UAV precise location landing method and system based on image recognition analysis

Abstract

The present invention discloses a method and system for accurately landing a UAV based on image recognition and analysis, which specifically relates to the technical field of aircraft landing, and aims to solve the problem of the influence of jitter interference caused by micro-airflow on landing accuracy; by real-time detection of the posture changes of the UAV during the landing process, the jitter signal strength is identified and it is judged whether it is abnormal; when the jitter is abnormal, multi-frame image superposition processing is performed to generate a stable image sequence; based on multi-resolution fractal dimension analysis, subtle texture disturbances between adjacent frames are detected to compensate for the nonlinear effects brought by micro-airflow; the jitter trajectory is recorded and the relative displacement of the landing target is calculated by inverse trajectory reconstruction; the convolutional neural network is combined to extract the calibrated landing area features, generate the offset compensation value of the landing point and dynamically correct the landing path; the present invention improves the landing accuracy of the UAV in complex environments and provides a reliable guarantee for the automatic landing of the UAV.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft landing, and more specifically, to a method and system for accurately landing an unmanned aerial vehicle (UAV) based on image recognition and analysis. Background Art

[0002] As the application of drones continues to expand, the need for precise landing of drones has become particularly important in multiple application scenarios, especially in unmanned, automated operations, and landing in confined spaces. Drones need to have high-precision landing capabilities. However, during the landing process of a drone, the attitude of the drone may change slightly and continuously due to the interference of micro-airflows. The irregular attitude fluctuations caused by these micro-airflow interferences will directly affect the camera image of the drone, causing slight image jitter.

[0003] In the prior art, image recognition and processing algorithms are usually used to assist drones in identifying the landing target location. However, the existing image recognition algorithms are difficult to filter out these subtle image jitter effects when faced with interference from tiny jitters, resulting in large positioning errors of target points in the image. Especially when the drone is landing slowly, the disturbance of micro-airflow will cause imperceptible jitters and changes in texture details in the image. This jitter will produce cumulative errors in the image superposition processing, affecting the stability and accuracy of drone image recognition.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method and system for accurately landing a UAV based on image recognition and analysis to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for accurately landing a UAV at a certain position based on image recognition analysis includes the following steps:

[0008] S1: Real-time detection of the attitude changes of the drone during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal;

[0009] S2: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence;

[0010] S3: Evaluate the texture changes of multi-frame superimposed images based on multi-resolution fractal dimension analysis, detect subtle texture disturbances between adjacent frames, and compensate for the nonlinear effects of micro-airflow on image features by evaluating the complexity of ground texture changes;

[0011] S4: Record the vibration trajectory caused by micro-airflow during the landing of the UAV, and analyze the relative displacement of the landing target based on the reverse trajectory reconstruction;

[0012] S5: Extract the features of the calibrated landing area through a convolutional neural network, correct the feature coordinate offset caused by the jitter in combination with the attitude sensor data, and generate the offset compensation value of the landing point;

[0013] S6: Combined with the generated offset compensation value, the landing path of the UAV is dynamically corrected in real time.

[0014] In a preferred embodiment, the posture change of the drone during landing is detected in real time, the jitter signal strength caused by the micro-airflow is identified, and whether the jitter signal strength is abnormal is determined, specifically including:

[0015] S101: collect the three-dimensional spatial attitude information of the UAV in the landing process in real time through the attitude sensor;

[0016] S102: Analyze the collected posture data to separate the posture change signal caused by the micro-airflow;

[0017] S103: calculating the amplitude value of the posture change based on the separated posture change signal, and identifying the intensity of the jitter signal caused by the micro-airflow based on the standard deviation of the amplitude value of the posture change;

[0018] S104: Based on the standard deviation of the amplitude value of the posture change, determine whether the jitter signal strength is abnormal.

[0019] In a preferred embodiment, when the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of the jitter on the image features and generate a stable image sequence, which specifically includes:

[0020] S201: after identifying that the jitter signal strength is abnormal, collecting continuous multiple frames of image data from the camera of the drone in real time;

[0021] S202: setting the time window length of time series superposition according to the jitter characteristics, and determining the number of frames used for image superposition;

[0022] S203: using an inter-frame alignment algorithm to perform time sequence alignment on the collected multiple frames of images;

[0023] S204: performing pixel-level averaging processing on the multiple frames of images after time sequence alignment;

[0024] S205: Outputting the processed image average result as a stable image sequence.

[0025] In a preferred embodiment, pixel-level averaging is performed on the time-aligned multiple-frame images, specifically: the average value of each pixel in all frames is selected, and the calculation formula is as follows:

[0026] ;in, represents the gray value of the generated stable image at the pixel position (x, y), N is the number of superimposed frames, For the The grayscale value of the frame at the pixel position (x,y), Indicates the frame number of the currently superimposed image, and x and y represent the pixel positions on the horizontal and vertical axes of the image, respectively.

[0027] In a preferred embodiment, texture changes of multiple superimposed images are evaluated based on multi-resolution fractal dimension analysis, subtle texture disturbances between adjacent frames are detected, and the nonlinear effect of micro-airflow on image features is compensated by evaluating the complexity of ground texture changes, specifically including:

[0028] S301: decomposing the multi-frame superimposed images into different resolution levels by a multi-resolution decomposition method to extract multi-scale texture information;

[0029] S302: using a fractal calculation method at each resolution level to obtain a fractal dimension value corresponding to the resolution level;

[0030] S303: Arrange the fractal dimension values ​​of each resolution level in chronological order to construct a time series of multi-resolution texture changes;

[0031] S304: comparing fractal dimension values ​​of adjacent frames in the time series to obtain fractal differences of adjacent frames at each resolution level;

[0032] S305: Calculating texture change compensation parameters for compensating texture changes according to the difference in fractal dimensions between adjacent frames.

