Airport panoramic 360° coordination and identification method based on remote tower

Abstract

This invention provides a method for collaborative linkage and recognition of a 360° panoramic view of an airport based on a remote control tower, comprising: acquiring a 360° panoramic video image of the airport; dividing the panoramic video image into two equal panoramic images and marking the pentagonal flight area; if the horizontal coordinate x of any pixel in the image is detected... r ≤W / 2, and P r Not within the five-sided flight area, control the first PTZ camera to move from its current position to P. r The corresponding position; if P r Within the pentagonal flight zone, the aircraft landing gear status is detected based on images captured by the first PTZ camera; if x r >W / 2, and P r Not within the five-sided flight area, control the second PTZ camera to move from its current position to P. r The corresponding position; if x r >W / 2, and P r Within the pentagonal flight zone, the landing gear status is detected based on images captured by a second PTZ camera. This invention enables the coordinated operation of two PTZ cameras and the detection of the landing gear status.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of remote tower optical systems, in particular to an airport panoramic 360° coordination linkage and identification method based on a remote tower. BACKGROUND

[0002] A remote tower is a facility that provides airport air traffic services by replacing visual observation with high-definition video images. Based on high-definition video image processing and low-latency data remote transmission, data is converted into graphics or enhanced intuitive images to support controller decision-making, and remote off-site command is realized. The system is not limited by location, saves labor costs while improving work efficiency, and truly realizes the leap from a "tower in the cloud" to a "digital tower". In the practical application of a remote tower, there are requirements for the installation position of the front-end camera group. A window view outside the tower needs to be constructed, so the 360° panoramic camera group needs to be constructed on the top of the tower.

[0003] The existing mature Hikvision, Dahua, etc. integrated 360° monitoring equipment is composed of two discrete 180° panoramic videos output, and does not realize the display of a 360° continuous panoramic picture in the true sense. Moreover, since it is integrated, the fixed camera cannot move left and right, but only up and down, resulting in large image distortion and low-quality spliced images. At the same time, only one rotatable spherical camera is configured, so it cannot simultaneously consider the linkage of the airport flight area and the parking area. Although the actual installation on the top of the tower is simple and fast, due to the complex environment on the top of the tower (including antenna, lightning rod and other equipment obstruction), it will cause great difficulty in splicing and low-quality spliced images.

[0004] Furthermore, the aircraft landing gear is a crucial support system during takeoff, landing, taxiing, and parking, and is also a primary load-bearing component. It enables braking and turning during takeoff and taxiing, and absorbs and dissipates the impact energy generated by the aircraft's contact with the ground during landing and taxiing. It is a vital safety component, and its condition directly affects the safety of passengers and the operational safety of the airport. Therefore, monitoring the landing gear's condition is particularly important for ensuring the safety of the aircraft during landing and takeoff. Although the aircraft's instrument panel can display the landing gear's status, it cannot avoid situations where instruments issue false alarms due to malfunctions. In the past decade, there have been numerous incidents both domestically and internationally of false alarms regarding landing gear status from aircraft instruments. In such cases, to confirm the landing gear's condition, the pilot often needs to perform multiple low-altitude flybys, allowing ground personnel to visually confirm the landing gear's status before taking further action. With the development and widespread application of computer vision technology, image-based landing gear status detection has become possible and will become a trend in the construction of future smart airports. This can be used not only for daily monitoring and early warning of abnormalities in the landing gear status of aircraft taking off and landing, but also as a reference for ground command and crew to judge whether landing conditions are met when other monitoring instruments malfunction. It has important research significance.

