A method and system for accurate landing of unmanned aerial vehicle based on visual recognition

Abstract

The present invention relates to the field of image processing technology, and more specifically, to a method and system for accurate landing of a drone based on visual recognition, including: obtaining the acceleration and angular velocity of the drone when collecting images, and calculating the attitude stability of the drone when collecting the current image, wherein the attitude stability is negatively correlated with the difference in acceleration and angular velocity between the current image collected by the drone and the previous frame of the image; and calculating the reliability coefficient of the image collected by the drone. The present invention calculates the attitude stability of the drone based on the acceleration data and angular velocity data when the drone collects images, and calculates the reliability coefficient of each image based on the attitude stability of the drone and the brightness of the image collected by the drone, and selects a guide map based on the reliability coefficient of the image, and can select a guide map that is more conducive to guided filtering based on the angle of the shooting conditions of the image and the conditions of the pixels in the image, thereby improving the filtering effect on the image.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology. More specifically, the present invention relates to a method and system for accurate landing of a drone based on visual recognition. Background Art

[0002] Drone equipment is used in many scenarios such as police security, power services and emergency rescue. In the application process of drone equipment, it is often faced with the problem of controlling the landing position. Landing in an inappropriate position may cause damage to the drone and cause certain economic losses. At present, there are two main aspects for the precise landing of drones. One is to navigate the real-time position of the drone and the landing airport location through GPS (Global Positioning System) to achieve precise landing in longitude and latitude. The second is to use the visual system to perform image recognition of the precise location of the landing airport (recognize the AprilTag logo in the image to assist in landing), and give the recognized landing airport location information to the drone flight control system to achieve precise landing of the drone.

[0003] In the related technology, for example, the Chinese patent application document with publication number CN115755985A discloses a method for landing a drone based on image processing. Compared with the visual fixed-point landing algorithm based on deep learning or other image recognition, the patent application document adopts the AprilTag algorithm, which can decode and locate more quickly and has a certain degree of complexity.

[0004] At present, when identifying the AprilTag logo, it is usually necessary to perform guided filtering on the image to remove noise in the image. However, the guided filtering calculates the filter coefficient in units of windows. Each pixel corresponds to multiple windows, and each pixel uses the average value of the filter coefficients corresponding to its multiple windows as the filter coefficient of the pixel. However, the above method does not take into account the differences in local features of the image, which may cause some features to disappear, affecting the recognition of the AprilTag logo, and may further affect the precise landing of the drone. Summary of the invention

[0005] The present invention provides a method and system for precise landing of a UAV based on visual recognition, aiming to solve the problem in the related art that when performing guided filtering, the average value of the filter coefficients corresponding to multiple windows corresponding to each pixel point is used as the filter coefficient of the pixel point, without considering the differences in local features of the image, which may cause some features to disappear, thereby affecting the precise landing of the UAV.

[0006] In the first aspect, the present invention provides a method for accurate landing of a drone based on visual recognition, comprising: obtaining the acceleration and angular velocity of the drone when collecting images, and calculating the attitude stability of the drone when collecting the current image, wherein the attitude stability is negatively correlated with the difference in acceleration and angular velocity between the current image collected by the drone and the previous frame image; calculating the reliability coefficient of the i-th frame image collected by the drone , where the reliability coefficient is negatively correlated with the average and standard deviation of the grayscale values ​​of all pixels in the image and positively correlated with the degree of posture stability; and the image with the largest reliability coefficient among the multiple images before the current image is selected as the guidance map; the filter coefficient of the window corresponding to each pixel in the current image is calculated using the guided filtering algorithm, and the filter coefficient is corrected to obtain the target filter coefficient of pixel m and , the correction formula is:

[0007] ; In the formula, is the average value of the gradient amplitude of the pixel points in window j corresponding to the pixel point m in the current image, is the pixel m in the current image corresponding to The maximum value among the average values ​​of the gradient amplitudes corresponding to the windows is is the number of filter windows corresponding to pixel m in the current image, and are the filter coefficients of the pixel point m in the current image corresponding to the window j, is an exponential function with the natural constant e as the base; based on the target filter coefficient, a target image after guidance filtering is obtained to guide the UAV to land.

[0008] The effect is: according to the flight state of the UAV when taking the image, a guidance map that is more conducive to guided filtering is selected, and the target filter coefficient of each pixel is calculated according to the reliability coefficient of the image and the gradient of each pixel. Finally, the image is guided filtered according to the target filter coefficient, and the UAV is accurately landed based on the result of the guided filtering.