[0033] In a preferred embodiment, the texture change compensation parameter can be expressed as: ; Where C represents the final texture change compensation parameter, M represents the total number of resolution levels, is the weight of the mth resolution level, It represents the difference value of the fractal dimension between adjacent frames at the mth resolution level, k represents the index of the adjacent frame in the time series, and m represents the number of the resolution level.

[0034] In a preferred embodiment, the jitter trajectory caused by the micro-airflow during the landing of the drone is recorded, and the relative displacement of the landing target is analyzed based on the reverse trajectory reconstruction, which specifically includes:

[0035] S401: real-time collection of three-dimensional attitude data and position data of the UAV during landing;

[0036] S402: Separating the vibration trajectory data caused by the micro-airflow from the three-dimensional posture data and the position data by using a filtering method;

[0037] S403: Arrange the jitter trajectory data in chronological order to generate a complete time series of the jitter trajectory;

[0038] S404: Based on the inverse trajectory reconstruction method, analyzing the change of the jitter trajectory in the complete time series of the jitter trajectory, and calculating the relative displacement of the landing target;

[0039] S405: Generate calibration parameters according to the relative displacement of the landing target, and perform position correction on the landing target point position.

[0040] In a preferred embodiment, the calibrated landing area features are extracted through a convolutional neural network, and the feature coordinate offset caused by the jitter is corrected in combination with the posture sensor data to generate an offset compensation value of the landing point, specifically:

[0041] S501: Compare the image features and posture data of the landing area, and analyze the impact of posture changes on image feature coordinates;

[0042] S502: Calculate the actual offset of the image feature coordinates according to the influence of the posture change on the image feature coordinates;

[0043] S503: Generate an offset compensation value of the landing point according to the actual offset of the image feature coordinates, which is used to correct the landing path of the drone.

[0044] In a preferred embodiment, the landing path of the UAV is dynamically corrected in real time in combination with the generated offset compensation value, specifically including:

[0045] S601: Calculate the spatial deviation between the current actual position and the target position according to the current drone position data;

[0046] S602: applying the offset compensation value to the landing path control to gradually adjust the position deviation of the UAV;

[0047] S603: updating the attitude angle and motion parameters of the UAV in real time according to the corrected landing path;

[0048] S604: Continuously obtain the deviation data and correct the landing path in real time until the UAV lands at the predetermined target location.

[0049] On the other hand, the present invention provides a drone precise location landing system based on image recognition analysis.

[0050] It includes a jitter intensity determination module, an image sequence generation module, an airflow impact analysis module, a reverse trajectory analysis module, an offset compensation correction module, and a real-time dynamic correction module;

[0051] Jitter strength determination module: real-time detection of the drone's attitude changes during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal;

[0052] Image sequence generation module: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence;

[0053] Airflow impact analysis module: It evaluates texture changes of multi-frame superimposed images based on multi-resolution fractal dimension analysis, detects subtle texture disturbances between adjacent frames, and compensates for the nonlinear effects of micro-airflow on image features by evaluating the complexity of ground texture changes;

[0054] Reverse trajectory analysis module: records the vibration trajectory caused by micro-airflow during the landing of the drone, and reconstructs and analyzes the relative displacement of the landing target based on the reverse trajectory;

[0055] Offset compensation correction module: extracts the features of the calibrated landing area through a convolutional neural network, corrects the feature coordinate offset caused by jitter in combination with the attitude sensor data, and generates the offset compensation value of the landing point;

[0056] Real-time dynamic correction module: Combined with the generated offset compensation value, the landing path of the drone is corrected in real time.

[0057] The technical effects and advantages of the method and system for accurately landing a drone based on image recognition and analysis of the present invention are as follows:

[0058] 1. The present invention detects the posture changes of the drone during the landing process in real time, identifies and determines whether the jitter signal strength caused by micro-airflow is abnormal. When the jitter signal reaches a certain intensity, it triggers the multi-frame image superposition algorithm and performs time-series superposition processing on the image sequence, thereby weakening the direct impact of tiny jitters on image features and generating a stable image sequence. Compared with the existing technology that is difficult to effectively handle tiny jitters, the present invention can accurately compensate for image jitters under the influence of micro-airflows through image superposition and fractal dimension analysis, further improving the stability of the drone landing image, making subsequent image recognition analysis more accurate and reliable, and effectively reducing recognition errors caused by jitters.

[0059] 2. The present invention combines a multi-resolution fractal dimension analysis method to evaluate the texture changes of the generated stable image, so as to detect subtle texture disturbances between adjacent frames and compensate for the nonlinear changes in image features caused by micro-airflows. The jitter trajectory of the drone caused by micro-airflows is recorded and analyzed through inverse trajectory reconstruction technology, the relative displacement of the landing target is calculated, and the convolutional neural network is further used to extract the calibrated landing area features, and the offset compensation value of the landing point is generated in combination with the attitude sensing data to achieve dynamic real-time correction of the landing path. This process significantly improves the accuracy of the drone's landing position, enabling the drone to effectively eliminate external interference and accurately land in the target area, providing a reliable guarantee for the high-precision automatic landing of the drone. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a method for accurately landing a UAV based on image recognition and analysis according to the present invention;

[0061] Figure 2 It is a structural schematic diagram of the UAV precise position landing system based on image recognition analysis of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0063] Figure 1 The present invention provides a method for accurately landing a UAV based on image recognition analysis, which includes the following steps:

[0064] S1: Real-time detection of the attitude changes of the drone during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal.

