A method, device, medium and system for optimizing automatic wake-up of a human face access control device

Abstract

This invention addresses the problem of resource waste and false alarms caused by frequent automatic wake-ups of facial recognition access control devices by passersby. It provides a method, apparatus, medium, and system for optimizing the automatic wake-up of facial recognition access control devices. The method involves acquiring the object distance corresponding to the application scenario and adjusting the default focal length of the camera. A preset face detection model is used to perform scanning detection at a preset cycle. When facial features are detected multiple times consecutively and the pixel position offset is less than a dynamically adjusted threshold, the target is determined to be stationary and a wake-up command is triggered for the access control device. After wake-up, the device can further perform liveness detection including infrared or 3D depth verification, as well as face recognition verification based on background removal and affine transformation alignment. Therefore, this invention combines physical static focusing with pixel offset intent determination to achieve accurate wake-up through pure vision without adding additional sensors, effectively eliminating interference from passersby, significantly reducing the ineffective power consumption of the access control device, and improving security.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically, to a method, apparatus, medium, and system for optimizing the automatic wake-up of facial recognition access control devices. Background Technology

[0002] In high-traffic areas such as corridors and roadsides, existing access control systems typically have high recognition sensitivity. Current access control wake-up technologies generally fall into two categories: one involves adding infrared or microwave radar for human detection wake-up, which not only increases hardware costs but also fails to distinguish between unintentional passing and intentional lingering by the radar; the other involves frequently driving the lens motor for autofocus (AF) after detecting a target to obtain a clear face. This not only results in high power consumption for the mechanical motor but also easily shortens the device's lifespan due to frequent focusing. When a passerby unintentionally passes by, the access control system is easily woken up frequently and actively supplements light for face capture. This causes the device to operate at high power for extended periods, consuming significant background computing resources and potentially leading to whitelisted individuals accidentally triggering the door, posing a serious security risk.

[0003] Therefore, how to provide an automatic wake-up technology for facial recognition access control devices that can accurately identify valid objects and eliminate interference from passersby has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for optimizing the automatic wake-up of a face recognition access control device. The method includes: acquiring a default focal length of a camera that matches the target distance in the application scenario, causing targets outside the target distance range to become blurred due to defocusing; performing face feature scanning and detection using a preset face detection model at a preset cycle; determining a valid target is detected when face features are detected multiple times consecutively and the position change feature parameters between adjacent detections meet a preset static condition; and triggering a wake-up command for the access control device when a valid target is detected. As can be seen, by combining physical-level static focal length matching with algorithm-level timed scanning and position change feature parameter detection, the stationary gaze map of the target can be accurately determined from the source without performing high-power recognition, significantly reducing invalid wake-up operations, lowering device power consumption, and improving the security of the access control system.

[0005] Furthermore, obtaining the camera's default focal length that matches the target distance in the application scenario includes: identifying the deployment scenario of the access control equipment and obtaining the target distance within the preset target detection range; configuring the corresponding sensitivity based on this target distance, with higher sensitivity for farther target distances and lower sensitivity for closer target distances; calculating and dynamically adjusting the camera's default focal length based on this target distance, so that targets outside the target distance range cannot have their facial features extracted due to defocus. It is evident that by utilizing the optical defocusing characteristics to shield background interference beyond the set distance, physical-level pre-filtering with zero computational power consumption is achieved, effectively avoiding the high power consumption and mechanical wear associated with traditional autofocus mechanisms.

[0006] Furthermore, the preset face detection model includes a multi-task cascaded convolutional neural network. The preset face detection model performs face feature scanning and detection according to a preset cycle, including: when no face feature is detected in the image, maintaining the current low-power state and entering a waiting state until the next preset cycle arrives; when a face feature is detected for the first time, extracting and temporarily storing the current face feature information and location information, and entering a waiting state until the next preset cycle arrives. It is evident that using a single-stage detector for periodic intermittent scanning and temporarily storing the initial detection information avoids invalid computation caused by false detections in a single frame, improving the reliability of state determination and energy efficiency.

