A method and system for optoelectronic scanning capture tracking unmanned aerial vehicles based on coarse orientation

Abstract

The application provides a kind of photoelectric scanning capture tracking unmanned aerial vehicle method and system based on rough orientation, method includes: equipment mark is relatively;Unmanned aerial vehicle is searched using search strategy, search strategy includes: general search, extension search and depth search;Unmanned aerial vehicle is searched using manual or automatic mode;Carry out far and near airspace U-shaped search;For the characteristics of unmanned aerial vehicle target and system, the YOLOv7 intelligent detection network based on deep learning is improved and optimized, the forward inference is executed to each search area image, the coordinate information of unmanned aerial vehicle target in image is obtained, and the image detection and recognition result is obtained;Iterate image detection and recognition result, and according to detection and recognition result, half recursive approximation unmanned aerial vehicle real position is approached, and unmanned aerial vehicle real position is approached;Unmanned aerial vehicle is locked and tracked.The application solves the technical problems of complex logic, low error adaptability and high application cost.

Description

Technical Field

[0001] This invention relates to the field of photoelectric reconnaissance, and more specifically to a method and system for capturing and tracking unmanned aerial vehicles (UAVs) based on coarse orientation photoelectric scanning. Background Technology

[0002] In recent years, the civilian drone market, especially the consumer drone market, has experienced explosive growth, with numerous "black flight" incidents and increasing uncontrollable risks posed by drones. This underscores the growing importance of drone detection and countermeasures. Common drone detection methods include radar, passive detection, and electro-optical solutions. Among these, a combination of passive detection and electro-optical vision verification is a popular approach.

[0003] The existing invention patent document CN105353772A, entitled "A Visual Servo Control Method for Mobile Target Positioning and Tracking in Unmanned Aerial Vehicles," establishes a geodetic coordinate system, a body coordinate system, a camera coordinate system, an image coordinate system, and a body-geodetic transition coordinate system. Based on the relationships between these coordinate systems and the target's imaging sequence, it calculates the target's positioning and the attitude angle setpoints for target tracking and flight path tracking, thus completing visual servo control. As can be seen from the specific implementation of this prior art, it uses a tracking camera and leverages positioning information, a globe model, and a central perspective projection model to process the target's coordinate system and trajectory data to obtain the servo control information required for target locking. This prior art algorithm involves numerous model architectures and various coordinate systems, resulting in a complex data processing process and a large number of parameters, which reduces the robustness and applicability of the existing solution.

[0004] The existing invention patent document CN106154262A, entitled "Anti-UAV Detection System and Control Method Thereof," includes a camera with an optical lens, a display, an automatic tracking servo mechanism for controlling the camera's movement, a radar device, a main controller, a jammer, and an image recognition decoder storing UAV image information. It employs an optoelectronic surveillance radar device combining a radar device and an optical tracking system, achieving integration of radar detection, optical image recognition, and electromagnetic interference. From the specific implementation details of this existing solution, it is clear that this technology combines UAV tracking radar and an optical tracking system, using methods such as phase accumulation to detect and match radar detection signals and optoelectronic recognition signals for UAV detection. Servo control and UAV signal detection are then performed based on the detection and matching data. This existing technology requires multiple types of devices and specific components and processing logic to fuse UAV signals acquired by different detection devices into a single signal. This technology has high operating costs, and using different types of equipment reduces the error adaptability of the directional control data.

[0005] In summary, existing technologies suffer from technical problems such as complex logic, low error adaptability, and high application costs. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to solve the problems of complex logic, low error adaptability and high application cost.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solution: A method for capturing and tracking unmanned aerial vehicles based on coarse orientation photoelectric scanning includes:

[0008] S1. Perform calibration operation on the photoelectric scanning equipment according to the preset calibration angle, calibration rotation direction and angle error parameters;

[0009] S2. Using no less than two preset search strategies, and based on the search angle range and tolerance angle error parameters of each preset search strategy, search for UAV targets to obtain the number of UAV targets and the distance between UAV targets. The preset search strategies include: general search strategy, extended search strategy and depth search strategy.

[0010] S3. Select the search method based on the number of drone targets, and then perform a round-robin search and capture of drone targets. The search methods include: automatic search and manual search.