[0009] Furthermore, the reliability coefficient of the image collected by the drone is calculated using the following formula: ; In the formula, is the reliability coefficient of the i-th frame image collected by the UAV, is the attitude stability of the UAV when collecting the i-th frame image, is the average grayscale value of all pixels in the i-th frame image collected by the drone, is the standard deviation of the grayscale values ​​of all pixels in the i-th frame image collected by the drone, It is an exponential function with the natural constant e as its base.

[0010] Further, the attitude stability of the drone when collecting the current image is calculated, including: calculating the acceleration value of the attitude stability of the drone when collecting the i-th frame image And the value of angular velocity , the acceleration value It is positively correlated with the sum of the differences in acceleration in multiple dimensions when collecting two adjacent frames of images. The angular velocity is It is positively correlated with the sum of the differences in angular velocity in multiple dimensions when two adjacent frames of images are collected, wherein the multiple dimensions are the X-axis dimension, the Y-axis dimension and the Z-axis dimension; the embodiment value of the acceleration And the angular velocity The weighted sum is used to obtain the attitude stability of the drone when collecting the i-th frame image. The attitude stability is related to the performance value , embodied value Both are negatively correlated.

[0011] The effect is: the flight state of the drone when taking images is measured based on the acceleration data and angular velocity data during the flight, and the attitude stability of the drone when collecting images is calculated based on this.

[0012] Furthermore, the embodied value is calculated , the calculation formula is: ; In the formula, The acceleration of the UAV in the kth dimension when collecting the i-th frame image, The acceleration of the UAV in the kth dimension when collecting the i-1th frame image, where k represents the x-axis dimension, y-axis dimension, and z-axis dimension when it is 1, 2, and 3, respectively. is the standard normalization function.

[0013] Furthermore, the embodied value is calculated , the calculation formula is: ; In the formula, is the angular velocity of the UAV in the kth dimension when collecting the i-th frame image, is the angular velocity of the UAV in the kth dimension when collecting the i-th frame image, is the standard normalization function.

[0014] Furthermore, the acceleration value And the angular velocity Weighted sum, including: ; In the formula, is the attitude stability of the UAV when collecting the i-th frame image, is the acceleration value of the attitude stability when the UAV collects the i-th frame image. is the angular velocity value of the attitude stability of the drone when it collects the i-th frame image. To reflect the value The preset weight size, It is an exponential function with the natural constant e as its base.

[0015] The effect is that: since the angular velocity and acceleration have different influences on the attitude stability of the UAV during the calculation process, the angular velocity and acceleration are given corresponding weights to improve the accuracy of the calculation results.

[0016] Furthermore, the gradient amplitude of the pixel points in the window j corresponding to the pixel point m is obtained, including: using the Sobel algorithm to calculate the gradient amplitude of each pixel point in the image.

[0017] The effect is that the Sobel operator is used not only to calculate the size of the gradient, but also to determine the direction of the edge by combining the horizontal and vertical gradients, making the Sobel operator more accurate in locating the edge.

[0018] Furthermore, The experience value is 2 / 3.

[0019] The effect is that drones usually fly at a low and stable speed. Compared with changes in angular velocity, changes in acceleration will cause more image jitter. Therefore, when measuring the stability of the drone's attitude, a larger weight can be given to acceleration to improve the accuracy of the calculation results.

[0020] In a second aspect of the present invention, a UAV precision landing system based on visual recognition is also provided, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned UAV precision landing methods based on visual recognition.

[0021] Beneficial effects:

[0022] (I) The attitude stability of the drone is calculated based on the acceleration data and angular velocity data when the drone collects images, and the reliability coefficient of each image is calculated based on the attitude stability of the drone and the brightness of the images collected by the drone. The guidance map is selected based on the reliability coefficient of the image. Based on the angle of the image shooting conditions and the conditions of the pixels in the image, a guidance map that is more conducive to guidance filtering can be selected to improve the filtering effect on the image.

