A dual-channel automatic selection and recognition face recognition system

Abstract

The application discloses a kind of dual-channel automatic selection identification face recognition system, it is related to face recognition field, including image acquisition processing module, face recognition module and output module, core in in that through obstruction detection unit intelligent analysis face stain or obstruction degree, and accordingly automatic switching identification channel: when obstruction is lighter, using the first identification channel based on face feature comparison, when obstruction is serious, then enable the second identification channel based on interactive eyeball tracking identification, in second channel, system guides user to gaze and follow dynamic movement on screen identification object, by high-precision comparison preset trajectory and actual eyeball movement trajectory, and fusion unobstructed face local feature, iris texture and other biological characteristics carry out multimodal judgment and realize identity recognition, effectively solve the problem of recognition failure caused by face obstruction under complex environment, by innovative active eyeball tracking provide very strong living body anti-counterfeiting ability, significantly improve the recognition rate of system.

Description

Technical Field

[0001] This invention relates to the field of facial recognition, specifically to a dual-channel automatic selection facial recognition system. Background Technology

[0002] With the development of the construction industry, the requirements for personnel management on construction sites are becoming increasingly stringent. Traditional methods such as manual roll call, card swiping, and fingerprint attendance are inefficient, prone to proxy attendance, and involve complex data statistics. In recent years, facial recognition technology has been widely used in attendance management, but existing facial recognition systems face many challenges in special environments such as construction sites. For example, workers' faces may be covered in mud, dust, or paint due to work requirements, or partially obscured by safety helmets, scarves, or masks; lighting conditions on construction sites are complex and variable, with strong light, backlight, or low light environments being common; and there is the risk of using photos or videos to deceive workers into clocking in. These factors seriously affect the accuracy and reliability of existing facial recognition systems, resulting in a high failure rate and failing to meet the actual needs of construction sites. Summary of the Invention

[0003] The purpose of this invention is to provide a dual-channel automatic selection and recognition face recognition system, solving the problems of low recognition rate, susceptibility to facial obscuration by dirt, and insufficient liveness detection capability in existing face recognition systems under complex construction site environments. By introducing dual-track recognition channels and a unique eye-tracking recognition mechanism, the system improves the recognition success rate, accuracy, and anti-spoofing capability under harsh conditions, thereby achieving efficient and accurate attendance management for construction site personnel.

[0004] To achieve the above objectives, the present invention provides the following technical solution: 1. A dual-channel automatic selection and recognition face recognition system, characterized in that it comprises:

[0005] The image acquisition and processing module is used to acquire facial images of the user to be identified;

[0006] The facial recognition module executes a dual-channel facial recognition strategy based on the occlusion analysis results of the facial image. The dual-channel facial recognition strategy includes a first recognition channel based on facial feature comparison and a second recognition channel based on interactive eye tracking. The second recognition channel performs the following steps:

[0007] S1: Generate and present a dynamic object that moves along a preset motion trajectory;

[0008] S2: Real-time capture and recording of the actual eye movement trajectory of the user to be identified as their eye follows the movement of the dynamic identification object;

[0009] S3: Compare the actual eye movement trajectory with the preset movement trajectory of the dynamic recognition object to obtain the trajectory matching degree.

[0010] S4: The trajectory matching degree is fused with at least one additional biometric feature extracted from the user to be identified to complete the identity recognition. The additional biometric features include: unobstructed facial features, iris texture features, and eye movement habit features.

[0011] The output recognition result module outputs identity recognition information based on the recognition results of the dual-channel face recognition strategy.

[0012] Preferably, the occlusion analysis results of the facial image are used to implement a dual-channel face recognition strategy, which includes performing pixel-level or region-level analysis on the facial image using a deep learning model to determine whether there are stains or occlusions on the face and the degree of occlusion, and automatically triggering the first recognition channel or the second recognition channel based on the comparison result of the degree of occlusion with a preset threshold.

[0013] Preferably, the first recognition channel based on facial feature comparison includes extracting a high-dimensional facial biometric vector from the facial image and comparing it with a pre-registered worker facial database.

[0014] As a preferred method, real-time capture of the user's eye movement trajectory specifically includes: using a high-definition camera to collect eye images, locating key facial points and the center of the pupil through a cascaded convolutional neural network, and calculating the user's real-time gaze point on the screen in combination with head posture information, thereby generating the actual eye movement trajectory.

[0015] Preferably, the method for comparing the actual eye movement trajectory with the preset movement trajectory includes using a dynamic time warping algorithm or a similarity calculation method based on Fraser distance, and evaluating the smoothness, response delay and acceleration characteristics of the eye movement to calculate the trajectory matching degree.

