Linkage calling pass checking method of clearance camera and gate equipment

Abstract

The invention discloses a linkage calling passing inspection method of a customs clearance camera, which is used for controlling calling of a first camera and a plurality of second cameras, the first camera is used for shooting an image of a pedestrian at a first visual angle, and the second cameras are used for shooting an image of the pedestrian at a second visual angle; the method comprises the following steps: a first camera tracks and records a plurality of first view angle images when a pedestrian approaches a gate; forming a pedestrian prediction track in a preset detection area of the first camera according to the first view angle images at the multiple different moments, and calling at least one second camera on the pedestrian prediction track; a second camera obtains a second visual angle image of the pedestrian and recognizes the second visual angle image and preset face information, and if recognition succeeds, a preset gate is controlled to be opened; and if the identification is wrong, closing the gate, and collecting the biological information of the pedestrian for verification. According to the linkage calling passing checking method, the calling efficiency of the camera module is effectively improved, and the passing efficiency of pedestrians is improved.

Description

Technical Field

[0001] This invention relates to the field of security detection technology, and in particular to a method for linking and calling access control cameras for passage inspection and a gate device. Background Technology

[0002] In densely populated and complex environments such as exhibitions and train stations, turnstiles are often used to verify the identities of people entering the venue or station hall to ensure public safety and control the number of people. Most turnstile devices on the market use cameras mounted on pillars. When people arrive at a designated area, their facial data is captured and recognized by the camera. After facial recognition is complete, the turnstile opens to allow passage. With this type of turnstile verification, people have to wait in front of the camera for facial recognition, resulting in a relatively long passage time per person and low efficiency. Furthermore, external environmental interference (such as being pushed by others) or subjective factors affecting pedestrians can cause people to move in a less than straight line, potentially resulting in tilting or other non-standard postures. In such cases, the fixed-position camera may not effectively capture the person's face, leading to facial recognition failure. When verification fails, people will choose to re-verify, further prolonging the time they spend in front of the camera and impacting the turnstile's efficiency. Summary of the Invention

[0003] Therefore, it is necessary to address the aforementioned problem of low efficiency in acquiring and verifying pedestrian facial information by providing a method for linking and calling access cameras for passage verification and a gate device that can improve the efficiency of gate passage.

[0004] A method for linking and activating access control cameras for access verification, used to control the activation of a first camera and multiple second cameras, wherein the first camera is used to capture images from a pedestrian's first-person perspective, and the second cameras are used to capture images from a pedestrian's second-person perspective; the method includes the following steps: The first camera tracks and records multiple first-person perspective images as pedestrians approach the gate; Based on first-view images at multiple different times, a pedestrian prediction trajectory is formed within a preset detection area of ​​the first camera, and at least one second camera on the pedestrian prediction trajectory is invoked. The second camera acquires a second-view image of the pedestrian and identifies it with preset facial information. If the identification is successful, the preset gate is opened; if the identification is incorrect, the gate is closed and the pedestrian's biometric information is collected for verification.

[0005] In one embodiment, the second camera acquires a second-view image of a pedestrian by: performing a preliminary screening of human images that match the model, and after meeting the conditions, acquiring a specific human image for identification and comparison.

[0006] In one embodiment, the first camera predicts the direction, speed and posture of the pedestrian based on the predicted pedestrian trajectory, and calls the second camera to collect a second-view image of the pedestrian based on the prediction result; The second-view image captured by the second camera is substituted into the local model for integrity verification. If the verification is successful, the second-view image is compared with the information on the remote server; otherwise,... If the verification fails, another second camera is called to re-acquire the second-view image of the pedestrian who failed the verification. The re-acquired second-view image is substituted into the local model for integrity verification. If the re-verification is successful, the re-acquired second-view image is compared with the information of the remote server. If the verification still fails, select multiple second cameras to simultaneously collect second-view images of the pedestrians who failed the verification, and then substitute them into the local model for integrity verification. If the second-view image collected by any second camera is successfully verified, compare the information of the successfully verified second-view image with the remote server. If all verifications fail, multi-image recognition is used to combine and stitch together the overlapping positions of the second-view images captured by multiple second cameras, and then the stitched image is substituted into the local model for integrity verification; if the stitched image verification fails, the pedestrian is classified as an abnormal person.

[0007] In one embodiment, the step of combining and stitching the overlapping positions of the second-view images captured by multiple second cameras using multi-image recognition includes: combining the installation positions of the second cameras and the calibrated relative position parameters between each second camera to stitch together multiple second-view images of the same pedestrian within a preset detection area of ​​the first camera.

[0008] In one embodiment, after selecting multiple second-view images of pedestrians whose verification failed, the system first selects one of the second-view images based on the image quality and pedestrian posture, and then improves the image recognition rate; or it first improves the image recognition rate and then selects one of the second-view images based on the image quality and pedestrian posture, and then performs integrity verification with the local model.

[0009] In one embodiment, improving the image recognition rate includes: based on the position of the face in the second-view image, performing distortion correction when the face is at the edge of the second-view image.

[0010] In one embodiment, the first camera continuously acquires first-view images and extracts pedestrian coordinates and pedestrian IDs from the first-view images to form a pedestrian movement trajectory. The acceleration is calculated based on the pedestrian movement trajectory to determine the time and posture of the pedestrian arriving at the preset image acquisition area, and the corresponding second camera is pre-called. When there are multiple pedestrians, multiple second cameras within a preset range are called to continuously acquire second-view images, or a single second camera is used to continuously shoot in the direction of pedestrian movement to pre-acquire images.

[0011] In one embodiment, at least one second camera on the pedestrian prediction trajectory is invoked using one of the following strategies: acquisition area level invocation strategy, proximity invocation strategy, and resolution invocation strategy; or multiple second cameras on the pedestrian prediction trajectory are invoked using a step-by-step invocation strategy. The collection area level calling strategy includes: the second camera with the largest collection area collects human images frame by frame, and the other second cameras collect human images every other frame; or the second camera with the smallest collection area collects human images frame by frame, and the other second cameras collect human images every other frame. The proximity-based call strategy includes: prioritizing the use of the second camera closest to the pedestrian to capture their image; The resolution retrieval strategy includes: when a pedestrian is outside the preset area of ​​the second camera, the second camera captures the image of the pedestrian at a first resolution; when a pedestrian enters the preset area of ​​the second camera, the second camera captures the image of the pedestrian at a second resolution, which is lower than the first resolution. The step-by-step invocation strategy includes: setting up multiple second cameras along the pedestrian's movement direction, and invoking each of the second cameras step-by-step along the pedestrian's predicted trajectory.

[0012] This invention also discloses a method for linking and calling access control cameras for passage verification, applied to a turnstile device for rapid passage. The turnstile device includes a moving channel, a buffer channel connected to the side of the moving channel near its exit, a straight-through gate at the exit of the moving channel, self-service verification gates respectively located at the entrance and exit of the buffer channel, a first camera, and multiple second cameras invoked within the moving channel. The straight-through gate is initially in an open state, and the self-service verification gate at the entrance of the buffer channel is initially in a closed state. The method for linking and calling access control cameras for passage verification includes the following steps: S1. The first camera detects whether there are pedestrians in the moving channel. When there are pedestrians, it tracks and records multiple first-view images of the pedestrians as they approach the straight-through gate. The pedestrian coordinates and pedestrian ID are extracted from the first-view images, and then the process proceeds to step S2. S2. Based on multiple first-view images at different times, a pedestrian prediction trajectory is formed within the preset detection area of ​​the first camera. At least one second camera on the pedestrian prediction trajectory is called. The second camera queries whether the pedestrian's ID is in the passage whitelist. If it is, the pedestrian passes directly through the straight-through gate; otherwise, proceed to step S3. S3. The second camera acquires a second-view image of the pedestrian and recognizes it with preset facial information. If the recognition is successful, the pedestrian's ID is updated to the access whitelist so that the pedestrian can pass directly through the straight gate. If the recognition is incorrect, the straight gate is closed and the self-service verification gate is opened so that the pedestrian who fails the recognition can enter the buffer channel. The buffer channel is used for the secondary verification of the pedestrian.