[0023] (ii) Based on the reliability coefficient of the image collected by the UAV and the gradient of the pixel points in the window corresponding to each pixel point, the target filter coefficient used for each pixel point is calculated, the image is guided filtered according to the target filter coefficient, and the UAV is accurately landed based on the result of the guided filtering. From the perspective of AprilTag logo recognition, the landing process of the UAV is optimized, the edge preservation capability of the guided filtering is improved, and the AprilTag logo is easier to be recognized, thereby improving the accuracy of the UAV landing position to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The following detailed description is read with reference to the accompanying drawings, which illustrate several embodiments of the present invention in an exemplary and non-limiting manner, and in which like or corresponding reference numerals represent like or corresponding parts, wherein:

[0025] Figure 1 is a flowchart schematically illustrating image guided filtering according to an embodiment of the present invention;

[0026] Figure 2 is a schematic diagram schematically illustrating a filter window of guided filtering according to an embodiment of the present invention;

[0027] Figure 3 is a schematic diagram showing a gamma change principle according to an embodiment of the present invention;

[0028] Figure 4 FIG. 1 is a schematic diagram schematically showing a system structure according to an embodiment of the present invention.

[0029] Reference numerals

[0030] 1. Processor; 2. Communication bus; 3. Memory; 4. Communication interface. DETAILED DESCRIPTION

[0031] 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 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0032] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0033] In one embodiment, the application process of drone equipment often faces the problem of controlling the landing position. Landing in an inappropriate position may cause damage to the drone, thereby causing certain economic losses. Usually when the drone lands, it can assist in landing by identifying the AprilTag logo in the image collected by the drone. When identifying the AprilTag logo, it is necessary to perform guided filtering on the image and enhance the image to facilitate more accurate identification of the AprilTag logo. However, when using the guided filtering algorithm, it depends on the manual subjective selection of the guided map. If the quality of the selected guided map is not high, it will cause certain inconveniences for the guided filtering, and under certain circumstances, it will also affect the precise landing of the drone. Therefore, this embodiment provides a method for selecting a high-quality guided map for guided filtering, which provides a favorable basis for the precise landing of the drone. The specific steps are as follows.

[0034] like Figure 1 As shown, S101: obtaining the acceleration and angular velocity of the drone when collecting images.

[0035] In one embodiment, since the images captured by the drone are affected by the flight state of the drone when capturing the images, and the images captured by the drone in different postures have different qualities, the acceleration data and angular velocity data of the drone during flight can be collected to measure the flight state of the drone when taking images, and the quality of the images taken by the drone can be judged based on the flight state.

[0036] S102: Calculate the reliability coefficient of the image collected by the drone.

[0037] In one embodiment, the acceleration data and angular velocity data of the drone during flight are collected to measure the flight state of the drone when taking images, including: calculating the attitude stability of the drone according to the acceleration data and angular velocity data when the drone collects images, wherein the attitude stability is negatively correlated with the difference in acceleration and angular velocity between the current image and the previous frame of image collected by the drone. In other words, when the difference in acceleration between the current image and the previous frame of image collected by the drone is larger, the drone is more likely to have a certain change in the flight direction, and therefore, the lower the attitude stability when the drone collects the current image, the lower the quality of the captured image; when the difference in angular velocity between the current image and the previous frame of image collected by the drone is larger, the drone is more likely to have a rotational motion, and therefore, the smaller the attitude stability when the drone collects the current image, the lower the quality of the captured image.

[0038] In one embodiment, calculating the attitude stability of the drone when collecting the current image specifically includes: first calculating the acceleration value of the attitude stability of the drone when collecting the i-th frame image And the value of angular velocity , the acceleration value It is positively correlated with the sum of the differences in acceleration in multiple dimensions when collecting two adjacent frames of images. The angular velocity is It is positively correlated with the sum of the differences in angular velocity in multiple dimensions when two adjacent frames of images are collected, wherein the multiple dimensions are respectively an X-axis dimension, a Y-axis dimension, and a Z-axis dimension.

[0039] In one embodiment, the acceleration value of the attitude stability of the drone when collecting the i-th frame image is calculated. , the calculation formula is: ; In the formula, The acceleration of the UAV in the kth dimension when collecting the i-th frame image, The acceleration of the UAV in the kth dimension when collecting the i-1th frame image, where k represents the x-axis dimension, y-axis dimension, and z-axis dimension when it is 1, 2, and 3, respectively. is the standard normalization function.

[0040] in, The value of reflects the difference between the acceleration in each dimension when the UAV collects the i-th frame image and the acceleration in each dimension when collecting the i-1-th frame image. This value measures whether the UAV is stable when collecting images in the dimension of acceleration. The larger the value, the more likely the UAV is to have a certain change in the flight direction when collecting the i-th frame image, and the lower the attitude stability of the UAV when collecting the i-th frame image; the smaller the value, the less likely the UAV is to have a change in the flight direction when collecting the i-th frame image, and the higher the attitude stability of the UAV when collecting the i-th frame image.