[0007] Furthermore, the position change feature parameters include pixel position offset; determining that a valid target has been detected includes: extracting multiple positional information of the facial features obtained from multiple consecutive scans; calculating the pixel position offset corresponding to the positional information of adjacent detections; when the pixel position offset is continuously less than the pixel position offset threshold, it is determined that the current target is in a stationary state and confirmed as a valid target; wherein, the pixel position offset threshold is positively correlated with the default focal length of the camera. It can be seen that by establishing a dynamic mapping relationship between the default focal length and the pixel position offset threshold, the local imaging magnification effect brought about by the long focal length can be adapted to, ensuring accurate judgment of the person's stationary gaze in different depth-of-field environments.

[0008] Furthermore, triggering wake-up commands for the access control device includes: triggering a mild wake-up command to perform liveness detection; if the liveness detection passes, triggering a full wake-up command to perform facial recognition verification. It is evident that performing tiered wake-up after determining the target is stationary allows for a more rational allocation of system computing resources.

[0009] Furthermore, liveness detection is performed, including: measuring the round-trip time of light reflected from the surface of the target using an infrared light sensor or a three-dimensional structured light sensor to generate depth map information; determining whether the target has a three-dimensional structure based on the depth map information; and if it has a three-dimensional structure, then the liveness detection is deemed successful. It is evident that using the time-of-flight principle to obtain three-dimensional depth information during the mild wake-up phase can intercept two-dimensional plane forgery attacks such as photos and screens, thus improving the anti-counterfeiting security of the access control system.

[0010] Further, facial recognition verification is performed, including: acquiring the verification image corresponding to the valid target, and using image segmentation technology to remove the background from the verification image and extract the foreground face region; using a key point detection algorithm to locate the feature points of the foreground face region and obtain the coordinates of the facial key points; performing an affine transformation based on the facial key point coordinates to align the foreground face region to a standard preset position; and extracting the feature data of the aligned foreground face region and comparing it with a whitelist database. It can be seen that the integrated processing of image segmentation and geometric correction eliminates deformation deviations caused by complex backgrounds and face distortion, improving the accuracy of the final access permission judgment.

[0011] To address the aforementioned technical problems, this invention also provides an optimized device for automatic wake-up of facial recognition access control equipment, comprising: a focus adjustment module for acquiring a default camera focus that matches the target distance in the application scenario, causing targets outside the target distance range to become blurred due to defocusing; a timed scanning module for performing facial feature scanning detection using a preset facial detection model at a preset cycle; a target determination module for determining a valid target has been detected when facial features are detected multiple times consecutively and the positional change feature parameters between adjacent detections meet a preset static condition; and a device wake-up module for triggering a wake-up command for the access control equipment when a valid target is detected. It is evident that through the coordinated work of these functional modules, integrated processing from physical focus adjustment to intent judgment is achieved, improving the overall operational efficiency of the access control equipment.

[0012] To address the aforementioned technical problems, this invention also provides a system for optimizing the automatic wake-up of facial recognition access control devices, comprising: a memory for storing a computer program; a processor for executing the computer program to implement the method described above; and a camera module controlled by the processor for acquiring a default camera focal length that matches the target distance in the application scenario. It is evident that by providing a specific system hardware architecture, a physical basis is provided for the implementation of the aforementioned methods to prevent false wake-up.

[0013] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in any of the preceding claims. It is evident that, through software instruction configuration, general-purpose hardware devices can execute optimized automatic wake-up mechanisms, possessing broad industrial applicability and ease of deployment. Attached Figure Description

[0014] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0015] Figure 1 A flowchart illustrating an optimized method for automatic wake-up of a facial recognition access control device, provided by an embodiment of the present invention; Figure 2 This is a structural block diagram of a device for optimizing the automatic wake-up of a facial recognition access control system, provided in an embodiment of the present invention. Detailed Implementation

