response to an access control authentication request, blows a directional airflow laterally along the surface of the transparent eye protection component.

[0026] Access control authentication requests can be generated when the person to be identified initiates the authentication process or when the person enters the image acquisition location. In response to the access control authentication request, an initial visible light image of the person to be identified is acquired. This initial visible light image is acquired before the directional airflow begins and is used to record the initial condensation state on the surface of the transparent eye protection device; the transparent eye protection device can be goggles or the transparent area corresponding to the eyes within a transparent face shield; when at least one key eye point can be extracted from the initial visible light image, the initial visible light image also serves as a data source for determining the initial airflow direction.

[0027] Subsequently, the outline and key points of the transparent eye protection device are extracted from the initial visible light image. Key points are image feature points representing the position of the eyes and the contour of the eyelids, including the inner corner of the eye, outer corner of the eye, upper eyelid contour points, and lower eyelid contour points. Key points can be obtained using existing face alignment methods. When there are areas in the initial visible light image that reveal periocular texture or from which key points can be extracted, the initial periocular region is determined based on the key points located within the outline of the transparent eye protection device, and the aforementioned visible local area is defined as the initial clear region; the initial clear region represents the local area already visible before directional airflow blowing, used to determine the starting blowing side.

[0028] After obtaining the initial sharp region, the distances from the initial sharp region to the left and right lateral sides of the initial peri-eye region are calculated. The distance can be taken as the pixel interval between the circumscribed contour of the initial sharp region and the corresponding lateral side along the binocular alignment direction. The lateral side corresponding to the smallest of the two distances is determined as the starting blowing side, and a directional airflow is blown in the direction from the starting blowing side towards the initial peri-eye region. If the two distances are equal, either distance cannot be calculated from image coordinates, or the initial sharp region does not exist, the pre-registered lateral side is used as the starting blowing side. The pre-registered lateral side can be recorded during the gating configuration stage, for example, recorded as the left or right side of the image, to provide a definite blowing direction in cases where no initial texture is available.

[0029] Directional airflow can be achieved by changing the lateral blowing side. At the start of certification, air is blown from the initial blowing side towards the initial periorbital area; when verification airflow is needed, blowing in the current direction is stopped, and air is blown from the opposite lateral side towards the direction of the initial airflow, or the airflow direction is directly changed so that the verification airflow is opposite to the direction of the initial airflow. The above switching is based on the reversal of the airflow direction and is not limited to a specific air supply mechanism.

[0030] A directional airflow travels laterally across the surface of the transparent eye protection, first dissipating condensation on the incoming side, and then passing through the adjacent space between the transparent eye protection and the respirator, causing the tail end of the exhaled heat trail on the outer edge of the respirator to deflect with the airflow. Before certification, a test blowing method can be used to confirm that the condensation dissipation boundary can move continuously and that the exhaled heat trail can deflect with the airflow, thereby determining the blowing direction and image acquisition cycle.

[0031] S102, during the application of directional airflow, acquires multiple frames of visible light images and multiple frames of thermal imaging images of the person to be identified in chronological order.

[0032] After the directional airflow begins, visible light and thermal images are continuously acquired, and a timestamp is recorded for each frame. Multiple visible light images are a sequence arranged chronologically and showing changes in the condensation area, while multiple thermal images are a sequence arranged chronologically during the same airflow period and showing continuous changes in the position of the thermal region. The timestamps are used to map both types of images to the same time period: the start of the directional airflow, the movement of the condensation dissipation boundary, and the deflection of the exhaled heat trail.

[0033] To ensure a common reference for the positions in both types of images, the visible light imaging range and the thermal imaging range are spatially mapped before authentication. For example, existing multimodal image registration methods can be used to transform the outline of the transparent eye protection, key eye points, the edge of the respiratory protection, and continuous thermal areas to the same coordinate plane. When the person to be identified moves their head during acquisition, the positions of adjacent images are corrected based on the outlines of the transparent eye protection and the respiratory protection, ensuring that subsequent positional changes are referenced to the face of the person to be identified.

