Laser positioning emission method based on YOLO and micro-motion target detection

Abstract

The present invention belongs to the field of laser positioning and emission technology, and specifically relates to a laser positioning and emission method based on YOLO and micro-motion target detection, which includes: S1, acquiring target image data, and using the YOLO target detection model to identify the target graphic anchor frame; S2, using the micro-motion target detection algorithm to identify the micro-motion target in the image data acquired by the camera, and obtaining the coordinates of the anchor frame of the target to be measured; S3, integrating the number of target image anchor frames obtained by the micro-motion target detection algorithm and the number of target image anchor frames of the YOLO target detection model, and completing the first recognition of the target to be measured; S4, establishing a world coordinate system transformation model and an adaptive anchor frame variable magnification and zoom model, completing the second recognition of the target to be measured, and controlling the laser positioning and emission device. The present invention combines machine learning with traditional image processing and hardware control, can achieve high-precision target detection, intelligent laser control, high degree of automation and large scalability.

Description

Technical Field

[0001] The invention belongs to the technical field of laser positioning and emission, and in particular relates to a laser positioning and emission method based on YOLO and micro-motion target detection. Background Art

[0002] Airports and power equipment sites are often affected by target image flight activities or plastic waste. Target images nest, stay or fly through near airports and power equipment, which may cause a series of safety issues. For example, debris may cause insulator failure, which may further seriously affect the safe and stable operation of power facilities and cause huge losses to the power system. In the aviation field, incidents of aircraft colliding with debris during takeoff and landing or flight have become the number one hidden danger to aviation safety, which may delay flights at the least and endanger aviation safety at the worst. Therefore, how to effectively drive away target images to remove garbage hanging on wires and ensure the safe operation of power equipment has become an urgent problem to be solved.

[0003] In order to avoid the possible threats caused by target image flight activities to airports and power equipment sites, the patent application document with publication number CN117958245A in the related art provides an intelligent directional target removal method, which obtains video stream data, inputs the video stream data into the YOLOv5 detection model, and obtains the recognition result output by the model; based on the recognition result, obtains the position information corresponding to the target image, and instructs the laser emission device or the sound expulsion device to work according to the recognition result and the position information. However, the target image detection model used in this method is too simple, and it is easy to miss and misdetect. Therefore, it is difficult for the related art to effectively realize the expulsion of target images near airports and power equipment sites. The patent application document with publication number CN118823689A provides an active monitoring and visual recognition technology, which can select a suitable SVM classification model for image classification according to the target scene type near the transmission tower, so as to quickly identify the target image and drive the laser emission actuator. However, this method is easily limited by the accuracy of the SVM classification model, and the recognition effect is not good for some rare or difficult to identify target image types or scenes. Therefore, the present invention proposes a laser positioning and emission method based on YOLO and micro-motion target detection, which can greatly improve the detection accuracy of suspicious targets, improve the recognition accuracy, and can be used for the safety protection of airports and power facilities. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a laser positioning and emitting method based on YOLO and micro-motion target detection, which effectively realizes the determination of micro-animal bodies and gives a target anchor frame through a micro-motion detection algorithm combining a pyramid Lucas-Kanade optical flow method and an improved DBSCAN clustering algorithm; integrates the detection results of the YOLO detection model and the micro-motion detection algorithm to complete a more accurate moving target detection task; the field of view of the camera at different magnifications and the orientation coordinate representation of the camera center realize the conversion from a pixel coordinate system to a world coordinate system under any camera posture, realizes high-precision detection of the target image and intelligent control of a laser emitting device, and has a high degree of automation and great scalability.

[0005] To achieve the above object, the present invention discloses the following technical solution:

[0006] A laser positioning and emission method based on YOLO and micro-motion target detection, comprising:

[0007] S1: Obtain target image data and use the YOLO target detection model to identify the target graphic anchor frame;

[0008] Use an optical zoom camera to capture the target image data as the input of the YOLO target detection model, and identify the coordinates and size of the target image anchor frame as follows:

[0009] ;

[0010] in, is the central horizontal coordinate of the predicted anchor box; is the center coordinate of the predicted anchor box; is the width of the predicted anchor box; is the height of the predicted anchor box; It is the central horizontal coordinate of the unit where the current detection frame is located; The center vertical coordinate of the unit where the current detection frame is located; is the horizontal axis offset predicted by the network; is the vertical axis offset predicted by the network; is the logarithmic horizontal scale value predicted by the network; is the logarithmic vertical scale value predicted by the network;

[0011] S2: Use the micro-motion target detection algorithm to identify the micro-motion targets in the image data acquired by the camera and obtain the anchor frame coordinates of the target to be detected; the micro-motion detection algorithm includes the optical flow pyramid method and the improved DBSCAN clustering algorithm;

[0012] S21: Obtain the anchor frames of the two adjacent target image frames obtained in step S1, and use the Shi-Tomasi corner detection algorithm to detect the feature points in the first target image frame; combine the optical flow pyramid method to use the feature points in the two adjacent target image frames and the first target image frame to calculate the optical flow information of the feature point set between two adjacent continuous target image frames:

[0013] ;

[0014] in, The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; is the horizontal coordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the ordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the horizontal coordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system; is the ordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system;

[0015] S22: Setting the displacement judgment threshold , screen and obtain effective feature points of the target image data; cluster the effective feature points of the target image data, and determine the coordinates of the anchor frame of the target to be measured by using the improved DBSCAN clustering algorithm;

[0016] S3: Integrate the target image anchor frames obtained by the micro-motion target detection algorithm in step S2 and the number of target image anchor frames of the YOLO target detection model in step S1 to complete the first recognition of the target to be detected; the integration rule is:

[0017] ;

[0018] in, For the A micro-motion detection anchor box; For the YOLO detection anchor boxes; For the The micro-motion detection anchor box and the The intersection area of ​​the YOLO detection anchor boxes; For the The area of ​​the micro-motion detection anchor box; For the The micro-motion detection anchor frame and the The intersection-over-union ratio of YOLO detection anchor boxes; For the delete operation; is a counting symbol; is the first determination threshold; is the second determination threshold; Number the YOLO detection anchor box; It is a conditional judgment function; The total number of anchor boxes detected by YOLO;