[0016] Preferably, step S4 further includes evaluating at least one live feature of the user's eye movements, including but not limited to: the smoothness of eye movements, the response speed to an object, or changes in pupil size.

[0017] Preferably, the unobstructed facial features are extracted from the user's eyebrow contour, forehead area, or eye corner area during the second recognition process.

[0018] Preferably, the iris texture features are captured and extracted from the eye area by a high-resolution camera during the second recognition process, when the user's eyeball is relatively still or moving slowly.

[0019] Preferably, the dynamic identification object includes, but is not limited to: light spots, patterned icons, continuously changing number sequences, or randomly generated graphics.

[0020] Preferably, the preset motion trajectory includes, but is not limited to: horizontal reciprocating movement, vertical lifting and lowering, circular trajectory, S-shaped trajectory, Z-shaped trajectory, or a combination of multiple types of trajectory.

[0021] In summary, the beneficial effects of this invention are:

[0022] This invention effectively addresses the core pain point of low recognition rates caused by facial dirt and occlusion in complex environments such as construction sites by introducing an "occlusion detection-dual-channel intelligent switching" mechanism. Its innovative interactive eye-tracking mode not only provides industry-leading, highly spoof-resistant liveness detection through dynamic trajectory following, but also integrates eye movement trajectories, unoccluded local facial features, and iris features for multimodal recognition when the face is severely occluded. This significantly improves the recognition success rate and system robustness under extreme conditions. The entire system has a high level of environmental adaptability and intelligence, and can complete reliable identity verification without human intervention. At the same time, it lays a solid foundation for the future expansion of biometric technology in extreme environments. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the invention 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 only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the overall process framework of a dual-channel automatic selection and recognition face recognition system according to the present invention;

[0025] Figure 2 This is a schematic diagram of the interface for dynamic object guidance in a dual-channel automatic selection and recognition face recognition system according to the present invention;

[0026] Figure 3 This is a schematic diagram illustrating successful recognition in a dual-channel automatic selection and recognition face recognition system according to the present invention;

[0027] Figure 4 This is a schematic diagram illustrating the recognition failure and retry process in a dual-channel automatic selection and recognition face recognition system according to the present invention.

[0028] Figure 5 This is a schematic diagram of the backend data interface in a dual-channel automatic selection and recognition face recognition system according to the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0030] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0031] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0032] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0033] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; they can refer to the internal communication of at least two elements or the interaction relationship of at least two elements, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0034] Please see Figures 1-5 Embodiment 1 provided by the present invention: a dual-channel automatic selection and recognition face recognition system, the core of which is to intelligently switch the recognition channel according to the worker's facial state, and to introduce innovative eye-tracking recognition technology for the situation of facial dirt obscuring the face, mainly composed of hardware modules and software modules;

[0035] The hardware modules include:

[0036] High-definition camera module: Includes at least one visible light camera, with optional infrared camera or 3D depth camera, for capturing high-definition video streams and depth information.

[0037] High-brightness display screen: used to display recognition prompts and results, and to carry the recognition object in specific modes.

[0038] Multi-sensor modules, such as ambient light sensors and distance sensors, assist the system in making environmental adaptive adjustments.

[0039] Network communication module: Supports Wi-Fi / 4G / 5G / wired Ethernet for data uploading and remote management.

[0040] Access control interface: Optional, used for linkage with turnstiles or access control systems.

[0041] The software modules include:

[0042] Image acquisition and processing module: Performs noise reduction, brightness / contrast adjustment, and distortion correction on camera data to acquire facial images of the user to be identified. When a worker approaches the recognition device, the high-definition camera module begins to acquire video streams in real time and preprocesses the acquired video frames, such as image enhancement, noise reduction, and illumination compensation, to improve the image quality for subsequent recognition.

[0043] The facial recognition module includes a face detection and localization unit, a liveness detection unit, an occlusion detection unit, and a recognition unit. Specifically:

[0044] The face detection and localization unit can quickly and accurately detect faces in images and locate key points such as eyes, nose, and mouth.

[0045] The liveness detection unit makes a preliminary judgment on whether the face is real, preventing deception by photos and videos;

[0046] The occlusion detection unit, based on a deep learning model, analyzes facial images in real time to determine whether there are stains or occlusions and the degree of occlusion.

[0047] The recognition unit executes a dual-channel face recognition strategy based on the occlusion analysis results of the occlusion detection unit. The dual-channel face recognition strategy includes a first recognition channel based on facial feature comparison and a second recognition channel based on interactive eye tracking.