[0034] After completing the temporal arrangement and spatial correspondence, the multiple visible light images undergo condensation fading processing, and the multiple thermal imaging images undergo exhaled thermal trace processing. These two processes can be performed separately, but they use the same directional airflow duration and the same lateral coordinate direction, thus providing a consistent benchmark for subsequent comparisons of the tail-end movement direction and the directional airflow direction.

[0035] S103, determine the peri-ocular area covered by the transparent eye protection device based on multiple frames of visible light images, and determine the condensation area, the clear target area, and the condensation fading boundary between the condensation area and the clear target area within the image area of ​​the transparent eye protection device.

[0036] In a multi-frame visible light image, the first frame, capable of completely extracting the same key eye point, is selected. The periocular region is then determined based on the distribution of this key eye point within that frame. This periocular region is the target area used to extract local facial features during the authentication process. Subsequently, based on the positional changes of the transparent eye protection device's outline in adjacent frames, the periocular region is mapped to the preceding visible light image before the first frame, providing a fixed reference for whether the condensation fading boundary in the preceding frame enters the periocular region. The initially clear area corresponds to the local visible state before blowing, while the target clear area corresponds to the local visible state formed during the application of directional airflow as the condensation fades.

[0037] After determining the periocular region, the condensation fading process is tracked within adjacent image blocks along the outline of the transparent eye protection component. For any image block, the periocular texture changes from being obscured by condensation to being discernible, or the eye key point detection result falls within the image block, indicating that the condensation occlusion has weakened to a state usable for periocular recognition; image blocks satisfying either of these states are identified as the target clear region. Image blocks adjacent to the target clear region that still obscure the periocular texture and do not show eye key point detection results are identified as condensation regions; the boundary line between the image blocks of the target clear region and the condensation region is identified as the condensation fading boundary.

[0038] In the transparent eye protection device, the area covering the eye periphery is defogging starting from the incoming airflow side under the influence of directional airflow. The incoming airflow side first reveals a clear target area capable of displaying the periphery texture or extracting key eye features; adjacent areas still obscuring the periphery texture are designated as condensation areas. The boundary between the clear target area and the condensation area moves with the airflow direction and is defined as the condensation fading boundary. When the condensation fading boundary moves into the periphery area, it indicates that the eye contour has transitioned from a condensation-obscured state to a recognizable state. At this point, local facial features are determined from the clear target area within the periphery region.

[0039] S104, when the condensation disappearance boundary moves to the peri-eye area, determine the local facial features based on the clear target area located within the peri-eye area.

[0040] First, identify the overlapping area between the clearly defined target area and the periocular region. When all keypoints corresponding to the same eye are located within the overlapping area at adjacent acquisition times, the region enclosed by all keypoints within the overlapping area is defined as the feature extraction region. All keypoints are a set of keypoints used to form the closed contour of the same eye. The fact that all keypoints enter the overlapping area at adjacent acquisition times indicates that the feature extraction region has covered the complete eye contour.

[0041] After obtaining the feature extraction region, the region is aligned and normalized in scale. Then, the eye contour and periorbital texture are encoded to form local facial features. Local facial features are feature data obtained from the currently clear target region and used to represent the local identity of the person to be identified. For example, an existing Local Binary Pattern can be used to encode the periorbital texture, or a trained facial feature extraction network can be used to process the same feature extraction region. The local facial features also record the position of either the left or right eye for subsequent selection of the corresponding pre-stored periorbital facial features.

[0042] S105, based on multi-frame thermal imaging images, determines the exhaled thermal trace extending outward from the edge of the respiratory protective device, as well as the source location and tail direction of the exhaled thermal trace.

