Four-eye AI face recognition passing equipment special for subway

Abstract

The invention belongs to the technical field of face recognition, and discloses four-eye AI face recognition passing equipment special for subways, which comprises a processing unit, a first double-camera module inclined upwards and a second double-camera module inclined downwards. The processing unit is configured to judge the number of faces in an image after the image is acquired; if there are a plurality of human faces, carrying out three-dimensional space distance measurement by using a four-eye system, and determining and only identifying the nearest target human face; and if the number is single, direct identification is carried out. According to the invention, through the up-and-down inclined four-eye camera structure, the vertical field angle is greatly expanded, and passengers with different heights can be adapted; meanwhile, by means of distance measurement and screening logic built in the processing unit, the problem of misrecognition during short-distance queuing is effectively solved, and the passing efficiency and accuracy of the subway turnstile are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of facial recognition technology, specifically relating to a four-eye AI facial recognition access control device for subway use. Background Technology

[0002] In recent years, facial recognition technology has been widely used in turnstile systems in public transportation scenarios such as subways and train stations due to its convenience and contactless operation. However, existing facial recognition access control devices still face some problems and challenges in practical applications.

[0003] First, during rush hour, subway stations experience massive passenger flows, often forming dense queues as passengers pass through turnstiles. Existing facial recognition devices are prone to missed or incorrect detections when handling such large-scale, high-density queues. For example, many subway turnstiles have a 1-meter queue line warning because when a passenger in a back row is too close (less than 1 meter) to the passenger in front who is being recognized, the camera may simultaneously capture two or more faces. In this situation, the device struggles to accurately determine the order and distance of the faces, potentially misidentifying the face of a passenger in the back row while opening the turnstile for the passenger in front, leading to chaotic passage and severely impacting efficiency and safety.

[0004] Secondly, existing facial recognition devices typically use horizontally mounted camera modules, which have a limited vertical field of view. This makes it difficult for the devices to simultaneously accommodate passengers with significant height differences, such as tall adults, short children, or people with special needs. These passengers often need to consciously adjust their posture, such as bending over or standing on tiptoe, to align their faces with the camera's effective recognition area, which not only reduces passage efficiency but also affects the user experience of seamless passage.

[0005] Therefore, how to solve the problem of interference from multiple faces in close-range queuing scenarios and expand the vertical recognition range of the device to accommodate users of different heights are the technical challenges that urgently need to be solved in the field of subway facial recognition access control. Summary of the Invention

[0006] The present invention aims to at least partially solve the aforementioned technical problems. Therefore, the objective of the present invention is to provide a four-eye AI facial recognition access control device specifically for subway systems.

[0007] The technical solution adopted in this invention is as follows:

[0008] A subway-specific four-eye AI facial recognition access control device includes: a device housing; a processing unit disposed inside the device housing; a first dual-camera module disposed on the upper part of the device housing and facing the area to be recognized at a preset first upward tilt angle; and a second dual-camera module disposed on the lower part of the device housing and facing the area to be recognized at a preset second downward tilt angle. The processing unit is configured to: acquire image data from the first dual-camera module and the second dual-camera module respectively, and perform facial recognition processing based on the image data to control access.

[0009] To more effectively handle complex queuing scenarios, the processing unit is specifically configured to perform the following operations: determine the number of faces contained in the image data; if there are multiple faces, perform three-dimensional spatial ranging on the multiple faces to determine the target face closest to the device, and perform recognition processing on the target face; if there is only one face, directly perform recognition processing on the single face.

[0010] To achieve accurate distance measurement, the processing unit is configured to perform three-dimensional spatial distance measurement by combining the four cameras contained in the first dual-camera module and the second dual-camera module in pairs and using different camera baseline distances.

[0011] Furthermore, in order to take into account the ranging accuracy at different distances, the processing unit is configured to: use two cameras in the first dual-camera module or the second dual-camera module to form a short baseline for short-range ranging; and use the cameras of the first dual-camera module and the cameras of the second dual-camera module to form an ultra-long baseline for long-range ranging.