[0016] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0017] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0018] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0019] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0020] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0021] In practical applications, the method provided in this invention relies on a corresponding hardware operating environment. The system is primarily deployed in high-traffic access scenarios, using pure machine vision to accurately wake up valid targets, thereby reducing device power consumption and minimizing security risks associated with invalid identification. Specifically, the access control device's motherboard uses a main control chip with edge-side neural network computing power, such as the Hi3519AV100 model chip, to support local inference operations for a multi-task cascaded CNN single-stage detector. Simultaneously, the access control device's image acquisition unit uses a camera module with focusing driven by a VCM (Voice Coil Motor). Combined with... Figure 1 The flowchart of the method shown and Figure 2 The device structure diagram shown illustrates that the system's automatic wake-up and control logic, combined with specific scenarios, executes in the following stages: In the optical pre-filtering stage, such as Figure 1 and Figure 2 As shown, the focus adjustment module 101 acquires the object distance data set for the current application scenario. For example, in a corridor scenario with a width of 1.5 meters, considering the interference from passersby, the system limits the effective detection of objects to within 0.5 meters of the access control machine. The processor generates a drive signal based on this 0.5-meter object distance data, controlling the VCM motor of the camera module to adjust to the corresponding default focus (generally set to 2.8mm), and maintains this focus parameter physically locked in the subsequent standby state. This process differs from the traditional real-time autofocus (AF) mechanism, which maintains a statically fixed default focus during the scanning waiting phase, actively utilizing optical defocus characteristics to shield against background interference from a distance. Specifically, when an ordinary pedestrian passes by outside the 0.5-meter effective range (e.g., at the far end of the corridor), limited by the 2.8mm focal length range locked by the access control camera, the image of their face on the image sensor (such as CMOS) appears severely blurred and out of focus. Because the gradient changes of pixel feature points in such blurry images are extremely gradual, they cannot meet the confidence detection requirements of subsequent CNN algorithms. Thus, the initial interception and filtering of distant pedestrians is completely completed at the physical acquisition node without consuming any algorithm computing power.

[0022] During the intent analysis phase, continue to refer to Figure 1 and Figure 2The timed scanning module 102 calls the neural network model to perform low-power image capture tasks at preset time intervals. In the device's sleep / standby state, the camera outputs a single frame image every 200 milliseconds. When the model first detects a face region in a frame, the system does not immediately wake up the main program, but temporarily stores the coordinate information of the facial feature key points (such as the center of the eyes or the tip of the nose). In subsequent consecutive scanning cycles, the target determination module 103 extracts the corresponding facial feature key point location information and calculates the pixel position offset of adjacent cycle coordinate points on the two-dimensional image. Considering specific behavioral scenarios: when an ordinary passerby walks horizontally across the access control's field of vision at a normal pace, the inter-frame pixel position offset of their facial feature key points within a 200-millisecond interval is usually extremely large (e.g., greater than 20 pixels); while when an internal employee intentionally stops in front of the access control to prepare for facial recognition, the inter-frame pixel position offset caused by the natural slight shaking of their face is extremely small. Therefore, if the calculated pixel position offset is less than the preset pixel position offset threshold in multiple consecutive frames, it is determined that the current detected target is not passing by unintentionally, but is in a state of intentional stillness or lingering, and is confirmed as a valid target.

[0023] It should be noted that the above embodiment, which uses the "pixel position offset between adjacent detections" to determine whether the target is stationary or stationary, is only a preferred embodiment of the present invention. In practical applications, those skilled in the art can also use other computer vision algorithms to extract positional change feature parameters between adjacent detections without departing from the core concept of the present invention, in order to determine whether the target meets the preset stationary condition. For example: (1) Optical Flow: The amplitude of the optical flow vector is obtained by calculating the optical flow field of the pixels in the face region between adjacent scanning detection cycles and used as the position change feature parameter. When the amplitude of the optical flow vector in the face region is continuously lower than the preset optical flow threshold, it is determined that the current target is in a stationary state.

[0024] (2) Frame Difference: Perform difference operations on images from adjacent scanning detection cycles to obtain the number of non-zero pixels or the sum of grayscale changes in the face region difference image as the positional change feature parameter. When the sum of these changes is less than a set difference threshold, it is confirmed as a valid target.