[0013] The present invention also discloses a turnstile device for implementing the above-mentioned method for linking and calling access control cameras for passage verification, the turnstile device comprising: The gate mechanism includes a moving channel, a buffer channel connected to the side of the moving channel near the exit end, a straight gate located in the moving channel and adjacent to the entrance of the buffer channel, and self-service verification gates respectively set at the entrance and exit of the buffer channel. The self-service verification gate located at the exit of the buffer channel is equipped with a biometric device and / or a document recognition device. The first camera is used to detect whether there are pedestrians in the moving channel. When there are pedestrians, it tracks and records the first-view image of the pedestrians as they approach the straight gate. The pedestrian coordinates and the distance from the pedestrians to the first camera are extracted from the first-view image, and each pedestrian is assigned an identification ID. Multiple second cameras are used to obtain pedestrian identification IDs and pedestrian coordinates from the first camera, and to acquire second-view images of pedestrians according to the calling strategy, pedestrian identification IDs and pedestrian coordinates, calculate pedestrian head poses and crop the images; The remote server stores facial information and a whitelist of access points. It is used to compare and identify the identifier ID obtained by the second camera with the whitelist of access points, or to compare and identify the second-view image obtained by the second camera with the facial information, and to update the whitelist of access points after successful identification. The control unit is used to keep the straight-through gate open and the self-service verification gate closed when the pedestrian at the front of the straight-through gate is on the whitelist and there is no tailgating, or when the pedestrian at the front of the straight-through gate is on the whitelist, there is tailgating and all the tailgating people are on the whitelist; otherwise, the control unit closes the straight-through gate and opens the self-service verification gate.

[0014] The present invention implements a method and gate device for the linkage and access control of cameras. It uses a first camera to capture multiple first-view images at different times to form a predicted pedestrian trajectory, and then calls a second camera on that predicted trajectory to capture a second-view image of the pedestrian. Before the pedestrian reaches the gate, the second-view image is captured for identification, eliminating the need for the pedestrian to stop at the gate, saving time waiting for facial verification, shortening the time for a single person to pass through the gate, and improving passage efficiency. Furthermore, by acquiring the predicted pedestrian trajectory and calling the second camera on that trajectory to capture the second-view image, the time of the pedestrian's arrival at the designated location is predicted. This allows the use of the second camera to adapt to the predicted pedestrian trajectory, i.e., selecting and calling the second camera corresponding to the pedestrian based on the predicted trajectory. This improves resource utilization efficiency, accuracy of information collection, and makes it easier to capture second-view images that can identify pedestrian information. It achieves efficient resource scheduling and utilization, improves pedestrian identification efficiency and pass rate, shortens the time pedestrians spend at the gate, and further improves passage efficiency. Attached Figure Description

[0015] Figure 1 This is a flowchart of the access control camera linkage and access verification method in Embodiment 1 of the present invention; Figure 2 This is a simplified scenario diagram illustrating the linkage between a gate camera and a passage verification method in one embodiment of the present invention. Figure 3 This is a schematic diagram of a camera arrangement method in one embodiment of the present invention; Figure 4 This is a schematic diagram of a second camera arrangement method in one embodiment of the present invention; Figure 5 This is a schematic diagram of a scenario in which a passage camera is linked to trigger a passage verification method according to one embodiment of the present invention. Figure 1 ; Figure 6 This is a schematic diagram of a scenario in which a passage camera is linked to trigger a passage verification method according to one embodiment of the present invention. Figure 2 ; Figure 7 This is a schematic diagram of a scenario in which a passage camera is linked to trigger a passage verification method according to one embodiment of the present invention. Figure 3 ; Figure 8 This is a flowchart of the method for linking and calling the access control camera in Embodiment 2 of the present invention; Figure 9 This is a structural block diagram of a gate device in one embodiment of the present invention; Figure 10This is a simplified hardware connection diagram of the gate device in one embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0017] Example 1 Please combine Figures 1 to 3 , Figures 5 to 7 as well as Figure 9 This invention discloses a method for linking and verifying access control cameras. This method controls the use of a first camera 100 and multiple second cameras 200. Both the first camera 100 and the second cameras 200 are positioned on a pedestrian movement path to capture images of pedestrians within the path. This movement path can be a physical path defined by barriers or other facilities, or a virtual path defined by setting the image capture range of the first camera 100 and the second camera 200. The movement path is equipped with gates. The first camera 100 captures images from a pedestrian's first-person perspective, and the second cameras 200 capture images from a pedestrian's second-person perspective. In this embodiment, the first camera 100 is suspended above the movement path, capturing images of pedestrians from above their heads (first-person perspective) to detect their movement trajectory. The second cameras 200 capture facial images of pedestrians from the second-person perspective, providing a basis for facial feature recognition.

[0018] The method for linking and calling access control cameras in this embodiment includes the following steps: S01. The first camera 100 tracks and records multiple first-view images as a pedestrian approaches the gate. These multiple first-view images are images captured by the first camera 100 at different times during the pedestrian's approach to the gate. By capturing these first-view images at different times, positional information can be extracted from the images. Based on the relationship between positional information and time, a movement trajectory of the pedestrian is formed. This trajectory represents the path formed by the positions the pedestrian has already passed, and it provides a basis for predicting the pedestrian's next movement trajectory.

[0019] S02. Based on multiple first-view images at different times, a pedestrian prediction trajectory is formed within a preset detection area of ​​the first camera 100, and at least one second camera 200 on the pedestrian prediction trajectory is invoked. In other words, in this embodiment, the pedestrian's movement trajectory is generated based on the position information extracted from the first-view images at different times during the pedestrian's approach to the gate. Based on the current movement trajectory and the pedestrian's movement parameters (e.g., calculating the pedestrian's speed, acceleration, direction, and trend based on their different positions at different times), the location the pedestrian will reach at a certain time or at certain future times is predicted, thus forming a predicted pedestrian trajectory. At least one second camera 200 on the predicted pedestrian trajectory is then invoked to achieve advance and accurate invocation of the second camera 200. For example, if it is predicted that the pedestrian will move to the left side of the passage at the next moment, the second camera 200 on the left side of the passage can be invoked in advance to collect the pedestrian's second-view image. This avoids the resource waste caused by all second cameras 200 simultaneously collecting pedestrian second-view images under a common camera acquisition strategy, and the problem that the second camera 200 cannot accurately collect pedestrian second-view images when the same second camera 200 is continuously used due to changes in the pedestrian's trajectory. In this way, by determining the second camera 200 that needs to be called based on the pedestrian prediction trajectory, and calling the second camera 200 on the pedestrian prediction trajectory separately, the second camera 200 can be activated in advance, which can optimize the recognition efficiency and allow other second cameras 200 that are not on the pedestrian prediction trajectory to be in standby mode, thereby saving resources.

[0020] S03. The second camera 200 acquires a second-view image of the pedestrian and recognizes it against preset facial information. If the recognition is successful, the preset gate is opened. If the gate device used in this linkage access verification method is initialized to a normally open state by default, then it is not necessary to control the gate opening again; it is only necessary to confirm that the gate is open. If the recognition fails, the gate is closed, and the pedestrian's biometric information is collected for verification. In this embodiment, by recognizing the second-view image acquired by the second camera 200 against the preset facial information, it can be determined whether the pedestrian is a person allowed to enter the area where the gate is located, thus preventing unauthorized "illegal" personnel (including those not registered in the system and those on the blacklist) from entering the relevant area. After a recognition failure, in addition to closing the gate and preventing forced passage, a second verification is performed by collecting the pedestrian's biometric information, which can reduce the error of misjudging the pedestrian as a person who does not meet the access conditions and improve the reliability of the recognition.

[0021] In step S03, the second camera 200 acquires the second-view image of the pedestrian, including: performing preliminary screening on images that conform to the model; and after meeting the conditions, acquiring specific images for recognition and comparison. In other words, this embodiment combines coarse and fine screening, first filtering out images that do not conform to the model or differ significantly from the actual object to be identified (pedestrian), and then comparing the filtered images. This reduces resource consumption, achieves secondary screening of the second-view images, thereby reducing the data processing volume for subsequent image recognition and comparison, and improving recognition efficiency.