[0012] To expand the detection range, the processing unit is configured to: correct and stitch together the images acquired by the first dual-camera module and the second dual-camera module before determining the number of faces, so as to generate a combined image with a larger vertical field of view, and display the combined image on the screen; at the same time, the screen can also display corresponding prompts based on the recognition results.

[0013] Preferably, the device further includes at least one infrared fill light or at least one white light fill light, disposed near the first dual-camera module and / or the second dual-camera module, for providing supplementary lighting for faces in low-light environments. When an infrared fill light is provided, the characteristics of infrared images can also be used to assist in liveness detection, preventing attacks using photos, videos, etc.

[0014] Preferably, both the preset first upward tilt angle and the preset second downward tilt angle are within the range of 6° to 10°.

[0015] In one specific instance, when both the first upward tilt angle and the second downward tilt angle are 6.37°, the vertical field of view of the device can be extended from the conventional 65° to more than 77.75°.

[0016] To ensure processing efficiency, the processing unit is preferably a high-performance processing chip that integrates artificial intelligence (AI) algorithms.

[0017] After the identification is completed, the processing unit is also configured to upload the information of the successfully identified personnel to the background system and output control signals to control the gate to complete the entire passage process.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention effectively increases the overall vertical field of view by arranging dual camera modules that are tilted upwards and downwards on the upper and lower parts of the device, respectively. This allows the device to easily capture the faces of passengers who are too tall or too short (such as 1.0 meters to 2.2 meters, or even a wider range) without requiring passengers to adjust their posture, thus truly achieving seamless passage and improving the passage experience and efficiency for all passengers.

[0020] The processing unit of this invention is configured to activate a precise three-dimensional spatial ranging function when multiple faces are detected. By accurately calculating and selecting the passenger closest to the device for identification, the problem of misidentification caused by queuing too closely is fundamentally solved, ensuring the accuracy of one person, one gate.

[0021] This invention flexibly combines four cameras to form baselines of different lengths (such as the short baseline inside the dual-camera module and the ultra-long baseline between the upper and lower modules). The device can select the optimal baseline for ranging as needed, achieving equally accurate depth perception for both near and far targets. Attached Figure Description

[0022] Figure 1 This is a hardware structure block diagram of a subway-specific four-eye AI face recognition access control device according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the installation position of the dual-camera module on the device in an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the vertical field of view after the upper and lower dual cameras are combined in an embodiment of the present invention.

[0025] Figure 4 This is a flowchart of the face recognition operation performed by the processing unit in an embodiment of the present invention. Detailed Implementation

[0026] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention. The components of the embodiments of the invention described and illustrated herein can generally be arranged and designed in various different configurations.

[0027] Please see Figures 1 to 3 This embodiment provides a four-eye AI facial recognition access control device specifically for subways. This device is primarily installed on subway turnstiles to replace traditional card-swiping or QR code scanning modules.

[0028] like Figure 1 As shown, the core hardware structure of the device includes a processing unit 101, a first dual-camera module 102, a second dual-camera module 103, and supplementary lights that cooperate with each dual-camera module. In this embodiment, the supplementary light is an infrared supplementary light board 104. The processing unit 101 is preferably a main control chip with high computing power and capable of running artificial intelligence algorithms such as deep learning to meet the needs of complex image processing and real-time recognition. The first dual-camera module 102 and the second dual-camera module 103 each include an RGB camera and an infrared camera, which are connected to the processing unit 101 through I / O interfaces and are responsible for collecting image data. The infrared supplementary light board 104 has the following main functions: first, it actively illuminates the face when there is insufficient light (such as at night or in a dark indoor place) to ensure that the camera can capture a clear image; second, it cooperates with the infrared camera to perform liveness detection, and by analyzing the reflection characteristics of infrared light on the skin of the face, it can effectively distinguish real people from photos, videos, and other forgeries. After successful recognition, the processing unit 101 uploads information to the background system and sends instructions to the gate controller.