[0043] For each frame of thermal imaging, the edge of the respiratory protective device is first determined. The respiratory protective device can be a mask or a respirator. Adjacent background areas are adjacent image regions outside the continuous thermal area that are not connected to the edge of the respiratory protective device; continuous thermal areas connected to the edge of the respiratory protective device and with a temperature higher than the adjacent background area are identified as exhaled thermal traces. Exhaled thermal traces represent the extended shape of exhaled moist and hot gas in the thermal imaging image and can be obtained from the thermal imaging image using existing connected component labeling.

[0044] After determining the exhaled thermal trace, the location where the continuous thermal area connects to the edge of the respirator is defined as the source end position. The source end position is the starting point where the exhaled thermal trace extends outward from the edge of the respirator. When the continuous thermal area forms multiple connected pixels with the edge of the respirator, the middle pixel after the connected pixels are arranged along the edge of the respirator is taken as the source end position. If the source end position is continuously located at the edge of the respirator in multiple adjacent thermal imaging frames arranged in time, the source end position is determined to remain at the edge of the respirator.

[0045] Next, the location furthest from the source end within the continuous thermal region is defined as the tail end position. The tail end position represents the outer end of the exhaled thermal trace furthest from the edge of the respirator. The tail end positions from multiple consecutive data collection times are connected according to timestamps, and the tail end movement direction is determined based on the cumulative positional change of the tail end positions. If the tail end movement direction is the same as the directional airflow direction, then the tail end movement direction is determined to be consistent with the directional airflow direction.

[0046] During the above process, the continuously escaping hot and humid gas from the edge of the respirator replenishes the source end of the exhaled heat trail, keeping the source end positioned at the edge of the respirator. After leaving the edge, the tail end of the exhaled heat trail enters the space traversed by the directional airflow, and moves downstream with the airflow. The condensation disappearance boundary in the visible light image and the direction of tail end movement in the thermal imaging image both correspond to the same period of directional airflow action. The two observations are correlated temporally using timestamps.

[0047] S106 If the source end position remains at the edge of the respiratory protective device and the direction of movement of the tail end is consistent with the direction of the directional airflow, then the local facial features are identified as liveness identification features.

[0048] Among these, the liveness identification feature is a local facial feature that simultaneously meets the conditions of periocular identity recognition and real-time expiratory response. The source end is positioned at the edge of the respirator to define the expiratory heat trace as originating from the continuously escaping warm, moist gas at the edge of the respirator; the tail end moves in the same direction as the directional airflow to indicate that the expiratory heat trace responds synchronously to the applied directional airflow. Once both conditions are met, the local facial feature obtained from a clearly defined target area during the same airflow period is identified as the liveness identification feature.

[0049] To further verify the response of the expiratory thermal trace to real-time airflow, when the above two conditions are met for the first time, the current directional airflow direction is recorded as the initial airflow direction, and the moment when the conditions are met is determined as the verification start time. The verification start time represents the moment when the initial expiratory thermal trace has formed a directional response to the directional airflow, and it is also the starting point for randomly selecting the verification airflow switching moment.

[0050] Subsequently, one of the at least two predetermined thermal imaging image acquisition times, occurring after the verification start time and before the directional airflow has switched, is randomly selected as the verification switching time. The at least two predetermined thermal imaging image acquisition times constitute the minimum number of candidates required for random selection. Increasing the number of candidates creates more possible switching points; the specific number can be determined by the image acquisition frequency and the single access control authentication time.

[0051] For example, if the verification start time is 12.0 seconds and the predetermined thermal imaging image acquisition period is 0.1 seconds, then the three predetermined acquisition times during the period before the directional airflow has switched can be 12.1 seconds, 12.2 seconds and 12.3 seconds respectively; when 12.2 seconds is randomly selected, the verification switching time is 12.2 seconds, and then the airflow reverse switching is performed only once according to that time.