[0011] S4. Based on the distance to the UAV target, the UAV target is processed in a hierarchical manner to obtain a U-shaped search target classification. The U-shaped search target classification includes: UAVs in the far airspace and UAVs in the near airspace. According to the U-shaped search target classification, U-shaped searches are performed on UAVs in the far airspace and UAVs in the near airspace respectively to obtain image detection and recognition results.

[0012] S5. Based on the two-dimensional image information in the search area, use a deep learning-based artificial intelligence algorithm to perform forward reasoning on the area and obtain the detection and recognition results of the UAV target in each search area.

[0013] S6. Traverse the image detection and recognition results. Based on the detection and recognition results, use the binary recursion method to approximate the actual position of the drone to overcome the problem of inconsistent drone recognition results and gimbal status acquisition time caused by image delay.

[0014] S7. Lock onto and track the drone target based on its actual location.

[0015] This invention, guided by the coarse orientation of a passive detection device for unmanned aerial vehicles (UAVs), enables a gimbal camera to automatically search for and lock onto UAV images in the relevant airspace. It also performs round-robin searches and capture of multiple UAVs simultaneously. This invention allows optoelectronic devices to search for and capture UAV images and perform lock-on tracking under the guidance of the detection device. The method has significant adaptability to errors in the coarse orientation angle of the UAV, eliminates the need for high-precision positioning equipment such as radar, and offers high cost-effectiveness.

[0016] In a more specific technical solution, in step S1, a coarse orientation angle is obtained using a preset passive orientation device.

[0017] In more specific technical solutions, the angle error parameters include: the angle measurement error of the direction finding equipment, the calibration error, and the angle synchronization difference between linked devices.

[0018] In a more specific technical solution, in step S2, the tolerance angle error range of the search angle range of the general search strategy includes: [-10°, +10°]; the tolerance angle error range of the extended search strategy includes: [-20°, +20°]; and the tolerance angle error range of the depth search strategy includes: [-30°, +30°].

[0019] Within a specific horizontal distance between the UAV and the optoelectronic device, and within an error of ±30° between the coarse orientation angle and the actual direction angle of the UAV, the optoelectronic device can search and capture UAV images in the relevant airspace under the guidance of the coarse orientation angle and perform lock-on tracking, thereby achieving automatic confirmation of optoelectronic vision.

[0020] In a more specific technical solution, in step S3, the system autonomously selects the drone with the strongest signal strength for automatic search. The automatic search methods include photoelectric search.

[0021] In a more specific technical solution, step S3 includes: in the manual search operation, manually selecting the UAV target to be detected by the passive detection device, guiding the preset optoelectronic device to perform optoelectronic search in the direction of the airspace where the UAV target is located.

[0022] This invention enables photoelectric equipment to patrol, search, capture, and track multiple drones in the airspace by selecting automatic or manual methods to detect drone targets, thus meeting the photoelectric confirmation requirements for multi-target aerial intrusion.

[0023] In a more specific technical solution, step S4 includes:

[0024] S41. Perform a U-shaped search on the near-airspace UAV to obtain the near-airspace UAV identification results;

[0025] S42. Perform a U-shaped search on the UAV in the distant airspace to obtain the identification results of the UAV in the distant airspace.

[0026] This invention employs a near-to-far search and recognition strategy, searching for UAV images in a U-shaped pattern, sequentially across different airspaces. During near-range and long-range searches, the camera magnification, number of search rounds, and search time are adapted to the UAV targets in the near and far airspaces, respectively. This multi-level near-far search mechanism ensures full coverage of both near and far airspaces within the search angle.

[0027] In a more specific technical solution, step S5 includes:

[0028] S51. Improve and optimize the YOLOv7 target detection network based on deep learning, taking into account the target characteristics and system characteristics of UAVs;

[0029] S52. Load the optimized detection network structure into the system and load the weight parameters after iterative training into the target detection network.

[0030] S53. Extract the two-dimensional image within the search area and perform image preprocessing according to network requirements;

[0031] S54. Feed the preprocessed image into the object detection network and perform forward inference;

[0032] S55. Analyze the features output by the network to obtain the coordinates and confidence level of the UAV target, and obtain the image detection and recognition results for each region.