[0019] S4: Establish a world coordinate system transformation model and an adaptive anchor frame zoom model, complete the second recognition of the target to be measured, and control the laser positioning and launching device;

[0020] The anchor frame identified for the first time in step S3 is converted into the orientation coordinates in the world coordinate system, and is sorted from low to high according to the horizontal angle of the orientation coordinates. The angle required for the laser positioning and transmitting device to point to each anchor frame and the camera magnification ratio are calculated through the world coordinate system transformation model and the adaptive anchor frame magnification and zoom model. The anchor frames obtained by the first identification are sequentially magnified and identified in step S1 according to the horizontal angle of the anchor frame to further determine whether there is a target to be measured at the location of the anchor frame. If the secondary identification result of the anchor frame is the target to be measured, the laser is emitted, otherwise the next anchor frame position is turned to continue the judgment until all the anchor frames identified for the first time complete the secondary identification, the second joint rotation angle of the gimbal is reset to zero and the camera zoom factor becomes 1, the first joint rotation angle of the gimbal is randomly assigned, and the gimbal performs regular monitoring.

[0021] Preferably, the YOLO target detection model in step S1 includes: a feature extraction module, a feature enhancement and fusion module and an output module;

[0022] The feature extraction module consists of convolutional layers, activation layers, and pooling layers; it can convert the input image from raw pixel information into feature representation and extract the hierarchical features of the image from the image;

[0023] The feature enhancement and fusion module realizes feature fusion by introducing the bottom-up and top-down PAN feature pyramid modules to enhance the perception ability of the target;

[0024] The output module adopts the anchor-free detection head without predefined anchor box method to directly predict the position and size of the target to be tested; at the same time, the classification and detection processes are decoupled using the head network, and each head specializes in one task, including target category classification and target image anchor box screening.

[0025] Preferably, in step S21, the Shi-Tomasi corner point detection algorithm is used to detect feature points in the target image of the previous frame, specifically:

[0026] Calculate the change in brightness around the target pixel to be measured to determine the corner point object. The calculation formula is:

[0027] ;

[0028] in, =1 means the pixel is a corner point; =0 means the pixel is a non-corner point; is the set threshold; The minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point;

[0029] The minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point for:

[0030] ;

[0031] in, is the feature matrix The first eigenvalue of is the feature matrix The second eigenvalue of is the minimum value function; For the image in the sliding window Direction and Characteristic matrix of the second-order derivatives of the direction;

[0032] Feature Matrix for:

[0033] ;

[0034] in, For The sliding window of corner point detection centered on is the weighted function; The target image is on the horizontal axis Directional derivatives; The target image is on the vertical axis Directional derivative.

[0035] Preferably, in step S22, the displacement judgment threshold is set , filter out the effective feature points of the target image data, specifically:

[0036] Use optical flow information to filter out static feature points in the target scene and feature points with large tracking errors. It is necessary to set a displacement judgment threshold. , used for feature point filtering, the method for determining whether a feature point is valid is:

[0037] ;

[0038] in, is the displacement judgment threshold.

[0039] Preferably, in step S22, the effective feature points of the target image data are clustered, and the coordinates of the anchor frame of the target to be measured are determined by an improved DBSCAN clustering algorithm, specifically:

[0040] Set the neighborhood distance threshold of the target image data , the distance of the target image data is The threshold value of the number of target image data in the neighborhood of ; The effective feature points are clustered into clusters; each cluster is determined as an object to be tested, and it is determined that it has only one anchor box. The calculation method of cluster center point, anchor box width and height is:

[0041] ;

[0042] in, For the The central abscissa of the cluster; For the The central ordinate of each cluster; For the The width of the cluster anchor box; For the The height of the cluster anchor box; For the The number of feature points of a cluster; Indicates The first The horizontal coordinate movement value of each point; For the The first The vertical coordinate movement value of each point; For the The number of feature points of each cluster, ; is the maximum value function; is the minimum value function;

[0043] Further, the outlier point itself is regarded as a cluster. The coordinates of the center point of the cluster are the coordinates of the outlier point. The width and height of the cluster anchor box are:

[0044] ;

[0045] in, For the The width of the outlier; For the The height of the outlier point; is the neighborhood distance threshold of the target image data of the clustering algorithm; Number the outliers;

[0046] Since the clusters determined by the DBCSAN clustering algorithm have , outliers Then the target micro-animal bodies to be detected determined by the micro-motion detection algorithm are The final number of anchor boxes is determined indivual.

[0047] Preferably, the micro-motion detection anchor frame in step S3 And YOLO detection anchor box The specific method for obtaining the vertex coordinates is:

[0048] Micro motion detection anchor box The coordinates of the four vertices are:

[0049] ;

[0050] in, The coordinates of the upper left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the upper right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; is the horizontal coordinate of the center point of the anchor frame; is the ordinate of the center point of the anchor box; is the width of the anchor box; is the height of the anchor frame;

[0051] YOLO detection anchor box The coordinates of the four vertices are:

[0052] ;

[0053] in, The coordinates of the upper left vertex in the target image anchor box identified by the YOLO model; The coordinates of the upper right vertex in the target image anchor box identified by the YOLO model; The coordinates of the lower left vertex in the target image anchor box identified by the YOLO model; The coordinates of the lower right vertex in the target image anchor box identified by the YOLO model.