[0029] like Figure 2 As shown, the key structural innovation of this invention lies in the layout of the two dual-camera modules. The first dual-camera module 102 is installed on the upper part of the device housing, with its lens optical axis pointing obliquely upward, forming a preset first upward tilt angle α with the horizontal plane. The second dual-camera module 103 is installed on the lower part of the device housing, with its lens optical axis pointing obliquely downward, forming a preset second downward tilt angle β with the horizontal plane.

[0030] In one specific example, the vertical field of view of a conventional horizontally mounted single dual-camera module is 65°. In this embodiment, the first upward tilt angle α and the second downward tilt angle β are both set to 6.37°. Figure 3As shown, the upward-tilted camera better captures the faces of taller passengers, while the downward-tilted camera easily covers children or shorter passengers. Through the fusion processing of the two images by the processing unit, the overall effective vertical field of view is expanded to 77.75°. According to actual tests, when the passenger is 0.9 meters away from the device, the device can easily identify passengers with a height ranging from 1 meter to 2.2 meters.

[0031] In another, more optimized example, if the first upward tilt angle α and the second downward tilt angle β are both increased to 10°, the recognizable height range at a distance of 0.9 meters from the device can be extended to 0.3 meters to 2.4 meters, covering almost all possible passenger heights, greatly improving the universality of the device and the user experience.

[0032] Based on the hardware structure of this embodiment, the internal operation logic of its processing unit 101 will be described in detail. Please refer to [link to relevant documentation]. Figure 4 The figure illustrates the process by which processing unit 101 is configured to perform face recognition operations.

[0033] Step S401: Image Acquisition

[0034] When a passenger enters the recognition area, the device's first dual-camera module 102 and second dual-camera module 103 are activated, simultaneously capturing images of the area to be recognized. The processing unit 101 receives raw image data from the four cameras (two RGB and two infrared).

[0035] Step S402: Image correction and stitching

[0036] Because the four cameras are installed at different positions and angles, their baselines are not parallel. Therefore, the processing unit 101 first corrects the images. Through a pre-calibration process, the intrinsic parameter matrices K1, K2, K3, and K4 of each camera, as well as the extrinsic parameter matrices (rotation matrix R and translation vector T) describing their relative positional relationships, are obtained. Using these calibration parameters, the processing unit 101 projects the images captured by the four cameras onto a common virtual plane and performs image correction, ensuring that the epipolar lines of all images are horizontally aligned. Simultaneously, the processing unit stitches the corrected images from the top and bottom paths together to form a combined image with an ultra-wide vertical field of view. The combined image is directly displayed on the screen, providing the passenger with visual feedback and facilitating observation, allowing the passenger to adjust their posture based on the displayed content.

[0037] Step S403: Face count determination

[0038] The processing unit 101 runs an efficient face detection algorithm (such as YOLO, MTCNN, etc.) on the corrected and stitched combined image to determine the number of faces in the field of view.

[0039] Steps S404 and S405: Single / Multiple Face Branch Processing

[0040] The processing unit executes different logical branches based on the number of faces detected:

[0041] If only a single face is recognized (step S404): This indicates that the scenario is simple, and the processing unit directly takes that face as the target and proceeds to step S406 for recognition.

[0042] If multiple faces are detected (step S405): This indicates that the queue is too close together. At this time, the processing unit is configured to activate the three-dimensional spatial ranging function. It uses a four-eye vision system to perform depth calculations, calculating the precise distance between each detected face and the device.

[0043] Distance measurement principle: Using triangulation, the depth Z of the object is calculated by calculating the difference in pixel position (i.e., parallax d) of the same object in the images of different cameras, combined with the baseline distance B between the cameras: Z = fB / d, where f is the focal length of the camera.

[0044] Variable baseline advantage: The processing unit can flexibly combine cameras to obtain different baselines. For example, the two cameras inside the first dual-camera module (with a baseline distance of about 2cm) can be used to form a short baseline for high-precision ranging of near faces; or the cameras of the first dual-camera module and the cameras of the second dual-camera module (with a baseline distance of up to 20cm) can be used to form an ultra-long baseline for accurate ranging of distant faces.

