Unmanned aerial vehicle tracking system and method based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to an unmanned aerial vehicle tracking system and method based on image recognition, and the system comprises a spectrum optimization module, a trajectory analysis module, a form adjustment module, a flight anomaly detection module and a tracking optimization module. According to the method, the brightness information of the ambient light and the reflected light of the unmanned aerial vehicle is finely analyzed and the spectral response of the image is adjusted, so that the image definition is greatly improved, the unmanned aerial vehicle can be accurately identified under various illumination conditions, and the flight path of the unmanned aerial vehicle is calculated in real time by accurately analyzing the image bounding box and timestamp information of the unmanned aerial vehicle. According to the method and the system, the tracking process is ensured to be continuous and accurate, any form change is effectively captured by continuously monitoring the form of the unmanned aerial vehicle in continuous image frames, the behavior mode change of the unmanned aerial vehicle is responded in time, and in the aspect of flight abnormity monitoring, behaviors inconsistent with a conventional flight mode can be accurately analyzed, and potential security threats can be identified in time.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a drone tracking system and method based on image recognition. Background Art

[0002] Image recognition technology is an important branch of computer vision. It mainly involves the use of algorithms and models to interpret and understand information extracted from images or video sequences. This technology enables computers to identify objects, scenes and activities by simulating the recognition ability of the human eye. Image recognition usually includes steps such as image acquisition, preprocessing, feature extraction, feature matching and classification. This technology has a wide range of applications in many fields, such as automatic monitoring, driverless cars, medical imaging analysis, robot navigation, and security and monitoring systems.

[0003] Among them, the image recognition drone tracking system is a system that uses image recognition technology to monitor and track drones. The system uses a camera to capture the image of the drone, and then uses image recognition technology to identify and track the location and movement trajectory of the drone. The main uses of this system include but are not limited to security monitoring, traffic supervision, military reconnaissance and disaster response. By tracking drones in real time, it can more effectively manage and control airspace safety, and it can also respond quickly in emergencies.

[0004] Existing technologies have difficulty accurately identifying drones in low-light or strong-light conditions. Due to the inability to effectively process changes in ambient and reflected light, tracking of high-speed moving targets is often lost due to data processing delays or insufficient accuracy. This is especially true in emergency situations that require a quick response. These limitations can easily lead to the inability of monitoring to effectively perform tasks at critical moments, such as maintaining continuous tracking of targets in complex backgrounds or tracking escaping drones. Furthermore, there are also deficiencies in the automatic detection of drone morphology and flight anomalies, and a lack of in-depth analysis of kinematic details, which limits its application effectiveness in complex environments. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a drone tracking system and method based on image recognition.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solution: a drone tracking system based on image recognition, the system comprising:

[0007] The spectral optimization module extracts the brightness information of the ambient light and the reflected light from the drone based on the captured drone image. It compares the brightness with the surrounding ambient light intensity pixel by pixel, identifies key areas of brightness deviation, and adjusts the response parameters of the infrared and visible light channels to optimize image accuracy and obtain an updated image sequence.

[0008] The trajectory analysis module obtains the image bounding box and corresponding timestamp of the drone based on the updated image sequence, compares the spatial position data of each frame, analyzes the position change of the center of the bounding box, calculates the real-time movement path of the drone, and generates drone trajectory data;

[0009] The morphology adjustment module locks the position of the drone in consecutive image frames based on the drone trajectory data, compares the position changes of the drone edge pixels between each two frames, reconfigures the edge pixel positions, and obtains the morphology adjustment offset index;

[0010] The flight anomaly detection module uses the UAV trajectory data to calculate the speed and acceleration changes of the UAV based on the time interval and position data, compares them with the normal flight data, identifies the trajectory segments that differ from the standard flight pattern, and obtains the abnormal flight segment index.

[0011] The improvements of the present invention are that the updated image sequence includes brightness level, color saturation, and contrast adjustment information; the drone trajectory data includes flight altitude change, direction offset data, and speed change graph; the morphological adjustment offset index includes shape alignment, edge clarity, and pixel adjustment range; and the abnormal flight segment index includes flight altitude abnormality segment, speed mutation segment, and navigation deviation index.

[0012] The present invention is improved in that the spectrum optimization module includes:

[0013] The brightness extraction submodule extracts the reflected light brightness value from the infrared channel and visible light channel of each pixel based on the captured drone image, calculates the deviation between the pixel brightness and the average brightness of the ambient light, and obtains the brightness offset value distribution;

[0014] The difference recognition submodule sets an ambient brightness deviation threshold based on the brightness offset value distribution, screens pixel areas with brightness higher than the threshold, analyzes pixel aggregation and brightness direction consistency, and identifies deviation abnormal areas with high aggregation and consistent direction;

[0015] The response adjustment submodule calls the offset abnormal area and analyzes the brightness difference between the infrared and visible light channels using the formula:

[0016]

[0017] Calculate the response channel adjustment coefficient ΔRC and adjust the brightness response parameters to optimize the image clarity to obtain the updated image sequence, where: represents the infrared channel brightness of the i-th pixel, Represents the visible light channel brightness of the i-th pixel, n rc Indicates the total number of pixels in the critical area of brightness deviation, Represents the brightness value of the visible light channel of the j-th pixel, Indicates the infrared channel brightness value of the jth pixel, m rc Indicates the number of pixels involved in the channel response difference calculation.

[0018] The present invention is improved in that the trajectory analysis module includes:

[0019] The bounding box extraction submodule extracts the drone image bounding box in each frame of the image based on the updated image sequence, detects the center coordinate information of the bounding box and the corresponding timestamp, and constructs a time positioning coordinate set by combining the image frame number and acquisition time;

[0020] The position change calculation submodule calls the time positioning coordinate set and uses the formula:

[0021]

[0022] Calculate the path change rate between every two frames, where VX o Indicates the path change rate per unit time between the oth frame and the o+1th frame, Qx o Represents the horizontal coordinate of the center of the bounding box of the oth frame, Qy o Represents the vertical coordinate of the center of the bounding box of the oth frame, Qt o Represents the timestamp of the oth frame, Qx o+1 Represents the horizontal coordinate of the center of the bounding box of the o+1th frame, Qy o+1 Represents the vertical coordinate of the center of the bounding box of the o+1th frame, Qt o+1 Represents the timestamp of the o+1th frame;

[0023] The path construction submodule calibrates the continuous connection relationship of the trajectory points frame by frame according to the path change rate, establishes the connection order between the nodes according to the path rate, and generates the UAV trajectory data.