[0025] (3) Intersection over Union (IoU): Extract the bounding boxes of two adjacent detected faces and calculate the intersection-over-union ratio of the bounding boxes in the two frames as the position change feature parameter. When the intersection-over-union ratio is consistently greater than the set overlap threshold (e.g., 0.95), it is determined that the relative position of the face has not moved significantly and is confirmed as a stationary valid target.

[0026] All of the above alternative solutions can achieve accurate determination of the target's stationary gaze based on pure visual algorithms, and should be included within the protection scope of this invention.

[0027] To adapt to environments with varying depths of field and widths, the memory stores a mapping configuration table containing the deployment scene, object distance, default focal length, and pixel position offset threshold. Sensitivity configuration is specifically implemented at the system level through a dynamic, linked mapping configuration table (lookup table method) between object distance, camera default focal length, and pixel position offset threshold. A set of preferred configurations is shown in Table 1: Table 1 Focal Length and Offset Settings Camera distance focal length Pixel fluctuation 0.5 meters 2.8mm 1-2 pixels 1 meter 3mm 1-2 pixels 1.5 meters 3.2mm 2 pixels 2 meters 3.4mm 2 pixels 2.5 meters 3.5mm 2 pixels 3 meters 3.6mm 2 pixels Before making a judgment, the system calls the corresponding parameters using a lookup table. For example, in the corridor scenario mentioned above, with the object distance limit set to within 0.5 meters, the focal length adjustment module 101 sets the default focal length to 2.8 mm. At this time, the system configures the pixel position offset threshold of the facial feature key points for determining whether a person is "stationary / stationary" to be less than 2 pixels. When the access control device is deployed in a more spacious plaza or lobby scenario, for example, when the recognition distance is relaxed to 3 meters, the focal length adjustment module 101 correspondingly increases the focal length to 3.6 mm. Furthermore, when applied to outdoor sidewalks (with an object distance set to within 5 meters) or parking lot entrances (with an object distance set to within 8 meters), due to the significant increase in recognition distance, the system will call the configuration table to significantly increase the focal length to 6 mm or 8 mm respectively. Because the long focal length brings a significant local imaging magnification effect, the same small shaking in the real physical space will be magnified in terms of pixel displacement in the image. To ensure the same tolerance for lingering attention, the offset threshold should be relatively relaxed. Therefore, the system uses a lookup table to dynamically increase the pixel position offset threshold to 3 pixels and 5 pixels respectively. This parameter linkage mechanism perfectly matches the actual optical characteristics, not only verifying the dynamic logic that "the larger the default focal length value, the larger the corresponding pixel position offset threshold setting" through detailed data, but also ensuring the dynamic adaptability of the intent determination standard.

[0028] In the tiered wake-up mechanism, once the target is confirmed to be in a valid stationary state, the device wake-up module 104 issues a wake-up command for the access control device. Specifically, the device wake-up module 104 first issues a mild wake-up command. After receiving the command, the infrared light sensor or three-dimensional structured light component emits a detection beam, calculates the round-trip time of the beam reaching the target surface and reflecting off, and then generates a depth map containing surface depth information. The system compares the depth map data with a standard three-dimensional human face model; if they match and have a three-dimensional structure, the liveness detection is deemed successful.

[0029] In the precise recognition phase, liveness detection triggers a full wake-up command via an event, activating the system's feature comparison engine. The processor runs an image segmentation algorithm to remove background pixels from the current image, retaining the foreground face pixel region. Subsequently, a keypoint detection model outputs the specific coordinates of the eye, nose, and mouth regions. Based on the coordinate data, an affine transformation matrix is ​​constructed to rotate and translate the deflected or tilted face image, correcting it to a standard frontal view position. The system extracts the feature vector of the standard face image and calculates its Euclidean distance or cosine similarity with pre-stored feature vectors in a whitelist database. When the similarity exceeds a threshold, an access control unlocking signal is output to an external relay.