[0005] Currently, there are relatively few studies on landing gear state detection based on image recognition and detection technology. From the perspective of the type of optical image used, some use visible light images, some use infrared images, and some use infrared and visible light composite images. Some of them are based on traditional image processing algorithms. For example, the technical solution disclosed in patent document 1 (CN111898444A, an aircraft landing gear state determination method based on image recognition) includes: first, image noise reduction and segmentation preprocessing are performed, then a multi-view image matching method is used to match the sample library to realize three-dimensional target recognition and pose estimation for the aircraft, then the relative invariance of each point position on the rigid body is used to determine the rectangular search area with the highest appearance probability of the landing gear, and finally the area is processed through image basic operations such as segmentation and edge detection to determine the landing gear state. The technical solution disclosed in patent document 2 (CN106203353A, a detection system and method for an aircraft landing gear) includes: based on the target gray scale distribution characteristics of the infrared image, the segmentation threshold is calculated, the image is binarized to obtain a segmented image, then the engine position and size are determined, then the relative position relationship is used to determine the landing gear position and size, and then the landing gear state is determined through the gradient characteristics and differences of the local area. In addition, some methods use deep learning algorithms, for example, Vibe algorithm and YOLO algorithm are used to locate the specific position of the aircraft in the image, and the aircraft image is extracted, and then SSD algorithm is used to identify and locate the landing gear. The above methods are carried out in multiple stages, which need to first locate the aircraft target or the landing gear search area, and then locate the landing gear in the extracted aircraft image or search area in different ways to determine the landing gear state. The process is relatively complicated, and usually involves image noise reduction, image segmentation, image correction and other preprocessing steps, resulting in low efficiency. On the other hand, these methods usually have strict requirements for the installation angle and position of the image acquisition device, such as being specified to be installed at certain positions on the runway accessory, and have weak generalization ability for images with complex backgrounds and different collection angles. SUMMARY

[0006] To solve the above technical problems, the technical solution adopted by the present application is:

[0007] The embodiment of the present application provides an airport panoramic 360° cooperative linkage and identification method based on a remote tower, which comprises the following steps:

[0008] S100, at a current calculation time, m images taken by a 360° panoramic camera group installed on the top of a remote tower are acquired, and a 360° panoramic video picture corresponding to the current calculation time is obtained based on the acquired m images; m≥2;

[0009] S200, divide the airport 360° panorama video picture into two equal panorama pictures and mark an airport five-side flight area in the airport 360° panorama video picture, wherein the panorama pictures comprise a first panorama picture and a second panorama picture; wherein the first panorama picture is associated with a first ball machine, and the second panorama picture is associated with a second ball machine, and the first ball machine and the second ball machine are respectively arranged on two sides of a 360° panorama camera group;

[0010] S300, if it is detected that any pixel point P r =(x r , y r ) in the airport 360° panorama video picture is clicked, execute S400; wherein the value of r is 1 to n, n is the number of pixel points in the airport 360° panorama video picture, x r and y r are the horizontal coordinate and the vertical coordinate of P r ;

[0011] S400, if x r ≤W / 2 and P r does not belong to the airport five-side flight area, control the first ball machine to move from the current position to the position corresponding to P r , and display the image captured by the first ball machine at the position corresponding to P r in a preset magnification size; if x r ≤W / 2 and P r belongs to the airport five-side flight area, execute S500; if x r >W / 2 and P r does not belong to the airport five-side flight area, control the second ball machine to move from the current position to the position corresponding to P r , and display the image captured by the second ball machine at the position corresponding to P r in a preset magnification size; if x r >W / 2 and P r belongs to the airport five-side flight area, execute S600; wherein W is the width of the airport 360° panorama video picture;

[0012] S500, detect the state of the aircraft landing gear based on the image captured by the first ball machine at the position corresponding to P i ;

[0013] S600, detect the state of the aircraft landing gear based on the image captured by the second ball machine at the position corresponding to P i .

[0014] The present application has at least the following beneficial effects:

[0015] The airport panoramic 360° cooperative linkage and identification method based on a remote tower provided by the embodiment of the present application can solve the linkage installation limitation and the shielding problem of one sphere machine, can quickly and synchronously display the flight area and the parking area picture, and can be mutually backed up, so that if one sphere machine has a problem, the other sphere machine can temporarily manually rotate to compensate for the blind spot. In addition, the method can realize the identification and detection of the aircraft landing gear state, can improve the monitoring ability of the controller to the safe operation of the aircraft, and can replace the controller to monitor the state of the aircraft landing gear through a telescope. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 The flowchart of the airport panoramic 360° cooperative linkage and identification method based on a remote tower provided by the embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0019] Figure 1 The flowchart of the airport panoramic 360° cooperative linkage and identification method based on a remote tower provided by the embodiment of the present application.