[0065] S2: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of the jitter on the image features and generate a stable image sequence.

[0066] S3: Texture changes of multi-frame superimposed images are evaluated based on multi-resolution fractal dimension analysis to detect subtle texture disturbances between adjacent frames. The nonlinear effect of micro-airflow on image features is compensated by evaluating the complexity of ground texture changes.

[0067] S4: Record the vibration trajectory caused by micro-airflow during the landing of the UAV, and analyze the relative displacement of the landing target based on the reverse trajectory reconstruction.

[0068] S5: Extract the features of the calibrated landing area through a convolutional neural network, and correct the feature coordinate offset caused by the jitter in combination with the attitude sensor data to generate an offset compensation value for the landing point.

[0069] S6: Combined with the generated offset compensation value, the landing path of the UAV is dynamically corrected in real time.

[0070] Real-time detection of the attitude changes of the drone during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal, including:

[0071] S101: The attitude sensor collects the three-dimensional attitude information of the drone in real time during the landing process:

[0072] During the landing process of the drone, the attitude sensor continuously collects the drone's three-dimensional spatial attitude information. The three-dimensional spatial attitude information records the spatial orientation and attitude angle of the drone in the three-dimensional coordinate system, including the roll angle, pitch angle, and yaw angle, which respectively represent the angle changes of the drone along the front-back axis, left-right axis, and vertical axis. These data are based on time to form a sequence of attitude changes over time, which is used to capture the impact of micro-airflow on the instantaneous attitude of the drone, and provide raw data support for the subsequent analysis of attitude changes and the identification of jitter signals.

[0073] S102: Analyze the collected posture data and separate the posture change signal caused by the micro-airflow:

[0074] The acquired attitude data sequence is input into the data analysis module for multi-level dynamic change analysis. In order to distinguish the attitude changes caused by the UAV's own motion from the subtle disturbances caused by micro-airflow, the attitude data is first trend-smoothed to remove the attitude changes caused by the stable motion of the UAV, so as to highlight the fast and short-period jitter characteristics. The upper limit of the dynamic change rate of each attitude angle is set, and the attitude changes exceeding this rate can be regarded as possible signals of micro-airflow disturbance. Next, the high-frequency change component of each attitude angle is extracted through a high-pass filter, which represents the rapid change of the attitude in a small time interval. This fast-changing signal can be used as a possible manifestation of micro-airflow interference, thereby preliminarily realizing the separation of the attitude disturbance signal caused by micro-airflow.

[0075] S103: Based on the separated posture change signal, the amplitude value of the posture change is calculated, and based on the standard deviation of the amplitude value of the posture change, the intensity of the jitter signal caused by the micro-airflow is identified:

[0076] Assuming that the roll angle, pitch angle and yaw angle are quantized in the form of rotation matrices and recorded as three-dimensional vectors, where the roll angle is recorded as R, the pitch angle is recorded as P, and the yaw angle is recorded as Y, the amplitude value of the attitude change at any moment can be expressed by the modulus of the three-dimensional vector, and the formula is: ; Where A is the amplitude value of the posture change.

[0077] Calculate the standard deviation of the amplitude value of the attitude change during the landing process of the UAV. The interval corresponding to the standard deviation calculation is determined according to the actual situation.

[0078] The larger the standard deviation of the amplitude value of the attitude change during the landing process of the UAV, the greater the intensity of the jitter signal caused by the micro-airflow, indicating that the volatility of the attitude change is strong, which means that the jitter is unstable and changes frequently, reflecting a higher jitter intensity.

[0079] The calculation range of the standard deviation can be determined according to the specific stage of the drone landing process or the actual operating environment. Usually, the appropriate range can be selected according to the following actual conditions:

[0080] Using a fixed-length time window, such as calculating the standard deviation every 1 or 5 seconds, can maintain a consistent analysis rhythm. This method is suitable for stable, continuous micro-airflow environments to regularly monitor the volatility of posture changes.

[0081] If the drone is particularly susceptible to micro-air currents at certain critical stages of the landing process (such as when close to the ground), separate intervals can be set for these stages. This allows the evaluation of attitude changes at critical moments to allow for timely adjustment of the landing attitude and compensation actions.

[0082] S104: Based on the standard deviation of the amplitude value of the posture change, determine whether the jitter signal strength is abnormal:

[0083] When the standard deviation of the amplitude value of the attitude change is greater than the corresponding preset threshold, the jitter signal strength is judged to be abnormal; this situation indicates that the attitude change of the drone fluctuates greatly, which means that the disturbance of the micro-airflow during the landing process is strong and unstable.

[0084] When the standard deviation of the amplitude value of the attitude change is less than or equal to the corresponding preset threshold, the jitter signal strength is determined to be normal; this situation indicates that the amplitude of the attitude change of the drone is within a controllable range, that is, the intensity of the jitter during the landing process is small and relatively stable.

