Wind turbine side intrusion target detection and alarm method based on Leiyu fusion

Abstract

The invention relates to intrusion target detection, in particular to a wind turbine side intrusion target detection alarm method based on thunder-vision fusion, which comprises the following steps: acquiring track information returned by a current frame radar, and screening a target track for thunder-sphere linkage; calculating the linkage position of a dome camera according to the target track, and recognizing an intrusion target in a thunder-dome linkage mode; pedestrian re-identification is carried out on the invasion targets according to the thunder ball linkage result, and whether the invasion targets are the same or not is judged; optimizing an intrusion alarm strategy according to a pedestrian re-identification result, and storing an intrusion image; the technical scheme provided by the invention can effectively overcome the defect that the intrusion target on the fan side of the new energy power station is difficult to accurately detect in the prior art.

Description

Technical Field

[0001] The present invention relates to intrusion target detection, and in particular to an intrusion target detection and alarm method on a wind turbine side based on radar and vision fusion. Background Art

[0002] With the rapid development of information technology, the transformation of the global energy structure and the growing demand for clean energy, unattended power stations have become a major demand for power intelligence in recent years, especially unattended new energy power stations. However, new energy power stations are usually located in remote areas, occupy a large area, and are in a complex and changeable environment, which brings many challenges to the operation and maintenance of new energy power stations, especially the safety of new energy power stations, which has become a key factor restricting its development.

[0003] In order to realize unmanned operation of new energy power stations, it is essential to sense intrusion targets on the wind turbine side. Traditional methods mostly use video recognition technology to sense and drive away intrusion targets. However, a single technical means is difficult to cope with the complex and changeable security scenarios of new energy power stations. There is a lack of effective environmental perception and early warning mechanisms. Relying solely on video recognition technology cannot meet the needs of intelligent perception, large-scale detection, and front-end and back-end intelligent linkage. At the same time, the lack of a reasonable alarm strategy will cause security personnel to be exhausted. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method for detecting and alarming intrusion targets on the wind turbine side based on radar and visual fusion, which can effectively overcome the defect of the prior art that it is difficult to accurately detect intrusion targets on the wind turbine side of a new energy power station.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A method for detecting and alarming intrusion targets on a wind turbine side based on radar and visual fusion comprises the following steps:

[0009] S1. Obtain the track information returned by the radar in the current frame and select the target track for the radar-ball linkage;

[0010] S2. Calculate the ball-camera linkage position according to the target track, and use the thunder-ball linkage method to identify the intruding target;

[0011] S3, perform pedestrian re-identification on the intruder target based on the result of the thunderball linkage, and determine whether the intruder target is the same;

[0012] S4. Optimize the intrusion alarm strategy based on the pedestrian re-identification results and save the intrusion image.

[0013] Preferably, S1 obtains the track information returned by the radar in the current frame, and screens the target track for the radar-ball linkage, including:

[0014] S11. Obtain the i-th track T in the current k-th frame according to the echo signal received by the millimeter-wave radar i (k) = {X i (k),Y i (k),V i (k),θ i (θ),N i (k),F i (k)};

[0015] Among them, X i (k), Y i (k) are the track T i (k) X, Y axis distance, V i (k) is the track T i (k) speed, the speed in the direction close to the radar is positive, θ i (k) is the track T i (k) Angle, N i (k) is the track T i (k) radar tracking number, F i (k) is the track T i (k) signal strength;

[0016] S12. According to the acquired track information, select the intermediate track T within the specified defense area. i ′(k):

[0017] T i ′(k)={X i ′(k),Y i ′(k)∈T i (k)|X L ≤Y i ′(k)≤X R And Y L ≤Y i ′(k)≤Y R};

[0018] Among them, X i '(k), Y i '(k) are the intermediate tracks T i '(k)'s distance in the X and Y axes, X L , X R They are the left and right boundary values ​​of the defense zone in the X-axis direction, and the Y L , Y R They are the lower boundary value and upper boundary value of the defense zone in the Y-axis direction respectively;