[0020] The embodiment of the present application provides an airport panoramic 360° cooperative linkage and identification method based on a remote tower, as shown in the figure, the method can include the following steps: Figure 1

[0021] S100, m images photographed by a 360° panoramic camera group installed on the top of a remote tower are acquired at a current calculation time, and a 360° panoramic video picture corresponding to the current calculation time is obtained based on the acquired m images; m is greater than or equal to 2.

[0022] In the embodiment of the present application, the m images photographed by the 360° panoramic camera group can be acquired according to a set calculation period, that is, the interval between two calculation times. The set calculation period can be determined based on actual conditions.

[0023] ​In the embodiment of the present application, the 360° panoramic camera group can include m fixed gun cameras, wherein each fixed gun camera is independently installed. In an illustrative embodiment, m = 12. In the embodiment of the present application, since each camera is independent, the installation is flexible, and in the splicing and fusion stage, each camera can be flexibly rotated according to the actual tower top antenna, lightning rod and other equipment influence, so as to avoid the influence of the antenna, lightning rod and other equipment in the camera splicing and fusion area. Therefore, the image splicing distortion is small, the spliced image quality is high, and the actual remote control of the airport scene can be met.

[0024] As known by those skilled in the art, any method for obtaining a current airport 360° panoramic video picture based on the m images obtained at the current calculation time is within the protection scope of the present application. For example, the Harris corner point detection method is used to extract image feature points, the Euclidean distance is used for feature vector matching, the RANSAC algorithm is used for purification processing of the registration points, a nonlinear weighted mask with high frequency is obtained at one time, a two-level pyramid Gaussian smoothing is used, and finally the original image and the smoothed image are directly used for difference and accumulation to finally realize the splicing and fusion processing of the image.

[0025] S200, dividing the airport 360° panoramic video picture into two equal panoramic pictures and marking an airport five-side flight area in the airport 360° panoramic video picture, wherein the panoramic pictures include a first panoramic picture and a second panoramic picture with the same angle; wherein the first panoramic picture is associated with a first ball machine, and the second panoramic picture is associated with a second ball machine, and the first ball machine and the second ball machine are respectively arranged on the two sides of the 360° panoramic camera group.

[0026] In the embodiment of the present application, the first panoramic picture can be a 0-180° panorama, and the second panoramic picture can be a 180-360° panorama. In a specific example, the first panoramic picture is an airport living area, and the second panoramic picture is an airport flight area.

[0027] In the embodiment of the present application, the airport five-side flight area is a prior art.

[0028] S300, if it is detected that any pixel point P r = (x r , y r ) in the airport 360° panoramic video picture is clicked, S400 is executed; wherein the value of r is 1 to n, n is the number of pixel points in the airport 360° panoramic video picture, x r and y r are the horizontal coordinate and the vertical coordinate of P r .

[0029] In this embodiment of the invention, the 360° circular panorama is unfolded, which is equivalent to a continuous panoramic image from 0 to 360°, thereby obtaining the coordinates of each pixel in the panoramic image.

[0030] In this embodiment of the invention, any pixel in the 360° panoramic video image displayed on the screen can be clicked using a mouse or similar means. Based on the panoramic image divided into equal parts in S200, pixel statistics show that the horizontal and vertical pixel sizes of the airport's 360° panoramic video image are W and H respectively, meaning the width and height of the airport's 360° panoramic video image are W and H respectively. Each pixel in the airport's 360° panoramic image corresponds to the absolute angle of the corresponding PTZ camera.