[0022] Furthermore, in this embodiment, the first camera 100 predicts the pedestrian's direction, speed, and posture based on the predicted pedestrian trajectory. Based on the prediction result, it calls the second camera 200 to capture the pedestrian's second-view image. That is, it calls the second camera 200 on the predicted pedestrian trajectory to capture the pedestrian's second-view image, thereby ensuring image acquisition and recognition effects while reducing the occupancy of second cameras 200 not on the predicted pedestrian trajectory. After the second camera 200 captures the second-view image, it substitutes the captured second-view image into the local model for integrity verification. Integrity verification includes verifying the pedestrian contour in the second-view image. If the pedestrian contour does not conform to the human contour in the local model, the integrity of the second-view image is insufficient, and the verification fails. For example, if a non-human contour is captured (such as a pedestrian placing a pet on their shoulder, resulting in a captured second-view image of the pet), or if the pedestrian is at the edge of the second camera 200's image acquisition range, causing the captured human contour to be incomplete, or if some facial features of the pedestrian are not within the second-view image, the second-view image verification is deemed to have failed. If the verification is successful, the second-view image is compared with the information on the remote server 500. The remote server 500 has pre-stored the facial information of people authorized to enter the relevant place. By comparing the information on the second-view image with the information on the remote server 500, the identity of the pedestrian can be verified to determine whether the pedestrian is a person authorized to enter the place.

[0023] Otherwise, if the verification fails, another second camera 200 is invoked to re-capture the second-view image of the pedestrian who failed the verification, thereby supplementing the second-view image of the pedestrian. The re-acquired second-view image is then substituted into the local model for integrity verification. If the re-verification is successful, the re-acquired second-view image is compared with the information from the remote server 500. The other second camera 200 referred to here is the second camera 200 located at the same position as the second camera 200 whose second-view image integrity verification failed (i.e., the same position along the length of the movement channel on the pedestrian prediction trajectory, but different positions along the width of the movement channel), or it can be the second camera 200 located at the next position on the pedestrian prediction trajectory after the second camera 200 whose integrity verification failed.

[0024] If the verification still fails, select multiple second cameras 200 to simultaneously collect second-view images of the pedestrians who failed the verification, and then substitute them into the local model for integrity verification. If the second-view image collected by any second camera 200 is successfully verified, compare the information of the successfully verified second-view image with the remote server 500.

[0025] If all verifications fail, multi-image recognition is used to combine and stitch together the overlapping positions of multiple second-view images captured by the second cameras 200. The stitched image is then substituted into the local model for integrity verification. If the stitched image verification fails, the pedestrian is classified as an abnormal person. That is, if the verification of a single second-view image fails, multiple second-view images are stitched together at overlapping positions to form a single stitched image, which is then used for integrity verification. This allows for the pre-identification of pedestrians before they reach the gate, thereby reducing the time pedestrians spend at the gate.

[0026] In this embodiment, the integrity of the acquired second-view images is first verified. Based on the pedestrian's form in the second-view images and a local model, the images captured by the second camera 200 are initially identified as pedestrian images, achieving preliminary detection of the pedestrian's second-view images. At this stage, it is unnecessary to perform remote server 500 recognition verification on all second-view images; instead, verification is performed only if the integrity verification is successful, reducing the amount of data processing. If the integrity verification fails, other second cameras 200 are used to acquire second-view images of the pedestrian, enabling dynamic compensation shooting of the pedestrian and improving the accuracy of second-view image acquisition and recognition.

[0027] It should be noted that in this solution, the local model is a model set according to the actual working conditions. The local model can be a collection of multiple human body outlines, a collection of multiple human upper body outlines, or a collection of multiple human head outlines. That is, the local model is a human body outline model, a human upper body model, or a head outline model. The outline information requires head and face images.

[0028] In this embodiment, multi-image recognition is used to combine and stitch together the overlapping positions of the second-view images captured by multiple second cameras 200. Specifically, the overlapping positions of the second-view images captured by multiple second cameras 200 are identified, and multiple different second-view images are combined and stitched together. This combination and stitching includes: Based on the installation position of the second camera 200 and the calibrated relative position parameters between each second camera 200, multiple second-view images of the same pedestrian within the preset detection area of ​​the first camera 100 are stitched together. In this embodiment, when multiple second-view images of the same pedestrian are simultaneously acquired by multiple second cameras 200, multiple second cameras 200 installed in close proximity should be used to capture the same pedestrian. This avoids the problem of large angular differences in the acquired second-view images due to significant differences in the installation positions of the multiple second cameras 200, which would lead to difficulties in image stitching. When stitching the second-view images, it is advisable to select images that are within the detection area of ​​the first camera 100 (i.e., located on the pedestrian's predicted trajectory) and simultaneously within the acquisition range of multiple second cameras 200 for stitching. This avoids invalid captures caused by a single second-view image having a shooting position that differs too much from other second-view images, making it difficult to stitch with other second-view images, thus reducing resource waste.

[0029] In one embodiment, after simultaneously acquiring second-view images of pedestrians whose verification failed, multiple second-camera 200 are selected. First, one of these second-view images is selected based on image quality and pedestrian posture to improve the image recognition rate; alternatively, the image recognition rate is improved first, and then one second-view image is selected based on image quality and pedestrian posture. The processed or selected second-view image is then compared with the local model for integrity verification. In this embodiment, image quality refers to indicators such as sharpness, brightness, sharpness, and signal-to-noise ratio of the second-view image. Pedestrian posture refers to the angle values ​​of the acquired second-view image of the pedestrian's face in the three axes of roll, rotation, and yaw. Brightness is a pixel grayscale value indicator, reflecting the overall brightness of the image; sharpness refers to the sharpness of edges and details in the image, reflecting the degree of detail features; signal-to-noise ratio reflects the intensity ratio of signal to noise, and is an objective indicator of image purity; sharpness is an overall evaluation of the image based on brightness, sharpness, and signal-to-noise ratio.

[0030] In other words, when performing integrity verification between multiple second-view images and the local model, one approach is to first select second-view images with a certain range of sharpness, brightness, sharpness, signal-to-noise ratio, and angle values ​​in the three axes of roll, rotation, and yaw based on the image quality and pedestrian posture of each second-view image (i.e., selecting the image with the best quality and posture). Under the condition of this second-view image, the image recognition rate of the second-view image is improved, and then the image with the improved image recognition rate is verified against the local model for integrity. Alternatively, one approach is to first improve the image recognition rate of each second-view image, and then select the second-view image with the best quality and posture from the improved images, and then verify the integrity of the selected second-view image against the local model. Preferably, in this embodiment, the second-view image is selected by first selecting one second-view image based on image quality and pedestrian posture, and then improving the image recognition rate, in order to reduce the amount of image processing.

[0031] Furthermore, improving the image recognition rate includes: based on the position of the face in the second-view image, distortion correction is performed when the face is located at the edge of the second-view image. Specifically, for a typical second camera 200 (camera lens), there is a characteristic of greater distortion around the lens periphery. Therefore, when a pedestrian's face is detected at the edge of the second-view image or around the second camera 200, the second-view image can be corrected. Distortion correction can adopt a common camera calibration method in the industry, that is, by photographing a calibration template of known size (such as a checkerboard or dot array), the radial and tangential distortion coefficients of the lens are calculated using a calibration algorithm, and then substituted into the distortion correction formula to map the pixel coordinates of the distorted image to ideal, distortion-free coordinates, thereby achieving distortion correction of the second-view image.