[0041] In one embodiment, the angular velocity value of the attitude stability of the drone when collecting the i-th frame image is calculated. , the calculation formula is: ; In the formula, is the angular velocity of the UAV in the kth dimension when collecting the i-th frame image, is the angular velocity of the UAV in the kth dimension when collecting the i-th frame image, is the standard normalization function.

[0042] in, The value of reflects the difference between the angular velocity in each dimension when the UAV collects the i-th frame image and the angular velocity in each dimension when collecting the i-1-th frame image. This value measures whether the UAV is stable when collecting images in the dimension of angular velocity. The larger the value, the more likely the UAV is to produce rotational motion when collecting the i-th frame image, and the lower the attitude stability of the UAV when collecting the i-th frame image; the smaller the value, the less likely the UAV is to be in rotational motion when collecting the i-th frame image, and the higher the attitude stability of the UAV when collecting the i-th frame image.

[0043] In one embodiment, according to the acceleration value calculated above, And the angular velocity , we can calculate the stability of the drone’s attitude when collecting the current image. The calculation formula is: ; In the formula, is the attitude stability of the UAV when collecting the i-th frame image, is the acceleration value of the attitude stability when the UAV collects the i-th frame image. is the angular velocity value of the attitude stability of the drone when it collects the i-th frame image. It is an exponential function with the natural constant e as its base.

[0044] In another embodiment, since the present invention is directed to the landing process of a drone, during the landing process, the drone usually flies at a low speed and in a stable state (even if it is flying steadily, the drone will also change its attitude due to factors such as airflow during the flight). Compared with the change in angular velocity, the change in acceleration will cause more image jitter. Therefore, when measuring the stability of the drone's attitude, a greater weight can be given to acceleration. Then, the acceleration value is calculated. And the angular velocity The weighted sum is used to obtain the attitude stability of the drone when collecting the i-th frame image. The attitude stability is related to the performance value , embodied value Both are negatively correlated.

[0045] In one embodiment, the acceleration value And the angular velocity Weighted summation, the weighted calculation formula is: ; In the formula, is the attitude stability of the UAV when collecting the i-th frame image, is the acceleration value of the attitude stability when the UAV collects the i-th frame image. is the angular velocity value of the attitude stability of the drone when it collects the i-th frame image. To reflect the value The preset weight size, is an exponential function with the natural constant e as the base. In this embodiment, due to the large weight of acceleration, The value of is set to 2 / 3. In other embodiments, implementers can determine the corresponding weight according to actual conditions.

[0046] S103: Determine a guidance graph based on the posture stability.

[0047] In one embodiment, the reliability coefficient of the image collected by the drone is calculated, wherein the reliability coefficient is negatively correlated with the average value and standard deviation of the grayscale values ​​of all pixels in the image and positively correlated with the attitude stability; and the image with the largest reliability coefficient among the multiple images before the current image is selected as the guide map of the current image. In this embodiment, the multiple images can be 5 or 7, which can be adjusted according to actual conditions.

[0048] In one embodiment, the reliability coefficient of the image collected by the drone is calculated using the following formula: ; In the formula, is the reliability coefficient of the i-th frame image collected by the UAV, is the attitude stability of the UAV when collecting the i-th frame image, is the average grayscale value of all pixels in the i-th frame image collected by the drone, is the standard deviation of the grayscale values ​​of all pixels in the i-th frame image collected by the drone, It is an exponential function with the natural constant e as its base.

[0049] Indicates the overall grayscale value level of the image. Since the image with a larger grayscale value is visually brighter, The smaller it is, the less likely it is that there is a reflective area in the image, and the greater the reliability coefficient of the i-th frame image collected by the drone; The larger the value is, the more likely it is that there is a reflective area in the image or the overall brightness of the image is large, and the smaller the reliability coefficient of the i-th frame image collected by the drone. It should be noted that since images with high overall brightness do not affect the recognition of the AprilTag logo, images with high overall brightness also have a high reliability coefficient, while the reflective area in the image with reflective areas may affect the recognition of the AprilTag logo and have a low reliability coefficient. In order to avoid the influence of the overall brightness of the image on the judgment of the image reliability coefficient, right An upward correction is performed so that images with a relatively high overall brightness have a higher reliability coefficient. The smaller the value, the more concentrated the grayscale distribution of the image is and the smaller the brightness change is, which means that the possibility of the image having a reflective area with significant brightness change is smaller. The larger the value, the more likely it is that the overall brightness of the image is high. In this case, it is more likely that there is no reflective area in the image, and the more important it is to An upward correction is performed to make the reliability coefficient of the image larger. When the value is larger, the grayscale distribution of the image is more dispersed and the brightness change is relatively large, which means that the image is more likely to have reflective areas with significant brightness changes, and the image should not be processed. Make an upward correction. right The smaller the correction degree is, the smaller the reliability coefficient of the image is.