[0031] S400, if x r ≤W / 2, and P r Not belonging to the five-sided flight area of ​​the airport, control the first PTZ camera to move from its current position to P. r The corresponding position, and the first PTZ camera at P r Images captured at the corresponding locations are displayed at a preset magnified size; if x r ≤W / 2, and P r If it falls within the five-sided flight area of ​​the aforementioned airport, execute S500; if x r >W / 2, and P r Not within the five-sided flight area of ​​the airport, control the second PTZ camera to move from its current position to P. r The corresponding position, and the second PTZ camera at P r Images captured at the corresponding locations are displayed at a preset magnified size; if x r >W / 2, and P r It falls within the five-sided flight area of ​​the aforementioned airport and operates under S600.

[0032] Specifically, in S400, the first PTZ camera rotates at an angle a1 in the lateral direction. x =(θ 1x *x r ) / (W / 2), the rotation angle a1 of the first PTZ camera in the longitudinal direction. y =(θ 1y *y r ) / H; The second PTZ camera's rotation angle a2 in the lateral direction x =(θ 2x *x r ) / W, the rotation angle a2 of the second PTZ camera in the longitudinal direction y =(θ 2y *y r ) / H; where θ 1x and θ 2xrespectively are maximum rotation angles of the first and second ball cameras in the transverse direction, respectively 1y and 2y respectively are maximum rotation angles of the first and second ball cameras in the longitudinal direction. In one illustrative embodiment, 1x = 90°, 2x = 180°, 1y = 90°, 2y = 60°.

[0033] The skilled in the art knows that P r Whether P belongs to the five-edge flight area of the airport can be determined by the prior art.

[0034] Further, in the embodiments of the present application, the image taken by the first ball camera at the position corresponding to P r is displayed according to a preset magnification size, and the image taken by the second ball camera at the position corresponding to P r is displayed according to a preset magnification size. The prior art can be used. For example, the ball camera obtains the zoomed-in and zoomed-out depth of field value Zx = cosθ * Zn by using the known vertical panoramic positive direction depth of field size (assuming 100), presetting the vertical positive depth of field value Zn, and specifying the angle position θ, and after calculating Zx, the ball camera can be controlled to move to the position of the specified depth of field absolute value Zx.

[0035] S500, detecting the state of the aircraft landing gear based on the image taken by the first ball camera at the position corresponding to P r .

[0036] S600, detecting the state of the aircraft landing gear based on the image taken by the second ball camera at the position corresponding to P r .

[0037] In the embodiments of the present application, the image for detection is an image that has been enlarged by a preset magnification size.

[0038] In the embodiments of the present application, the state of the aircraft landing gear can be detected based on a YOLO-V3 target detection model. The core idea of the YOLO-V3 target detection model is to extract multi-scale feature maps of the input picture by using its CNN backbone network, and then directly regress the boundary box position of the detection target and its belonging category based on the multi-scale feature maps, and the final result is obtained after non-maximum suppression optimization.

[0039] In the embodiment of the present application, the YOLO-V3 target detection model is trained by adding a landing gear state output channel to the network, designing a training sample label and a loss function based on the original algorithm, so that the network has the ability to directly regress the landing gear state attribute and the position of the aircraft and landing gear bounding box, and truly realizes the end-to-end training and real-time detection of the combination of landing gear state recognition and target detection. Specifically, the YOLO-V3 target detection model can be obtained based on the following steps:

[0040] S10, an initial YOLO-V3 network is constructed, the output channel parameters of the initial YOLO-V3 network include target bounding box coordinates, landing gear state attributes, bounding box confidence and object class attributes, the object class attributes include an aircraft and a landing gear, that is, each prediction is an 8-dimensional vector; the loss function of the initial YOLO-V3 network is loss=coodErr+gearErr+confErr+clsErr, wherein coodErr is a bounding box coordinate error, gearErr is a landing gear state error, confErr is a bounding box confidence error, and clsErr is a classification error.