[0024] The present invention is improved in that the morphology adjustment module includes:

[0025] The trajectory locking submodule extracts coordinate points of the drone position in consecutive image frames based on the drone trajectory data, screens stable data points by comparing the coordinate changes of the drone pixel center of gravity in adjacent image frames, and locates the real-time trajectory of the drone to obtain the drone image displacement trajectory;

[0026] The edge change analysis submodule calls the displacement trajectory of the drone image, monitors the edge pixels of each frame of the drone, calculates the coordinate changes of the pixels on the image plane, identifies the offset anomalies of the edge pixels, and obtains the abnormal edge displacement difference;

[0027] The offset index extraction submodule reconfigures the edge pixel position according to the abnormal edge displacement difference and the changes beyond the normal range, using the formula:

[0028]

[0029] Get the morphological adjustment offset indicator OU, where OE o Represents the abnormal edge displacement difference of the oth frame, OL o Indicates the expected deviation position of the edge in the oth frame, n ou Represents the total number of image frames processed.

[0030] The present invention is improved in that the flight anomaly detection module includes:

[0031] The speed calculation submodule uses the UAV trajectory data to extract continuous timestamps and position coordinates, calculates the time difference for each pair of continuous timestamps, determines the distance between the position points corresponding to the two time points, and generates a speed change sequence;

[0032] The acceleration analysis submodule calls the speed change sequence, compares the continuous speed data, matches the time intervals of each pair of continuous speed data in the sequence, calculates the acceleration value of each time period, and obtains the acceleration change trend;

[0033] The state recognition submodule compares the acceleration change trend with the acceleration distribution characteristics in the standard flight trajectory sample segment by segment, analyzes the difference between acceleration and speed in each segment, and uses the formula:

[0034]

[0035] Obtain the state deviation of each trajectory segment, judge the flight state of the UAV, and obtain the abnormal flight segment index, where SN z Represents the state deviation of the zth segment, VN zc represents the actual velocity of the cth point in the zth segment, Represents the standard speed value of the cth point in the zth segment, AN zc Indicates the actual acceleration value of the cth point in the zth segment, n sn represents the number of trajectory segments, and λ is the unit correction coefficient.

[0036] The present invention is improved in that the system further comprises:

[0037] The tracking optimization module synchronously retrieves the corresponding image sequence frames based on the abnormal flight segment index and morphological adjustment offset index, and adjusts the UAV tracking process in combination with the current image capture parameters to match the flight anomaly or emergency situation, and obtains the dynamic tracking parameter adjustment result;

[0038] The dynamic tracking parameter adjustment result includes a tracking sensitivity adjustment result, a response time optimization result, and a tracking accuracy optimization result.

[0039] The present invention is improved in that the tracking optimization module includes:

[0040] The index extraction submodule synchronously retrieves the UAV image data corresponding to the index based on the abnormal flight segment index and the morphological adjustment offset index, analyzes the correlation between the flight state and the image sequence, identifies the key parameters affecting the flight trajectory, and obtains the state number matching segment;

[0041] The image frame synchronization submodule calls the state number matching section to perform time axis matching on the image sequence frames, and at the same time uses the image capture parameters to match the frame numbers, compares the frame numbers with the timing offsets of the original image frames, selects frames that meet the image number consistency, and generates a corresponding sequence set of image frames;

[0042] The dynamic adjustment submodule calls the corresponding sequence set of the image frames and performs dynamic optimization analysis using the formula:

[0043]

[0044] Adjust the UAV flight tracking process and obtain the dynamic tracking parameter adjustment result PL d , where VL b Indicates the capture parameters of the bth image sequence frame, SL b Indicates the adjustment parameter of the bth flight state, HL h represents the hth offset index, α b represents the adjustment factors affecting the capture parameters of the b-th image sequence frame, β b represents the adjustment factor affecting the b-th flight state adjustment parameter, γ h Represents the adjustment influencing factor of the h-th offset indicator.

[0045] The image recognition-based drone tracking method is performed based on the above-mentioned image recognition-based drone tracking system and includes the following steps:

[0046] S1: Based on the captured drone image, the brightness information of the ambient light and the drone's reflected light is extracted, the brightness is compared with the surrounding ambient light intensity pixel by pixel, the key areas of brightness deviation are identified, and the response parameters of the infrared and visible light channels are adjusted to obtain the updated image sequence;

[0047] S2: Based on the updated image sequence, obtain the image bounding box and the corresponding timestamp of the drone, compare the spatial position data of each frame, analyze the position change of the center of the bounding box, calculate the real-time movement path of the drone, and generate drone trajectory data;

[0048] S3: Based on the drone trajectory data, lock the position of the drone in consecutive image frames, compare the position changes of the drone edge pixels between every two frames, reconfigure the edge pixel positions for changes that exceed the normal range, and obtain a morphological adjustment offset index;

[0049] S4: Using the UAV trajectory data, calculate the speed and acceleration changes of the UAV according to the time interval and position data, compare the calculated speed and acceleration changes with the normal flight data, identify the trajectory segments that differ from the standard flight pattern, and obtain the abnormal flight segment index;

[0050] S5: Based on the abnormal flight segment index and the morphological adjustment offset index, the corresponding image sequence frames are synchronously retrieved, and the tracking process of the UAV is adjusted in combination with the current image capture parameters to obtain a dynamic tracking parameter adjustment result.

[0051] Compared with the prior art, the advantages and positive effects of the present invention are:

[0052] In the present invention, by performing a detailed analysis of the brightness information of the ambient light and the reflected light of the drone and adjusting the spectral response of the image, the image clarity is greatly improved, ensuring that the drone can be accurately identified under various lighting conditions. By accurately analyzing the image bounding box and timestamp information of the drone, the flight path of the drone is calculated in real time to ensure the continuity and accuracy of the tracking process. In addition, by continuously monitoring the drone's morphology in consecutive image frames, any morphological changes can be effectively captured and changes in the drone's behavioral patterns can be responded to immediately. In terms of flight anomaly monitoring, it can accurately analyze behaviors that are inconsistent with conventional flight patterns and promptly identify potential security threats, thereby improving monitoring efficiency and reducing reliance on human intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The module diagram of the drone tracking system based on image recognition is proposed for the present invention;

[0054] Figure 2 This is a flow chart of the spectrum optimization module in the present invention;

[0055] Figure 3 This is a flow chart of the trajectory analysis module in the present invention;

[0056] Figure 4 This is a flow chart of the morphology adjustment module in the present invention;

[0057] Figure 5 This is a flow chart of the flight anomaly detection model in the present invention;

[0058] Figure 6 This is a flowchart of the tracking optimization module in the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0061] Example

[0062] See also Figure 1 The present invention provides a technical solution: a drone tracking system based on image recognition includes:

[0063] The spectral optimization module extracts the brightness information of the ambient light and the reflected light from the drone based on the captured drone image. It compares the brightness with the surrounding ambient light intensity pixel by pixel, identifies key areas of brightness deviation, and adjusts the response parameters of the infrared and visible light channels to optimize image clarity and obtain an updated image sequence.