[0033] In a more specific embodiment, step S51 includes:

[0034] S511. Remove the 32x downsampling large target detection head from the original YOLOv7 network, while retaining the feature fusion of deep features in the neck area;

[0035] S512, with the addition of a 4x downsampling detection head for extremely small targets;

[0036] S513. Delete the direct connection between shallow features and the small target detection head.

[0037] Based on deep learning, this invention optimizes the current best-performing YOLOv7 target detection network structure. Taking into account the small proportion and indistinct features of drone targets in images, the feature extraction strategy of the backbone structure and the structure of the neck and head parts of the network are improved in turn, thereby improving the detection rate of drone targets and the robustness of drone detection under different backgrounds and features.

[0038] In a more specific technical solution, step S6 includes:

[0039] S61. Position the UAV detection gimbal;

[0040] S62. Utilize the drone to detect the gimbal, and based on the detection and identification results, use a binary recursive method to approximate the actual position of the drone.

[0041] In a more specific technical solution, step S62 includes:

[0042] S621. Use an electro-optical gimbal to start searching from the first point in the nearby airspace and stop at the second point. Wait in the distant airspace for a preset time and determine whether the drone image was detected during the search phase from the first point to the second point.

[0043] S622. If not, wait for the end and proceed to the next stage of the search;

[0044] S623. If so, wait for the end, perform a binary recursive search operation on the drone, and move the gimbal camera from the second point to the third point;

[0045] S624. Determine whether a drone image was detected during the search phase from the second point to the third point;

[0046] S625. If so, recursively search from point 3 to point 4.

[0047] S626. If not, perform a recursive search during the search phase from the first point to the second point until the drone is found.

[0048] After the gimbal is in place, the present invention waits for and traverses the image detection and recognition results, and then uses a binary recursive approach to approximate the actual position of the UAV based on the detection and recognition results, thereby improving the accuracy of UAV position detection.

[0049] In a more specific technical solution, a photoelectric scanning capture and tracking UAV system based on coarse orientation includes:

[0050] The calibration module is used to perform calibration operations on the photoelectric scanning equipment according to preset calibration angle, calibration rotation direction and angle error parameters;

[0051] The difference strategy search module is used to search for UAV targets by utilizing no less than two preset search strategies, based on the search angle range and tolerance angle error parameters of each preset search strategy, thereby obtaining the number of UAV targets and the distance between UAV targets. The preset search strategies include: general search strategy, extended search strategy and deep search strategy. The difference strategy search module is connected to the calibration module.

[0052] The polling capture module is used to select the search method according to the number of UAV targets, and to perform polling search and capture of UAV targets. The search methods include automatic search and manual search. The polling capture module is connected to the differential strategy search module.

[0053] The U-shaped search module is used to process UAV targets in a hierarchical manner according to the distance of the UAV target to obtain U-shaped search target classification. The U-shaped search target classification includes: UAVs in the far airspace and UAVs in the near airspace. According to the U-shaped search target classification, U-shaped search is performed on UAVs in the far airspace and UAVs in the near airspace respectively to obtain image detection and recognition results. The U-shaped search module is connected to the difference strategy search module.

[0054] The image detection module is used to improve and optimize the YOLOv7 intelligent detection network based on deep learning, taking into account the characteristics of UAV targets and systems. It performs forward inference on the images of each search area to obtain information such as the coordinates of UAV targets in the image and obtains the image detection and recognition results.

[0055] The binary approximation module is used to traverse the image detection and recognition results. Based on the detection and recognition results, it uses a binary recursive method to approximate the actual position of the UAV. The binary approximation module is connected to the U-shaped search module and the image detection module.

[0056] The lock-on tracking module is used to lock onto and track the drone target based on the actual position of the drone. The folded approximation module is connected to the folded approximation module.

[0057] Compared with existing technologies, this invention has the following advantages: Guided by the coarse orientation of the UAV passive detection device, the gimbal camera automatically searches for and locks onto UAV images in the relevant airspace. It simultaneously performs round-robin searches and captures multiple UAVs. This invention enables photoelectric devices to search for and capture UAV images and perform lock-on tracking under the guidance of the detection device. The method has a large tolerance for errors in the coarse orientation angle of the UAV, does not require guidance from high-precision positioning equipment such as radar, and is cost-effective.