[0041] In the embodiment of the present application, the backbone network of the initial YOLO-V3 network is DarkNet-53. The network of the present application obtains three kinds of feature maps of 52×52, 26×26 and 13×13 scales for each input image with a resolution of 416×416, respectively, and performs prediction on the three scales, respectively, and sets 3 kinds of prior boxes (the design method of the prior boxes is the same as that of the original network) on each grid of each scale feature map, a total of 10647 predictions.

[0042] In the embodiment of the present application, Wherein, S 2 is the number of feature map grids, B is the number of prior boxes, x i , y i , w i , h i respectively correspond to the training label values of the horizontal and vertical coordinates of the bounding box center point and the width and height of the bounding box, G i , C i , P i (c) respectively correspond to the training label values of the landing gear state, the confidence and the class; x i , y i , w i , h i , G i , C i , P i (c) are the corresponding prediction values; λ coord , λ noobj , λgear I is a weighting constant used to weigh the contribution of each loss component; obj ij This indicates whether the j-th prior box in the i-th grid is responsible for predicting the target. Specifically, it determines whether the prior box has the largest Intersection Over Union (IOU) with the ground truth bounding box among the nine prior boxes in that grid. If it has the largest IOU, then it is responsible for predicting the target. obj ij The value is 1 if it is 1, otherwise it is 0; I noobj ij Z indicates whether the j-th prior box in the i-th grid is not responsible for predicting the target; plane ij This indicates whether the target predicted by the j-th prior box of the i-th grid is an aircraft target. If it is an aircraft target, its value is 1, otherwise it is 0. This ignores the landing gear state loss predicted for the landing gear target.

[0043] S20, acquire sample images, wherein each sample image is labeled with a corresponding status identification label, the status identification label including a first label indicating that the landing gear is not deployed and a second label indicating that the landing gear is deployed.

[0044] Since there are very few publicly available datasets with landing gear status annotations, a self-made dataset was used in this embodiment of the invention. Using aircraft takeoff and landing videos recorded at airports and publicly available videos downloaded from the internet as data sources, 1668 basic images were obtained through video frame slicing and frame skipping filtering. Bounding boxes and landing gear status annotations were then created for both the aircraft and the landing gear, generating annotation files. 1192 images were selected and divided into training and validation sets in a 6:4 ratio, with the remainder used as the test set. This dataset covers aircraft targets with two landing gear statuses from different angles (videos of aircraft performing free flight demonstrations were specifically selected for this purpose).

[0045] In this embodiment of the invention, the first label can be represented by 0 and the second label can be represented by 1. Of course, it is not limited to this and other identifiers can also be used.

[0046] S30, the sample image is input into the initial YOLO-V3 network for training, and the model when the loss function satisfies the convergence condition is used as the YOLO-V3 object detection model.

[0047] Specifically, the training set in the sample image is input into the initial YOLO-V3 network for training, and the specific training process can be the prior art. For example, in the training process of the YOLO-V3 model, 416x416 images are taken as input, the initial learning rate is set to 10-4, the batch size is set to 12, all layers are directly trained, the learning rate is controlled by a callback function, the learning rate is reduced by 10% if the loss on the validation set does not decrease for 6 consecutive epochs, 200 epochs are trained, and the network model with the lowest loss on the validation set is saved as the final model, i.e., the YOLO-V3 target detection model.

[0048] For the trained YOLO-V3 model, the target detection accuracy on the test set and the state recognition accuracy for the aircraft landing gear are as shown in Table 1. The model gives the boundary box for all targets, and displays the class name, class score and landing gear state field (for example, "Landing gears opened!" represents "landing gear opened"; "Landing gears not found!" represents "landing gear not opened") in the upper left corner of the boundary box for the aircraft target, and only displays the class name and class score in the upper left corner of the boundary box for the landing gear target. That is, the detection result of the YOLO-V3 target detection model provided by the embodiment of the present application can include the aircraft target detection result and the landing gear target detection result.