[0019] S13, calculate the intermediate track T i '(k)'s threat level W i (k):

[0020]

[0021] Among them, V max 、F max are the maximum speed and maximum signal strength of the threat target, V max =2m / s, F max =150, α, β, γ, δ are the unified weighting coefficients of the distance, speed, signal strength, and driving direction of the threat target, respectively, α=0.3, β=0.2, γ=0.3, δ=0.2;

[0022] S14, obtain the track list C that has been tracked by the radar and the ball linkage, according to the number of radar tracking times N i (k) For the intermediate track T i '(k) Sort from large to small, select the middle track with the largest number of radar tracking times and compare it with the track list C. If there is a track that has not been linked to the radar, then this track is used as the target track T i ”(k) Perform thunder and ball linkage, otherwise, select the track with the greatest threat as the target track T i ” (k) Carry out thunder and ball linkage.

[0023] Preferably, in S11, the i-th track T in the current k-th frame is obtained according to the echo signal received by the millimeter wave radar. i (k) including:

[0024] Based on the echo signal received by the millimeter-wave radar, the point information is obtained, the DBSCAN algorithm is used to condense the point, and the condensed points are used to participate in the point-to-flight association;

[0025] Among them, the millimeter wave radar is a 24GHz millimeter wave radar with a distance resolution of 1.34m and a speed resolution of 0.138m / s.

[0026] Preferably, in S2, the ball camera linkage position is calculated according to the target track, and the intrusion target is identified by using the thunder-ball linkage method, including:

[0027] S21, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i” (k)}, calculate the horizontal angle increment ΔP:

[0028]

[0029] Among them, X i ”(k), Y i ” (k) are the target track T i ” (k) distance in the X and Y axes, V i ”(k) is the target track T i ” (k) speed, the speed close to the radar is positive, θ i ”(k) is the target track T i ” (k) angle, N i ”(k) is the target track T i ”(k) radar tracking number, F i ”(k) is the target track T i ” (k) signal strength, (X c ,Y c ) is the position coordinate of the preset ball camera illumination center point in the radar coordinate system, which is obtained during the radar-ball linkage calibration;

[0030] S22, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i ” (k)}, calculate the zoom factor increment ΔT:

[0031]

[0032] Among them, At is the focal length variable parameter of the ball camera, which is a fixed value and is calculated according to the zoom multiple increment ΔT of the ball camera during the calibration of the thunder-ball linkage;

[0033] S23, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i ” (k)}, calculate the vertical angle increment ΔZ:

[0034]

[0035] Among them, Az is the pitch angle parameter of the ball camera, which is a fixed value and is calculated according to the vertical angle increment ΔZ of the ball camera during the calibration of the thunder-ball linkage;

[0036] S24, calculate the updated PTZ parameters of the ball camera:

[0037] P1=P0+ΔP;

[0038] T1=T0+ΔT;

[0039] Z1=Z0+ΔZ;

[0040] Among them, (P0, T0, Z0) is the preset PTZ parameters of the center point of the ball camera, and (P1, T1, Z1) is the updated PTZ parameters of the ball camera;

[0041] S25. The ball camera rotates the camera according to the updated PTZ parameters (P1, T1, Z1) of the ball camera, and uses the thunder-ball linkage method to identify the intrusion target.