[0085] Among them, the preset threshold value corresponding to the standard deviation of the amplitude value of the attitude change is set based on the standard deviation of the amplitude value of the attitude change of the drone in a stable landing environment to determine the normal jitter range. First, under ideal landing conditions without micro-airflow interference, a large amount of attitude data is collected and its standard deviation is calculated to obtain a reference value. Then, based on the risk level and jitter tolerance of different landing environments, a certain safety margin is added to the reference value to set the threshold. Usually, a lower threshold is set in high-risk areas (such as areas with high wind speeds or more obstacles) to more sensitively detect abnormal jitters; in environments with lower wind speeds, the threshold can be appropriately increased to reduce the probability of misjudgment.

[0086] When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence, including:

[0087] S201: After identifying that the jitter signal strength is abnormal, continuous multi-frame image data is collected from the drone's camera in real time:

[0088] After confirming that the jitter signal strength fed back by the attitude sensor is abnormal, the drone starts the image acquisition mode and continuously obtains multiple frames of image data from the camera. These image data are closely adjacent in time, forming a time series, which is designed to capture multi-dimensional image information under the jitter state. These frames not only contain the landing area of ​​the drone, but also record the influence of attitude changes under micro-airflow disturbances in the area in real time, providing continuous image input for subsequent image processing steps.

[0089] S202: Setting the time window length of the time sequence superposition according to the jitter characteristics, and determining the number of frames used for image superposition:

[0090] After the image acquisition is completed, according to the characteristics of the identified jitter signal, an appropriate time window length is set to determine the number of frames used for image superposition. The length of the time sequence superposition window is determined by factors such as the frequency and intensity of the jitter and the rate of change of the drone's attitude. For example, in the case of strong micro-airflow, the time window can be shortened to quickly capture and process the instantaneous impact of attitude changes on the image; when the jitter frequency is low, a longer window is selected to average the jitter. Finally, the timeliness of the superposition process and the stability of the image are ensured by setting the number of frames.

[0091] S203: Using an inter-frame alignment algorithm to perform time sequence alignment on the collected multiple frames of images:

[0092] In order to ensure that multiple frames of images are not blurred or misaligned during the superposition process, the inter-frame alignment algorithm is used to align the collected images in time. The inter-frame alignment algorithm can be based on methods such as image feature points or histogram matching, so that each frame has the same spatial reference point during the superposition process. For example, by selecting the significant feature points of the landing area as reference points, the position of each frame is corrected to eliminate the displacement error caused by the shaking of the drone. This time-series alignment process ensures the spatial consistency of each image, making the subsequent superposition calculation more accurate.

[0093] S204: Perform pixel-level averaging on the multiple frames of images after time sequence alignment:

[0094] After the time sequence alignment is completed, the aligned multi-frame images are processed by pixel-level averaging to generate a stable image. Specifically, the average value of each pixel in all frames is selected, and the calculation formula is as follows: ;in, represents the gray value of the generated stable image at the pixel position (x, y), N is the number of superimposed frames, For the The grayscale value of the frame at the pixel position (x,y), Indicates the frame number of the currently superimposed image, and x and y represent the pixel positions on the horizontal and vertical axes of the image, respectively.

[0095] S205: Output the processed image average result as a stable image sequence:

[0096] After the pixel-level averaging process is completed, a stable image is obtained, which represents the current state of the drone landing area. This stable image is used as the input image for the subsequent steps to ensure the integrity and clarity of the image features in the case of abnormal jitter signals. This stable image sequence can be continuously generated, and the new jitter effect is removed in each frame update to form a dynamic stable image sequence during the landing process.

[0097] Based on multi-resolution fractal dimension analysis, texture changes of multi-frame superimposed images are evaluated to detect subtle texture disturbances between adjacent frames. By evaluating the complexity of ground texture changes, the nonlinear effect of micro-airflow on image features is compensated. Specifically, it includes:

[0098] S301: Decomposing the multi-frame superimposed images into different resolution levels by a multi-resolution decomposition method to extract multi-scale texture information:

[0099] After obtaining multiple frames of superimposed images, the image is subjected to multi-resolution decomposition. Multi-resolution decomposition decomposes the original image into multiple different resolution levels, each level corresponding to image features at a scale. The purpose of multi-resolution decomposition is to extract texture information at different scales to capture the impact of micro-airflow on image texture at different scales. The image generated at each resolution level represents the texture features of that level and can be used to analyze the changes in texture complexity at different resolutions. The implementation method of multi-resolution decomposition can be based on decomposition algorithms such as wavelet transform and Laplace pyramid to obtain image matrices at each resolution level.

[0100] S302: Using a fractal calculation method at each resolution level to obtain a fractal dimension value corresponding to the resolution level:

[0101] Fractal dimension is a parameter used to quantify the complexity of image texture, reflecting the self-similarity and texture distribution characteristics of the image at the current resolution level.

[0102] Suppose the image matrix of a certain resolution level is , where m represents the number of the resolution level.

[0103] The fractal dimension can be calculated by the box dimension method. Specifically, in the image matrix Set boxes of different scales above, and get the fractal dimension by counting the number of pixels in the box: ;in, Indicates the fractal dimension value of the mth resolution level; Indicates the number of boxes required at scale b; b represents the scale parameter of the box. The fractal dimension value is used to quantify the texture complexity at the current resolution level.