[0049]

[0050]

[0051] wherein AP is the average precision of a single class, and its value is equal to the area enclosed by the P-R (precision-recall) curve. mAP refers to the mean value of the average precision of multiple classes, and the larger the value, the higher the overall detection accuracy of the model.

[0052] Comparative experiment

[0053] The embodiment of the present application is compared with the existing two-stage method of landing gear state recognition. Specifically, the training set of the above sample image is input into the Faster-RCNN model trained on the open source data set such as VOC. The model is used to detect the aircraft target of the entire data set, and the image blocks in all detected aircraft target bounding boxes are extracted and saved, and the landing gear state value in the original annotation file is used as the corresponding state category label value. The division of the image block data set is consistent with the original data set (that is, the image blocks from the training set are still used as the training set, and so on). Then in the second stage, the image blocks are resized to 224x224 and input into the VGG16 network for training. The loss function uses binary cross entropy, the batch size is set to 32, the learning rate is set to 10-4, all layers are trained, and the network model with the lowest loss on the validation set is saved as the final model when training for 100 epochs.

[0054] After training, the recognition accuracy of the landing gear state recognition classification network in the second stage on the test set is 99.94%, which proves that without landing gear detection, only using the classification network can also predict the landing gear state of the aircraft target.

[0055] From Table 1 and the comparative experiment, it can be seen that the YOLO-V3 target detection model used in the embodiment of the present application can not only better perform target detection, but also simultaneously regress the landing gear state of the aircraft target. The landing gear state recognition accuracy is slightly higher than that of the two-stage method, and the method is more efficient and accurate than the two-stage method. The method also has good robustness for multi-view and different attitude aircraft images, and can continue to enhance its generalization ability by expanding samples. That is, compared with the commonly used multi-stage algorithm, the method does not need complex preprocessing and post-processing stages, is more efficient, and only needs to increase the images collected under different angles and different backgrounds as training samples to easily improve the robustness of the network to different angles and complex background images, and is more convenient for engineering application.

[0056] Further, the method provided by the embodiment of the present application further comprises:

[0057] S700, the detection results obtained by S500 and S600 are displayed.

[0058] The detection results obtained by S500 and S600 can include aircraft target detection results and landing gear target detection results.

[0059] The embodiment of the present application further provides a non-transitory computer readable storage medium, which can be arranged in an electronic device to store at least one instruction or at least one program for implementing a method related to the method in the method embodiment, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided by the above embodiment.

[0060] The embodiment of the present application further provides an electronic device, comprising a processor and the aforementioned non-transitory computer readable storage medium.

[0061] The embodiment of the present application further provides a computer program product, which comprises program codes, and the program codes are used for causing an electronic device to execute the steps in the method according to various exemplary embodiments of the present application described in the specification when the program product is run on the electronic device.

[0062] Although some specific embodiments of the present application have been described in detail by examples, those skilled in the art should understand that the above examples are only for illustration, but not for limiting the scope of the present application. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.

Claims

1. A method for collaborative linkage and recognition of a 360° panoramic view of an airport based on a remote control tower, characterized in that, The method includes the following steps: S100: At the current calculation time, acquire m images captured by the 360° panoramic camera group installed on the top of the remote control tower, and obtain the airport 360° panoramic video image corresponding to the current calculation time based on the acquired m images. m≥2； S200, the airport 360° panoramic video image is divided into two equal panoramic images and the airport five-sided flight area is marked in the airport 360° panoramic video image, wherein the panoramic image includes a first panoramic image and a second panoramic image; wherein the first panoramic image is associated with a first PTZ camera and the second panoramic image is associated with a second PTZ camera, and the first PTZ camera and the second PTZ camera are respectively set on both sides of the 360° panoramic camera group. S300, if a click is detected on any pixel P in the 360° panoramic video of the airport... r = (x r y r ), execute S400; where r takes values ​​from 1 to n, n is the number of pixels in the 360° panoramic video image of the airport, x r and y r For P r The x and y coordinates; S400, if x r ≤W / 2, and P r Not belonging to the five-sided flight area of ​​the airport, control the first PTZ camera to move from its current position to P. r The corresponding position, and the first PTZ camera at P r Images captured at the corresponding locations are displayed at a preset magnified size; if x r ≤W / 2, and P r If it falls within the five-sided flight area of ​​the aforementioned airport, execute S500; if x r >W / 2, and P r Not within the airport's five-sided flight area, control the second PTZ camera to move from its current position to P. r The corresponding position, and the second PTZ camera at P r Images captured at the corresponding locations are displayed at a preset magnified size; if x r >W / 2, and P r Within the five-sided flight area of ​​the airport, S600 is executed; where W is the width of the 360° panoramic video image of the airport; S500, based on the first PTZ camera in P r Images taken at the corresponding locations are used to detect the condition of the aircraft landing gear. S600, based on the second PTZ camera in P i Images taken at the corresponding locations are used to detect the condition of the aircraft landing gear.