[0042] Preferably, in S3, pedestrian re-identification is performed on the intruder target according to the thunderball linkage result to determine whether the intruder target is the same, including:

[0043] S31, obtain the real-time video stream of the ball camera, and use YOLOv7 to perform real-time target detection;

[0044] In order to improve efficiency, the real-time target detection is performed by using the frame-by-frame processing method, that is, two frames of images are extracted from the real-time video stream of the ball camera every second for real-time target detection;

[0045] S32, obtaining the position of the target bounding box, cropping the target bounding box, and inputting it into the OSNet network to extract target features to obtain a target feature vector;

[0046] S33, calculating the Euclidean distance D between the intrusion targets in the two frames of images according to the target feature vector:

[0047]

[0048] in, is the target feature vector f in image a a The value in the jth dimension, a=1,2,…,N, is the target feature vector f in image b b The value in the jth dimension, b = 1, 2, ..., N, where N is the dimension of the target feature vector;

[0049] S34. Calculate the similarity Sim according to the Euclidean distance D:

[0050]

[0051] The value range of similarity Sim is [0,1], Sim=1 means completely similar, Sim=0 means completely dissimilar, D max = max {D};

[0052] S35, performing continuity judgment on the tracked radar track ID number, if the radar track ID numbers of the previous and next frames change, the continuity value Con is set to 0, otherwise the continuity value Con is set to 1;

[0053] S36, perform pedestrian re-identification according to the similarity Sim and the continuity value Con, and calculate the re-identification value Sam of the intruder target:

[0054] Sam=α1·Sim+β1*Con;

[0055] Among them, α1 and β1 are the unified weighting coefficients of the similarity Sim and the continuity value Con, α1 = 0.4, β1 = 0.6;

[0056] S37. When the re-identification value Sam of the intrusion target is greater than 0.75, it is considered that the intrusion target is the same.

[0057] Preferably, in S4, the intrusion alarm strategy is optimized according to the pedestrian re-identification result, and the intrusion image is saved, including:

[0058] S41, if the intrusion target is the same, only one alarm report is made, and the threat level of the intrusion target is continuously updated according to the radar information, and the threat level of the intrusion target is judged by multiple frames using the sliding window method. If the threat level of N frames out of M frames is greater than the preset threat level threshold, repeated alarm reports are made at a certain interval;

[0059] Among them, M=5, N=4, the preset threat level threshold is 0.75, and the interval time is 20s;

[0060] S42. For two frames of images with a re-identification value Sam>0.75 of the intrusion target, perform Laplace transformation on the two frames of images respectively, retain only the clearest photo, and update the database photo and the front-end alarm display photo in real time to optimize the front-end display effect and reduce the performance pressure of the back-end server;

[0061] S43. When the alarm is reported, the video during the period of Thunderball linkage is intercepted and saved. Security personnel can quickly understand the situation on the scene by viewing the video records, make more accurate and effective response decisions, and facilitate later review and evidence collection.

[0062] (III) Beneficial effects

[0063] Compared with the prior art, the wind turbine side intrusion target detection and alarm method based on radar and visual fusion provided by the present invention has the following beneficial effects:

[0064] 1) Combined with millimeter-wave radar, it can detect over a large area, use YOLOv7 for real-time target detection, and use OSNet network for target feature extraction. By combining image recognition technology, it can eliminate the influence of radar false alarm on false target tracking in thunder and ball linkage, effectively reduce the false alarm rate of the system, and continuously track static or micro-moving targets to ensure the continuity and stability of safety monitoring of new energy power stations;

[0065] 2) Optimize the intrusion alarm strategy based on the pedestrian re-identification results to reduce the alarm fatigue of security personnel caused by frequent alarms and false alarms, while reducing the operating pressure of the system and improving the accuracy and robustness of the security system. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0067] Figure 1 It is a schematic diagram of the process of the present invention;

[0068] Figure 2 It is a schematic diagram of the installation of the radar and the ball camera in the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] A wind turbine side intrusion target detection and alarm method based on radar and visual fusion, such as Figure 1 As shown, S1, obtain the track information returned by the radar in the current frame, and filter the target track for the radar-ball linkage, specifically including:

[0071] S11. Obtain the i-th track T in the current k-th frame according to the echo signal received by the millimeter-wave radar i (k) = {X i (k),Y i(k),V i (k),θ i (θ),N i (k),F i (k)};