[0104] S303: Arrange the fractal dimension values ​​of each resolution level in chronological order to construct a time series of multi-resolution texture changes:

[0105] Set the fractal dimension value represents the fractal dimension value of the mth resolution level at time point u, then the time series It can be expressed as: ;in, Represents multiple adjacent time points, Represents the texture change time series of the mth resolution level, and n represents the total number of frames in the time series.

[0106] Through time series The temporal variation trend of the fractal dimension value can be observed, further characterizing the dynamic changes of the texture at this resolution level.

[0107] S304: Compare the fractal dimension values ​​of adjacent frames in the time series to obtain the fractal differences of adjacent frames at each resolution level:

[0108] In time series In , the fractal dimension values ​​of adjacent frames are compared, and the fractal difference of each pair of adjacent time points is calculated. Assume that at the mth resolution level, the adjacent time points and The fractal dimension values ​​are and , then the fractal difference between adjacent frames can be expressed as: ;in, represents the fractal dimension difference value of adjacent frames at the mth resolution level, and k represents the index of the adjacent frame in the time series.

[0109] The degree of texture change between adjacent frames can be quantified to support the subsequent compensation steps.

[0110] S305: Calculate texture change compensation parameters based on the difference in fractal dimensions between adjacent frames to compensate for texture changes:

[0111] In obtaining Finally, the fractal difference results of all resolution levels are combined to generate the compensation parameters of texture changes. The compensation parameters can be calculated by weighted average method, in which the fractal difference values ​​of each resolution level are weighted according to the weight, and the texture change compensation parameters can be expressed as: ; Where C represents the final texture change compensation parameter; M represents the total number of resolution levels; is the weight of the mth resolution level.

[0112] The texture variation compensation parameter represents the overall variation degree of the texture and is used to adjust the image features to compensate for the nonlinear effect of micro airflow on the texture.

[0113] The compensation effect of the texture change compensation parameter is to adjust the pixel values ​​or image features in multiple frames of images to weaken the nonlinear disturbance caused by micro-airflow, so as to make the texture features of the image more stable. The specific method is: according to the size of the texture change compensation parameter, the multi-frame superimposed image is adjusted pixel by pixel or the contrast and brightness are fine-tuned. For example, when the texture change compensation parameter is large, it indicates that the texture disturbance is obvious. The edge instability caused by the disturbance can be weakened by reducing the brightness difference between adjacent pixels; when the texture change compensation parameter value is small, the current image features are kept unchanged. In addition, the texture change compensation parameter can also be used to control the intensity of the smoothing filter. By adjusting the filter parameters, the sudden texture is further weakened, thereby realizing the texture change compensation of the image and ensuring the stability of the image features.

[0114] Record the vibration trajectory caused by micro-airflow during the landing of the drone, and reconstruct and analyze the relative displacement of the landing target based on the reverse trajectory, including:

[0115] S401: Real-time collection of 3D attitude data and position data of the drone during landing:

[0116] During the landing process of the drone, the attitude sensor and positioning system collect the drone's three-dimensional attitude data and position data in real time. The attitude data includes roll angle, pitch angle and yaw angle, which respectively represent the angle changes of the drone around its three axes, and are used to reflect the angle attitude changes of the drone in space.

[0117] The position data is the spatial position of the drone in a three-dimensional coordinate system, which usually includes position coordinates in the horizontal and vertical directions, representing the real-time position of the drone on the landing path.

[0118] The collected attitude data and position data form a time series, recording the attitude and position change information of the drone under the influence of micro-airflow during landing. These data provide the basis for subsequent jitter trajectory extraction and reverse trajectory reconstruction.

[0119] S402: Separate the vibration trajectory data caused by the micro-airflow from the three-dimensional posture data and the position data using a filtering method:

[0120] The filtering method is used to separate the jitter trajectory caused by micro-airflow. Specifically, the attitude data and position data are processed using a high-pass filter to filter out the low-frequency motion changes caused by the autonomous landing action of the drone, and retain the high-frequency attitude and position fluctuations, which are the jitter trajectory under the interference of micro-airflow.

[0121] The setting of the filtering method includes setting the frequency cutoff value of the filter to separate the high frequency jitter data.

[0122] S403: Arrange the jitter trajectory data in chronological order to generate a complete time series of the jitter trajectory:

[0123] The time series is represented as a dynamic record of the jitter trajectory data during the entire landing process, including the attitude changes and position changes at each moment.

[0124] The time series records the changes in the jitter trajectory at different time points and is the basic data for inverse trajectory reconstruction, reflecting the dynamic impact of micro-airflow disturbances on the attitude and position of the UAV.

[0125] S404: Based on the reverse trajectory reconstruction method, the change of the jitter trajectory in the complete time series of the jitter trajectory is analyzed to calculate the relative displacement of the landing target:

[0126] The purpose of reverse trajectory reconstruction is to infer the initial offset of the landing target through the change of the jitter trajectory. Assume that the preset position of the UAV under ideal interference-free conditions is , then the relative displacement caused by the jitter trajectory can be expressed as: ;in, is the relative displacement caused by the jitter trajectory, Indicates at time The preset target position at all times, Indicates at time The actual position affected by the jitter at the moment, c represents the frame number in the complete time series of the jitter trajectory, and C represents the total number of frames in the complete time series of the jitter trajectory. Represents the time point corresponding to the cth frame in the complete time series of the shaking trajectory.

[0127] Through the inverse trajectory reconstruction process, the relative displacement caused by the jitter during the landing of the UAV is calculated to provide data support for the calibration of the landing target position.