[0052] When the verification switching moment is reached, the current lateral blowing direction is switched to the opposite direction of the initial airflow to form a verification airflow, and the thermal imaging images acquired since the verification switching moment are determined as multi-frame verification thermal imaging images. The verification airflow can be formed by blowing from the opposite lateral side or by changing the blowing direction of the same airflow path; the verification airflow is used to verify whether the exhaled heat trace can change its movement direction with random reverse airflow.

[0053] For each frame of the verification thermal imaging image, a continuous thermal region connected to the edge of the respirator is identified, and this continuous thermal region is designated as the verification exhalation thermal trace. The location where the verification exhalation thermal trace connects to the edge of the respirator is designated as the verification source position, and the location in the continuous thermal region farthest from the verification source position is designated as the verification tail position. The direction of movement of the verification tail is determined based on the changes in the verification tail position at different acquisition times.

[0054] When an airflow switch occurs, the tail end of the already formed expiratory heat trail may retain its original movement trend. After the verification airflow reaches the space where the expiratory heat trail is located, the newly escaped hot and humid gas and the outer part of the original expiratory heat trail begin to deflect towards the verification airflow direction. The moment when the movement direction of the verification tail end first aligns with the verification airflow direction is defined as the heat trail turning moment, and the time interval between the verification switch moment and the heat trail turning moment is defined as the verification response duration. The verification response duration represents the time taken for the expiratory heat trail to respond directionally to the reverse operation of the airflow.

[0055] For example, the preset response time limit is the upper limit of the time allowed for the verification expiratory thermal trace to complete its turn, and is used to determine whether the verification expiratory thermal trace responds to the verification airflow in a timely manner. The preset response time limit can be set according to the airflow propagation time and image acquisition cycle measured during the pre-certification test blow; if the value is lower than the airflow propagation time, the actual expiratory thermal trace may only complete its turn after the time limit expires; if the value is too long, delayed changes in the thermal trace may also be included in this response.

[0056] For example, if the verification switching time is 12.2 seconds and the heat trace turning time is 12.9 seconds, then the verification response time is 12.9 seconds minus 12.2 seconds, which is 0.7 seconds. When the preset response time limit obtained by test blowing calibration is 1.0 seconds, 0.7 seconds does not exceed 1.0 seconds. Therefore, it can be determined that the verification of the exhaled heat trace is effective in verifying the direction switching response of the airflow.

[0057] If the verification response time does not exceed the preset response time limit, the verification of the expiratory thermal trace is deemed valid for the direction switching response of the verification airflow. Therefore, a temporal relationship is established between randomly occurring airflow switching events and subsequent thermal trace reversal events; the verification of the expiratory thermal trace requires the completion of the direction change within the corresponding time limit.

[0058] Next, edge reference lines for the respirator are determined based on multiple frames of thermal imaging images and multiple frames of verification thermal imaging images. The edge reference lines are the common edge positions formed after image registration of the respirator edges before and after the airflow switch, used to compare and verify the source-end positions. When determining the edge reference lines, the positions of the respirator edges before and after the airflow switch are corrected, and the corrected edges are used as a unified mapping benchmark.

[0059] The source and verification source positions are mapped to the edge reference line along the direction shortest to the edge reference line, resulting in two edge-mapped positions. Each edge-mapped position is the corresponding image sampling position of the source position on the unified edge reference line. The mapping error of the control points during image registration is determined, and the allowable positional deviation along the edge reference line is determined based on this error. When the distance between the two edge-mapped positions along the edge reference line is no greater than the allowable positional deviation, the source and verification source positions are determined to remain at the same edge position of the respiratory protective device.

[0060] When the source and verification source positions are located at the same edge of the respiratory protective device, the tail end moves in the same direction as the initial airflow, and the verification exhalation thermal trace responds effectively to the change in the direction of the verification airflow, the local facial features are identified as liveness detection features. Before and after the airflow direction changes randomly, the exhalation thermal trace is generated from the same edge position, and the tail end moves sequentially along two opposite airflow directions, thus forming a liveness detection result where the source and tail response complement each other.