[0072] Among them, X i (k), Y i (k) are the track T i (k) X, Y axis distance, V i (k is the track T i (k) speed, the speed in the direction close to the radar is positive, θ i (k) is the track T i (k) Angle, N i (k) is the track T i (k) radar tracking number, F i (k) is the track T i (k) signal strength;

[0073] S12. According to the acquired track information, select the intermediate track T within the specified defense area. i ′(k):

[0074] T i ′(k)={X i ′(k),Y i ′(k)∈T i (k)|X L ≤Y i ′(k)≤X R And Y L ≤Y i ′(k)≤Y R};

[0075] Among them, X i '(k), Y i '(k) are the intermediate tracks T i '(k)'s distance in the X and Y axes, X L , X R They are the left and right boundary values ​​of the defense zone in the X-axis direction, and the Y L , Y R They are the lower and upper boundary values ​​of the defense zone in the Y-axis direction, respectively. Figure 2 As shown;

[0076] S13, calculate the intermediate track T i '(k)'s threat level W i (k):

[0077]

[0078] Among them, Vmax 、F max are the maximum speed and maximum signal strength of the threat target, V max =2m / s, F max =150, α, β, γ, δ are the unified weighting coefficients of the distance, speed, signal strength, and driving direction of the threat target, respectively, α=0.3, β=0.2, γ=0.3, δ=0.2;

[0079] S14, obtain the track list C that has been tracked by the radar and the ball linkage, according to the number of radar tracking times N i (k) For the intermediate track T i '(k) Sort from large to small, select the middle track with the largest number of radar tracking times and compare it with the track list C. If there is a track that has not been linked to the radar, then this track is used as the target track T i ”(k) Perform thunder and ball linkage, otherwise, select the track with the greatest threat as the target track T i ” (k) Carry out thunder and ball linkage.

[0080] Specifically, in S11, the i-th track T in the current k-th frame is obtained according to the echo signal received by the millimeter-wave radar. i (k) including:

[0081] Based on the echo signal received by the millimeter-wave radar, the point information is obtained, and the DBSCAN algorithm is used to condense the point (because the resolution of the 24GHz millimeter-wave radar is high, the target has multiple radar reflection points, so the DBSCAN algorithm is needed to condense the point), and the condensed points are used to participate in point-to-flight association;

[0082] Among them, the millimeter wave radar is a 24GHz millimeter wave radar with a distance resolution of 1.34m and a speed resolution of 0.138m / s.

[0083] S2. Calculate the ball-camera linkage position based on the target track, and use the thunder-ball linkage method to identify the intrusion target, including:

[0084] S21, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i ” (k)}, calculate the horizontal angle increment ΔP:

[0085]

[0086] Among them, X i ”(k), Yi ” (k) are the target track T i ” (k) distance in the X and Y axes, V i ”(k) is the target track T i ” (k) speed, the speed in the direction close to the radar is positive, θ i ”(k) is the target track T i ” (k) angle, N i ”(k) is the target track T i ”(k) radar tracking number, F i ”(k) is the target track T i ” (k) signal strength, (X c ,Y c ) is the position coordinate of the preset ball camera illumination center point in the radar coordinate system, which is obtained during the radar-ball linkage calibration;

[0087] S22, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i ” (k)}, calculate the zoom factor increment ΔT:

[0088]

[0089] Among them, At is the focal length variable parameter of the ball camera, which is a fixed value and is calculated according to the zoom multiple increment ΔT of the ball camera during the calibration of the thunder-ball linkage;

[0090] S23, according to the target track T i ”(k)={X i ”(k),Y i ”(k),V i ”(k),θ i ”(k),N i ”(k),F i ” (k)}, calculate the vertical angle increment ΔZ:

[0091]

[0092] Among them, Az is the pitch angle parameter of the ball camera, which is a fixed value and is calculated according to the vertical angle increment ΔZ of the ball camera during the calibration of the thunder-ball linkage;

[0093] S24, calculate the updated PTZ parameters of the ball camera:

[0094] P1=P0+ΔP;

[0095] T1=T0+ΔT;

[0096] Z1=Z0+ΔZ;

[0097] Among them, (P0, T0, Z0) is the preset PTZ parameters of the center point of the ball camera, and (P1, T1, Z1) is the updated PTZ parameters of the ball camera;

[0098] S25. The ball camera rotates the camera according to the updated PTZ parameters (P1, T1, Z1) of the ball camera, and uses the thunder-ball linkage method to identify the intrusion target.