[0054] Preferably, the world coordinate system transformation model in step S4 is specifically:

[0055] The camera coordinate system is rotated and translated to obtain the six degrees of freedom of the camera in the gimbal-based coordinate system. The six degrees of freedom are converted to the camera orientation and expressed as an angle vector. At this time, the camera angle vector ( , ),in is the horizontal movement angle of the camera, is the vertical movement angle of the camera, the angle vector is an absolute coordinate, the horizontal movement angle of the camera is determined by the first rotation joint of the gimbal, and the vertical movement angle of the camera can be determined by the second rotation joint of the gimbal; the camera orientation vector is the center vector of the field of view, and the change in focal length maps the change in the camera field of view; the camera position is set at the origin of the coordinate system, and the image formed can be mapped on the spherical surface centered on the origin, then the field of view of the determined anchor frame is:

[0056] ;

[0057] in, For the The horizontal field of view of the anchor box; For the The vertical field of view of the anchor box; For the The minimum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The maximum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The vertical maximum angle of the orientation coordinate in the world coordinate system of the anchor frame; Respectively The vertical minimum angle of the orientation coordinate in the world coordinate system of the anchor frame;

[0058] The method for obtaining the horizontal minimum angle, horizontal maximum angle, vertical maximum angle, and vertical minimum angle of the orientation coordinates in the anchor frame world coordinate system is as follows:

[0059] ;

[0060] in, It is the horizontal field of view of the camera at the current magnification; is the vertical field of view of the camera at the current magnification; For the The minimum horizontal axis value of the anchor box in the pixel coordinate system; For the The maximum value of the horizontal axis of the anchor box in the pixel coordinate system; For the The minimum vertical axis value of the anchor box in the pixel coordinate system; For the The maximum value of the vertical axis of the anchor box in the pixel coordinate system; is the width in the camera pixel coordinate system; is the height in the camera pixel coordinate system; is the horizontal angle of the camera’s orientation; is the vertical angle of the camera;

[0061] The orientation coordinates of the center point of the anchor frame are:

[0062] ;

[0063] in, For the The horizontal angle of the anchor frame; For the The vertical angle of the anchor box; For the The orientation coordinates of the anchor box in the world coordinate system.

[0064] Preferably, the adaptive anchor frame variable magnification and zoom model in step S4 is specifically:

[0065] Camera horizontal field of view Perpendicular to the camera's field of view Can be adjusted according to focal length and the sensor size is determined as:

[0066] ;

[0067] in, is the horizontal field of view of the camera; is the vertical field of view of the camera; is the sensor width; is the sensor height; is the focal length, when the focal length When increasing, the viewing angle and Reduce, when the focal length When it decreases, the viewing angle increases;

[0068] The initial focal length is , the camera's current zoom ratio is , then the current focal length for:

[0069] ;

[0070] in, is the zoom ratio; is the initial focal length;

[0071] The zoom ratio The affected field of view is:

[0072] ;

[0073] in, is the width of the field of view; is the visual field height;

[0074] With the zoom ratio Increase, the focal length becomes larger, the field of view becomes smaller; when the zoom ratio When it decreases, the focal length decreases and the field of view angle increases;

[0075] For the determined anchor frame, the required vertical field of view and horizontal field of view are brought into the fitting function to determine the optimal zoom ratio of the anchor frame. for:

[0076] ;

[0077] in, is the best zoom ratio; is the optimal zoom factor of the anchor frame obtained by fitting the function within the vertical viewing angle range; They are the best zoom factors of the anchor frame obtained by fitting the function under the horizontal viewing angle range; The vertical field of view determined for the anchor box; The horizontal field of view determined for the anchor frame; is the fitted vertical field of view-magnification function; is the fitted horizontal field of view-magnification function.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] (1) The present invention combines machine learning with traditional image processing algorithms and hardware control algorithms, and can achieve high-precision detection, intelligent control, data recording and analysis, with a high degree of automation and great scalability.

[0080] (2) The present invention realizes dual-mode joint detection, performs preliminary screening through the YOLO detection model, and combines the micro-motion detection algorithm to detect dynamic objects to be detected farther away in the image. The results of the micro-motion detection algorithm are added to the YOLO detection results, and the detection results of the two are integrated to improve the accuracy of the detection results.

[0081] (3) The present invention converts the six-degree-of-freedom representation of the camera in world coordinates into an orientation vector, establishes a corresponding world spherical coordinate system, and realizes the conversion from the pixel coordinate system to the world coordinate system under any camera posture. At the same time, with the help of the data of the camera's different field of view ranges of 1-30 times, the optimal zoom factor under any camera field of view is fitted, effectively improving the working efficiency of the laser emitting device. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1It is a control block diagram of the laser positioning and emission method based on YOLO and micro-motion target detection of the present invention;

[0083] Figure 2 This is a specific application process diagram of the laser emitting device of the present invention;

[0084] Figure 3 is a functional relationship diagram of the horizontal and vertical viewing angle ranges of the camera of the present invention and the camera magnification;

[0085] Figure 4 It is a network structure diagram of the YOLOv8 target detection model of the present invention. DETAILED DESCRIPTION

[0086] The exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0087] The embodiment of the present invention takes the airport clearing target task as a specific analysis scenario and provides a laser positioning and emission method based on YOLO and micro-motion target detection, such as Figure 1 As shown, the target image data is obtained, and the target graphic anchor frame is identified by using the YOLO target detection model; the target image data in the target image anchor frame is identified by using the micro-motion target detection algorithm, and the coordinates of the target anchor frame to be measured are obtained; the target image anchor frame obtained by the micro-motion target detection algorithm and the number of target image anchor frames of the YOLO target detection model are integrated to complete the first recognition of the target to be measured; a world coordinate system transformation model and an adaptive anchor frame variable magnification and zoom model are established to complete the second recognition of the target to be measured, and a laser positioning and launching device is controlled; it includes:

[0088] Step S1: Obtain target image data and use the YOLOv8 target detection model to identify the target graphic anchor box.

[0089] Use an optical zoom camera to capture target image data as input to the YOLOv8 target detection model. The YOLOv8 target detection model is as follows: Figure 4 The figure shows the network structure diagram of the YOLOv8 target detection model of the present invention; it specifically includes: a feature extraction module, a feature enhancement and fusion module and an output module.