[0061] The photos and displays only provide visible light images around the eyes, and the continuously generated exhaled heat traces are missing from the edges of the respiratory protective gear. A continuous thermal area formed by an independent heat source also needs to satisfy the requirement that the source end position before and after airflow switching maps to the same edge position, and that the tail end changes direction accordingly with randomly switched airflow. Pre-recorded thermal imaging images have fixed heat trace turning times; the randomly generated verification switching times in this certification are used to examine the temporal relationship between heat trace turning and real-time airflow.

[0062] S107, Match the liveness identification features with the pre-stored periorbital facial features to generate access control recognition results.

[0063] The pre-stored periorbital facial features are periorbital feature data generated during the personnel registration phase and stored in association with the personnel identification. This implementation uses 1:1 authentication as an example, where the access control authentication request carries the personnel identification. During registration, a visible light image containing the complete periorbital region is acquired. Using the same eye keypoint localization, position alignment, scale normalization, and feature encoding process as in the authentication phase, left and right periorbital facial features are obtained. The left periorbital facial feature records the left eye position marker, and the right periorbital facial feature records the right eye position marker, thus forming pre-stored periorbital facial features with local positional correspondences.

[0064] After liveness detection is completed, pre-stored periorbital facial features are selected based on the personnel identification carried in the access control authentication request and the corresponding eye positions. For example, when condensation recedes from the left side and all key eye points corresponding to the left eye enter the target clear area, the local facial feature record shows the left eye position, and the pre-stored periorbital facial features recording the left eye position under the personnel identification are selected accordingly. If the right eye region first meets the feature extraction conditions, the corresponding selection is completed according to the right eye position.

[0065] Subsequently, the feature similarity between the liveness identification features and the selected pre-stored periorbital facial features is determined. Feature similarity is a numerical value used to represent the degree of proximity between two periorbital feature codes. For example, existing cosine similarity can be used to calculate the directional proximity between two feature codes; before calculation, the lengths of the two feature vectors are normalized. If it is necessary to map the cosine result to the interval between 0 and 1, the cosine result can be incremented by 1 and then divided by 2. A value approaching 1 indicates that the two feature codes are directionally close, and a value approaching 0 indicates that the difference between the two feature codes is increasing.

[0066] As an example, the preset matching threshold can be determined from the verification image samples of registered personnel. Specifically, the preset matching threshold is selected by utilizing the same person's similarity distribution formed by different collections of the same person, and the different person's similarity distribution formed by different people, combined with the allowable false recognition rate and false rejection rate in the access control scenario. When the false recognition rate requirement is lower, the preset matching threshold is increased accordingly; when the false rejection rate requirement is lower, the preset matching threshold is decreased accordingly.

[0067] For example, if a preset matching threshold of 0.78 is selected after sample calibration, and the two normalized periorbital feature codes are mapped by cosine similarity to obtain 0.81, then 0.81 reaches 0.78, generating an access control recognition result that allows passage; if the mapping result is 0.70, then 0.70 does not reach 0.78, generating an access control recognition result that denies passage.

[0068] Finally, when the feature similarity reaches a preset matching threshold, an access control recognition result is generated allowing passage; when the feature similarity does not reach the preset matching threshold, an access control recognition result is generated denying passage. The access control recognition result is a joint decision formed by liveness detection and identity matching. When the liveness detection fails, local facial features are not included in the identity matching process, and an access control recognition result denying passage is directly generated, thus subjecting identity matching to a pre-limitation based on real-time breathing response.