[0058] Within a specific horizontal distance between the UAV and the optoelectronic device, and within an error of ±30° between the coarse orientation angle and the actual direction angle of the UAV, the optoelectronic device can search and capture UAV images in the relevant airspace under the guidance of the coarse orientation angle and perform lock-on tracking, thereby achieving automatic confirmation of optoelectronic vision.

[0059] This invention enables photoelectric equipment to patrol, search, capture, and track multiple drones in the airspace by selecting automatic or manual methods to detect drone targets, thus meeting the photoelectric confirmation requirements for multi-target aerial intrusion.

[0060] This invention employs a near-to-far search and recognition strategy, searching for UAV images in a U-shaped pattern, sequentially across different airspaces. During near-range and long-range searches, the camera magnification, number of search rounds, and search time are adapted to the UAV targets in the near and far airspaces, respectively. This multi-level near-far search mechanism ensures full coverage of both near and far airspaces within the search angle.

[0061] This invention optimizes the current state-of-the-art deep learning-based object detection network YOLOv7, and improves the structure of each part of the network to take into account the characteristics of drone targets in images. For each region of the scene, its two-dimensional image is acquired and fed into the improved and optimized object detection network to obtain accurate image detection and recognition results for each region.

[0062] This invention, after the gimbal is in place, waits for and iterates through the image detection and recognition results, and then uses a binary recursive approach to approximate the actual position of the UAV based on the detection and recognition results, thereby improving the accuracy of UAV position detection. This invention solves the technical problems of complex logic, low error adaptability, and high application cost existing in the prior art. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the basic steps of a photoelectric scanning method for capturing and tracking a drone based on coarse orientation, according to Embodiment 1 of the present invention.

[0064] Figure 2 This is a schematic diagram of the multi-level near and far search of the photoelectric gimbal in Embodiment 1 of the present invention;

[0065] Figure 3 This is a schematic diagram of the UAV photoelectric search, capture, lock, and tracking interface in Embodiment 1 of the present invention.

[0066] Figure 4 This is a schematic diagram of the improved UAV target detection and recognition network structure according to Embodiment 1 of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example 1

[0069] like Figure 1 As shown, the present invention provides a method for capturing and tracking unmanned aerial vehicles (UAVs) based on coarse orientation photoelectric scanning, which includes the following basic steps:

[0070] S1, Equipment Standards;

[0071] In this embodiment, the passive direction finding device has a built-in compass, which defaults to true north as the detection 0 degrees, and rotates clockwise one full circle to 360 degrees. The photoelectric device has user-customizable horizontal and vertical zero-degree angles. The horizontal zero-degree angle needs to be set to true north, and the vertical zero-degree angle to horizontal, to complete the calibration of the two devices. Compass error, direction finding angle error, and calibration error will all directly affect the linkage effect between the devices.

[0072] S2. Use search strategies to search for drones. Search strategies include: general search, extended search and deep search.

[0073] S3. Search for drones using manual or automatic methods;

[0074] S4. Conduct a U-shaped search in both near and far airspace;

[0075] S5. Based on the two-dimensional image information in the search area, use a deep learning-based artificial intelligence algorithm to perform forward reasoning on the area and obtain the detection and recognition results of the UAV target in each search area.

[0076] S6. Traverse the image detection and recognition results, and approximate the true position of the drone by binary recursion based on the detection and recognition results.

[0077] S7. Lock onto and track the drone.

[0078] In this embodiment, the operation of electro-optical search and capture of the UAV is carried out by a passive direction-finding device providing a coarse orientation angle. The electro-optical device searches for the UAV within the relevant airspace of the coarse orientation angle. Angle error refers to the angle difference caused by the angle measurement error of the direction-finding device, calibration error, and angle synchronization between linked devices, assuming that the angular coordinate systems of the direction-finding device and the electro-optical device have been calibrated, using the angular coordinate system of the electro-optical device as a reference. This error generally does not exceed 10 degrees and has a maximum of 30 degrees.