[0090] The feature extraction module is composed of multiple convolutional layers, activation layers, and pooling layers. It can convert the input image from raw pixel information to feature representation and extract the hierarchical features of the image from the image. The feature extraction module is one of the core components of the YOLOv8 target detection model. This module is composed of multiple convolutional layers, activation layers, and pooling layers. The main function of this module is to convert the input image from raw pixel information to feature representation with high-level semantics. Through this process, YOLOv8 can extract features of different levels from the image, gradually from low-level features (such as edges, corners, etc.) to high-level features (such as the outline and shape of the object). This provides rich contextual information for subsequent target classification and positioning. The feature extraction module is a crucial part of YOLOv8 target detection and provides the basis for accurate detection of the entire model.

[0091] The main function of the feature enhancement and fusion module is to further improve the detection performance of YOLOv8 in complex scenes. By introducing the bottom-up and top-down PAN feature pyramid modules to achieve multi-scale feature fusion, YOLOv8 can process large and small objects at the same time, enhancing the perception of objects of different scales. The accuracy of target detection is improved.

[0092] The output module uses an anchor-free detection head without a predefined anchor method, which directly predicts the position and size of the target to be tested and simplifies the model structure. At the same time, the head network is used to decouple the classification and detection processes, including loss calculation and target image anchor box screening. The coordinates and size of the target image anchor box are identified as:

[0093] ;

[0094] in, is the central horizontal coordinate of the predicted anchor box; is the center coordinate of the predicted anchor box; is the width of the predicted anchor box; is the height of the predicted anchor box; It is the central horizontal coordinate of the unit where the current detection frame is located; The center vertical coordinate of the unit where the current detection frame is located; is the horizontal axis offset predicted by the network; is the vertical axis offset predicted by the network; is the logarithmic horizontal scale value predicted by the network; The logarithmic vertical scale values ​​predicted by the network.

[0095] In special scenarios such as airports and power grids, the appearance of small animal targets may pose a major safety hazard, so the accuracy of the detection results is the top priority. However, for targets farther away, they are small and difficult to detect through ordinary target detection models. In order to solve the problem of YOLOv8 detecting smaller objects.

[0096] Step S2: Use the micro-motion target detection algorithm to identify the target image data acquired by the camera in step S1, and obtain the anchor frame coordinates of the target to be measured; the micro-motion detection algorithm is composed of the optical flow pyramid method and the improved DBSCAN clustering algorithm.

[0097] Step S21: Obtain the anchor frames of the two adjacent target image frames obtained in step S1, and use the Shi-Tomasi corner point detection algorithm to detect the feature points in the previous target image frame, specifically:

[0098] Calculate the change in brightness around the target pixel to be measured to determine the corner point object. The calculation formula is:

[0099] ;

[0100] in, =1 means the pixel is a corner point; =0 means the pixel is a non-corner point; is the set threshold; It is the minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point.

[0101] The minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point for:

[0102] ;

[0103] in, is the feature matrix The first eigenvalue of is the feature matrix The second eigenvalue of is the minimum value function; For the image in the sliding window Direction and Characteristic matrix of the second-order derivatives in the direction.

[0104] Feature Matrix for:

[0105] ;

[0106] in, For The sliding window of corner point detection centered on is the weighted function; The target image is on the horizontal axis Directional derivatives; The target image is on the vertical axis Directional derivative.

[0107] Combined with the optical flow pyramid technology, the optical flow information between two adjacent target image frames is calculated by using the feature points in two adjacent target image frames and the feature points in the previous target image:

[0108] ;

[0109] in, The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; The horizontal coordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the ordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the horizontal coordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system; It is the ordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system.

[0110] Step S22: Setting the displacement judgment threshold , filter out the effective feature points of the target image data, specifically: use the optical flow information to filter out the static feature points in the target scene to be measured and the feature points with large tracking errors, and set the displacement judgment threshold , used for feature point filtering, the method for determining whether a feature point is valid is:

[0111] ;

[0112] in, is the displacement judgment threshold.

[0113] Then, the effective feature points of the target image data are clustered, and the coordinates of the anchor frame of the target to be measured are determined by the improved DBSCAN clustering algorithm, specifically:

[0114] Set the neighborhood distance threshold of the target image data , the distance of the target image data is The threshold value of the number of target image data in the neighborhood of ; The effective feature points are clustered into clusters; each cluster is determined as a target to be detected, and it is estimated that it has only one anchor box. The center point of the cluster and the width and height of the anchor box are calculated as follows:

[0115] ;

[0116] in, For the The central abscissa of the cluster; For the The central ordinate of each cluster; For the The width of the cluster anchor box; For the The height of the cluster anchor box; For the The number of feature points of a cluster; Indicates The first The horizontal coordinate movement value of each point; For the The first The vertical coordinate movement value of each point; For the The number of feature points of each cluster, ; is the maximum value function; To find the minimum value function.

[0117] Further, the outlier point itself is regarded as a cluster, the coordinates of the center point of the cluster are the coordinates of the point, and the width and height of the anchor box of the cluster are:

[0118] ;

[0119] in, For the The width of the outlier; For the The height of the outlier point; is the neighborhood distance threshold of the target image data of the clustering algorithm; Number the outliers.

[0120] Since the clusters determined by the DBCSAN clustering algorithm have , outliers Then the target micro-animal bodies to be detected determined by the micro-motion detection algorithm are The final number of anchor boxes is determined indivual.

[0121] Step S3: Integrate the target image anchor frames obtained by the micro-motion target detection algorithm in step S2 and the number of target image anchor frames of the YOLOv8 target detection model in step S1 to complete the first recognition of the target to be detected; the integration rule is:

[0122] ;

[0123] in, For the A micro-motion detection anchor box; For the YOLOv8 detection anchor boxes; For the The micro-motion detection anchor box and the The intersection area of ​​​​YOLOv8 detection anchor boxes; For the The area of ​​the micro-motion detection anchor box; For the The micro-motion detection anchor frame and the The intersection-over-union ratio of YOLOv8 detection anchor boxes; For the delete operation; It is a counting symbol. When the condition in the brackets is met, the number increases by 1. is the first determination threshold; is the second determination threshold; Number the anchor boxes for YOLOv8 detection; It is a conditional judgment function; The total number of anchor boxes detected by YOLOv8.