[0069] Through steps S101 to S107, the initial visible light image determines the distance from the initial clear area to the left and right lateral sides, and the starting blowing side is determined by the distance value or the pre-registered lateral sides; the directional airflow pushes the condensation dissipation boundary from the transparent eye protection image area into the reflected periocular area, and multiple frames of visible light images form local facial features; multiple frames of thermal imaging images during the same airflow action form the source end position and the tail end movement direction, and multiple frames of verification thermal imaging images during the verification airflow action form the verification source end position and the verification tail end movement direction, and determine whether the verification response is valid; after the liveness condition is met, the local facial features are converted into liveness identity features and matched with the pre-stored periocular facial features at the corresponding position. The input of each step comes from the image or processing result formed in the previous step, making condensation dissipation, face recognition, liveness detection, and access control anti-counterfeiting form a continuous authentication process.

[0070] The above provides a detailed description of a face recognition access control liveness detection and anti-counterfeiting method for complex environments. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for liveness detection and anti-counterfeiting in complex environments using facial recognition access control, used for access control identity authentication of personnel wearing transparent eye protection and respiratory protection, characterized in that, include: In response to an access control authentication request, a directional airflow is blown laterally along the surface of the transparent eye protection component; During the application of the directional airflow, multiple frames of visible light images and multiple frames of thermal imaging images of the person to be identified are acquired in chronological order. The periocular area covered by the transparent eye protection device is determined based on the multi-frame visible light images, and the condensation area, the clear target area, and the condensation fading boundary between the condensation area and the clear target area are determined within the image area of ​​the transparent eye protection device. When the condensation disappearance boundary moves to the peri-eye area, local facial features are determined based on the clear target area located within the peri-eye area; Based on the multi-frame thermal imaging images, the exhaled thermal trace extending outward from the edge of the respiratory protective device is determined, as well as the source end position and tail end movement direction of the exhaled thermal trace. If the source end position remains at the edge of the respiratory protective device, and the movement direction of the tail end is consistent with the direction of the directional airflow, then the local facial features are identified as liveness identification features; The liveness identification features are matched with pre-stored periorbital facial features to generate access control recognition results.

2. The method according to claim 1, characterized in that, The step of identifying the local facial features as liveness identification features includes: In response to the source end position remaining at the edge of the respiratory protective device and the tail end moving in the same direction as the directional airflow, the direction of the directional airflow is recorded as the initial airflow direction; The directional airflow is switched to a verification airflow with a direction opposite to that of the initial airflow; During the application of the verification airflow, multiple frames of verification thermal imaging images of the person to be identified are acquired in chronological order. The verification exhalation thermal trace extending outward from the edge of the respiratory protective device, as well as the verification source end position and verification tail end movement direction of the verification exhalation thermal trace are determined based on the multi-frame verification thermal imaging images. If the source end position and the verification source end position remain at the same edge position of the respiratory protective device, the movement direction of the tail end is consistent with the initial airflow direction, and the verification exhalation heat trace responds effectively to the direction switching of the verification airflow, then the local facial feature is determined as a liveness identification feature.

3. The method according to claim 2, characterized in that, The step of switching the directional airflow to a verification airflow with a direction opposite to the initial airflow direction includes: The moment when the source end position is maintained at the edge of the respiratory protective device and the direction of movement of the tail end is consistent with the initial airflow direction is determined as the verification start moment; Randomly select one acquisition time from at least two predetermined thermal imaging image acquisition times during the period after the verification start time and before the directional airflow has switched, and determine the selected acquisition time as the verification switching time; When the verification switching time is reached, the directional airflow is switched to the verification airflow, and the thermal imaging image collected from the verification switching time is determined as the verification thermal imaging image.

4. The method according to claim 3, characterized in that, The step of determining the verification exhalation thermal trace extending outward from the edge of the respiratory protective device, and the verification source position and verification tail movement direction of the verification exhalation thermal trace based on the multi-frame verification thermal imaging images, includes: Identify the continuous thermal region connected to the edge of the respiratory protective device from each frame of the verification thermal imaging image; The location where the continuous thermal area connects to the edge of the respiratory protective device is determined as the verification source location; The location furthest from the verification source end in the continuous hot region is determined as the verification tail end location; The direction of movement of the verification tail end is determined based on the position of the verification tail end at different acquisition times, and the moment when the direction of movement of the verification tail end first aligns with the direction of the verification airflow is determined. The verification response duration is determined based on the verification switching time and the thermal trace turning time. If the verification response time does not exceed the preset response time limit, then the verification exhalation heat trace is determined to be effective in responding to the direction switching of the verification airflow.