[0045] Target filtering: After calculating the distance to all visible faces, the processing unit filters out the face with the smallest distance value and identifies it as the only valid target face at the moment.

[0046] Step S406: Target Face Recognition

[0047] The processing unit 101 extracts the features of the unique target face determined in the previous step (whether it is the face in the case of a single face or the most recent face selected in the case of multiple faces) and compares it with the authorized face information pre-stored in the database in a 1:N or 1:1 ratio.

[0048] Step S407: Information Upload and Control

[0049] If the facial recognition is successful, processing unit 101 is configured to perform the final access control operation: it packages the passenger's identity information, passage time, and other data, and uploads it to the subway's central back-end system via a network interface for recording entry or exit information, calculating fares, etc. Simultaneously, processing unit 101 outputs a high / low level signal to the gate controller to open the gate and allow the passenger to pass. If the recognition fails, the process ends, and the gate remains closed.

[0050] Through the above-described implementation methods, this device, with its unique hardware structure and built-in logic in its processing unit, not only solves the problem of recognizing passengers of different heights, but also effectively overcomes recognition interference caused by close-range queuing through a precise multi-face ranging and filtering mechanism. This significantly improves the passage efficiency, accuracy, and user experience of subway turnstiles. The technical solution of this invention is also applicable to other scenarios requiring facial recognition access, such as high-speed rail stations, airports, office buildings, and residential communities.

[0051] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.

Claims

1. A four-eye AI facial recognition access control device specifically for subways, characterized in that: include: Equipment housing; The processing unit is located inside the device housing; The first dual-camera module is disposed on the upper part of the device housing and faces the area to be identified at a preset first upward tilt angle; The second dual-camera module is located at the lower part of the device housing and faces the area to be identified at a preset second downward tilt angle; The processing unit is configured to acquire image data from the first dual-camera module and the second dual-camera module respectively, and perform face recognition processing based on the image data to control passage.

2. The device according to claim 1, characterized in that, The processing unit is specifically configured to perform the following operations: Determine the number of faces contained in the image data; If there are multiple faces, then three-dimensional spatial ranging is performed on the multiple faces to determine the target face closest to the device, and the target face is then identified. If the number of faces is a single face, then the recognition process is performed directly on that single face.

3. The device according to claim 2, characterized in that, The processing unit is configured to perform three-dimensional spatial ranging by combining the four cameras contained in the first dual-camera module and the second dual-camera module in pairs and using different camera baseline distances.

4. The device according to claim 3, characterized in that, The processing unit is configured to: use two cameras in the first dual-camera module or the second dual-camera module to form a short baseline for close-range distance measurement; and use the cameras in the first dual-camera module and the cameras in the second dual-camera module to form an ultra-long baseline for long-range distance measurement.

5. The device according to claim 2, characterized in that, Before determining the number of faces, the processing unit is further configured to: The images acquired by the first dual-camera module and the second dual-camera module are corrected and stitched together to generate a combined image with a larger vertical field of view.

6. The device according to claim 5, characterized in that, Also includes: At least one screen is used to display the combined image generated after correction and stitching by the processing unit, and to display corresponding prompts based on the recognition results.

7. The device according to claim 1, characterized in that, Also includes: At least one infrared fill light is disposed near the first dual-camera module and / or the second dual-camera module to provide fill light for the face in low-light environments and to assist in liveness detection; And / or at least one white light fill light, disposed near the first dual-camera module and / or the second dual-camera module, for providing fill light for the face in low-light environments.

8. The device according to claim 1, characterized in that, Both the preset first upward tilt angle and the preset second downward tilt angle are within the range of 6° to 10°.

9. The device according to claim 8, characterized in that, When both the first upward tilt angle and the second downward tilt angle are 6.37°, the vertical field of view of the device extends to more than 77.75°.

10. The device according to claim 2, characterized in that, After completing the identification process, the processing unit is also configured to upload the information of successfully identified personnel to the background system and output control signals to control the gate's operation.

Images