[0032] In one embodiment, the first camera 100 continuously acquires first-view images and extracts pedestrian coordinates and pedestrian identification IDs from the first-view images to form a pedestrian movement trajectory. The identification ID is an identity information or number generated by the first camera 100 based on the pedestrian coordinates extracted from the first-view image of the same pedestrian, associated with the pedestrian coordinates and the time of coordinate acquisition. The pedestrian coordinates include the pedestrian's position information in the width direction and the position information in the length direction of the movement channel. Acceleration is calculated based on the pedestrian movement trajectory to determine the time and posture of the pedestrian arriving at the preset image acquisition area. The corresponding second camera 200 is pre-activated. Pre-activating the corresponding second camera 200 includes: continuously taking pictures with one second camera 200 at a time along the pedestrian's predicted trajectory (i.e., the direction of travel) to pre-activate the corresponding second camera 200, achieving accurate acquisition of second-view images of the pedestrian and reducing the number of second cameras 200 required; or using multiple second cameras 200 within a certain range to continuously take pictures, so as to select the best image from multiple second-view images. When multiple pedestrians are present, multiple second cameras 200 within a preset range are used to continuously capture second-view images, or a single second camera 200 is used to continuously capture images in the direction of pedestrian movement to pre-acquire images. Preferably, when multiple pedestrians are present, multiple second cameras 200 within a preset range are used to continuously capture second-view images to reduce the difficulty in capturing some pedestrian face images caused by obstruction between pedestrians or shooting from a single perspective.

[0033] In one embodiment, at least one second camera 200 on the pedestrian prediction trajectory is invoked using one of the following strategies: acquisition area-level invocation strategy, proximity invocation strategy, or resolution invocation strategy; or multiple second cameras 200 on the pedestrian prediction trajectory are invoked using a hierarchical invocation strategy. In other words, in this embodiment, the second cameras 200 are invoked using one of the following strategies: acquisition area-level invocation strategy, proximity invocation strategy, resolution invocation strategy, or hierarchical invocation strategy. Specifically, for short movement channels, one or a combination of the acquisition area-level invocation strategy, proximity invocation strategy, and resolution invocation strategy can be used; for long movement channels, the hierarchical invocation strategy combined with the acquisition area-level invocation strategy, proximity invocation strategy, and resolution invocation strategy can be used, or a combination thereof.

[0034] The acquisition area-based allocation strategy includes: first, the second camera 200 with the largest acquisition area acquires images frame-by-frame, while the other second cameras 200 acquire images every other frame; or second, the second camera 200 with the smallest acquisition area acquires images frame-by-frame, while the other second cameras 200 acquire images every other frame. Since the acquisition areas of multiple second cameras differ, if one second camera 200 is centered, it acquires a larger area and has a higher probability of capturing a suitable image, while another second camera 200 is positioned towards the edge of the moving channel, acquiring a smaller area and having a lower probability of capturing a suitable image. By using the acquisition area-based allocation strategy, fewer images are acquired by the second cameras 200 that are difficult to capture, allowing more system resources to be allocated to the second cameras 200 with larger acquisition areas, thus optimizing the acquisition of the best camera. Alternatively, the second camera 200 with a larger acquisition area can acquire pedestrian images every other frame, while the second camera 200 with a smaller acquisition area can acquire pedestrian images frame by frame. In this way, the second camera 200, which has a difficult acquisition area, can acquire more images, thus allocating more system resources to the second camera 200 with a smaller acquisition area. This provides more resources to the "weaker" camera, enabling it to acquire more pedestrian images for subsequent image processing.

[0035] The proximity-based mobilization strategy includes prioritizing the use of the second camera 200 closest to the pedestrian to capture their image. For example, when it is predicted that a pedestrian will move from the right side of the passageway towards the gate, the second camera 200 located on the right side of the passageway can be activated to more accurately capture the pedestrian's second-view image, thus achieving proximity-based mobilization of the second camera 200.

[0036] The resolution allocation strategy includes: when a pedestrian is outside the preset area of ​​the second camera 200, the second camera 200 captures the image at a first resolution; when a pedestrian enters the preset area of ​​the second camera 200, the second camera 200 captures the image at a second resolution, which is lower than the first resolution. That is, based on the predicted pedestrian movement, after a pedestrian approaches the second camera 200 and enters the preset area of ​​its capture, the resolution can be reduced to speed up the image acquisition speed of the second camera 200, thus capturing more faces. Since the second camera 200 captures images in 2D, the image size varies depending on distance. When a pedestrian is within a certain distance of the second camera 200, although the image resolution decreases, the proportion of the image occupied by the face is still large, and the effective pixel change of the face is not significant. Reducing the resolution can reduce the complexity of subsequent processing.

[0037] The hierarchical invocation strategy includes: setting up multiple second cameras 200 along the pedestrian's movement direction, and hierarchically invoking each second camera 200 along the predicted pedestrian trajectory (e.g., ...). Figure 4 (As shown). This can be understood as follows: as the pedestrian moves along the predicted trajectory, images of the pedestrian from a second perspective are acquired by sequentially activating each of the second cameras 200, thereby reducing the consumption of equipment resources. Under the step-by-step activation strategy, each second camera 200 uses different pedestrian features for step-by-step identification and confirmation, or along the pedestrian's movement trajectory, if the previous second camera 200 fails to identify the pedestrian's features, the next second camera 200 identifies and confirms the same pedestrian feature. For example, in this embodiment, three second cameras 200 are set up along the pedestrian's movement trajectory. Each of the three second cameras 200 collects features 1, 2, and 3 of the same pedestrian ID. During the step-by-step identification process, feature 1 of the pedestrian collected by the first second camera 200 is identified first, then feature 2 of the pedestrian collected by the second second camera 200 is identified, and finally feature 3 of the pedestrian collected by the third second camera 200 is identified. When one or more of the different features collected by different second cameras 200 meet preset requirements (such as only meeting eye features, or simultaneously meeting eye, mouth, and nose features), the pedestrian is considered successfully identified. Alternatively, if the first second camera 200 fails to identify feature 1 of the pedestrian, the second second camera 200 can be called to continue identifying feature 1 of the pedestrian, thereby reducing the error of misjudging the pedestrian as someone who does not meet the crossing conditions.

[0038] The above strategies can be used in sequence or one of them can be chosen. Among them, the step-by-step calling strategy is only applicable to long moving channels and situations with multiple groups (note that not multiple, a group can have multiple 200) of cameras. Here, multiple groups of cameras means that each group of cameras contains a large number of cameras.

[0039] It should be noted that in this embodiment, when a second-view image and facial information recognition error occurs and biometric verification is performed, multiple interchangeable methods can be used for biometric verification. Specifically, the biometric verification of pedestrians is conducted within a self-service channel located next to the exit of the mobile channel. When the second camera acquires a second-view image of the pedestrian and recognizes it against preset facial information, if the recognition is successful, the gate at the exit of the mobile channel is opened; if the recognition is incorrect, the gate at the exit of the mobile channel is closed, and the entrance gate of the self-service channel at the exit of the mobile channel is opened, guiding the pedestrian to the self-service channel, and the pedestrian's biometric information is collected for verification. The collection of pedestrian biometric information for verification includes verification using either manual assistance or image fusion verification.

[0040] Among them, manual assistance confirmation includes providing pedestrian images for manual confirmation. For example, one image is selected from multiple second-view images captured by multiple second cameras 200 and sent to the backend for manual confirmation by staff.

[0041] Image fusion verification includes: capturing images of a person from different angles using multiple cameras within a biometric device deployed in the self-service channel; and verifying the pedestrian's image when the images from different angles meet preset rules. The preset rules include: verification is successful and a preset gate is opened when all images from different angles pass verification, or when any one image from any angle passes verification, or when at least N images from different angles pass verification, where N is an integer greater than or equal to 1. For example, when images of the left and right sides of the same pedestrian are captured simultaneously, if the left image successfully verifies the image information of the preset face, or the right image successfully verifies the right side image information of the preset face, or both left and right images successfully verify the left and right side images of the preset face, then even if a complete frontal image of the pedestrian is not obtained, the pedestrian is still determined to be a person eligible to enter the relevant area.