[0099] S3: Re-identify the intruder according to the result of the thunderball linkage to determine whether the intruder is the same, including:

[0100] S31, obtain the real-time video stream of the ball camera, and use YOLOv7 to perform real-time target detection;

[0101] In order to improve efficiency, the real-time target detection is performed by using the frame-by-frame processing method, that is, two frames of images are extracted from the real-time video stream of the ball camera every second for real-time target detection;

[0102] S32, obtaining the position of the target bounding box, cropping the target bounding box, and inputting it into the OSNet network to extract target features to obtain a target feature vector;

[0103] S33, calculating the Euclidean distance D between the intrusion targets in the two frames of images according to the target feature vector:

[0104]

[0105] in, is the target feature vector f in image a a The value in the jth dimension, a=1,2,…,N, is the target feature vector f in image b b The value in the jth dimension, b = 1, 2, ..., N, where N is the dimension of the target feature vector;

[0106] S34. Calculate the similarity Sim according to the Euclidean distance D:

[0107]

[0108] The value range of similarity Sim is [0,1], Sim=1 means completely similar, Sim=0 means completely dissimilar, D max =max{D};

[0109] S35, performing continuity judgment on the tracked radar track ID number, if the radar track ID numbers of the previous and next frames change, the continuity value Con is set to 0, otherwise the continuity value Con is set to 1;

[0110] S36, perform pedestrian re-identification according to the similarity Sim and the continuity value Con, and calculate the re-identification value Sam of the intruder target:

[0111] Sam = α1*Sim+β1*Con;

[0112] Among them, α1 and β1 are the unified weighting coefficients of the similarity Sim and the continuity value Con, α1 = 0.4, β1 = 0.6;

[0113] S37. When the re-identification value Sam of the intrusion target is greater than 0.75, it is considered that the intrusion target is the same.

[0114] S4. Optimize the intrusion alarm strategy based on the pedestrian re-identification results and save the intrusion image, including:

[0115] S41, if the intrusion target is the same, only one alarm report is made, and the threat level of the intrusion target is continuously updated according to the radar information, and the threat level of the intrusion target is judged by multiple frames using the sliding window method. If the threat level of N frames out of M frames is greater than the preset threat level threshold, repeated alarm reports are made at a certain interval;

[0116] Among them, M=5, N=4, the preset threat level threshold is 0.75, and the interval time is 20s;

[0117] S42. For two frames of images with a re-identification value Sam>0.75 of the intrusion target, perform Laplace transformation on the two frames of images respectively, retain only the clearest photo, and update the database photo and the front-end alarm display photo in real time to optimize the front-end display effect and reduce the performance pressure of the back-end server;

[0118] S43. When the alarm is reported, the video during the period of Thunderball linkage is intercepted and saved. Security personnel can quickly understand the situation on the scene by viewing the video records, make more accurate and effective response decisions, and facilitate later review and evidence collection.

[0119] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind turbine side intrusion target detection and alarm method based on radar and visual fusion, characterized in that: The following steps are involved: S1. Obtain the track information returned by the radar in the current frame and select the target track for the radar-ball linkage; S2. Calculate the ball-camera linkage position according to the target track, and use the thunder-ball linkage method to identify the intruding target; S3, perform pedestrian re-identification on the intruder target based on the result of the thunderball linkage, and determine whether the intruder target is the same; S4. Optimize the intrusion alarm strategy based on the pedestrian re-identification results and save the intrusion image.