[0048] Specifically, the occlusion detection unit is based on a pre-trained deep learning model, such as a convolutional neural network (CNN), which can perform pixel-level or region-level analysis of the face region, identify and quantify facial stains (such as dirt, dust, and paint) and common occlusions (such as helmet rims, masks, and scarves).

[0049] Based on the detection results, the system intelligently determines the clarity and degree of occlusion of the face:

[0050] If the face is clear and there are no obvious stains or obstructions, or the degree of obstruction is less than the preset threshold T1, the system will automatically switch to the first recognition channel: normal face recognition mode.

[0051] If the face has obvious dirt or obstruction, or the degree of obstruction is higher than the preset threshold T1, the system will automatically switch to the second recognition channel: interactive eye-tracking recognition mode.

[0052] First recognition channel: Normal face recognition mode;

[0053] The system extracts high-dimensional facial biometric vectors from clear facial images.

[0054] The extracted feature vectors are quickly compared with the registered worker face database in the system.

[0055] If the similarity reaches the preset threshold T2, the recognition is successful.

[0056] The system uses voice and display screen prompts to identify the results (name, attendance status), records entry and exit times, and uploads the attendance data to the attendance management system.

[0057] Second recognition channel: Interactive eye-tracking recognition mode;

[0058] 1. Object recognition generation and motion guidance:

[0059] The system immediately generates a virtual identifier on the display screen, such as a constantly flashing small light spot, a small icon with a specific pattern, or a continuously changing sequence of numbers.

[0060] The system sets a predefined or randomly generated complex motion trajectory for the object to be identified, such as: horizontal reciprocating movement, vertical lifting and lowering, circular trajectory, S-shaped trajectory, Z-shaped trajectory, or a combination of multiple types of trajectory. The complexity, speed, and duration of the trajectory can be adjusted according to actual needs, aiming to ensure that the user needs to actively and continuously follow it with their eyes.

[0061] The system guides the user through voice prompts and on-screen text: "Please look at and follow the moving point on the screen."

[0062] 2. High-precision eye tracking and trajectory matching:

[0063] The system utilizes high-definition cameras and professional eye-tracking algorithms, such as algorithms that combine deep learning for human eye key point detection, pupil center localization, and gaze direction estimation, to capture and analyze the user's eye area in real time with high precision.

[0064] The system detects the center position of the user's pupils and the direction of eye movement in real time, and calculates the user's gaze point and movement trajectory accordingly.

[0065] The system continuously compares the user's real-time eye movement trajectory with the preset object movement trajectory to determine whether the user's eyes accurately, smoothly and continuously follow the object.

[0066] During this process, the system will also detect liveness characteristics such as the smoothness of eye movements, response speed, and changes in pupil size to enhance its anti-spoofing capabilities.

[0067] 3. Multimodal feature fusion recognition:

[0068] Liveness and attention verification: Accurately following the identified object is itself a very strong liveness detection mechanism. It proves that the user is a real person who is awake and can actively interact, effectively preventing deception by photos, videos, etc.

[0069] Local facial feature supplementation: During eye tracking, even if most of the face is obscured or smudged, the system will try to extract facial features from unaffected local areas, such as the eyebrow outline, forehead, corners of the eyes, and ear edges.

[0070] Eye biometrics assistance is optional: if the camera resolution is sufficient, the system can attempt to capture part of the texture of the user's iris during eye tracking and compare it with the iris features recorded during registration, or compare the micro-habit features of the user's eye movements.

[0071] The system performs multimodal fusion judgment based on eye-tracking trajectory matching, liveness features, and extracted local facial features. Fusion algorithms can employ weighted averaging, decision trees, or deep learning classifiers, among others.

[0072] 4. Judgment of recognition results:

[0073] If the eye-tracking trajectory matching degree reaches the threshold T3, and the liveness feature verification is passed, and the auxiliary feature (such as local face or iris) comparison is successful, then the overall judgment is that the recognition is successful.

[0074] The system provides recognition results via voice and display screen, records entry and exit times and eye-tracking recognition methods, and uploads attendance data to the attendance management system.

[0075] If eye tracking fails, such as failing to follow or being interrupted, the system will prompt you to retry or guide you to a manual confirmation process.

[0076] It is worth mentioning that this embodiment also includes an attendance management module, which records attendance data, manages personnel information, performs data statistics and generates reports. The system stores all attendance records locally, including timestamps, personnel IDs, recognition results, recognition modes, and on-site captured images, and uploads them to the cloud attendance management platform in real time or on a scheduled basis via a network communication module.