2. The method according to claim 1, characterized in that, In S400, the first PTZ camera rotates at an angle a1 in the lateral direction. x =(θ 1x ×x r ) / (W / 2), the first PTZ camera rotates at an angle a1 in the longitudinal direction. y =(θ 1y ×y r ) / H; The second PTZ camera's rotation angle a2 in the lateral direction x =(θ 2x ×x r ) / W, the rotation angle a2 of the second PTZ camera in the longitudinal direction y =(θ 2y ×y r ) / H; where θ 1x and θ 2x θ represents the maximum rotation angle in the lateral direction for the first and second PTZ cameras, respectively. 1y and θ 2y These are the maximum rotation angles of the first and second PTZ cameras in the longitudinal direction, respectively, and H is the height of the 360° panoramic video image of the airport.

3. The method according to claim 1, characterized in that, The first panoramic view is of the airport living area, and the second panoramic view is of the airport flight area.

4. The method according to claim 1, characterized in that, The 360° panoramic camera group includes m fixed bullet cameras, each of which is installed independently.

5. The method according to claim 1, characterized in that, The YOLO-V3 target detection model is used to detect the status of aircraft landing gear.

6. The method according to claim 5, characterized in that, The YOLO-V3 object detection model is obtained based on the following steps: S10, Construct an initial YOLO-V3 network. The output channel parameters of the initial YOLO-V3 network include target bounding box coordinates, landing gear state attributes, bounding box confidence, and object category attributes. The object category attributes include aircraft and landing gear. The loss function of the initial YOLO-V3 network is loss = coodErr + gearErr + confErr + clsErr, where coodErr is the bounding box coordinate error, gearErr is the landing gear state error, confErr is the bounding box confidence error, and clsErr is the classification error. S20, acquire sample images, wherein each sample image is labeled with a corresponding status identification label, the status identification label including a first label indicating that the landing gear is not deployed and a second label indicating that the landing gear is deployed; S30, the sample image is input into the initial YOLO-V3 network for training, and the model when the loss function satisfies the convergence condition is used as the YOLO-V3 object detection model.

7. The method according to claim 6, characterized in that, The backbone network of the initial YOLO-V3 network is DarkNet-53.

8. The method according to claim 6, characterized in that, in, ， in, S 2 The number of grid cells in the feature map. B The number of prior boxes. x i , y i , w i , h i The training label values ​​correspond to the horizontal and vertical coordinates of the center point of the bounding box and the width and height of the bounding box, respectively. G i , C i , P i ( c These correspond to the training label values ​​for landing gear status, confidence level, and category, respectively. , , , , , , They are respectively x i , y i , w i , h i、 G i , C i , P i ( c The corresponding predicted value; λ coord , λ noobj , λ gear This is a weighting constant used to weigh the contribution of each loss component; I obj ij Indicates the first i The first grid j Does each prior box account for the prediction of the target? I noobj ij Indicates the first i The first grid j Does each prior bounding box not contribute to the prediction of the target? plane ij Then it means the first i The first grid j Each prior box is responsible for predicting whether the target is an aircraft target.

Images