[0128] S405: Generate calibration parameters according to the relative displacement of the landing target, and perform position correction on the landing target point:

[0129] After the relative displacement caused by the shaking trajectory is obtained, the corresponding calibration parameters are generated according to the relative displacement caused by the shaking trajectory to correct the landing target point position in the image.

[0130] The calibration parameters are set to , whose value is equal to the opposite number of the relative displacement caused by the jitter trajectory, is used to offset the deviation caused by the micro-airflow disturbance. Assuming that the calibration position of the landing target point after adjustment is L, then: ; Where T represents the target point position before calibration.

[0131] By adjusting the target point position and compensating for the offset caused by jitter, the landing target point in the corrected image is more accurate. This calibration process ensures the consistency of the final landing position with the preset landing point, improving the landing accuracy.

[0132] The calibrated landing area features are extracted through the convolutional neural network, and the feature coordinate offset caused by the jitter is corrected in combination with the attitude sensor data to generate the offset compensation value of the landing point, specifically:

[0133] S501: Compare the image features and posture data of the landing area, and analyze the impact of posture changes on the image feature coordinates:

[0134] A convolutional neural network is used to extract features from the calibrated landing area image to generate key features such as edges, contours, and textures in the area. After feature extraction, these image features are compared with the drone attitude sensor data. The attitude data includes the specific position and angle information of the drone in three-dimensional space, such as roll angle, pitch angle, and yaw angle. The convolutional neural network converts image features into quantifiable coordinates through feature mapping.

[0135] To ensure the correspondence between image features and posture data, the influence of posture changes on each feature point is analyzed in this step.

[0136] The attitude data provided by the attitude sensor includes three angle changes: roll angle, pitch angle, and yaw angle. After the convolutional neural network feature extraction, the impact component of each attitude change is compared and analyzed to clarify the impact of attitude angle changes on the offset of feature point coordinates.

[0137] S502: Calculate the actual offset of the image feature coordinates according to the influence of the posture change on the image feature coordinates:

[0138] After confirming the influence of posture change on the image feature coordinates, the specific coordinate offset is calculated. Based on the influence of posture change on each feature point, the rotation matrix model is used to calculate the actual offset of the feature point.

[0139] Let the rotation matrix of the posture change be R, and apply the rotation matrix to correct the offset of the feature coordinates. The rotation matrix is ​​defined as: ;in, is the posture change angle.

[0140] The rotation matrix is ​​applied to the feature point coordinates to generate the corrected coordinates of the feature points. , and calculate the difference between the original feature coordinates and the corrected feature coordinates, that is, the actual offset of the feature point: ;in, Indicates the actual offset of the feature point (the coordinate change caused by the posture change). Represents the original image feature point coordinates (the initial position of the feature point before the posture change).

[0141] S503: Generate an offset compensation value of the landing point based on the actual offset of the image feature coordinates, which is used to correct the landing path of the drone:

[0142] Assume that the offset compensation value of the landing point is , which is calculated based on the offset of all feature points. The specific calculation method is to take the average of the offset of each feature point to obtain the offset compensation value of the landing point: ; Where S represents the total number of feature points in the image, Indicates the offset of the sth feature point, where s is the number of the feature point.

[0143] The offset compensation value of the landing point is used to correct the path during the landing of the drone; this compensation value is input into the control system of the drone to adjust its landing trajectory to ensure that the drone can land accurately at the target location.

[0144] Combined with the generated offset compensation value, the landing path of the drone is dynamically corrected in real time, including:

[0145] S601: Calculate the spatial deviation between the current actual position and the target position based on the current drone position data:

[0146] During the landing process of the drone, the current position data of the drone is first obtained in real time through the positioning system. The current position data contains the precise position coordinates of the drone in three-dimensional space, and these coordinate information reflects the real-time position status of the drone. At the same time, the system stores the preset target position of the drone, that is, the coordinates of the final landing point of the drone.

[0147] Compare the current position with the target position to get the current spatial deviation of the drone. The spatial deviation value represents the relative position difference between the current position of the drone and the target position, including the horizontal displacement difference and the vertical displacement difference, which are used to identify the deviation direction and distance of the drone during the landing process.

[0148] S602: Apply the offset compensation value to the landing path control and gradually adjust the position deviation of the drone:

[0149] After calculating the spatial deviation value, the offset compensation value generated in the previous step is applied to the control system of the landing path. The offset compensation value represents the comprehensive correction amount of the position deviation caused by external interference such as micro-airflow, and can be used to make instant compensation on the landing path of the drone. Specifically, the offset compensation value is used to fine-tune the path control parameters of the drone to gradually reduce the spatial deviation value.

[0150] For example, the horizontal and vertical motion parameters of the drone are corrected in combination with the offset compensation value, so that the drone gradually approaches the target position. This process alleviates or eliminates the offset effect caused by attitude changes through path adjustment, ensuring the stability and accuracy of the drone's landing.

[0151] S603: Update the attitude angle and motion parameters of the drone in real time according to the corrected landing path:

[0152] After applying the offset compensation value to correct the path, the landing path of the drone will be adjusted. To ensure accurate landing after the path correction, the system needs to update the attitude angle and motion parameters of the drone in real time. The attitude angle includes roll angle, pitch angle and yaw angle, which correspond to the rotation attitude of the drone around different axes and are used to control the motion posture of the drone in space.