[0064] For the key areas of brightness deviation, after they are extracted, the continuous image frames are first subjected to temporal consistency analysis, and only the deviation areas with continuous positions and smooth trajectories in multiple frames are retained, excluding the interference of static high-brightness background. Secondly, the brightness changes in the infrared and visible light channels are synchronously analyzed, and the brightness deviation areas with high spatial overlap are fused and verified to optimize the target authenticity judgment. Combined with the regional contour features, the highlight areas with incomplete structure, disconnected edges or irregular shapes are eliminated. In order to further adapt to the time-varying characteristics of the background environment, the difference between the current image frame and the background is dynamically compared to enhance the foreground target response and suppress periodic or local background reflection interference. The above mechanisms work together to reduce the impact of background noise on the recognition of key areas of brightness deviation, thereby improving the extraction accuracy of drone targets in image sequences.

[0065] The trajectory analysis module obtains the image bounding box and corresponding timestamp of the drone based on the updated image sequence, compares the spatial position data of each frame, analyzes the position change of the center of the bounding box, calculates the real-time movement path of the drone, and generates the drone trajectory data;

[0066] The morphology adjustment module locks the position of the drone in consecutive image frames based on the drone trajectory data, compares the position changes of the drone's edge pixels between every two frames, and reconfigures the edge pixel positions for changes that exceed the normal range to obtain the morphology adjustment offset index;

[0067] The flight anomaly detection module uses the drone trajectory data to calculate the drone's speed and acceleration changes based on time intervals and position data, compares it with normal flight data, identifies trajectory segments that differ from the standard flight pattern, determines the drone's flight status, and obtains the abnormal flight segment index;

[0068] The tracking optimization module synchronously retrieves the corresponding image sequence frames based on the abnormal flight segment index and morphological adjustment offset index, and adjusts the UAV's tracking process in combination with the current image capture parameters to match flight anomalies or emergencies, and obtains the dynamic tracking parameter adjustment results.

[0069] The updated image sequence includes brightness level, color saturation, and contrast adjustment information; the drone trajectory data includes flight altitude changes, direction offset data, and speed change graphs; the morphological adjustment offset indicators include shape alignment, edge clarity, and pixel adjustment range; the abnormal flight segment index includes abnormal flight altitude segments, speed mutation segments, and navigation deviation indicators; the dynamic tracking parameter adjustment results include tracking sensitivity adjustment results, response time optimization results, and tracking accuracy optimization results.

[0070] See also Figure 2 , the spectrum optimization module includes:

[0071] The brightness extraction submodule extracts the reflected light brightness value from the infrared channel and visible light channel of each pixel based on the captured drone image, calculates the deviation between the pixel brightness and the average brightness of the ambient light, and obtains the brightness offset value distribution;

[0072] The image recognition system extracts the reflected light brightness value from the infrared channel and visible light channel of each pixel. The image is captured by a camera on the ground or other platforms. The captured target is a flying drone. The brightness value of each pixel is calculated by the image sensor or video stream data. The infrared channel is used to obtain the thermal imaging data of the target, while the visible light channel is used to obtain conventional visual images. In image processing, for a 5×5 pixel sub-block of an image, it is assumed that its infrared brightness is 123, 115, 110, 118, and 121 respectively, and the visible light brightness is 127, 120, 113, 116, 125, first collect a portion of pixel points without target objects from the background area of the image, calculate the average brightness of the ambient light, assuming that the ambient light brightness is 119, and then calculate the deviation of the infrared and visible light brightness of each pixel from the ambient light brightness. For example, if the brightness of the infrared channel is 123, the offset value is 123-119=4; if the visible light brightness is 127, the offset value is 127-119=8. In this way, the offset values of all pixels are calculated to form a brightness offset value distribution, which can be used for subsequent abnormal area identification and image optimization.

[0073] The difference recognition submodule sets the ambient brightness deviation threshold based on the brightness offset value distribution, filters pixel areas with brightness higher than the threshold, analyzes pixel aggregation and brightness direction consistency, and identifies deviation anomaly areas with high aggregation and consistent direction;

[0074] Setting the ambient brightness deviation threshold is used to filter pixel areas with large offsets. The threshold can be set by setting a fluctuation range of 5% to 10% of the average ambient light brightness value. If the ambient light is 119 and the deviation threshold is 3, then when the brightness offset value of a pixel is 8 in the visible light channel and 4 in the infrared channel, this pixel meets the condition that the offset is greater than 3 in both channels and enters the primary screening set. Then, its degree of aggregation with adjacent pixels in space is analyzed, that is, how many of its four adjacent pixels above, below, left and right belong to the primary screening set. If more than two of the adjacent pixels are offset pixels, the pixel is included in the primary screening set. The pixel aggregation degree is high, and the consistency of the brightness offset direction is further judged, that is, the infrared channel and the visible light channel offset values have the same sign. If the infrared channel offset of the current pixel is positive and the infrared channel offset value of the adjacent pixel is also positive, the direction is determined to be consistent. If three consecutive pixels in a certain area meet the conditions of an offset value greater than 3, an aggregation degree not less than 2, and the same offset direction, they are identified as an offset abnormal area. For example, the infrared offset of pixel 1 is +4, pixel 2 is +5, and pixel 3 is +3. The three are in the same direction and arranged continuously, then the three constitute an abnormal area. Finally, all areas that meet the conditions are marked and the offset abnormal area is output.