[0079] In this embodiment, with true north on the horizontal plane as 0 degrees, a clockwise rotation is recorded as 360 degrees. After the equipment calibration is completed, even if both the direction-finding device and the optoelectronic gimbal device point north at 0 degrees, there is still an angular error between the UAV's coarse orientation angle and its actual direction. This error mainly includes the UAV angle measurement error of the direction-finding device, the equipment calibration error, and the synchronization error of the linked angles of the two devices, generally within ±30 degrees.

[0080] In this embodiment, the electro-optical search and capture method for drones is divided into three strategies: general search, extended search, and depth search. The general search strategy has a search angle range of 20°, meaning it tolerates an angle error of ±10°. The extended search strategy has a search angle range of 40°, meaning it tolerates an angle error of ±20°. The depth search strategy has a search angle range of 60°, meaning it tolerates an angle error of ±30°. In this embodiment, the electro-optical search and capture process for tracking drones first determines the search strategy, defaulting to general search, which has a search angle range of 20°.

[0081] In this embodiment, the present invention can simultaneously perform round-robin search and capture of multiple UAVs. This embodiment includes both automatic and manual search. Automatic search involves the system autonomously selecting the UAV with the strongest signal strength (i.e., the closest UAV) for photoelectric search. Manual search involves manually selecting a UAV detected by a passive detection device, thereby guiding the photoelectric device to perform a photoelectric search in that airspace. In this embodiment, the second step of the photoelectric search and capture tracking process for UAVs defaults to automatic search; in the case of a single UAV, automatic search is equivalent to manual search.

[0082] In this embodiment, the photoelectric search for capturing the UAV involves a multi-level search at both near and far distances, and a U-shaped, airspace-by-air search for UAV images. During near-range searches, the camera magnification is low, the number of search rounds is small, and the time taken is short. During long-range searches, the camera magnification is high, the number of search rounds is large, and the time taken is long. This multi-level search ensures full coverage of the near and far airspace within the search angle.

[0083] In this embodiment, due to camera decoding, transmission, algorithm delays, etc., the completion time of the image detection and recognition results is delayed compared to the gimbal positioning time. After the gimbal is in place, it is necessary to wait for and traverse the image detection and recognition results, and then recursively approximate the actual position of the drone based on the detection and recognition results.

[0084] In this embodiment, after the system searches for and captures the drone, it can lock the drone, causing the gimbal to move with the drone's position, ensuring that the drone is always centered in the frame at a specific size.

[0085] In this embodiment, the search for the UAV is divided into multi-level searches at near and far distances, using a U-shaped, airspace-by-air search method to examine the UAV image. During near-range searches, the camera magnification is low, resulting in fewer search rounds and shorter search time. During long-range searches, the camera magnification is high, resulting in more search rounds and longer search time. This multi-level search ensures full coverage of both near and far airspace within the search angle.

[0086] In this embodiment, a 6.6-350mm focal length, 53x zoom laser gimbal camera, and a two-stage near-far search strategy are used as an example. During the near-far search, the zoom level is set to 5. At this point, the camera's field of view is 10°. To cover the 20° search range, two searches are required. The search begins from the upper left corner of the near-far airspace at a vertical angle of 40° and proceeds downwards until the vertical angle reaches 0° (the horizontal angle). The gimbal then moves horizontally 10° to the right and begins searching upwards from the lower right corner at a vertical angle of 0° (the horizontal angle) until the vertical angle reaches 40°.

[0087] When searching at long distances, with zoom level 10, the camera's field of view is 5°. To cover a 20° search range, four searches are required. Start searching from the upper left corner of the nearby airspace at a vertical angle of 40°, moving downwards until reaching 0° (the horizontal angle). Then, move the gimbal horizontally 5° to the right, starting from 0° (the horizontal angle) and searching upwards until reaching 40° (the vertical angle). Repeat this process twice, for a total of four searches.

[0088] The search strategy described in the above embodiments can meet the task of UAV photoelectric search and capture within a horizontal distance of 300 meters from the photoelectric device. If it is necessary to cover a longer distance, the two-level search can be changed to a multi-level search.

[0089] In this embodiment, due to camera decoding, transmission, algorithm delays, etc., the image detection and recognition results are delayed compared to the gimbal positioning time. After the gimbal is in place, it is necessary to wait for and traverse the image detection and recognition results, and then recursively approximate the actual position of the drone based on the detection and recognition results.