[0077] The cloud service interface module enables remote server data synchronization, model updates, and remote management. It supports attendance data query, statistics, and report generation, and can interface with construction site project management systems, payroll systems, etc.

[0078] Example 2 differs from Example 1 in that it optimizes the first recognition channel: normal face recognition mode, specifically including:

[0079] 1. Multi-angle facial image acquisition: During the initial registration, the system can guide users to acquire facial images from multiple angles (front view, left and right tilted 15-30 degrees) and different expressions to build a more robust facial feature library.

[0080] 2. Enhanced Liveness Detection: In addition to basic liveness detection methods such as blinking and mouth opening, a 3D depth sensor can be introduced to acquire facial depth information and combine it with 2D images to determine if it is a genuine face. Alternatively, an infrared camera can be used to collect near-infrared images and analyze skin texture and blood flow characteristics to further enhance anti-counterfeiting capabilities.

[0081] 3. Adaptive Illumination Compensation: Combined with an ambient light sensor, the camera's exposure, gain, and white balance are adjusted in real time. Through image processing algorithms such as local histogram equalization and gamma correction, high-quality face images can be generated even in strong light, backlight, and low light environments, providing reliable input for face feature extraction.

[0082] 4. Incremental learning and model update: As workers' appearance changes, such as wearing glasses or growing beards, the system can support incremental learning and update their facial recognition model with the help of a small number of samples, without the need for re-registration.

[0083] Example 3 differs from Example 1 in that it optimizes the second recognition channel: the interactive eye-tracking recognition mode, specifically including:

[0084] 1. Diverse object styles and trajectories:

[0085] Style: The identifying object can be a graphic with variable color, shape, and size, or a dynamically changing sequence of numbers or letters, or even a complex pattern generated by a random algorithm.

[0086] Trajectory: In addition to the aforementioned geometric trajectory, it can also be a pseudo-random walk trajectory, a trajectory that simulates natural movement (such as insect flight), or a composite trajectory that combines sound and visual cues to further increase the difficulty of imitation.

[0087] Personalized Trajectory: For registered users, the system can learn their unique eye movement habits and generate slightly personalized object recognition trajectories to improve the specificity of recognition.

[0088] 2. High-precision eye positioning and gaze point calculation:

[0089] Using a cascaded convolutional neural network (CNN) model, the face region is first detected and key points are located, including the corners of the eyes and the center of the pupil.

[0090] The pupil center can be accurately extracted using image processing techniques such as Hough transform and star filtering.

[0091] By combining the positional relationship between the pupil and the corner of the eye with head posture information (estimated through facial key points), the user's real-time gaze point on the display screen is calculated.

[0092] 3. Trajectory matching algorithm:

[0093] The algorithm uses dynamic time warping (DTW) or distance-based similarity algorithms, such as Euclidean distance and Frescher distance, to compare the similarity between the user's actual eye movement trajectory and the preset object trajectory.

[0094] Time series analysis techniques were introduced to evaluate dynamic characteristics of eye movements, such as smoothness, response delay, and acceleration.

[0095] Set multiple threshold levels: for example, high matching degree passes directly, medium matching degree requires further judgment based on local facial features or iris features, and low matching degree fails to identify.

[0096] 4. Iris feature-assisted recognition:

[0097] During eye tracking, when the eye is relatively still or moving slowly, the system can attempt to capture the eye area using a high-resolution camera, and extract iris texture features after image enhancement, noise reduction, and normalization.

[0098] The extracted iris features are compared with the iris features collected during registration (or local iris features extracted from high-quality face images) as auxiliary recognition information.

[0099] Example 4 differs from Example 1 in its system fault tolerance and user interaction, specifically including:

[0100] 1. Retry Mechanism for Recognition Failure: If the initial recognition fails, the system will prompt the user via voice and display to adjust their posture, distance, or retry. If the recognition fails X times consecutively, the user will be guided to a manual verification process.

[0101] 2. Optimized prompts and guidance: In eye-tracking mode, in addition to voice prompts, guide lines or arrows can be dynamically displayed on the screen to help users understand how to follow the identified object.

[0102] 3. Data security and privacy protection: Store facial feature templates instead of original facial images, encrypt transmitted data, and comply with relevant data protection regulations.