[0075] The response adjustment submodule calls the offset abnormal area and analyzes the brightness difference between the infrared and visible light channels using the formula:

[0076]

[0077] Calculate the response channel adjustment coefficient ΔRC and adjust the brightness response parameters to optimize the image clarity to obtain the updated image sequence, where: represents the infrared channel brightness of the i-th pixel, Represents the visible light channel brightness of the i-th pixel, n rc Indicates the total number of pixels in the critical area of brightness deviation, Represents the brightness value of the visible light channel of the j-th pixel, Indicates the infrared channel brightness value of the jth pixel, m rc Indicates the number of pixels involved in the calculation of channel response differences;

[0078] Obtain the brightness values of the pixels in the area under the infrared channel and the visible light channel. First, identify the position of the drone in the image and calculate the possible deviation abnormal area based on its motion trajectory, such as the brightness difference between the drone's fuselage reflected light and the background. Assume that the area contains 3 pixels, and the infrared brightness values are and The corresponding visible light brightness values are and Pixels are extracted from the drone’s body or its surrounding area, and the brightness difference between the infrared and visible light channels is calculated for each pixel:

[0079]

[0080] Then, the sum of the brightness differences of all pixels in the statistical area is calculated Right now:

[0081] (125-120)+(119-115)+(121-118)=5+4+3=12;

[0082] Substitute into the formula for calculation:

[0083]

[0084] e -12 ≈6.1442×10 -6 ;

[0085]

[0086] Get the response adjustment coefficient:

[0087] ΔRC=-4×1≈-4;

[0088] The results show that there is a significant brightness deviation around the drone body, and the brightness of the infrared channel is lower than that of the visible light channel. It is necessary to adjust the brightness of the infrared channel in the corresponding area of the image to achieve response consistency between the two channels, enhance the grayscale consistency and edge contrast of the local area, and thus optimize image clarity and improve tracking accuracy, obtaining an updated image sequence.

[0089] See also Figure 3 , the trajectory analysis module includes:

[0090] The bounding box extraction submodule extracts the drone image bounding box in each frame based on the updated image sequence, detects the center coordinate information of the bounding box and the corresponding timestamp, and constructs a time positioning coordinate set by combining the image frame number and acquisition time;

[0091] For the updated image sequence, the image data frame needs to be read frame by frame. The image data is arranged in numerical order, starting from 0. The acquisition frequency is set to 25 frames per second, and the inter-frame time interval is 0.04 seconds. For example, the acquisition time of frame 0 is 1.00 seconds, frame 1 is 1.04 seconds, frame 2 is 1.08 seconds, and so on. In each frame image, the bounding box area of the drone target is extracted. The bounding box consists of the pixel coordinates of the upper left corner and the lower right corner. For example, in frame 0, the coordinates of the upper left corner are (100, 200) and the lower right corner are (110, 210). The horizontal coordinate of the center point of the bounding box can be calculated as The vertical axis is This method can be applied to subsequent frame images. For example, if the detection results of the first frame are (108, 208) in the upper left corner and (113, 213) in the lower right corner, the horizontal and vertical coordinates of the center point are Qx1 = 110.5 and Qy1 = 210.5, respectively. Similarly, for the second frame, Qx2 = 114.7 and Qy2 = 215.3 are obtained. In addition, the acquisition timestamp of the center point of each frame must also be matched. The 0th frame is 1.00 seconds, the 1st frame is 1.04 seconds, and the 2nd frame is 1.08 seconds. The frame number, center point coordinates, and timestamp are mapped one by one to establish a time positioning mapping relationship. The constructed center point time positioning information should be saved in an array structure for subsequent operations, such as [105, 205, 1.00], [110.5, 210.5, 1.04], [114.7, 215.3, 1.08], etc., which is the time positioning coordinate set.

[0092] The position change calculation submodule calls the time positioning coordinate set and uses the formula:

[0093]

[0094] Calculate the path change rate between every two frames, where VX oIt represents the rate of change of the path per unit time between the oth frame and the o+1th frame, and is obtained by calculating the ratio of the Euclidean distance between the two frames to the time difference. o Represents the horizontal coordinate of the center of the bounding box of the oth frame, Qy o Represents the vertical coordinate of the center of the bounding box of the oth frame, Qt o Represents the timestamp of the oth frame, Qx o+1 Represents the horizontal coordinate of the center of the bounding box of the o+1th frame, Qy o+1 Represents the vertical coordinate of the center of the bounding box of the o+1th frame, Qt o+1 Represents the timestamp of the o+1th frame;

[0095] The center point information of each group of adjacent frames is selected in turn to perform the rate calculation task. For example, using the data between frame 0 and frame 1, the horizontal and vertical coordinate differences are extracted as Qx1-Qx0=110.5-105=5.5, Qy1-Qy0=210.5-205=5.5, and the time difference is Qt1-Qt0=1.04-1.00=0.04 seconds. Substituting them into the calculation, we get:

[0096]

[0097] This value is the path change rate per unit time between frame 0 and frame 1. Processing the rate from frame 1 to frame 2 in the same way, we can get Qx2-Qx1=114.7-110.5=4.2, Qy2-Qy1=215.3-210.5=4.8, and the time difference Qt2-Qt1=0.04. Substituting into the formula:

[0098]

[0099] The above processing method is applicable to all frame pairs in the image sequence. By performing similar operations on each frame pair, a continuous sequence of path change rate values, that is, a sequence of path rates per unit time, can be obtained.

[0100] The path construction submodule calibrates the continuous connection relationship of trajectory points frame by frame according to the path change rate, establishes the connection sequence between nodes according to the path rate, and generates the UAV trajectory data;

[0101] Starting from frame 0, its center point is used as the first trajectory node. Combined with its corresponding center point of the next frame, a trajectory segment connecting line is constructed. For example, if the center point of frame 0 is (105, 205) and that of frame 1 is (110.5, 210.5), the direction of the connecting line is tilted to the upper right, indicating that the drone is moving in the upper right direction. Then, from frame 1 to frame 2, the connecting point (110.5, 210.5) to (114.7, 215.3) is connected, and the trajectory continues to extend in the upper right direction. All frame pairs are connected in sequence to form a complete set of path segments. A continuous trajectory point chain is obtained by accumulating continuous path segments. The speed value of each segment in the trajectory point chain can be annotated according to the aforementioned rate result. For example, the speed from frame 0 to frame 1 is 194.38 pixels / s, and from frame 1 to frame 2 is 159.48 pixels / s. The path continuity or mutation can be further analyzed according to the size of the speed value. Finally, after completing the connection of all nodes and the generation of path shape, continuous spatial movement path information is obtained, which is the drone trajectory data.