[0090] like Figure 2 As shown, in this embodiment, the photoelectric gimbal starts searching from point a in the nearby airspace and stops at point b. At this point, it does not immediately turn right to execute the next stage of the search, but waits at a distance, such as for 200ms. The waiting time is determined based on the overall delay time. While waiting, it checks whether a drone image has been detected in the stage from a to b. If not, the wait ends and the next stage of the search begins. If so, the wait ends and a binary recursive search begins. In the recursive search stage, the gimbal camera moves from b to c, stops waiting, and checks whether a drone image has been detected in the stage from b to c. If so, the recursive search continues from c to d. If not, the recursive search continues in the segment from a to c until a drone is found. If the recursive search does not find a drone, the next stage of the search begins.

[0091] The photoelectric detection and identification method for drones is based on an AI-powered intelligent recognition algorithm for drone images. This algorithm can automatically identify and select multi-rotor drones within a scene. In this embodiment, the drone image intelligent recognition algorithm can be a drone intelligent recognition algorithm based on the deep learning YOLOv7 model.

[0092] like Figure 3 As shown, in this embodiment, an improved YOLOv7 model is used to detect and identify drone targets. The network consists of four parts, from top to bottom: input part, backbone part, neck part, and head part. The input part represents the images of each search region transmitted to the network; the backbone part represents the main network, used to extract image features of different sizes at different stages. The closer to the input, the shallower the feature; the neck part is used to fuse multi-stage, multi-scale information; the head part is used to output features with different downsampling rates, responsible for detecting targets of different sizes. The shallow features extracted from the backbone network correspond to the low downsampling rate small target detection head of the head part, used to detect targets that occupy a smaller proportion of the image, while the deep features correspond to the high downsampling rate large target detection head, used to detect targets that occupy a larger proportion of the image.

[0093] In this embodiment, considering the size of the drone target and the target tracking characteristics, the drone will not occupy too large a proportion in the input image, and it is also necessary to distinguish it from interference factors such as fog and leaves. Therefore, after comprehensively comparing various deep learning-based target detection networks, the YOLOv7 network, which currently has the best performance, was selected, and targeted structural improvements and optimizations were made to adapt to the specific drone target detection requirements.

[0094] In this embodiment, the large target detection head with high downsampling rate of YOLOv7 is removed, but the deep features with high downsampling rate are still retained for feature fusion in the neck region. Under the premise of ensuring that the network can still perform sufficient information fusion, the computational resource overhead of the network can be effectively reduced and the operating efficiency of the system can be improved.

[0095] In this embodiment, compared to the original YOLOv7's 8x, 16x, and 32x downsampling detection heads, a new 4x downsampling rate microtarget detection head is added to detect drone targets that are too small in proportion to the target area due to occlusion, scaling, or other issues. The extracted shallow features are used in the neck feature fusion and head feature output; therefore, the direct connection between shallow features and the microtarget detection head is removed, improving the system's detection rate for micro-targets and increasing the system's robustness in complex environments.

[0096] In the above embodiments, the network structure deletion is indicated by a dashed rectangle with the text "delete" in the figure, and the network structure addition is indicated by a dashed rectangle with the text "add" in the figure, and the added part is marked with a thick solid line.

[0097] The improved and optimized network outputs features FEATURE1, FEATURE2, and FEATURE3 through three detection heads. By parsing these features, or obtaining information such as the category, confidence level, and coordinates of multiple targets in the input image, the image detection and recognition results within the search area can be finally obtained.

[0098] like Figure 4 As shown in this embodiment, photoelectric search, capture, lock, and track drone footage means that after the system searches for and captures the drone, it can lock onto it, causing the gimbal to move with the drone's position to ensure that the drone is always centered in the frame at a specific size.

[0099] In the state of drone lock-on tracking, the system detects and analyzes in real time the horizontal position, vertical position and size of the drone image in the gimbal camera's view, and then adjusts the gimbal angle and camera zoom value to ensure that the drone is always located in the exact center of the camera's view at a specific size.