[0153] The motion parameters include information such as the speed and acceleration of the drone, which are used to adjust its speed and path direction during landing. By updating the attitude angle and motion parameters, the drone is ensured to always move on the corrected path, gradually reducing the spatial deviation between the current position and the target position.

[0154] S604: Continue to obtain the deviation data and correct the landing path in real time until the UAV lands at the predetermined target location:

[0155] During the entire process of the drone landing, it continuously monitors and obtains new position data to update the spatial deviation value. Whenever the position data changes, the system dynamically adjusts the path according to the latest deviation data and corrects the landing path in real time.

[0156] The process of offset compensation and path correction will continue until the drone lands at the predetermined target location. This continuous path adjustment ensures the drone's precise landing and ensures that it eventually stays at the desired landing point.

[0157] Example 2

[0158] Embodiment 2 of the present invention introduces a UAV precise location landing system based on image recognition and analysis.

[0159] Figure 2 A structural schematic diagram of the UAV precise position landing system based on image recognition and analysis of the present invention is given. The UAV precise position landing system based on image recognition and analysis includes a jitter intensity determination module, an image sequence generation module, an airflow impact analysis module, a reverse trajectory analysis module, an offset compensation correction module and a real-time dynamic correction module.

[0160] Jitter intensity determination module: real-time detection of the drone's attitude changes during landing, identification of the jitter signal intensity caused by micro-airflow, and determination of whether the jitter signal intensity is abnormal.

[0161] Image sequence generation module: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence.

[0162] Airflow impact analysis module: It evaluates texture changes of multiple superimposed images based on multi-resolution fractal dimension analysis, detects subtle texture disturbances between adjacent frames, and compensates for the nonlinear impact of micro-airflow on image features by evaluating the complexity of ground texture changes.

[0163] Reverse trajectory analysis module: records the vibration trajectory caused by micro-airflow during the landing of the drone, and reconstructs and analyzes the relative displacement of the landing target based on the reverse trajectory.

[0164] Offset compensation correction module: The calibrated landing area features are extracted through a convolutional neural network, and the feature coordinate offset caused by jitter is corrected in combination with the attitude sensor data to generate the offset compensation value of the landing point.

[0165] Real-time dynamic correction module: Combined with the generated offset compensation value, the landing path of the drone is corrected in real time.

[0166] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0167] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0168] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0169] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0170] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0171] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0173] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0174] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0175] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for accurately landing a UAV based on image recognition analysis, characterized in that: The steps include: S1: Real-time detection of the attitude changes of the drone during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal; S2: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence; S3: Evaluate the texture changes of multi-frame superimposed images based on multi-resolution fractal dimension analysis, detect subtle texture disturbances between adjacent frames, and compensate for the nonlinear effects of micro-airflow on image features by evaluating the complexity of ground texture changes; S4: Record the vibration trajectory caused by micro-airflow during the landing of the UAV, and analyze the relative displacement of the landing target based on the reverse trajectory reconstruction; S5: Extract the features of the calibrated landing area through a convolutional neural network, correct the feature coordinate offset caused by the jitter in combination with the attitude sensor data, and generate the offset compensation value of the landing point; S6: Combined with the generated offset compensation value, the landing path of the UAV is dynamically corrected in real time; When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence, including: S201: after identifying that the jitter signal strength is abnormal, collecting continuous multiple frames of image data from the camera of the drone in real time; S202: setting the time window length of time series superposition according to the jitter characteristics, and determining the number of frames used for image superposition; S203: using an inter-frame alignment algorithm to perform time sequence alignment on the collected multiple frames of images; S204: performing pixel-level averaging processing on the multiple frames of images after time sequence alignment; S205: Outputting the processed image average result as a stable image sequence; Based on multi-resolution fractal dimension analysis, texture changes of multi-frame superimposed images are evaluated to detect subtle texture disturbances between adjacent frames. By evaluating the complexity of ground texture changes, the nonlinear effect of micro-airflow on image features is compensated. Specifically, it includes: S301: decomposing the multi-frame superimposed images into different resolution levels by a multi-resolution decomposition method to extract multi-scale texture information; S302: using a fractal calculation method at each resolution level to obtain a fractal dimension value corresponding to the resolution level; S303: Arrange the fractal dimension values ​​of each resolution level in chronological order to construct a time series of multi-resolution texture changes; S304: comparing fractal dimension values ​​of adjacent frames in the time series to obtain fractal differences of adjacent frames at each resolution level; S305: Calculating texture change compensation parameters for compensating texture changes according to the difference in fractal dimensions between adjacent frames.

2. The method for accurately landing a UAV based on image recognition analysis according to claim 1 is characterized in that: Real-time detection of the attitude changes of the drone during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal, including: S101: collect the three-dimensional spatial attitude information of the UAV in the landing process in real time through the attitude sensor; S102: Analyze the collected posture data to separate the posture change signal caused by the micro-airflow; S103: calculating the amplitude value of the posture change based on the separated posture change signal, and identifying the intensity of the jitter signal caused by the micro-airflow based on the standard deviation of the amplitude value of the posture change; S104: Based on the standard deviation of the amplitude value of the posture change, determine whether the jitter signal strength is abnormal.

3. The method for accurately landing a UAV based on image recognition analysis according to claim 2 is characterized in that: The pixel-level average processing is performed on the multi-frame images after time sequence alignment. Specifically, the average value of each pixel in all frames is selected. The calculation formula is as follows: ;in, Indicates that the generated stable image is at pixel position The gray value at is the number of superimposed frames, For the Frame at pixel location The gray value at Indicates the current superimposed image frame number, and Represents the pixel position on the horizontal and vertical axes of the image respectively.