[0102] See also Figure 4 , the morphology adjustment module includes:

[0103] The trajectory locking submodule extracts the coordinate points of the drone position in continuous image frames based on the drone trajectory data. By comparing the coordinate changes of the drone pixel center of gravity in adjacent image frames, it selects stable data points and locates the real-time trajectory of the drone to obtain the drone image displacement trajectory.

[0104] To extract the coordinate points of the drone position in continuous image frames, we first need to extract the pixel area of the drone target from the image sequence frame by frame. We can determine the drone target area by analyzing the grayscale distribution or color characteristics in the image. After uniformly preprocessing the image and standardizing its resolution to 1280×720 pixels, we can locate the drone area based on the brightness center of gravity or contour features, and then extract its center of gravity coordinates (x i ,y i ) as the position point of the frame, and then compare the centroid coordinates of adjacent frames, and define the inter-frame displacement vector as Calculate ΔP for multiple consecutive frames i sequence, forming a set of displacement sample sequences {ΔP1, ΔP2,…, ΔP n}, and then use this sample to calculate the average displacement μ P and standard deviation σ P , to determine whether each displacement falls within the stable interval [μ P -σ P ,μ P +σ P ], if the ΔP of a certain frame i Within this range, it is considered a stable point, otherwise it is considered an abnormal value and is removed. For example, if the frame displacement is 1.2, 1.4, 3.5, 1.1, 1.3 pixels, μ P=1.9,σ P =0.9, only 3.5 frames are out of the interval [1.0, 2.8], and the rest are stable values. Finally, the valid frames are retained to form the trajectory path of the drone.

[0105] The edge change analysis submodule calls the displacement trajectory of the drone image, monitors the edge pixels of each frame of the drone, calculates the coordinate changes of the pixels on the image plane, identifies the offset anomalies of the edge pixels, and obtains the abnormal edge displacement difference;

[0106] When monitoring the edge pixels of each drone frame, it is necessary to first identify the set of pixels in the image that are within 5 pixels of the target edge based on the target area pointed by the trajectory. j}, record its coordinates in the frame as (x j (t),y j (t)), then the pixels with the same number in the current frame and the previous frame are mapped one by one to calculate the displacement offset Δd of the edge pixels j =

[0107] Then construct the edge offset sequence {Δd1, Δd2, ..., Δd m}, and then calculate the mean μ of the sequence d and standard deviation σ d , according to the stability interval [μ d -2σ d ,μ d +2σ d ] Determine whether each pixel is an offset abnormal point. For example, if the offset of the edge point of a certain frame is 2.0, 3.1, 11.2, 4.5, and 3.0 pixels respectively, if μ d =4.7,σ d =2.5, the stable interval is [-0.3, 9.7]. 11.2 is beyond this range and is identified as an offset abnormal point. The offsets of the abnormal points in all frames are accumulated and averaged to obtain the abnormal edge displacement difference.

[0108] The offset index extraction submodule reconfigures the edge pixel position according to the abnormal edge displacement difference and the changes beyond the normal range, using the formula:

[0109]

[0110] Get the morphological adjustment offset indicator OU, where OE o Represents the abnormal edge displacement difference of the oth frame, OL o Indicates the expected deviation position of the edge in the oth frame, n ou Represents the total number of image frames processed;

[0111] First, identify all abnormal edge pixels and call the stable edge point positions of adjacent frames to build the regression boundary. Use the fitted straight line or curve trend to predict the ideal edge position. Let the predicted ideal edge position of a frame be OL o , the corresponding abnormal edge offset is OE o , the frame number is o, then the formula is used to calculate the morphological adjustment offset index. For example, if the offsets of the three abnormal points in a frame are 1.2, 2.0, and 2.6, then OE o =(1.2+2.0+2.6) / 3=1.93, while the predicted ideal position offset OL o It is an estimate of the position based on the edge fitting trend. For example, if the adjacent edge points in the previous and next frames are located at (120, 100) and (122, 103), the estimated deviation position can be predicted to be (121, 101.5). The difference between the offset and the actual value is calculated and then averaged over all frames. The total number of frames is n. ou , if the five frames in the sequence are: OE = {2.1, 3.4, 2.7, 4.0, 3.3}, OL = {2.0, 3.0, 2.5, 3.8, 3.1}, then the calculation is as follows:

[0112]

[0113] The morphological adjustment offset index OU = 0.22, which represents the average difference between the abnormal edge and its predicted position in all frames. This result shows that after the overall edge of the image is relocated, the average offset error is 0.22 pixels, which meets the judgment interval of the edge stability error benchmark of 0.5 pixels.

[0114] See also Figure 5 , the flight anomaly detection module includes:

[0115] The speed calculation submodule uses the UAV trajectory data to extract continuous timestamps and position coordinates, calculates the time difference for each pair of continuous timestamps, determines the distance between the position points corresponding to the two time points, and generates a speed change sequence;

[0116] The image recognition system collects the position information of the drone in the picture at a frame rate of 1 frame per second. The timestamps are 0 seconds, 1 second, 2 seconds, and 3 seconds respectively. The coordinates in the image are converted into the actual position points in the plane coordinate system through spatial calibration: (0.0, 0.0), (1.2, 0.5), (2.7, 1.8), (4.5, 3.6). Take the first pair of points and calculate the Euclidean distance between the two points: The corresponding time interval is Δt1=1s, from which the speed of the first section is Then process the second pair of points and get The time difference is still 1 second, so v2≈1.985m / s, and the speed of the third pair of points is ≈2.545 m / s, and the velocity sequence is {1.3, 1.985, 2.545} m / s, which is used as the basic data for subsequent analysis.

[0117] The acceleration analysis submodule calls the speed change sequence, compares the continuous speed data, matches the time intervals of each pair of continuous speed data in the sequence, calculates the acceleration value of each time period, and obtains the acceleration change trend;

[0118] Call the aforementioned velocity sequence {1.3, 1.985, 2.545} m / s, take the difference between the adjacent velocity pairs and divide it by the time interval Δt = 1s to calculate the acceleration. The velocity difference of the first segment is Δv1 = 1.985-1.3 = 0.685 m / s, so the acceleration is m / s 2 , the speed difference in the second section is Δv2 = 2.545-1.985 = 0.56 m / s, and the corresponding acceleration is The final acceleration change trend is {0.685, 0.56} m / s 2 ,All acceleration values are cached in time order, and an index is established that corresponds to the ,original velocity sequence frame segments one to one for trajectory segment ,state identification.