[0100] In summary, this invention, guided by the coarse orientation of a passive detection device for UAVs, enables a gimbal camera to automatically search for and lock onto UAV images in the relevant airspace. It also performs round-robin search and capture of multiple UAVs. This invention allows optoelectronic devices to search for and capture UAV images and perform lock-on tracking under the guidance of the detection device. The method has significant adaptability to errors in the coarse orientation angle of the UAV, does not require guidance from high-precision positioning equipment such as radar, and offers high cost-effectiveness.

[0101] Within a specific horizontal distance between the UAV and the optoelectronic device, and within an error of ±30° between the coarse orientation angle and the actual direction angle of the UAV, the optoelectronic device can search and capture UAV images in the relevant airspace under the guidance of the coarse orientation angle and perform lock-on tracking, thereby achieving automatic confirmation of optoelectronic vision.

[0102] This invention enables photoelectric equipment to patrol, search, capture, and track multiple drones in the airspace by selecting automatic or manual methods to detect drone targets, thus meeting the photoelectric confirmation requirements for multi-target aerial intrusion.

[0103] This invention employs a near-to-far search and recognition strategy, searching for UAV images in a U-shaped pattern, sequentially across different airspaces. During near-range and long-range searches, the camera magnification, number of search rounds, and search time are adapted to the UAV targets in the near and far airspaces, respectively. This multi-level near-far search mechanism ensures full coverage of both near and far airspaces within the search angle.

[0104] This invention, after the gimbal is in place, waits for and iterates through the image detection and recognition results, and then uses a binary recursive approach to approximate the actual position of the UAV based on the detection and recognition results, thereby improving the accuracy of UAV position detection. This invention solves the technical problems of complex logic, low error adaptability, and high application cost existing in the prior art.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do 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 method for coarse orientation based opto-electronic scanning capture tracking drone, characterized in that, The method comprises: S1, performing calibration operation on the photoelectric scanning device according to preset calibration angle, calibration rotation direction and angle error parameter; S2, searching for the unmanned aerial vehicle target according to the search angle range and the tolerance angle error parameter of each preset search strategy by using not less than two preset search strategies, so as to obtain the number and distance of the unmanned aerial vehicle target, wherein the preset search strategy comprises a general search strategy, an extended search strategy and a deep search strategy; S3, selecting a search mode according to the number of the unmanned aerial vehicle target, so as to perform round search and capture on the unmanned aerial vehicle target, wherein the search mode comprises automatic search and manual search; S4, processing the unmanned aerial vehicle target according to the distance of the unmanned aerial vehicle target, so as to obtain U-shaped search target classification, wherein the U-shaped search target classification comprises far airspace unmanned aerial vehicle and near airspace unmanned aerial vehicle, and the far airspace unmanned aerial vehicle and the near airspace unmanned aerial vehicle are respectively subjected to U-shaped search according to the U-shaped search target classification, so as to obtain image detection and recognition result; S5, improving and optimizing the YOLOv7 intelligent detection network based on deep learning according to the characteristics of the unmanned aerial vehicle target and the system, performing forward reasoning on each search area image, obtaining the coordinate information of the unmanned aerial vehicle target in the image, and obtaining the image detection and recognition result; S6, traversing the image detection and recognition result, and approaching the actual position of the unmanned aerial vehicle by using the bisection recursive method according to the detection and recognition result; S7, locking and tracking the unmanned aerial vehicle target according to the actual position of the unmanned aerial vehicle.

2. The method of claim 1, wherein, In the step S1, the preset passive direction finding device is used to obtain the coarse orientation angle.

3. The method of claim 1, wherein, The angle error parameter comprises direction finding device angle error, calibration error and angle synchronization difference between linkage devices.

4. The method of claim 1, wherein, In the step S2, the tolerance angle error interval of the search angle range of the general search strategy comprises [-10°, +10°], the tolerance angle error interval of the extended search strategy comprises [-20°, +20°], and the tolerance angle error interval of the deep search strategy comprises [-30°, +30°].

5. The method of claim 1, wherein, In the step S3, the system autonomously selects the unmanned aerial vehicle with the maximum signal strength to perform the automatic search, and the automatic search mode comprises photoelectric search.

6. The method of claim 1, wherein, In the step S3, the manual search mode comprises manually selecting the unmanned aerial vehicle target detected by the passive detection device and guiding the preset photoelectric device to perform photoelectric search on the airspace direction where the unmanned aerial vehicle target is located.