[0124] Micro motion detection anchor box And YOLOv8 detection anchor box The specific method for obtaining the vertex coordinates is:

[0125] Micro motion detection anchor box The coordinates of the four vertices are:

[0126] ;

[0127] in, The coordinates of the upper left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the upper right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; is the horizontal coordinate of the center point of the anchor frame; is the ordinate of the center point of the anchor box; is the width of the anchor box; is the height of the anchor box.

[0128] YOLOv8 detection anchor box The coordinates of the four vertices are:

[0129] ;

[0130] in, The coordinates of the upper left vertex in the target image anchor box identified by the YOLOv8 model; The coordinates of the upper right vertex in the target image anchor box identified by the YOLOv8 model; The coordinates of the lower left vertex in the target image anchor box identified by the YOLOv8 model; The coordinates of the lower right vertex in the target image anchor box identified by the YOLOv8 model.

[0131] Step S4: Establish a world coordinate system transformation model and an adaptive anchor frame zoom model, complete the second recognition of the target to be measured, and control the laser positioning and launching device.

[0132] The anchor frame identified for the first time in step S3 is converted into the orientation coordinates in the world coordinate system, and sorted from low to high according to the horizontal angle of the orientation coordinates. The angle required for the laser positioning transmitting device to point to each anchor frame and the camera magnification are calculated through the world coordinate system transformation model and the adaptive anchor frame zoom model. The anchor frame obtained by the first identification is subjected to steps S1 and S2 for secondary magnification and identification according to the horizontal angle of the anchor frame, so as to further determine whether there is a target to be measured at the location of the anchor frame.

[0133] The world coordinate system transformation model is as follows: rotating and translating the camera coordinate system can obtain the six degrees of freedom representation of the camera in the gimbal-based coordinate system. The six degrees of freedom are converted to the camera orientation and expressed as an angle vector. At this time, the camera angle vector ( , ),in is the horizontal movement angle of the camera, is the vertical movement angle of the camera. This vector is an absolute coordinate. The horizontal movement angle of the camera is determined by the first rotation joint of the gimbal. The vertical movement angle of the camera can be determined by the second rotation joint of the gimbal. The camera orientation vector is the center vector of the field of view. The change in focal length maps the change in the camera field of view. The camera position is set at the origin of the coordinate system. The image formed can be mapped on the spherical surface centered on the origin. Then, the field of view of the determined anchor frame is:

[0134] ;

[0135] in, For the The horizontal field of view of the anchor box; For the The vertical field of view of the anchor box; For the The minimum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The maximum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The vertical maximum angle of the orientation coordinate in the world coordinate system of the anchor frame; Respectively The vertical minimum angle of the orientation coordinate in the world coordinate system of the anchor frame.

[0136] The method for obtaining the horizontal minimum angle, horizontal maximum angle, vertical maximum angle, and vertical minimum angle of the orientation coordinates in the anchor frame world coordinate system is as follows:

[0137] ;

[0138] in, It is the horizontal field of view of the camera at the current magnification; is the vertical field of view of the camera at the current magnification; For the The minimum horizontal axis value of the anchor box in the pixel coordinate system; For the The maximum value of the horizontal axis of the anchor box in the pixel coordinate system; For the The minimum vertical axis value of the anchor box in the pixel coordinate system; For the The maximum value of the vertical axis of the anchor box in the pixel coordinate system; is the width in the camera pixel coordinate system; is the height in the camera pixel coordinate system; is the horizontal angle of the camera’s orientation; is the vertical angle the camera is facing.

[0139] The orientation coordinates of the center point of the anchor frame are:

[0140] ;

[0141] in, For the The horizontal angle of the anchor frame; For the The vertical angle of the anchor box; For the The orientation coordinates of the anchor box in the world coordinate system.

[0142] Adaptive anchor frame zoom model, specifically: camera horizontal field of view angle Perpendicular to the camera's field of view Can be adjusted according to focal length and the sensor size is determined as:

[0143] ;

[0144] in, is the horizontal field of view of the camera; is the vertical field of view of the camera; is the sensor width; is the sensor height; is the focal length, when the focal length When increasing, the viewing angle and Reduce, when the focal length When it decreases, the viewing angle increases.

[0145] The initial focal length is , the camera's current zoom ratio is , then the current focal length for:

[0146] ;

[0147] in, is the zoom ratio; is the initial focal length.

[0148] The zoom ratio The affected field of view is:

[0149] ;

[0150] in, is the width of the field of view; The height of the field of view.

[0151] With the zoom ratio Increase, the focal length becomes larger, the field of view becomes smaller; when the zoom ratio When it is reduced, the focal length decreases and the field of view angle increases.

[0152] like Figure 3 The figure shows the functional relationship between the horizontal and vertical viewing angles of the camera and the camera magnification of the present invention, which is mainly used to determine the optimal zoom ratio of the initial recognition anchor frame; for the determined anchor frame, the required vertical field of view range and horizontal field of view range are brought into the fitting function to determine the optimal zoom ratio of the anchor frame, and the optimal zoom ratio for:

[0153] ;

[0154] in, is the best zoom ratio; is the optimal zoom factor of the anchor frame obtained by fitting the function within the vertical viewing angle range; They are the best zoom factors of the anchor frame obtained by fitting the function under the horizontal viewing angle range; The vertical field of view determined for the anchor box; The horizontal field of view determined for the anchor frame; is the fitted vertical field of view-magnification function; is the fitted horizontal field of view-magnification function.

[0155] If the secondary recognition result is still the target to be measured, the laser is emitted. Otherwise, the next anchor frame position is turned to continue the judgment until all the anchor frames recognized for the first time complete the secondary recognition. The rotation angle of the second joint of the gimbal is reset to zero and the camera zoom factor is changed to 1. The rotation angle of the first joint of the gimbal is randomly assigned, and the gimbal performs regular monitoring.