[0042] Example 2 Please combine Figures 2 to 9 The present invention also discloses a method for linking and calling access control cameras for access verification. This method is applied to a gate device for fast passage. The gate device is provided with a moving channel 300, a buffer channel 400 connected to the side of the moving channel 300 near the exit end, a straight gate 310 located at the exit end of the moving channel 300, and self-service verification gates 410 respectively set at the entrance and exit of the buffer channel 400 (i.e., the self-service verification gate 410 set at the entrance of the buffer channel 400 (1) and the self-service verification gate 410 set at the exit of the buffer channel 400). The system includes a self-service verification gate 410(2), a first camera 100, and multiple second cameras 200 invoked within the mobile channel 300. The self-service verification gate 410(2) at the exit is typically equipped with a secondary verification dedicated channel verification terminal 420. This secondary verification dedicated channel verification terminal 420 includes a biometric device 421 and / or a document recognition device 422, used for pedestrian self-service verification. If the self-service verification information matches the database successfully, the self-service verification gate 410(2) at the exit opens to allow passage. Only when self-service verification fails does the administrator need to be notified for intervention. The straight-through gate 310 is initially in an open state, while the self-service verification gate 410 at the entrance of the buffer channel 400 is initially in a closed state. This allows passage without waiting for the straight-through gate 310 to open, thus reducing the time pedestrians spend at the straight-through gate 310, provided that pedestrian ID lookup or facial recognition is successful.

[0043] The method for linking and activating access control cameras for passage verification includes the following steps: S1. The first camera 100 detects whether there are pedestrians in the moving channel 300. When there are pedestrians, it tracks and records multiple first-view images of the pedestrian approaching the straight-through gate 310. The pedestrian coordinates and the pedestrian's identification ID are extracted from the first-view images, and the process proceeds to step S2. The identification ID is the identity information or number generated by the first camera 100 based on the pedestrian coordinates extracted from the first first-view image of the same pedestrian. The pedestrian coordinates include the pedestrian's position information in the width direction and the position information in the length direction of the moving channel 300.

[0044] S2. Based on multiple first-view images at different times, a pedestrian prediction trajectory is formed within a preset detection area of ​​the first camera 100. At least one second camera 200 on the pedestrian prediction trajectory is invoked. The second camera 200 queries whether the pedestrian's identification ID is in the passage whitelist. If it is, the pedestrian passes directly through the straight-through gate 310; otherwise, proceed to step S3. In this embodiment, the process of invoking at least one second camera 200 on the pedestrian prediction trajectory is exactly the same as in Embodiment 1. For details, please refer to the relevant description in Embodiment 1, which will not be repeated here.

[0045] S3. The second camera 200 acquires a second-view image of the pedestrian and recognizes it against preset facial information. If the recognition is successful, the pedestrian's ID is updated to the access whitelist, allowing the pedestrian to pass directly through the straight-through gate 310. If the recognition fails, the straight-through gate 310 is closed, and the self-service verification gate 410 is opened, allowing the unsuccessful pedestrian to enter the buffer channel 400, which is used for secondary verification of the pedestrian. In this embodiment, after recognizing the second-view image, the straight-through gate 310 is closed and the self-service verification gate 410 is opened in two situations: First, the remote server 500 storing pre-stored facial information does not contain content matching the second-view image, or the pedestrian's second-view image cannot be acquired or recognized due to obstruction. In this case, the pedestrian is determined to be a person who does not meet the access requirements. Second, the acquired second-view image contains illegal information, and the pedestrian is a blacklisted person, a person with poor credit, or other person not allowed to enter the gate equipment installation site. In other words, the pedestrian is determined to be a person who does not meet the access requirements.

[0046] In this embodiment, images of pedestrians within the mobile channel 300 are captured when the first camera 100 detects them, thus avoiding resource waste caused by the continuous operation of the first camera 100 when there are no pedestrians in the mobile channel 300. Since the straight-through gate 310 is located at the exit end of the mobile channel 300, the whitelist lookup or second-view image recognition of pedestrians within the mobile channel 300 occurs before the pedestrians reach the straight-through gate 310, thereby shortening the time pedestrians spend at the gate. Because the time required for the whitelist lookup is shorter than the time required for image recognition, after entering the mobile channel 300, the whitelist is checked first; if the check fails, second-view image recognition is then performed, further shortening the pedestrian recognition time. When the turnstile device is initialized, the system's whitelist does not store any identifiers. When a pedestrian first enters the moving channel 300 and is captured by the first camera 100, there is no need to query the whitelist; instead, the second-view image is directly captured and recognized. After successful recognition, the pedestrian's identifier is added to the whitelist. Once the pedestrian's identifier is added to the whitelist, facial recognition is no longer performed as the pedestrian moves forward in the moving channel 300 until the pedestrian approaches the straight-through gate 310, at which point a whitelist query is performed. Since the identifier is already stored in the whitelist, passage is possible after a whitelist query, achieving rapid query and passage. In this embodiment, when multiple pedestrians enter the moving channel 300 simultaneously, the process proceeds from front to back, starting with the pedestrian closest to the straight-through gate 310, querying each pedestrian's identifier one by one. If the identifier is not found, the second-view image recognition method is used to add or update the pedestrian's identifier to the whitelist. In this way, pedestrians whose second-view images can be accurately acquired and identified have their identification IDs added to the passage whitelist before reaching the straight-through gate 310. When approaching the straight-through gate 310, only the passage whitelist needs to be checked to speed up the passage. If, during the process of approaching the straight-through gate 310, the second-view image of a pedestrian is not accurately acquired due to reasons such as turning their head, clothing, hats, or people in front obstructing their view, and their identification ID is not added to the passage whitelist, when the pedestrian moves to the front of the queue, the obstruction factor is removed. After the passage whitelist query fails, passage can be achieved by acquiring and identifying the second-view image. If identification still fails, the straight-through gate 310 is closed and the self-service verification gate 410 is opened to allow pedestrians to enter the buffer channel 400 for self-service verification to avoid queue congestion.

[0047] In this embodiment, a first camera 100 detects pedestrian trajectories and extracts pedestrian coordinates and identification IDs from the acquired first-view images. The first camera 100 can be a 3D sensor or a 3D camera equipped with imaging capabilities and related MCU control circuitry, used to acquire pedestrian information (including pedestrian coordinates and identification IDs) from above the moving channel 300. The second camera 200 is a 2D camera module used to acquire second-view images of pedestrians from a horizontal direction. Since the pedestrian coordinates acquired by the first camera 100 are three-dimensional coordinates, while the second camera 200 uses two-dimensional coordinates for pedestrian image acquisition and positioning, the internal parameters of the second camera 200 need to be calibrated before acquiring the second-view image of the pedestrian based on the pedestrian coordinates. The second camera 200 calibrates the relative position parameters between the first camera 100 and the second camera 200 according to its installation position. When receiving pedestrian coordinates, the second camera 200 converts the pedestrian coordinates to its own coordinate system based on its internal parameters and relative position parameters (i.e., converts the three-dimensional coordinates to its own coordinate system). Two-dimensional coordinates are used to provide the coordinates for collecting pedestrian facial information (i.e., the coordinates in the width direction of the moving channel 300), thereby binding the facial information collection coordinates with the identifier ID. This allows the identifier ID acquired by the first camera 100 and the pedestrian coordinates associated with that identifier ID to be accurately acquired by the second camera 200. In this way, after the second camera 200 acquires and recognizes the second-view image using the converted facial information collection coordinates, the second-view image can correspond to the identifier ID associated with the pedestrian coordinates before the conversion, thus ensuring the accuracy and reliability of the second-view image recognition and the update of the access whitelist.