[0119] The state recognition submodule compares the acceleration change trend with the acceleration distribution characteristics in the standard flight trajectory sample segment by segment, analyzes the difference between acceleration and speed in each segment, and uses the formula:

[0120]

[0121] Obtain the state deviation of each trajectory segment, judge the flight state of the UAV, and obtain the abnormal flight segment index, where SN z Represents the state deviation of the zth segment, which is used to evaluate the deviation degree of the UAV flight in this segment. VN zc represents the actual velocity of the cth point in the zth segment, Represents the standard speed value of the cth point in the zth segment, AN zc Indicates the actual acceleration value of the cth point in the zth segment, n sn represents the number of trajectory segments, and λ is the unit correction coefficient, which is used to map the effect of acceleration on the flight state to the equivalent dimension range of velocity change;

[0122] If the first segment contains three frames, the actual velocity sequence is [1.3, 1.985, 2.1] m / s, the standard velocity sequence is [1.0, 2.0, 2.2] m / s, and the acceleration is [0.685, 0.56, 0.41] m / s 2, assuming λ is 1.0s, substitute into the formula to get:

[0123]

[0124] Set the state deviation judgment threshold to 0.6, that is, when SN z When the threshold is >0.6, the trajectory segment is considered to have flight anomalies. This value is calculated by adding 0.5 times the standard deviation to the mean of the deviation distribution obtained from the speed fluctuation statistics in the standard sample. For example, if the mean deviation of the standard sample is 0.3 and the standard deviation is 0.6, then the threshold is 0.3 + 0.5 × 0.6 = 0.6. Therefore, trajectory segment 1 is considered to be an abnormal segment, and its frame index is recorded.

[0125] See also Figure 6 , tracking optimization module includes:

[0126] The index extraction submodule synchronously retrieves the UAV image data corresponding to the index based on the abnormal flight segment index and morphological adjustment offset index, analyzes the correlation between the flight state and the image sequence, identifies the key parameters affecting the flight trajectory, and obtains the state number matching segment;

[0127] Collect multiple frames of drone images recorded in the ground-side image recognition system, establish an initial image number sequence in the order of timestamps, and obtain the abnormal flight segment index by setting the image recognition confidence curve fluctuation standard. If the recognition confidence change rate in continuous images is greater than the set difference threshold (such as Δ confidence ≥ 0.15 / frame), the corresponding image number is included in the abnormal segment index, and then the corresponding number image is offset analyzed. The offset index is calculated as follows: the Euclidean distance normalized value between the center point of the recognition frame and the center coordinate of the image frame. The morphological adjustment offset index is obtained by jointly calculating the size difference and posture change of the drone contour matching result. Subsequently, the correspondence between each frame image and the flight number is established. The image number is converted into the corresponding flight segment number by matching the image timestamp with the time field of the drone broadcast signal, and the image number is synchronized accordingly. The image sequence frames corresponding to the abnormal segment index are retrieved in the first step. For example, the time of image numbers 8 to 12 is 12.1 to 12.5 seconds respectively, which is synchronized with the UAV flight control recording time. The contour boundary, attitude angle, flight altitude estimation value and other data in the image frame are extracted and the difference is calculated with the flight state estimation value. The difference sequence is analyzed by sliding window. The frames with consistent change trends within the window (such as the image attitude angle and the flight control attitude angle trend in the same direction) are counted as consistent frames. If the number of consecutive consistent frames exceeds 3 frames, the consecutive numbered segment is used as the flight trajectory association segment. The correlation score of the contour change rate, the recognition boundary jump amplitude and the matching degree of the flight trajectory estimation path in the image sequence is further performed. The frame segment with a correlation coefficient exceeding 0.6 is taken as the valid tracking interval, and the consecutive number range, such as frames 8 to 12, is output as the state number matching segment.

[0128] The image frame synchronization submodule calls the state number matching section to perform time axis matching on the image sequence frames. It also uses the image capture parameters to match the frame numbers, compares the timing offsets of the frame numbers and the original image frames, selects frames that meet the image number consistency, and generates a corresponding sequence set of image frames.

[0129] Extract the timestamps of all image frames in the numbered segment and compare them frame by frame with the positioning timestamps broadcast in the drone communication signal. Assume that the time offset range is ±0.03 seconds. If the image frame time is 12.21 seconds and the matching drone trajectory broadcast time is 12.20 seconds, the offset value is 0.01 seconds, which meets the synchronization conditions and is judged as a valid time matching frame. Then call the image capture parameters of the synchronized frame, such as the image frame numbering sequence, the direction of change of the center coordinates of the contour recognition frame, and the trend of image angle change, to judge the numbering consistency. For example, the horizontal coordinate of the center of the recognition frame of image frame 10 increases from 240px to 245px, the image number changes from 10 to 11, and the rightward shift trend of the center point is consistent with the number growth direction, which is determined to be number consistent. If the number sequence jumps back or the positioning offset direction is opposite to the number growth direction, it is removed. After completing the number consistency judgment, the timing offset judgment is performed on the remaining frames, that is, the time difference between the image time and the flight trajectory signal is calculated again. If the time difference between a frame exceeds 0.05 seconds twice in a row, it is marked as an asynchronous frame. The frame number sequence retained after elimination is a valid frame segment that meets the image time synchronization, number progressive consistency, and trajectory trend matching. This sequence is the image frame corresponding sequence set.

[0130] The dynamic adjustment submodule calls the corresponding sequence set of image frames and performs dynamic optimization analysis using the formula:

[0131]

[0132] Adjust the UAV flight tracking process and obtain the dynamic tracking parameter adjustment result PL d , where VL b Indicates the capture parameters of the bth image sequence frame, including the frame timestamp, position, angle and other information, SL b Indicates the adjustment parameters of the bth flight state, involving the adjustment values of flight states such as speed, altitude, and direction, HL h represents the hth deviation index, which is a quantitative indicator to measure the deviation between the actual flight path and the expected path. b represents the adjustment factors affecting the capture parameters of the b-th image sequence frame, β b represents the adjustment factor affecting the b-th flight state adjustment parameter, γ h represents the adjustment influencing factor of the h-th offset indicator;

[0133] Taking the image frame corresponding sequence set consisting of frame 1 and frame 2 as the object, the image frame parameters captured by the image recognition system, the estimated flight state parameters, and the image offset index are extracted in sequence. The frame 1 parameters are set as follows:

[0134] The center coordinates of the drone in the image are (240px, 180px), the image resolution is 480×360, and the normalized values are: horizontal coordinate 240 / 480=0.5, vertical coordinate 180 / 360=0.5, the image attitude angle is 10°, and the normalized value is 10 / 20=0.5 within the set range 0°-20°. The synthetic image capture parameters The flight status is estimated as follows: speed 3.2m / s (normalized to 0.8 for the range 0-4), altitude 100.0m (normalized to (100-90) / 30=0.333 for the range 90-120), heading angle 20°

[0135] (0~40 normalized to 0.5), synthetic flight state parameters The image offset index HL1 = 4.2 (%), which indicates the offset percentage between the image recognition frame and the center of the actual flight estimated trajectory. The adjustment influence factors are set as: α1 = 0.6, β1 = 0.5, γ1 = 0.5;

[0136] Frame 2 parameter settings:

[0137] The center coordinates of the drone in the image (252px, 190px) are normalized to 252 / 480≈0.525, 190 / 360≈0.528. The attitude angle is 12°, normalized to 12 / 20=0.6. The flight status is estimated as follows: speed 3.4m / s (normalized to 0.85), altitude 100.2m (normalized to 0.4), heading angle 25° (normalized to 0.625), Deviation index HL2 = 4.5, α2 = 0.7, β2 = 0.6, γ2 = 0.5;

[0138] Numerator calculation:

[0139] Numerator 1 = 0.6 0.5 + 0.5 0.5443 = 0.3 + 0.2722 = 0.5722;

[0140] Numerator 2 = 0.7 0.551 + 0.6 0.625 = 0.3857 + 0.375 = 0.7607;

[0141] Total numerator = 0.5722 + 0.7607 = 1.3329;

[0142] Denominator calculation:

[0143] Denominator 1 = 0.5 4.2 = 2.1;

[0144] Denominator 2 = 0.5 4.5 = 2.25;

[0145] Total denominator = 2.1 + 2.25 = 4.35;

[0146] Get PL d :

[0147]

[0148] The result of dynamic adjustment parameters is PL d =0.3064, which falls within the system's preset response range of [0.3, 0.4]. This indicates that the tracking status of the current two frames of image for the tracked drone is within the adjustment stability range, and the offset between the image recognition center and the estimated flight trajectory is within an acceptable range. This value is used as the dynamic tracking parameter adjustment result to adjust the subsequent recognition frame position and update the frame capture strategy to ensure that the image recognition system continues to stably track the tracked drone.

[0149] The drone tracking method based on image recognition includes the following steps:

[0150] S1: Based on the captured drone image, the brightness information of the ambient light and the drone's reflected light is extracted, the brightness is compared with the surrounding ambient light intensity pixel by pixel, the key areas of brightness deviation are identified, and the response parameters of the infrared and visible light channels are adjusted to obtain the updated image sequence;

[0151] S2: Based on the updated image sequence, obtain the image bounding box and the corresponding timestamp of the drone, compare the spatial position data of each frame, analyze the position change of the center of the bounding box, calculate the real-time movement path of the drone, and generate the drone trajectory data;

[0152] S3: Based on the drone trajectory data, the position of the drone in the continuous image frames is locked, and the position changes of the drone's edge pixels between each two frames are compared. If the changes exceed the normal range, the edge pixel positions are reconfigured to obtain the morphological adjustment offset index;

[0153] S4: Using the drone trajectory data, calculate the drone's speed and acceleration changes based on the time interval and position data, compare them with the normal flight data, identify the trajectory segments that differ from the standard flight pattern, and obtain the abnormal flight segment index;

[0154] S5: Based on the abnormal flight segment index and the morphological adjustment offset index, the corresponding image sequence frames are synchronously retrieved, and the tracking process of the UAV is adjusted in combination with the current image capture parameters to obtain the dynamic tracking parameter adjustment result.

[0155] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. The drone tracking system based on image recognition is characterized by: The system comprises: The spectral optimization module extracts the brightness information of the ambient light and the reflected light from the drone based on the captured drone image. It compares the brightness with the surrounding ambient light intensity pixel by pixel, identifies key areas of brightness deviation, and adjusts the response parameters of the infrared and visible light channels to optimize image accuracy and obtain an updated image sequence. The trajectory analysis module obtains the image bounding box and corresponding timestamp of the drone based on the updated image sequence, compares the spatial position data of each frame, analyzes the position change of the center of the bounding box, calculates the real-time movement path of the drone, and generates drone trajectory data; The morphology adjustment module locks the position of the drone in consecutive image frames based on the drone trajectory data, compares the position changes of the drone edge pixels between each two frames, reconfigures the edge pixel positions, and obtains the morphology adjustment offset index; The flight anomaly detection module uses the UAV trajectory data to calculate the speed and acceleration changes of the UAV based on the time interval and position data, compares them with the normal flight data, identifies the trajectory segments that differ from the standard flight pattern, and obtains the abnormal flight segment index.

2. The UAV tracking system based on image recognition according to claim 1, characterized in that: The updated image sequence includes brightness level, color saturation, and contrast adjustment information; the drone trajectory data includes flight altitude changes, direction offset data, and speed change graphs; the morphological adjustment offset indicators include shape alignment, edge clarity, and pixel adjustment range; and the abnormal flight segment index includes flight altitude abnormality segments, speed mutation segments, and navigation deviation indicators.

3. The drone tracking system based on image recognition according to claim 1, characterized in that: The spectrum optimization module includes: The brightness extraction submodule extracts the reflected light brightness value from the infrared channel and visible light channel of each pixel based on the captured drone image, calculates the deviation between the pixel brightness and the average brightness of the ambient light, and obtains the brightness offset value distribution; The difference recognition submodule sets an ambient brightness deviation threshold based on the brightness offset value distribution, screens pixel areas with brightness higher than the threshold, analyzes pixel aggregation and brightness direction consistency, and identifies deviation abnormal areas with high aggregation and consistent direction; The response adjustment submodule calls the offset abnormal area and analyzes the brightness difference between the infrared and visible light channels using the formula: Calculate the response channel adjustment coefficient ΔRC and adjust the brightness response parameters to optimize the image clarity to obtain the updated image sequence, where: represents the infrared channel brightness of the i-th pixel, Represents the visible light channel brightness of the i-th pixel, n rc Indicates the total number of pixels in the critical area of brightness deviation, Represents the brightness value of the visible light channel of the j-th pixel, Indicates the infrared channel brightness value of the jth pixel, m rc Indicates the number of pixels involved in the channel response difference calculation.