[0156] The experiment of the embodiment of the present invention respectively compares the performance of the target detection model used in the YOLOv8 model under the data collected under the field of view of an ordinary camera and the data under the field of view of a pan-tilt zoom camera. The data split used for model evaluation is a training set: validation set ratio of 9:1. The experimental data is shown in Table 1.

[0157] Table 1 YOLOv8 model experimental data table

[0158] equipment Precision / % Recall / % mAP50 / % mAP50-95 / % Ordinary camera + gimbal 92 66 85 45 Zoom camera + gimbal 90 81 90 57

[0159] Experimental results show that although the accuracy of the dataset obtained by ordinary cameras in the YOLOv8 model is 2% higher than that of the gimbal camera, the dataset of the gimbal system performs better in other key performance indicators, especially in recall, mAP50 and mAP50-95, which are 15%, 5% and 12% higher than those of ordinary cameras, respectively.

[0160] Based on the above experiments, the embodiments of the present invention also compare the effects of the gimbal system embedded with the micro-motion detection algorithm of this article and the gimbal system not embedded with the micro-motion detection algorithm in the task of detecting moving targets. The experimental data are shown in Table 2.

[0161] Table 2 Experimental data of micro-motion detection algorithm

[0162] Model Total number of tests Detection accuracy Missed detection rate Actual total quantity With micro-movement 1345 0.957 0.043 1405 No micro-motion 1281 0.911 0.089 1405

[0163] In terms of detection accuracy, the accuracy of the PTZ system embedded with the micro-motion detection algorithm has increased from 0.911 to 0.957, an increase of 4.6%. This improvement shows that the micro-motion detection algorithm can effectively distinguish between moving targets and background interference by identifying the tiny motion characteristics of moving targets, reducing false detections, and thus significantly improving the recognition accuracy of the PTZ system. In addition, the missed detection rate has also dropped from 0.089 to 0.043, a decrease of 4.6%, further indicating that the algorithm is more reliable in capturing actual moving targets, especially those whose movements are not obvious or in complex backgrounds. This improvement ensures the integrity and accuracy of monitoring.

[0164] The second aspect of the present invention proposes a laser positioning and emitting device equipped with a laser positioning and emitting method based on YOLO and micro-motion target detection, which includes: a 30x optical zoom camera, a programmable secondary development gimbal and a laser emitting device.

[0165] The 30x optical zoom camera can magnify and zoom the external environment 1-30 times without losing focus, and capture global and local target image data in the external environment in high definition; the 30x optical zoom camera model is HM303818S5, the optical zoom ratio can reach 30 times, the camera pixel is 5 million pixels, the focal length range is 4.7-141mm, the zoom and control communication interface supports RS232 / 485, the camera communication interface is RJ45 network cable, the field of view angle range is 2.0°-66.1°, the camera frame rate is 20fps, and it can magnify and zoom the external environment 1-30 times without losing focus, and capture global and local image information in the environment in high definition.

[0166] Programmable secondary development pan / tilt, supports embedded secondary development and network communication, can accurately control the direction and angle of the camera, and realize all-round real-time monitoring; programmable secondary development pan / tilt, model BSL-C40-L, supports embedded secondary development, communication mode supports RS485 / USB / network communication, pan / tilt positioning accuracy is 0.1°, horizontal rotation speed is 10° / second, vertical rotation speed is 5° / second, supports rotation speed variation, has built-in angle query function, supports input angle control function, supports internal routing of conductive slip rings, can accurately control the direction and angle of the camera, and realize all-round real-time monitoring.

[0167] The laser emitting device is used to irradiate the target to be measured with laser to achieve the effect; the specific application process of the laser emitting device is as follows Figure 2 As shown, the specific steps include:

[0168] Step S51: The PTZ randomly moves and transmits scene video stream data to the PC.

[0169] Step S52: Detect suspicious moving targets through the YOLOv8 model and micro-motion detection algorithm, integrate anchor frame information, and integrate data to guide the gimbal to move in a directional manner.

[0170] Step S53: The gimbal and camera integrate data from the initially identified anchor frames to perform directional movement and magnification changes, obtain an enlarged high-definition picture of the suspicious moving target, and perform secondary recognition in combination with the YOLOv8 algorithm.

[0171] Step S54: If the recognition result of step S53 is a moving target, the laser is emitted to drive away the moving target. After driving away the moving target, step S55 is entered. If not, it is determined whether there are any remaining integrated anchor frames that have not been recognized for the second time. If so, it is transferred to step S53. Otherwise, the gimbal and camera are adjusted back to the initial state, and the gimbal is driven to move randomly for closed-loop detection.

[0172] Step S55: Determine whether there are any remaining integrated anchor frames that have not been re-identified. If so, proceed to step S53. Otherwise, adjust the gimbal and camera back to the initial state, drive the gimbal to move randomly, and perform closed-loop detection.

[0173] The beneficial effects of the embodiments of the present invention are as follows: compared with traditional physical, acoustic and chemical methods, the present invention combines machine learning with traditional image processing algorithms and hardware control algorithms, and can achieve high-precision detection, intelligent control, data recording and analysis, with a high degree of automation and greater scalability; it avoids the fact that traditional methods are effective in the short term, but the effect will be weakened in the long term. Further analysis avoids the ecological balance that may be destroyed by acoustic, chemical and other methods of removing targets. The embodiments of the present invention use the YOLOV8 detection model for preliminary screening, and at the same time combine the self-developed micro-motion detection algorithm to detect dynamic objects to be detected farther away in the image, and add the results of the micro-motion detection algorithm to the YOLOv8 detection results. The detection results of the two are effectively integrated to guide the subsequent secondary recognition tasks of the gimbal and camera. It is found through experiments that in this joint detection mode, the detection accuracy is improved to a certain extent, and the control laser emission device realizes the function, and the actual use effect is good.