[0048] The process of querying the access whitelist and performing second-view image recognition also includes determining whether pedestrians are tailgating. Specifically, determining whether pedestrians are tailing involves obtaining the coordinates of all pedestrians and calculating the distances from each other pedestrian to the person in front to determine if there are any people tailgating the person in front. If the distances from all other pedestrians to the person in front are greater than a set threshold, meaning there are no people tailgating the person in front, then only the person in front needs to be queried or identified individually. When the ID is not on the access whitelist or the second-view image recognition fails, the straight-through gate 310 is closed and the self-service verification gate 410 is opened so that pedestrians can proceed to the buffer channel 400 for self-service verification. When a pedestrian whose ID is not on the access whitelist or whose second-view image recognition fails enters the buffer channel 400, the self-service verification gate 410 at the entrance is closed to prevent other people from accidentally entering. If the distances from other pedestrians to the person in front are less than or equal to a set threshold, meaning there are people tailgating the person in front, then the people tailgating can also be queried or identified to improve passage efficiency and speed. In this scenario, the IDs of pedestrians whose distance is less than or equal to a set threshold (i.e., trailing pedestrians) are recorded. The system then checks if each pedestrian's ID is in the whitelist. In this case, only the whitelist query is needed, without requiring second-view image recognition, thus speeding up passage. If the IDs of the foremost pedestrian and the other recorded pedestrians are all in the whitelist, the straight-through gate 310 remains open while the self-service verification gate 410 remains closed, allowing the foremost pedestrian and their trailing pedestrians to pass directly through the straight-through gate 310 for rapid passage. Otherwise, pedestrians whose distance from the foremost pedestrian is less than or equal to a set threshold are defined as abnormal individuals, and their IDs are recorded. The straight-through gate 310 is closed while the self-service verification gate 410 is opened, allowing pedestrians to proceed in batches to the buffer channel 400 for verification, thus eliminating abnormal situations. Additionally, if the distance between the first and second person is within a threshold range (indicating tailgating), and either the first or second person is considered an abnormal individual, then both the first and second person are treated as the first person. The system then determines whether the distance between the second person and the person behind them is within the threshold range, i.e., whether multiple people are tailgating. Abnormal situations here include: multiple second cameras (200) failing to detect faces after coordinate conversion; not detecting faces matching the pose (e.g., head turning, clothing / hat obscuring the face); pedestrians walking backwards; faces being obscured; people following too closely; the first camera (100) failing to track pedestrians; pedestrians' second facial information not being uploaded; and overly exaggerated makeup.

[0049] In the above embodiments, the determination of whether a person is tailgating can occur before or after the access whitelist query, or it can be performed after the access whitelist query is successful. If the access whitelist query fails, the determination of whether a person is tailgating is performed after a second-view image is captured and successfully identified. The determination of tailgating is based on the distance of the person along the length of the moving channel 300 in the first-view image captured by the first camera 100. If this distance is less than or equal to a set threshold, tailgating exists; otherwise, tailgating does not exist.

[0050] When the second camera 200 acquires a second-view image, if multiple pedestrian images are present in the second-view image, further processing of the image is included to obtain a second-view image that can be recognized by facial information in the remote server 500. Specifically, the second camera 200 acquires images of people within the detection area of ​​the first camera 100 and performs image region overlap detection. If the overlap rate exceeds a threshold, and the distance from the current image information acquisition coordinates to the second camera 200 is a non-minimum value of the distance from the image information acquisition coordinates of all identified IDs to the second camera 200 that exceeds the overlap rate threshold, that is, the acquired image is a pedestrian who is not at the very front among several people with a high overlap rate and is obscured by people in front of them. In this case, the pedestrian's ID is recorded as the obscured area, and the ID of the next pedestrian is judged (the next pedestrian is first checked against the whitelist, and then the second-view image is acquired before image region overlap detection is performed); otherwise, it is recorded as an unobscured area and head pose is evaluated. The image information acquisition coordinates here are the coordinates after the pedestrian coordinates are transformed to the coordinate system of the second camera 200 itself. When the second camera 200 performs head pose assessment on a pedestrian's image, if the pose assessment result exceeds a threshold, the pedestrian's identifier ID is recorded as a non-frontal face pose. The identifier ID of the next pedestrian is then assessed (a whitelist query is performed on the next pedestrian, followed by second-view image acquisition and image region overlap detection). Otherwise, it is recorded as a frontal face pose and the image is cropped. When the second camera 200 crops the face image, it obtains the image information and records the pedestrian's identifier ID and pose assessment result into a fusion list (i.e., the fusion list includes image information, identifier ID, and pose assessment result). The identifier ID of the next pedestrian is then assessed (a whitelist query is performed on the next pedestrian, followed by second-view image acquisition and image region overlap detection). In this way, when only a single second camera 200 is used to acquire a face image, that face image can be used as a second-view image for recognition along with the face information.

[0051] In the above processing, by detecting overlap in the human image region, a high overlap rate indicates that someone is obstructing the view. In this case, it's necessary to determine which pedestrian's second-view image to acquire. The pedestrian closest to the second camera 200 among the multiple pedestrians with high overlap rates is the one whose second-view image needs to be acquired. The other pedestrians with high overlap rates do not need to be detected. It should be noted that if the straight-through gate 310 is wide, and there are multiple people walking side-by-side at the front of the moving passage 300 without obstruction, second-view images of these people can be acquired sequentially or simultaneously. The head posture assessment results are the angle values ​​in the three axes of roll, rotation, and yaw (equivalent to the angle values ​​in the X, Y, and Z axes of the spatial coordinate system). The angle values ​​are used to determine whether the image is in a frontal posture (frontal posture is easier to identify pedestrian facial features). The frontal posture is the facial posture defined in the debugging process that is close to or equal to the facial image state of a frontal shot of a face. That is, it analyzes the state of the face in the X, Y, and Z axes relative to the coordinate axes within the threshold.

[0052] Furthermore, when multiple second cameras 200 are used to acquire second-view images, while recording the pedestrian's identification ID and pose evaluation results to the fusion list, the device number of the specific second camera 200 and the cropped face image are also sent to the fusion list. This allows for further filtering of multiple face images acquired for the same pedestrian, selecting the best face image as the second-view image for recognition, thereby improving recognition accuracy. Specifically, multiple second cameras 200 acquire multiple image information from different angles when capturing the image of a pedestrian with the same identification ID. Subsequently, the fusion lists of each second camera 200 are obtained, and a unified fusion ID list is created (i.e., the ID list includes fusion lists corresponding to multiple identification IDs). The image information of the current identification ID in each fusion list is queried, and when multiple sets of image information exist, pose fusion is performed. The best image information is selected as the second-view image and compared with the pre-stored face information for recognition. During pose fusion, head pose assessment results can be used to select the best portrait based on the angle values ​​of the three axes of roll, rotation, and yaw. Alternatively, multiple portrait images can be stitched together to obtain a portrait with a complete frontal view as the best portrait. If recognition is successful, the pedestrian's ID is updated to the access whitelist, and the next ID in the unified list of IDs to be fused is used for portrait information query, pose fusion, and comparison. Otherwise, the portrait information query, pose fusion, and comparison are directly performed on the next ID in the unified list of IDs to be fused.

[0053] In this embodiment, three second cameras 200 are provided within the moving channel 300. Specifically, one second camera 200 is set on each of the left and right sides of the moving channel 300 as a slave device, and one second camera 200 is suspended above the middle of the moving channel 300 as a master device. The slave devices are used to capture dynamic human images at high speed, perform face detection, calculate the head pose, and crop the human image. The master device is used to capture dynamic human images at high speed, perform face detection, calculate the head pose, crop the human image, and perform multi-pose fusion (i.e., further processing of the image when there are multiple pedestrian images in the second-view image) to extract the best human image. In this way, the master device sequentially sends instructions to obtain the pedestrian identifier IDs from the list of IDs to be fused by the slave devices. If the identifier ID does not exist in the master device, it is added to create a unified list of IDs to be fused to achieve data synchronization. Subsequently, the master device sequentially performs selection and recognition processes on all identifier IDs in the unified ID list to be merged. The selection process includes: querying the cropped faces (i.e., facial information) of the current identifier ID in the merge lists of both the master and slave devices; if multiple sets exist, the poses are merged, and the best cropped face is selected. The recognition process includes: sending the cropped face to the face server for face recognition comparison; if recognition is successful, the pedestrian identifier ID is updated to the access whitelist, and the next ID in the unified ID list to be merged is either cropped or the best face is selected; otherwise, the next ID in the unified ID list to be merged is directly cropped or the best face is selected.