5. The method according to claim 4, characterized in that, Determining that the source end position and the verification source end position remain at the same edge position of the respiratory protective device includes: The edge reference line of the respiratory protective device is determined based on the multi-frame thermal imaging images and the verification thermal imaging images; The source end position and the verification source end position are respectively mapped to the edge reference line to obtain two edge mapping positions; The mapping error of the control points during image registration is determined, and the allowable position deviation along the edge reference line is determined based on the mapping error; if the distance between two edge mapping positions along the edge reference line is not greater than the allowable position deviation, then the source end position and the verification source end position are determined to remain at the same edge position of the respiratory protective device.

6. The method according to claim 1, characterized in that, The step of determining the periocular area covered by the transparent eye protection device and the condensation fading boundary between the condensation area and the target clear area based on the multiple frames of visible light images includes: The outline and key eye points of the transparent eye protection device are extracted from the multi-frame visible light images. The perianal region is determined based on the key eye points located within the outline. The condensation fading process is tracked in adjacent image blocks within the outline of the transparent eye protection device. Image blocks that reveal perianal texture or extract key eye points are identified as target clear regions. Image blocks that are adjacent to the target clear regions but still obscure perianal texture are identified as condensation regions. The boundary line between the image blocks of the target clear regions and the condensation regions is identified as the condensation fading boundary.

7. The method according to claim 6, characterized in that, The step of matching the liveness identification features with pre-stored periorbital facial features to generate access control recognition results includes: Based on the personnel identification carried in the access control authentication request and the location of the liveness identification feature in the periorbital area, pre-stored periorbital facial features of the same person and corresponding location are selected. The feature similarity between the liveness identification feature and the selected pre-stored periorbital facial features is determined. When the feature similarity reaches a preset matching threshold, an access control recognition result that allows passage is generated; otherwise, an access control recognition result that denies passage is generated.

8. The method according to claim 6, characterized in that, When the condensation disappearance boundary moves to the peri-eye area, determining local facial features based on the clear target area within the peri-eye area includes: The overlapping area between the target clear area and the eye periorbital area is determined. When all key points of the same eye are located in the overlapping area, the area enclosed by all key points of the eye in the overlapping area is determined as the feature extraction area, and local facial features are extracted from the feature extraction area.

9. The method according to claim 1, characterized in that, The step of determining, based on the multi-frame thermal imaging images, the exhaled thermal trace extending outward from the edge of the respirator, as well as the source position and tail direction of the exhaled thermal trace, includes: Extract continuous thermal regions that are connected to the edge of the respiratory protective device and have a temperature higher than the adjacent background area from each frame of thermal imaging images. Define the continuous thermal regions as exhaled thermal traces. Define the connection position between the continuous thermal regions and the edge of the respiratory protective device as the source end position. Determine the tail end movement direction based on the cumulative position change of the tail end position farthest from the source end position over multiple consecutive acquisition times.

10. The method according to claim 1, characterized in that, The method of laterally blowing a directional airflow along the surface of the transparent eye protection in response to an access control authentication request includes: Based on the initial visible light image at the time the access control authentication request occurs, determine the initial periorbital region and the initial clear region within it that reveals periorbital texture or extracts key eye points. The lateral side corresponding to the smallest of the two distances from the initial clear region to the left and right lateral sides of the initial periorbital region is determined as the starting blowing side. If the two distances are equal or the distance cannot be determined, the pre-registered lateral side is determined as the starting blowing side, and a directional airflow is blown in the direction from the starting blowing side to the initial periorbital region.

Images