[0174] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A laser positioning and emission method based on YOLO and micro-motion target detection, characterized in that: It includes: S1: Obtain target image data and use the YOLO target detection model to identify the target graphic anchor frame; Use an optical zoom camera to capture the target image data as the input of the YOLO target detection model, and identify the coordinates and size of the target image anchor frame as follows: ； in, is the central horizontal coordinate of the predicted anchor box; is the center coordinate of the predicted anchor box; is the width of the predicted anchor box; is the height of the predicted anchor box; It is the central horizontal coordinate of the unit where the current detection frame is located; The center vertical coordinate of the unit where the current detection frame is located; is the horizontal axis offset predicted by the network; is the vertical axis offset predicted by the network; is the logarithmic horizontal scale value predicted by the network; is the logarithmic vertical scale value predicted by the network; S2: Use the micro-motion target detection algorithm to identify the micro-motion targets in the image data acquired by the camera and obtain the anchor frame coordinates of the target to be detected; the micro-motion detection algorithm includes the optical flow pyramid method and the improved DBSCAN clustering algorithm; S21: Obtain the anchor frames of the two adjacent target image frames obtained in step S1, and use the Shi-Tomasi corner detection algorithm to detect the feature points in the first target image frame; combine the optical flow pyramid method to use the feature points in the two adjacent target image frames and the first target image frame to calculate the optical flow information of the feature point set between two adjacent continuous target image frames: ； in, The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; The feature point is located along The displacement difference of the coordinate axis before and after the frame motion; is the horizontal coordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the ordinate value of the corresponding feature point in the current frame image in the pixel coordinate system; is the horizontal coordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system; is the ordinate value of the corresponding feature point in the previous frame image in the pixel coordinate system; S22: Setting the displacement judgment threshold , screen and obtain effective feature points of the target image data; cluster the effective feature points of the target image data, and determine the coordinates of the anchor frame of the target to be measured by using the improved DBSCAN clustering algorithm; S3: Integrate the target image anchor frames obtained by the micro-motion target detection algorithm in step S2 and the number of target image anchor frames of the YOLO target detection model in step S1 to complete the first recognition of the target to be detected; the integration rule is: ； in, For the A micro-motion detection anchor box; For the YOLO detection anchor boxes; For the The micro-motion detection anchor box and the The intersection area of ​​the YOLO detection anchor boxes; For the The area of ​​the micro-motion detection anchor box; For the The micro-motion detection anchor frame and the The intersection-over-union ratio of YOLO detection anchor boxes; For the delete operation; is a counting symbol; is the first determination threshold; is the second determination threshold; Number the YOLO detection anchor box; It is a conditional judgment function; The total number of anchor boxes detected by YOLO; S4: Establish a world coordinate system transformation model and an adaptive anchor frame zoom model, complete the second recognition of the target to be measured, and control the laser positioning and launching device; The anchor frame identified for the first time in step S3 is converted into the orientation coordinates in the world coordinate system, and is sorted from low to high according to the horizontal angle of the orientation coordinates. The angle required for the laser positioning and transmitting device to point to each anchor frame and the camera magnification ratio are calculated through the world coordinate system transformation model and the adaptive anchor frame magnification and zoom model. The anchor frames obtained by the first identification are sequentially magnified and identified in step S1 according to the horizontal angle of the anchor frame to further determine whether there is a target to be measured at the location of the anchor frame. If the secondary identification result of the anchor frame is the target to be measured, the laser is emitted, otherwise the next anchor frame position is turned to continue the judgment until all the anchor frames identified for the first time complete the secondary identification, the second joint rotation angle of the gimbal is reset to zero and the camera zoom factor becomes 1, the first joint rotation angle of the gimbal is randomly assigned, and the gimbal performs regular monitoring.

2. The laser positioning and emission method based on YOLO and micro-motion target detection according to claim 1, characterized in that: The YOLO target detection model in step S1 includes: a feature extraction module, a feature enhancement and fusion module, and an output module; The feature extraction module consists of convolutional layers, activation layers, and pooling layers; it can convert the input image from raw pixel information into feature representation and extract the hierarchical features of the image from the image; The feature enhancement and fusion module realizes feature fusion by introducing the bottom-up and top-down PAN feature pyramid modules to enhance the perception ability of the target; The output module adopts the anchor-free detection head without predefined anchor box method to directly predict the position and size of the target to be tested; at the same time, the classification and detection processes are decoupled using the head network, and each head specializes in one task, including target category classification and target image anchor box screening.

3. The laser positioning and emission method based on YOLO and micro-motion target detection according to claim 1, characterized in that: In step S21, the Shi-Tomasi corner point detection algorithm is used to detect the feature points in the target image of the previous frame, specifically: Calculate the change in brightness around the target pixel to be measured to determine the corner point object. The calculation formula is: ； in, =1 means the pixel is a corner point; =0 means the pixel is a non-corner point; is the set threshold; The minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point; The minimum eigenvalue corresponding to the feature matrix of the sliding window corresponding to the pixel point for: ； in, is the feature matrix The first eigenvalue of is the feature matrix The second eigenvalue of is the minimum value function; For the image in the sliding window Direction and Characteristic matrix of the second-order derivatives of the direction; Feature Matrix for: ； in, For The sliding window of corner point detection centered on is the weighted function; The target image is on the horizontal axis Directional derivatives; The target image is on the vertical axis Directional derivative.

4. The laser positioning and emission method based on YOLO and micro-motion target detection according to claim 1, characterized in that: In step S22, the displacement judgment threshold is set , filter out the effective feature points of the target image data, specifically: Use optical flow information to filter out static feature points in the target scene and feature points with large tracking errors. It is necessary to set a displacement judgment threshold. , used for feature point filtering, the method for determining whether a feature point is valid is: ； in, is the displacement judgment threshold.