[0103] In summary, compared with existing facial recognition attendance systems, this invention has the following significant improvements and breakthroughs:

[0104] 1. Solved a core pain point on construction sites: Existing systems suffer from low recognition rates due to facial dirt and occlusion in construction site environments. This invention fundamentally solves this problem through a dual-track channel and eye-tracking mechanism, significantly improving the robustness of recognition.

[0105] 2. Innovative Liveness Detection Mechanism: Traditional liveness detection often relies on static or simple dynamic behaviors such as blinking, opening the mouth, and shaking the head, which are easily fooled by photos, videos, and even simulated masks. The eye-tracking recognition mode of this invention requires users to actively and accurately follow the dynamic object being detected. This interactive and highly challenging liveness detection mechanism exponentially increases the difficulty of spoofing, achieving industry-leading anti-spoofing capabilities.

[0106] 3. Advantages of multimodal fusion recognition: In eye-tracking mode, the system can fuse eye movement trajectories, local facial features (unoccluded parts), and even local iris features. This multimodal fusion decision-making mechanism can improve the accuracy and security of recognition even under extreme occlusion conditions.

[0107] 4. High intelligence and adaptability: The system can intelligently switch recognition modes according to the environment and facial conditions without manual intervention, which greatly improves the intelligence level of the system and the user experience.

[0108] 5. Laying the foundation for future technological development: The eye-tracking technology of this invention provides a new approach and technological foundation for biometric identification in extremely harsh environments (such as scenarios where the faces of firefighters and miners are completely obscured).

[0109] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

Claims

1. A dual-channel automatic selection and recognition face recognition system, characterized in that, include: The image acquisition and processing module is used to acquire facial images of the user to be identified; The facial recognition module executes a dual-channel facial recognition strategy based on the occlusion analysis results of the facial image. The dual-channel facial recognition strategy includes a first recognition channel based on facial feature comparison and a second recognition channel based on interactive eye tracking. The second recognition channel performs the following steps: S1: Generate and present a dynamic object that moves along a preset motion trajectory; S2: Real-time capture and recording of the actual eye movement trajectory of the user to be identified as their eye follows the movement of the dynamic identification object; S3: Compare the actual eye movement trajectory with the preset movement trajectory of the dynamic recognition object to obtain the trajectory matching degree. S4: The trajectory matching degree is fused with at least one additional biometric feature extracted from the user to be identified to complete the identity recognition. The additional biometric features include: unobstructed facial features, iris texture features, and eye movement habit features. The output recognition result module outputs identity recognition information based on the recognition results of the dual-channel face recognition strategy.

2. The dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The occlusion analysis results of the facial image are used to implement a dual-channel face recognition strategy, which includes performing pixel-level or region-level analysis on the facial image using a deep learning model to determine whether there are stains or occlusions on the face and the degree of occlusion, and automatically triggering the first recognition channel or the second recognition channel based on the comparison results of the degree of occlusion with a preset threshold.

3. The dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The first recognition channel based on facial feature comparison includes extracting a high-dimensional facial biometric vector from the facial image and comparing it with a pre-registered database of worker faces.

4. The dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: Real-time capture of user eye movement trajectory specifically includes: using a high-definition camera to collect eye images, locating key facial points and pupil centers through a cascaded convolutional neural network, and calculating the user's real-time gaze point on the screen in combination with head posture information, thereby generating the actual eye movement trajectory.

5. A dual-channel automatic selection and recognition face recognition system according to claim 4, characterized in that: The method for comparing the actual eye movement trajectory with the preset movement trajectory includes using a dynamic time warping algorithm or a similarity calculation method based on Fraser distance, and evaluating the smoothness, response delay and acceleration characteristics of the eye movement to calculate the trajectory matching degree.

6. A dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The S4 step also includes evaluating at least one liveness feature of the user's eye movements, including but not limited to: the smoothness of eye movements, the response speed to the object being identified, or changes in pupil size.

7. A dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The unobstructed facial features are extracted from the user's eyebrow contour, forehead area, or eye corner area during the second recognition process.

8. A dual-channel automatic selection and recognition face recognition system according to claim 7, characterized in that: The iris texture features are captured and extracted from the eye area by a high-resolution camera during the second recognition process, when the user's eyeball is relatively still or moving slowly.

9. A dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The dynamic identification objects include, but are not limited to: light spots, pattern icons, continuously changing number sequences, or randomly generated graphics.

10. A dual-channel automatic selection and recognition face recognition system according to claim 1, characterized in that: The preset motion trajectory includes, but is not limited to: horizontal reciprocating movement, vertical lifting and lowering, circular trajectory, S-shaped trajectory, Z-shaped trajectory, or a combination of multiple types of trajectory.

Images