[0030] In summary, the embodiments of the present invention achieve physical defocus filtering by setting a static default focal length, and dynamically adjust the offset threshold by combining neural network periodic scanning and table lookup method, thereby constructing a progressive intent recognition and hierarchical wake-up mechanism, realizing low power consumption and high precision recognition control of access control devices in high-person-density scenarios.

[0031] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A method for optimizing automatic wake-up of a human face access control device, characterized in that, include: Obtain the default focal length of the camera that matches the target distance in the application scenario, so that targets outside the target distance range become blurred due to defocusing; Facial feature scanning and detection are performed using a preset face detection model at a preset cycle. When the facial features are detected multiple times consecutively, and the positional change feature parameters between adjacent detections meet the preset static condition, it is determined that a valid target has been detected. When a valid target is detected, a wake-up command for the access control device is triggered.

2. The method according to claim 1, characterized in that, The process of obtaining the camera's default focal length that matches the target object distance in the application scenario includes: Identify the deployment scenario of the access control equipment and obtain the target distance within the preset target detection range; The sensitivity is configured according to the target object distance. The farther the target object distance is, the higher the sensitivity is configured, and the closer the target object distance is, the lower the sensitivity is configured. The default focal length of the camera is calculated and dynamically adjusted based on the target distance, so that the facial features of targets outside the target distance range cannot be extracted due to defocus.

3. The method according to claim 1, characterized in that, The preset face detection model includes a multi-task cascaded convolutional neural network; the step of using the preset face detection model to perform face feature scanning and detection according to a preset period includes: If the facial features are not detected in the image, maintain the current low-power state and enter a waiting state until the next preset cycle arrives. When the facial feature is detected for the first time, the current facial feature information and location information are extracted and temporarily stored, and the system enters a waiting state until the next preset period arrives.

4. The method according to claim 3, characterized in that, The position change feature parameters include pixel position offset; The determination that a valid target has been detected includes Extract multiple location information obtained from consecutive scans and detections; Calculate the pixel position offset corresponding to the position information of adjacent detections; When the pixel position offset is continuously less than the pixel position offset threshold, it is determined that the current target is in a stationary state and confirmed as a valid target; The pixel position offset threshold is positively correlated with the default focal length of the camera.

5. The method according to claim 1, characterized in that, The trigger command to wake up the access control device includes: Trigger a mild wake-up command to perform a liveness detection; If the liveness detection passes, a full wake-up command is triggered to perform facial recognition verification.

6. The method according to claim 5, characterized in that, The liveness detection includes: The round-trip time of light reflected from the surface of the effective target is measured using an infrared light sensor or a three-dimensional structured light sensor to generate depth map information; Based on the depth map information, it is determined whether the effective target has a three-dimensional structure; if it has the three-dimensional structure, the liveness detection is deemed successful.

7. The method according to claim 5, characterized in that, The facial recognition verification includes: Obtain the verification image corresponding to the valid target, and use image segmentation technology to perform background removal processing on the verification image to extract the foreground face region; A key point detection algorithm is used to locate feature points in the foreground face region to obtain the coordinates of facial key points; Perform an affine transformation based on the coordinates of the facial key points to align the foreground face region to a standard preset position; The feature data of the aligned foreground face region is extracted and compared with the whitelist database.

8. A device for optimizing the automatic wake-up of a facial recognition access control system, characterized in that, include: The focal length adjustment module is used to obtain the default focal length of the camera that matches the target distance of the application scenario, so that targets outside the target distance range become blurred due to defocusing. The timed scanning module is used to perform facial feature scanning and detection according to a preset period using a preset face detection model; The target determination module is used to determine that a valid target has been detected when the face features are detected multiple times consecutively and the position change feature parameters between adjacent detections meet the preset static condition. The device wake-up module is used to trigger a wake-up command for the access control device when the valid target is detected.

9. A system for optimizing the automatic wake-up of facial recognition access control devices, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7; The camera module, controlled by the processor, is used to obtain the default focal length of the camera that matches the target distance in the application scenario.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images