4. The drone tracking system based on image recognition according to claim 1, characterized in that: The trajectory analysis module includes: The bounding box extraction submodule extracts the drone image bounding box in each frame of the image based on the updated image sequence, detects the center coordinate information of the bounding box and the corresponding timestamp, and constructs a time positioning coordinate set by combining the image frame number and acquisition time; The position change calculation submodule calls the time positioning coordinate set and uses the formula: Calculate the path change rate between every two frames, where VX o Indicates the path change rate per unit time between the oth frame and the o+1th frame, Qx o Represents the horizontal coordinate of the center of the bounding box of the oth frame, Qy o Represents the vertical coordinate of the center of the bounding box of the oth frame, Qt o Represents the timestamp of the oth frame, Qx o+1 Represents the horizontal coordinate of the center of the bounding box of the o+1th frame, Qy o+1 Represents the vertical coordinate of the center of the bounding box of the o+1th frame, Qt o+1 Represents the timestamp of the o+1th frame; The path construction submodule calibrates the continuous connection relationship of the trajectory points frame by frame according to the path change rate, establishes the connection order between the nodes according to the path rate, and generates the UAV trajectory data.

5. The UAV tracking system based on image recognition according to claim 1, characterized in that: The morphology adjustment module includes: The trajectory locking submodule extracts coordinate points of the drone position in consecutive image frames based on the drone trajectory data, screens stable data points by comparing the coordinate changes of the drone pixel center of gravity in adjacent image frames, and locates the real-time trajectory of the drone to obtain the drone image displacement trajectory; The edge change analysis submodule calls the displacement trajectory of the drone image, monitors the edge pixels of each frame of the drone, calculates the coordinate changes of the pixels on the image plane, identifies the offset anomalies of the edge pixels, and obtains the abnormal edge displacement difference; The offset index extraction submodule reconfigures the edge pixel position according to the abnormal edge displacement difference and the changes beyond the normal range, using the formula: Get the morphological adjustment offset indicator OU, where OE o Represents the abnormal edge displacement difference of the oth frame, OL o Indicates the expected deviation position of the edge in the oth frame, n ou Represents the total number of image frames processed.

6. The drone tracking system based on image recognition according to claim 1, characterized in that: The flight anomaly detection module includes: The speed calculation submodule uses the UAV trajectory data to extract continuous timestamps and position coordinates, calculates the time difference for each pair of continuous timestamps, determines the distance between the position points corresponding to the two time points, and generates a speed change sequence; The acceleration analysis submodule calls the speed change sequence, compares the continuous speed data, matches the time intervals of each pair of continuous speed data in the sequence, calculates the acceleration value of each time period, and obtains the acceleration change trend; The state recognition submodule compares the acceleration change trend with the acceleration distribution characteristics in the standard flight trajectory sample segment by segment, analyzes the difference between acceleration and speed in each segment, and uses the formula: Obtain the state deviation of each trajectory segment, judge the flight state of the UAV, and obtain the abnormal flight segment index, where SN z Represents the state deviation of the zth segment, VN zc represents the actual velocity of the cth point in the zth segment, Represents the standard speed value of the cth point in the zth segment, AN zc Indicates the actual acceleration value of the cth point in the zth segment, n sn represents the number of trajectory segments, and λ is the unit correction coefficient.

7. The UAV tracking system based on image recognition according to claim 1, characterized in that: The system further comprises: The tracking optimization module synchronously retrieves the corresponding image sequence frames based on the abnormal flight segment index and morphological adjustment offset index, and adjusts the UAV tracking process in combination with the current image capture parameters to match the flight anomaly or emergency situation, and obtains the dynamic tracking parameter adjustment result; The dynamic tracking parameter adjustment result includes a tracking sensitivity adjustment result, a response time optimization result, and a tracking accuracy optimization result.

8. The image recognition-based drone tracking system according to claim 7, characterized in that: The tracking optimization module includes: The index extraction submodule synchronously retrieves the UAV image data corresponding to the index based on the abnormal flight segment index and the morphological adjustment offset index, analyzes the correlation between the flight state and the image sequence, identifies the key parameters affecting the flight trajectory, and obtains the state number matching segment; The image frame synchronization submodule calls the state number matching section to perform time axis matching on the image sequence frames, and at the same time uses the image capture parameters to match the frame numbers, compares the frame numbers with the timing offsets of the original image frames, selects frames that meet the image number consistency, and generates a corresponding sequence set of image frames; The dynamic adjustment submodule calls the corresponding sequence set of the image frames and performs dynamic optimization analysis using the formula: Adjust the UAV flight tracking process and obtain the dynamic tracking parameter adjustment result PL d , where VL b Indicates the capture parameters of the bth image sequence frame, SL b Indicates the adjustment parameter of the bth flight state, HL h represents the hth offset index, α b represents the adjustment factors affecting the capture parameters of the b-th image sequence frame, β b represents the adjustment factor affecting the b-th flight state adjustment parameter, γ h Represents the adjustment influencing factor of the h-th offset indicator.

9. The drone tracking method based on image recognition is characterized in that: The image recognition-based drone tracking system according to any one of claims 1 to 8 comprises the following steps: S1: Based on the captured drone image, the brightness information of the ambient light and the drone's reflected light is extracted, the brightness is compared with the surrounding ambient light intensity pixel by pixel, the key areas of brightness deviation are identified, and the response parameters of the infrared and visible light channels are adjusted to obtain the updated image sequence; S2: Based on the updated image sequence, obtain the image bounding box and the corresponding timestamp of the drone, compare the spatial position data of each frame, analyze the position change of the center of the bounding box, calculate the real-time movement path of the drone, and generate drone trajectory data; S3: Based on the drone trajectory data, lock the position of the drone in consecutive image frames, compare the position changes of the drone edge pixels between every two frames, reconfigure the edge pixel positions for changes that exceed the normal range, and obtain a morphological adjustment offset index; S4: Using the UAV trajectory data, calculate the speed and acceleration changes of the UAV according to the time interval and position data, compare the calculated speed and acceleration changes with the normal flight data, identify the trajectory segments that differ from the standard flight pattern, and obtain the abnormal flight segment index; S5: Based on the abnormal flight segment index and the morphological adjustment offset index, the corresponding image sequence frames are synchronously retrieved, and the tracking process of the UAV is adjusted in combination with the current image capture parameters to obtain a dynamic tracking parameter adjustment result.

Images