[0054] Example 3 Please combine Figures 1 to 10The present invention also discloses a turnstile device for implementing the access control and verification method of the access cameras in Embodiments 1 and 2 described above. Specifically, the turnstile device includes a gate mechanism, a first camera 100, multiple second cameras 200, a remote server 500, and a control unit 600. The gate mechanism includes a moving channel 300, a buffer channel 400 connected to the side of the moving channel 300 near the exit end, a straight gate 310 located within the moving channel 300 and adjacent to the entrance of the buffer channel 400, and a gate with multiple exits. Self-service verification gates 410 (including self-service verification gate 410(1) at the entrance of buffer channel 400 and self-service verification gate 410(2) at the exit of buffer channel 400) are set at the entrance and exit of buffer channel 400. A secondary verification dedicated channel verification terminal 420 is also set at the self-service verification gate 410(2) at the exit of buffer channel 400. The secondary verification dedicated channel verification terminal 420 can be set as a biometric identification device 421 and / or a document identification device 422. It can be understood that the straight gate 310 is located at the exit end of the moving channel 300. In this way, when a pedestrian moves from the moving channel 300 to the straight gate 310, the passage whitelist query or the identification ID after second-view image recognition has been performed, so as to avoid the problem of slow clearance efficiency caused by the pedestrian only performing image recognition after reaching the straight gate 310. The first camera 100 is used to detect whether there are pedestrians in the moving channel 300. When a pedestrian is present, it tracks and records a first-view image of the pedestrian approaching the straight-through gate 310. The pedestrian's coordinates and distance from the pedestrian to the first camera 100 are extracted from the first-view image, and each pedestrian is assigned an identification ID. The second camera 200 is used to obtain the pedestrian's identification ID and coordinates from the first camera 100, and to acquire a second-view image of the pedestrian according to the calling strategy, the pedestrian's identification ID, and the pedestrian's coordinates. The pedestrian's head posture is calculated and the image is cropped. The remote server 500 pre-stores facial information and a whitelist for access. It is used to compare and identify the identification ID obtained by the second camera 200 with the whitelist, or to compare and identify the second-view image obtained by the second camera 200 with facial information, and to update the whitelist after successful identification. In this embodiment, when the remote server 500 has a pre-stored access whitelist, when the remote server 500 is working, it first checks whether the identifier ID is in the access whitelist. If it exists, there is no need to update the access whitelist. If it does not exist, it then recognizes the second-view image obtained by the second camera 200 with the face information. If the recognition is successful, the access whitelist is updated.The control unit 600 is used to keep the straight-through gate 310 open and the self-service verification gate 410(2) closed when the pedestrian at the front of the straight-through gate 310 is on the whitelist and there is no tailing, or when the pedestrian at the front of the straight-through gate 310 is on the whitelist and there is tailing and all the tailings are on the whitelist; otherwise, the control unit 600 closes the straight-through gate 310 and opens the self-service verification gate 410(2). The control unit 600 controls the opening and closing of the corresponding gates based on the pedestrian whitelist query results or the second-view image recognition results, so as to realize the direct passage or diversion of personnel and ensure the clearance capacity of the mobile channel 300. The straight gate 310 remains open in the initialization state, and the self-service verification gate 410 (1) at the entrance of the buffer channel remains closed in the initialization state. In this way, if the pedestrian is successfully identified, they can pass directly through the open straight gate 310. Compared with the normally closed design of the straight gate 310, the time of waiting for the straight gate 310 to open can be saved, thereby improving the speed and efficiency of passage. By guiding the people who fail the identification to the buffer channel 400, the situation of pedestrians blocking the moving channel 300 can be avoided, and fast passage can be achieved.

[0055] In this embodiment, the first camera 100 integrates a 3D ToF module, an RGB camera module, a passage logic control board, and indicator lights. The first camera 100 applies ToF vision technology, using three-dimensional laser scanning to perceive the three-dimensional or near / far distance of a target. Its 3D vision technology collects and outputs three-dimensional vector information of the human body, objects, and space. Therefore, it can extract the pedestrian coordinate identifier ID from the first-view image collected by the first camera 100, providing data such as the position coordinates, depth data, shape, and pose information of objects within the field of view. This solves the shortcomings of traditional detection technologies in perceiving motion information in the gate passage space, realizing the three-dimensional data acquisition of the gate passage and the data-driven perception of target motion. After acquiring the first-view image, the first camera 100 can determine whether there is a pedestrian tailgating based on the distance between multiple pedestrians in the first-view image, thus assisting in the implementation of the still-common practice of preventing tailgating.

[0056] Furthermore, in this embodiment, three second cameras 200 are provided corresponding to the same first camera 100. One second camera 200 is positioned directly above the moving channel 300 as the master device, and the other two second cameras 200 are positioned on the left and right sides of the moving channel 300 as slave devices. The slave devices are used to capture dynamic human images at high speed, perform face detection, calculate head pose, and crop the images. The master device is used to capture dynamic human images at high speed, perform face detection, calculate head pose, crop the images, and acquire its own cropped images and the cropped images sent by the slave devices. Multi-pose fusion is performed on these cropped images to extract the best image as the first face information. That is, the multiple second cameras 200 are used to detect, identify, and select the best frontal face image of a pedestrian based on the real-time monitored pedestrian trajectory; that is, to identify and select images in the second-view information where the facial information is close to a frontal face, in order to identify the pedestrian. When the moving channel 300 is long, it can be divided into multiple segments, and a detection unit can be set up in each segment to detect pedestrians. The detection unit includes a first camera 100 and multiple second cameras 200 to collect as much pedestrian data as possible, thereby improving the probability of recognition.

[0057] In this embodiment, the moving channel 300 is divided into three segments along its length: a pedestrian access control area from the entrance of the moving channel 300 to the second camera 200 (this segment is the area where the moving channel is extended to form a long moving channel, and can be extended according to actual conditions); a control opening area between the second camera 200 moving towards the straight gate 310 and a preset point; and a turnstile passage area from the preset point to the straight gate 310. The pedestrian access control area is the area for pedestrian detection and recognition. When a pedestrian enters this area, the pedestrian access control area control process is triggered, i.e., the first camera 100 tracks the pedestrian, the second camera 200 detects and fuses the pedestrian image, and the remote server 500 recognizes the face within this area. The control opening area is the area that ultimately decides whether to open or close the gate. When a pedestrian enters this area, the control opening area control process is triggered, ultimately deciding whether to open or close the straight gate 310 or the self-service verification gate 410. The turnstile passage area is designed to prevent pedestrians from being pinched by the gate and does not involve any software processes or other operations. With a pedestrian design speed ≤1.5m / s, the pedestrian passage control area is 3m, and the pedestrian stay time within this area is 2s; the gate opening control area is 0.45m long, the estimated pedestrian stay time here is 0.2s, and the detection response time is 0.1s; the turnstile passage area is 0.9m long, the gate opening response time is 0.5s, and the image detection and recognition algorithm delay is 0.1s; the total length of the moving channel 300 is 4.35m. In practice, with multiple detection units, the length of the moving channel 300 can be further adjusted.

[0058] In practical applications, the number of second cameras 200 can be increased or decreased, the length and width of the moving channel 300 can be increased, and the installation position of the second cameras 200 can be adjusted. When setting up multiple moving channels 300 and corresponding buffer channels 400, the buffer channels 400 of multiple moving channels 300 can be merged to save layout space. Multiple second cameras 200 can also share the same computer for calculation and data processing. In order to save resources, multiple moving channels 300 can also share the same remote server 500. For multiple second-view images of the same pedestrian, no filtering is required, and they can be directly sent to the remote server 500 for comparison.

[0059] The aforementioned linkage between the access control cameras and the access control gate equipment utilizes multiple first-view images captured by the first camera 100 at different times to form a predicted pedestrian trajectory. Then, it calls upon the second camera 200 along this predicted trajectory to capture a second-view image of the pedestrian. Before the pedestrian reaches the gate, the second-view image is captured for identification, eliminating the need for the pedestrian to stop at the gate and saving time waiting for facial verification. This shortens the time for a single person to pass through the gate and improves the efficiency of passage. Furthermore, by acquiring the predicted pedestrian trajectory and calling the second-view image... The second camera 200 on the trajectory captures second-view images, enabling prediction of the time it takes for a pedestrian to reach a designated location. This allows the use of the second camera 200 to be adapted to the predicted pedestrian trajectory. Specifically, the second camera 200 corresponding to the pedestrian is selected and called based on the predicted trajectory, improving resource utilization efficiency and the accuracy of information collection. It also makes it easier to collect second-view images that can identify pedestrian information, achieving efficient resource scheduling and utilization. This improves the efficiency and pass rate of pedestrian identification, shortens the time pedestrians spend at the gate, and further enhances pedestrian clearance efficiency.