5. The laser positioning and emission method based on YOLO and micro-motion target detection according to claim 1, characterized in that: In step S22, the effective feature points of the target image data are clustered, and the coordinates of the anchor frame of the target to be measured are determined by the improved DBSCAN clustering algorithm, specifically: Set the neighborhood distance threshold of the target image data , the distance of the target image data is The threshold value of the number of target image data in the neighborhood of ; The effective feature points are clustered into clusters; each cluster is determined as an object to be tested, and it is determined that it has only one anchor box. The calculation method of cluster center point, anchor box width and height is: ； in, For the The central abscissa of the cluster; For the The central ordinate of each cluster; For the The width of the cluster anchor box; For the The height of the cluster anchor box; For the The number of feature points of a cluster; Indicates The first The horizontal coordinate movement value of each point; For the The first The vertical coordinate movement value of each point; For the The number of feature points of each cluster, ; is the maximum value function; is the minimum value function; Further, the outlier point itself is regarded as a cluster. The coordinates of the center point of the cluster are the coordinates of the outlier point. The width and height of the cluster anchor box are: ； in, For the The width of the outlier; For the The height of the outlier point; is the neighborhood distance threshold of the target image data of the clustering algorithm; Number the outliers; Since the clusters determined by the DBCSAN clustering algorithm have , outliers Then the target micro-animal bodies to be detected determined by the micro-motion detection algorithm are The final number of anchor boxes is determined indivual.

6. The laser positioning and emitting method based on YOLO and micro-motion target detection according to claim 1, characterized in that: The micro-motion detection anchor box in step S3 And YOLO detection anchor box The specific method for obtaining the vertex coordinates is as follows: Micro motion detection anchor box The coordinates of the four vertices are: ； in, The coordinates of the upper left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the upper right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower left vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; The coordinates of the lower right vertex in the anchor frame of the dynamic object to be detected identified by the micro-motion detection algorithm; is the horizontal coordinate of the center point of the anchor frame; is the ordinate of the center point of the anchor box; is the width of the anchor box; is the height of the anchor frame; YOLO detection anchor box The coordinates of the four vertices are: ； in, The coordinates of the upper left vertex in the target image anchor box identified by the YOLO model; The coordinates of the upper right vertex in the target image anchor box identified by the YOLO model; The coordinates of the lower left vertex in the target image anchor box identified by the YOLO model; The coordinates of the lower right vertex in the target image anchor box identified by the YOLO model.

7. The laser positioning and emission method based on YOLO and micro-motion target detection according to claim 1, characterized in that: The world coordinate system transformation model in step S4 is specifically: The camera coordinate system is rotated and translated to obtain the six degrees of freedom of the camera in the gimbal-based coordinate system. The six degrees of freedom are converted to the camera orientation and expressed as an angle vector. At this time, the camera angle vector ( , ),in is the horizontal movement angle of the camera, is the vertical movement angle of the camera, the angle vector is an absolute coordinate, the horizontal movement angle of the camera is determined by the first rotation joint of the gimbal, and the vertical movement angle of the camera can be determined by the second rotation joint of the gimbal; the camera orientation vector is the center vector of the field of view, and the change in focal length maps the change in the camera field of view; the camera position is set at the origin of the coordinate system, and the image formed can be mapped on the spherical surface centered on the origin, then the field of view of the determined anchor frame is: ； in, For the The horizontal field of view of the anchor box; For the The vertical field of view of the anchor box; For the The minimum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The maximum horizontal angle of the orientation coordinates in the world coordinate system of the anchor frame; For the The vertical maximum angle of the orientation coordinate in the world coordinate system of the anchor frame; Respectively The vertical minimum angle of the orientation coordinate in the world coordinate system of the anchor frame; The method for obtaining the horizontal minimum angle, horizontal maximum angle, vertical maximum angle, and vertical minimum angle of the orientation coordinates in the anchor frame world coordinate system is as follows: ； in, It is the horizontal field of view of the camera at the current magnification; is the vertical field of view of the camera at the current magnification; For the The minimum horizontal axis value of the anchor box in the pixel coordinate system; For the The maximum value of the horizontal axis of the anchor box in the pixel coordinate system; For the The minimum vertical axis value of the anchor box in the pixel coordinate system; For the The maximum value of the vertical axis of the anchor box in the pixel coordinate system; is the width in the camera pixel coordinate system; is the height in the camera pixel coordinate system; is the horizontal angle of the camera’s orientation; is the vertical angle of the camera; The orientation coordinates of the center point of the anchor frame are: ； in, For the The horizontal angle of the anchor frame; For the The vertical angle of the anchor box; For the The orientation coordinates of the anchor box in the world coordinate system.

8. The laser positioning and emitting method based on YOLO and micro-motion target detection according to claim 1, characterized in that: The adaptive anchor frame variable magnification and zoom model in step S4 is specifically: Camera horizontal field of view Perpendicular to the camera's field of view Can be adjusted according to focal length and the sensor size is determined as: ； in, is the horizontal field of view of the camera; is the vertical field of view of the camera; is the sensor width; is the sensor height; is the focal length, when the focal length When increasing, the viewing angle and Reduce, when the focal length When it decreases, the viewing angle increases; The initial focal length is , the camera's current zoom ratio is , then the current focal length for: ； in, is the zoom ratio; is the initial focal length; The zoom ratio The affected field of view is: ； in, is the width of the field of view; is the visual field height; With the zoom ratio Increase, the focal length becomes larger, the field of view becomes smaller; when the zoom ratio When it decreases, the focal length decreases and the field of view angle increases; For the determined anchor frame, the required vertical field of view and horizontal field of view are brought into the fitting function to determine the optimal zoom ratio of the anchor frame. for: ； in, is the best zoom ratio; is the optimal zoom factor of the anchor frame obtained by fitting the function within the vertical viewing angle range; They are the best zoom factors of the anchor frame obtained by fitting the function under the horizontal viewing angle range; The vertical field of view determined for the anchor box; The horizontal field of view determined for the anchor frame; is the fitted vertical field of view-magnification function; is the fitted horizontal field of view-magnification function.

Images