2. The wind turbine side intrusion target detection and alarm method based on radar and visual fusion according to claim 1 is characterized by: In S1, the track information returned by the radar in the current frame is obtained, and the target track for the radar-ball linkage is screened, including: S11. Obtain the i-th track T in the current k-th frame according to the echo signal received by the millimeter-wave radar i (k) = {X i (k), Y i (k), V i (k), θ i (k), N i (k), F i (k)}; Among them, X i (k), Y i (k) are the track T i (k) X, Y axis distance, V i (k) is the track T i (k) speed, the speed close to the radar is positive, θ i (k) is the track T i (k) Angle, N i (k) is the track T i (k) radar tracking number, F i (k) is the track T i (k) signal strength; S12. According to the acquired track information, select the intermediate track T within the specified defense area. i ′(k): T i ′(k) = {X i ′(k), Y i ′(k) ∈ T i (k)|X L ≤ X i ′(k) ≤ X R and Y L ≤ Y i ′(k) ≤ Y R}; Among them, X i ′(k), Y i ′(k) are the intermediate tracks T i ′(k) in the X and Y axis directions, X L , X R They are the left and right boundary values ​​of the defense zone in the X-axis direction, and the Y L , Y R They are the lower boundary value and upper boundary value of the defense zone in the Y-axis direction respectively; S13, calculate the intermediate track T i The threat level of ′(k) is W i (k): Among them, V max 、F max are the maximum speed and maximum signal strength of the threat target, V max =2m / s, F max =150, α, β, γ, δ are the unified weighting coefficients of the distance, speed, signal strength, and driving direction of the threat target, respectively, α=0.3, β=0.2, y=0.3, δ=0.2; S14, obtain the track list C that has been tracked by the radar and the ball linkage, according to the number of radar tracking times N i (k) For the intermediate track T i ′(k) are sorted from large to small, and the intermediate track with the largest number of radar tracking times is selected for comparison with the track list C. If there is a track that has not been linked to the radar, then this track is used as the target track T i ″(k) Perform thunder and ball linkage, otherwise, select the track with the greatest threat as the target track T i ″(k) Conduct thunder and ball linkage.

3. The wind turbine side intrusion target detection and alarm method based on radar and visual fusion according to claim 2 is characterized by: In S11, the i-th track T in the current k-th frame is obtained according to the echo signal received by the millimeter-wave radar. i (k) including: Based on the echo signal received by the millimeter-wave radar, the point information is obtained, the DBSCAN algorithm is used to condense the point, and the condensed points are used to participate in the point-to-flight association; Among them, the millimeter wave radar is a 24GHz millimeter wave radar with a distance resolution of 1.34m and a speed resolution of 0.138m / s.

4. The wind turbine side intrusion target detection and alarm method based on radar and visual fusion according to claim 2 is characterized by: In S2, the ball-camera linkage position is calculated according to the target track, and the intrusion target is identified by the thunder-ball linkage method, including: S21, according to the target track T i ″(k)={X i ″(k),Y i ″(k), V i ″(k),θ i ″(k), N i ″(k), F i ″(k)}, calculate the horizontal angle increment ΔP: Among them, X i ″(k), Y i ″(k) are the target track T i ″(k) distance in the X and Y axes, V i ″(k) is the target track T i ″(k) speed, the speed close to the radar is positive, θ i ″(k) is the target track T i ″(k) angle, N i ″(k) is the target track T i ″(k) radar tracking number, F i ″(k) is the target track T i ″(k) signal strength, (X c , Y c ) is the position coordinate of the preset ball camera illumination center point in the radar coordinate system, which is obtained during the radar-ball linkage calibration; S22, according to the target track T i ″(k)={X i ″(k),Y i ″(k), V i ″(k),θ i ″(k), N i ″(k), F i ″(k)}, calculate the zoom factor increment ΔT: Among them, At is the focal length variable parameter of the ball camera, which is a fixed value and is calculated according to the zoom multiple increment ΔT of the ball camera during the calibration of the thunder-ball linkage; S23, according to the target track T i ″(k)={X i ″(k),Y i ″(k), V i ″(k),θ i ″(k), N i ″(k), F i ″(k)}, calculate the vertical angle increment ΔZ: Among them, Az is the pitch angle parameter of the ball camera, which is a fixed value and is calculated according to the vertical angle increment ΔZ of the ball camera during the calibration of the thunder-ball linkage; S24, calculate the updated PTZ parameters of the ball camera: P1=P0+AP; T1=T0+ΔT; Z1=Z0+ΔZ; Among them, (P0, T0, Z0) are the preset PTZ parameters of the center point of the ball camera, and (P1, T1, Z1) are the updated PTZ parameters of the ball camera; S25. The ball camera rotates the camera according to the updated PTZ parameters (P1, T1, Z1) of the ball camera, and uses the thunder-ball linkage method to identify the intrusion target.