[0060] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0061] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for linking and calling access control cameras for access verification, used to control the use of a first camera and multiple second cameras, wherein the first camera is used to capture images from a pedestrian's first-person perspective, and the second cameras are used to capture images from a pedestrian's second-person perspective; characterized in that, The method for linking and activating access control cameras for passage verification includes the following steps: The first camera tracks and records multiple first-person perspective images as pedestrians approach the gate; Based on first-view images at multiple different times, a pedestrian prediction trajectory is formed within a preset detection area of ​​the first camera, and at least one second camera on the pedestrian prediction trajectory is invoked. The second camera acquires a second-view image of the pedestrian and identifies it with preset facial information. If the identification is successful, the preset gate is opened; if the identification is incorrect, the gate is closed and the pedestrian's biometric information is collected for verification.

2. The method for linking and calling access control cameras for passage verification according to claim 1, characterized in that, The second camera acquires second-view images of pedestrians by: performing preliminary screening of human images that match the model, and then, after meeting the conditions, acquiring specific human images for recognition and comparison.

3. The method for linking and calling access control cameras for passage verification according to claim 2, characterized in that, The first camera predicts the direction, speed and posture of the pedestrian based on the predicted trajectory, and calls the second camera to collect the second-view image of the pedestrian based on the prediction result; The second-view image captured by the second camera is substituted into the local model for integrity verification. If the verification is successful, the second-view image is compared with the information of the remote server. otherwise, If the verification fails, another second camera is called to re-acquire the second-view image of the pedestrian who failed the verification. The re-acquired second-view image is substituted into the local model for integrity verification. If the re-verification is successful, the re-acquired second-view image is compared with the information of the remote server. If the verification still fails, select multiple second cameras to simultaneously collect second-view images of the pedestrians who failed the verification, and then substitute them into the local model for integrity verification. If the second-view image collected by any second camera is successfully verified, compare the information of the successfully verified second-view image with the remote server. If all verifications fail, multi-image recognition is used to combine and stitch together the overlapping positions of the second-view images captured by multiple second cameras, and then the stitched image is substituted into the local model for integrity verification; if the stitched image verification fails, the pedestrian is classified as an abnormal person.

4. The method for linking and calling access control cameras for passage inspection according to claim 3, characterized in that, The method of combining and stitching the overlapping positions of the second-view images captured by multiple second cameras using multi-image recognition includes: combining the installation positions of the second cameras and the calibrated relative position parameters between each second camera to stitch together multiple second-view images of the same pedestrian within the preset detection area of ​​the first camera.

5. The method for linking and calling access control cameras for passage verification according to claim 3, characterized in that, After selecting multiple second-view images of pedestrians who failed verification to be captured simultaneously by multiple second cameras, first select one of the second-view images based on image quality and pedestrian posture and then improve the image recognition rate, or first improve the image recognition rate and then select one of the second-view images based on image quality and pedestrian posture. Then perform integrity verification between the processed or selected second-view images and the local model.

6. The method for linking and calling access control cameras for passage verification according to claim 5, characterized in that, The improvement of image recognition rate includes: based on the position of the face in the second-view image, distortion correction is performed when the face is at the edge of the second-view image.

7. The method for linking and calling access control cameras for passage verification according to claim 1, characterized in that, The first camera continuously acquires first-view images and extracts pedestrian coordinates and pedestrian IDs from the first-view images to form pedestrian movement trajectories. Acceleration is calculated based on the pedestrian movement trajectory to determine the time and posture of the pedestrian arriving at the preset image acquisition area, and the corresponding second camera is pre-called. When there are multiple pedestrians, multiple second cameras within the preset range are called to continuously acquire second-view images, or a single second camera is used to continuously shoot in the direction of pedestrian movement to pre-acquire images.

8. The method for linking and calling access control cameras for passage verification according to any one of claims 1 to 7, characterized in that, The system may use one of the following strategies to call at least one second camera on the pedestrian prediction trajectory: a data acquisition area level calling strategy, a proximity calling strategy, or a resolution calling strategy; or it may use a hierarchical calling strategy to call multiple second cameras on the pedestrian prediction trajectory. The collection area level calling strategy includes: the second camera with the largest collection area collects human images frame by frame, and the other second cameras collect human images every other frame; or the second camera with the smallest collection area collects human images frame by frame, and the other second cameras collect human images every other frame. The proximity-based call strategy includes: prioritizing the use of the second camera closest to the pedestrian to capture their image; The resolution retrieval strategy includes: when a pedestrian is outside the preset area of ​​the second camera, the second camera captures the image of the pedestrian at a first resolution; when a pedestrian enters the preset area of ​​the second camera, the second camera captures the image of the pedestrian at a second resolution, which is lower than the first resolution. The step-by-step invocation strategy includes: setting up multiple second cameras along the pedestrian's movement direction, and invoking each of the second cameras step-by-step along the pedestrian's predicted trajectory.

9. A method for linking and calling access control cameras for passage verification, applied to a turnstile device for fast passage, wherein the turnstile device is provided with a moving channel, a buffer channel connected to the side of the moving channel near the exit end, a straight gate located at the exit end of the moving channel, self-service verification gates respectively set at the entrance and exit of the buffer channel, a first camera, and multiple second cameras called within the moving channel, wherein, The initial state of the straight-through gate is open, and the initial state of the self-service verification gate at the entrance of the buffer channel is closed. The method for linking the passage camera to perform the passage verification includes the following steps: S1. The first camera detects whether there are pedestrians in the moving channel. When there are pedestrians, it tracks and records multiple first-view images of the pedestrians as they approach the straight-through gate. The pedestrian coordinates and pedestrian ID are extracted from the first-view images, and then the process proceeds to step S2. S2. Based on multiple first-view images at different times, a pedestrian prediction trajectory is formed within the preset detection area of ​​the first camera. At least one second camera on the pedestrian prediction trajectory is called. The second camera queries whether the pedestrian's ID is in the passage whitelist. If it is, the pedestrian passes directly through the straight-through gate; otherwise, proceed to step S3. S3. The second camera acquires a second-view image of the pedestrian and recognizes it with preset facial information. If the recognition is successful, the pedestrian's ID is updated to the access whitelist so that the pedestrian can pass directly through the straight gate. If the recognition is incorrect, the straight gate is closed and the self-service verification gate is opened so that the pedestrian who fails the recognition can enter the buffer channel. The buffer channel is used for the secondary verification of the pedestrian.

10. A turnstile device for implementing the access control and verification method based on the linkage of the access camera as described in any one of claims 1 to 9, characterized in that, The turnstile equipment includes: The gate mechanism includes a moving channel, a buffer channel connected to the side of the moving channel near the exit end, a straight gate located in the moving channel and adjacent to the entrance of the buffer channel, and self-service verification gates respectively set at the entrance and exit of the buffer channel. The self-service verification gate located at the exit of the buffer channel is equipped with a biometric device and / or a document recognition device. The first camera is used to detect whether there are pedestrians in the moving channel. When there are pedestrians, it tracks and records the first-view image of the pedestrians as they approach the straight gate. The pedestrian coordinates and the distance from the pedestrians to the first camera are extracted from the first-view image, and each pedestrian is assigned an identification ID. Multiple second cameras are used to obtain pedestrian identification IDs and pedestrian coordinates from the first camera, and to acquire second-view images of pedestrians according to the calling strategy, pedestrian identification IDs and pedestrian coordinates, calculate pedestrian head poses and crop the images; The remote server stores facial information and a whitelist of access points. It is used to compare and identify the identifier ID obtained by the second camera with the whitelist of access points, or to compare and identify the second-view image obtained by the second camera with the facial information, and to update the whitelist of access points after successful identification. The control unit is used to keep the straight-through gate open and the self-service verification gate closed when the pedestrian at the front of the straight-through gate is on the whitelist and there is no tailgating, or when the pedestrian at the front of the straight-through gate is on the whitelist, there is tailgating and all the tailgating people are on the whitelist; otherwise, the control unit closes the straight-through gate and opens the self-service verification gate.

Images