7. The method of claim 1, wherein, The step S4 comprises: S41, performing U-shaped search on the near airspace unmanned aerial vehicle to obtain near airspace unmanned aerial vehicle recognition result; S42, performing U-shaped search on the far airspace unmanned aerial vehicle to obtain far airspace unmanned aerial vehicle recognition result.

8. The method of claim 1, wherein, The step S5 comprises: S51, improving and optimizing the YOLOv7 target detection network based on deep learning according to the characteristics of the unmanned aerial vehicle target and the system, wherein the step S51 comprises: S511, deleting the 32 times down-sampling large target detection head in the original YOLOv7 network, while retaining the feature fusion of deep features in the neck part; S512, add a 4 times down-sampling small target detection head; S513, delete the direct connection between the shallow feature and small target detection head; S52, load the optimized detection network structure into the system, and load the weight parameters after iterative training into the target detection network; S53, extract the two-dimensional image in the search area, and perform image preprocessing according to the network requirements; S54, send the preprocessed image into the target detection network to perform forward reasoning; S55, analyze the features output by the network to obtain the coordinate value and confidence of the unmanned aerial vehicle target, and obtain the image detection and recognition result of each region.

9. The method of claim 1, wherein, The step S6 comprises: S61, position the unmanned aerial vehicle detection gimbal; S62, using the unmanned aerial vehicle detection gimbal, according to the detection and recognition result, using the bisection recursive method to approximate the actual position of the unmanned aerial vehicle, the step S62 comprises: S621, using the photoelectric gimbal, searching from a first point in the near space to a second point, stopping, waiting for a preset time in the far space, and judging whether the unmanned aerial vehicle image is detected in the search stage from the first point to the second point; S622, if not, waiting for the end, and performing the next stage search; S623, if yes, waiting for the end, and performing the bisection recursive search operation on the unmanned aerial vehicle, moving the gimbal camera from the second point to a third point; S624, judging whether the unmanned aerial vehicle image is detected in the search stage from the second point to the third point; S625, if yes, recursively searching in the search stage from the third point to a fourth point; S626, if not, recursively searching in the search stage from the first point to the second point until the unmanned aerial vehicle is searched.

10. A coarse orientation based opto-electronic scanning capture tracking drone system, characterized in that, The system comprises: a calibration module configured to perform calibration operation on the photoelectric scanning device according to preset calibration angle, calibration rotation direction and angle error parameter; a difference strategy search module configured to search for unmanned aerial vehicle targets according to search angle range and tolerance angle error parameter of each preset search strategy by using not less than two preset search strategies, so as to obtain number of unmanned aerial vehicle targets and distance of unmanned aerial vehicle targets, wherein the preset search strategies comprise general search strategy, extended search strategy and deep search strategy, and the difference strategy search module is connected with the calibration module; a round-patrol capture module configured to select search mode according to the number of unmanned aerial vehicle targets, so as to round-patrol search and capture the unmanned aerial vehicle targets, wherein the search mode comprises automatic search and manual search, and the round-patrol capture module is connected with the difference strategy search module; a U-shaped search module configured to process the unmanned aerial vehicle targets according to the distance of the unmanned aerial vehicle targets, so as to obtain U-shaped search target classification, wherein the U-shaped search target classification comprises far space unmanned aerial vehicle and near space unmanned aerial vehicle, and the far space unmanned aerial vehicle and the near space unmanned aerial vehicle are respectively subjected to U-shaped search according to the U-shaped search target classification, so as to obtain image detection and recognition result, and the U-shaped search module is connected with the difference strategy search module; An image detection module is used to improve and optimize a YOLOv7 intelligent detection network based on deep learning according to the characteristics of the UAV target and the system, perform forward inference on each search area image, obtain the coordinate information of the UAV target in the image, and obtain the image detection and recognition result; A bisection approximation module is used to traverse the image detection and recognition result, approximate the actual position of the UAV using a bisection recursive method according to the detection and recognition result, and the bisection approximation module is connected with the U-shaped search module and the image detection module; A locking tracking module is used to lock and track the UAV target according to the actual position of the UAV, and the bisection approximation module is connected with the bisection approximation module.

Images