5. The wind turbine side intrusion target detection and alarm method based on radar and visual fusion according to claim 4 is characterized in that: In S3, pedestrian re-identification is performed on the intruder target based on the results of the thunderball linkage to determine whether the intruder target is the same, including: S31, obtain the real-time video stream of the ball camera, and use YOLOv7 to perform real-time target detection; In order to improve efficiency, the real-time target detection is performed by using the frame-by-frame processing method, that is, two frames of images are extracted from the real-time video stream of the ball camera every second for real-time target detection; S32, obtaining the position of the target bounding box, cropping the target bounding box, and inputting it into the OSNet network to extract target features to obtain a target feature vector; S33, calculating the Euclidean distance D between the intrusion targets in the two frames of images according to the target feature vector: in, is the target feature vector f in image a a The value in the jth dimension, a = 1, 2, ..., N, is the target feature vector f in image b b The value in the jth dimension, b = 1, 2, ..., N, where N is the dimension of the target feature vector; S34. Calculate the similarity Sim according to the Euclidean distance D: The value range of similarity Sim is [0,1], Sim=1 means completely similar, Sim=0 means completely dissimilar, D max =max{D}; S35, performing continuity judgment on the tracked radar track ID number, if the radar track ID numbers of the previous and next frames change, the continuity value Con is set to 0, otherwise the continuity value Con is set to 1; S36, perform pedestrian re-identification according to the similarity Sim and the continuity value Con, and calculate the re-identification value Sam of the intruder target: Sam = α1*Sim+β1*Con; Among them, α1 and β1 are the unified weighting coefficients of the similarity Sim and the continuity value Con, α1 = 0.4, β1 = 0.6; S37. When the re-identification value Sam of the intrusion target is greater than 0.75, it is considered that the intrusion target is the same.

6. The wind turbine side intrusion target detection and alarm method based on radar and visual fusion according to claim 5 is characterized by: In S4, the intrusion alarm strategy is optimized based on the pedestrian re-identification results, and the intrusion image is saved, including: S41, if the intrusion target is the same, only one alarm report is made, and the threat level of the intrusion target is continuously updated according to the radar information, and the threat level of the intrusion target is judged by multiple frames using the sliding window method. If the threat level of N frames out of M frames is greater than the preset threat level threshold, repeated alarm reports are made at a certain interval; Among them, M=5, N=4, the preset threat level threshold is 0.75, and the interval time is 20s; S42. For two frames of images with a re-identification value Sam>0.75 of the intrusion target, perform Laplace transformation on the two frames of images respectively, retain only the clearest photo, and update the database photo and the front-end alarm display photo in real time to optimize the front-end display effect and reduce the performance pressure of the back-end server; S43. When the alarm is reported, the video during the period of Thunderball linkage is intercepted and saved. Security personnel can quickly understand the situation on the scene by viewing the video records, make more accurate and effective response decisions, and facilitate later review and evidence collection.

Images