4. The method for accurately landing a UAV based on image recognition analysis according to claim 3 is characterized in that: The texture change compensation parameter can be expressed as: ;in, Represents the final texture change compensation parameter, Indicates the total number of resolution levels, For the The weight of the resolution level, Indicates The difference in fractal dimensions between adjacent frames at the resolution level, represents the index of adjacent frames in the time series, Indicates the number of the resolution level.

5. The method for accurately landing a UAV based on image recognition analysis according to claim 4 is characterized in that: Record the vibration trajectory caused by micro-airflow during the landing of the drone, and reconstruct and analyze the relative displacement of the landing target based on the reverse trajectory, including: S401: real-time collection of three-dimensional attitude data and position data of the UAV during landing; S402: Separating the vibration trajectory data caused by the micro-airflow from the three-dimensional posture data and the position data by using a filtering method; S403: Arrange the jitter trajectory data in chronological order to generate a complete time series of the jitter trajectory; S404: Based on the inverse trajectory reconstruction method, analyzing the change of the jitter trajectory in the complete time series of the jitter trajectory, and calculating the relative displacement of the landing target; S405: Generate calibration parameters according to the relative displacement of the landing target, and perform position correction on the landing target point position.

6. The method for accurately landing a UAV based on image recognition analysis according to claim 5 is characterized in that: The calibrated landing area features are extracted through the convolutional neural network, and the feature coordinate offset caused by the jitter is corrected in combination with the attitude sensor data to generate the offset compensation value of the landing point, specifically: S501: Compare the image features and posture data of the landing area, and analyze the impact of posture changes on image feature coordinates; S502: Calculate the actual offset of the image feature coordinates according to the influence of the posture change on the image feature coordinates; S503: Generate an offset compensation value of the landing point according to the actual offset of the image feature coordinates, which is used to correct the landing path of the drone.

7. The method for accurately landing a UAV based on image recognition analysis according to claim 6 is characterized in that: Combined with the generated offset compensation value, the landing path of the drone is corrected dynamically in real time, including: S601: Calculate the spatial deviation between the current actual position and the target position according to the current drone position data; S602: applying the offset compensation value to the landing path control to gradually adjust the position deviation of the UAV; S603: updating the attitude angle and motion parameters of the UAV in real time according to the corrected landing path; S604: Continuously obtain the deviation data and correct the landing path in real time until the UAV lands at the predetermined target location.

8. A UAV precise location landing system based on image recognition analysis, used to implement the UAV precise location landing method based on image recognition analysis according to any one of claims 1 to 7, characterized in that: It includes a jitter intensity determination module, an image sequence generation module, an airflow impact analysis module, a reverse trajectory analysis module, an offset compensation correction module, and a real-time dynamic correction module; Jitter strength determination module: real-time detection of the drone's attitude changes during landing, identification of the jitter signal strength caused by micro-airflow, and determination of whether the jitter signal strength is abnormal; Image sequence generation module: When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence; Airflow impact analysis module: It evaluates texture changes of multi-frame superimposed images based on multi-resolution fractal dimension analysis, detects subtle texture disturbances between adjacent frames, and compensates for the nonlinear effects of micro-airflow on image features by evaluating the complexity of ground texture changes; Reverse trajectory analysis module: records the vibration trajectory caused by micro-airflow during the landing of the drone, and reconstructs and analyzes the relative displacement of the landing target based on the reverse trajectory; Offset compensation correction module: extracts the features of the calibrated landing area through a convolutional neural network, corrects the feature coordinate offset caused by jitter in combination with the attitude sensor data, and generates the offset compensation value of the landing point; Real-time dynamic correction module: Combined with the generated offset compensation value, the landing path of the drone is corrected in real time and dynamically; When the jitter signal strength is abnormal, a multi-frame image superposition algorithm is used to perform time-series superposition processing on the image sequence to weaken the direct impact of jitter on image features and generate a stable image sequence, including: S201: after identifying that the jitter signal strength is abnormal, collecting continuous multiple frames of image data from the camera of the drone in real time; S202: setting the time window length of time series superposition according to the jitter characteristics, and determining the number of frames used for image superposition; S203: using an inter-frame alignment algorithm to perform time sequence alignment on the collected multiple frames of images; S204: performing pixel-level averaging processing on the multiple frames of images after time sequence alignment; S205: Outputting the processed image average result as a stable image sequence; Based on multi-resolution fractal dimension analysis, texture changes of multi-frame superimposed images are evaluated to detect subtle texture disturbances between adjacent frames. By evaluating the complexity of ground texture changes, the nonlinear effect of micro-airflow on image features is compensated. Specifically, it includes: S301: decomposing the multi-frame superimposed images into different resolution levels by a multi-resolution decomposition method to extract multi-scale texture information; S302: using a fractal calculation method at each resolution level to obtain a fractal dimension value corresponding to the resolution level; S303: Arrange the fractal dimension values ​​of each resolution level in chronological order to construct a time series of multi-resolution texture changes; S304: comparing fractal dimension values ​​of adjacent frames in the time series to obtain fractal differences of adjacent frames at each resolution level; S305: Calculating texture change compensation parameters for compensating texture changes according to the difference in fractal dimensions between adjacent frames.

Images