[0050] S104: Calculate the target filter coefficient of each pixel.

[0051] In one embodiment, after obtaining the guidance map, the current image is subjected to guidance filtering. When the image is subjected to guidance filtering, each pixel corresponds to multiple filter windows (such as Figure 2 As shown in FIG. 1 , pixel 1 corresponds to 9 filter windows, and the dotted box in the figure represents the filter window). Preferably, the filter window size is 3×3. According to the principle of guided filtering, each filter window can calculate a set of filter coefficients, and the filter coefficients are corrected to obtain the target filter coefficients of each pixel.

[0052] In one embodiment, the filter coefficient is corrected, and the correction formula is: ; In the formula, is the reliability coefficient of the i-th frame image collected by the UAV, is the average value of the gradient amplitude of the pixel points in window j corresponding to the pixel point m in the current image, is the pixel m in the current image corresponding to The maximum value among the average values ​​of the gradient amplitudes corresponding to the windows is is the number of filter windows corresponding to pixel m in the current image, and are the filter coefficients of the pixel point m in the current image corresponding to the window j, It is an exponential function with the natural constant e as its base.

[0053] in, It is the relative size of the gradient in the jth window of the mth pixel in the image. The larger the value, the more likely it is that there is a more obvious edge in the window. If there is an obvious edge in the AprilTag logo, the target filter coefficient corresponding to the window should play a greater role in filtering the mth pixel. The smaller the value, the lower the possibility of an obvious edge in the window. In order to better retain the edge in the AprilTag logo, the target filter coefficient corresponding to the window should play a smaller role in filtering the mth pixel.

[0054] In one embodiment, is the reliability coefficient of the i-th frame image collected by the UAV, when The smaller the value, the lower the quality of the current image. In this case, the filter coefficients corresponding to windows with obvious edges should be given a greater weight, and the filter coefficients corresponding to windows with unclear edges should be given a smaller weight, so as to reduce the impact of low-quality areas in the image on AprilTag identification. The larger it is, the higher the quality of the current image is, which means that the filter coefficients of multiple windows corresponding to each pixel in the image are more likely to be equally important, and the filter weights corresponding to multiple windows corresponding to each pixel in the image are more average.

[0055] According to the principle of gamma transformation ( Figure 3 ) It can be seen that when the base is smaller, the adjustment effect of different exponents on the base is greater, and it can be seen that when the base is larger, the adjustment effect of different exponents on the base is smaller, so As the weight of the j-th window filter coefficient of the m-th pixel in the image, The smaller the size, the smaller the corresponding windows are. right The greater the difference in the correction effect, the greater the weight of the filter coefficient corresponding to the window with obvious edges, and the smaller the weight of the filter coefficient corresponding to the window with unclear edges. The larger the size, the corresponding right The closer the correction effect is, the more average the filtering weights corresponding to the multiple windows corresponding to the pixel points will be.

[0056] S105: Obtaining a target image after guidance filtering based on the target filtering coefficient to guide the UAV to land.

[0057] At this point, the target filter coefficient corresponding to each pixel point is calculated by the above method, and based on the target filter coefficient and the pixel point corresponding to each pixel point of the current image in the guidance map, the target image after guided filtering is obtained, and the target image is subjected to threshold segmentation, edge extraction, quadrilateral fitting and quadrilateral decoding processes, the AprilTag logo in the target image is identified, and the landing process of the UAV is guided according to the identification result.

[0058] According to the above steps, according to the flight state of the UAV when taking images, a guidance map that is more conducive to guided filtering is selected, and the target filter coefficient of each pixel is calculated according to the reliability coefficient of the image and the gradient of each pixel. Finally, the image is guided filtered according to the target filter coefficient, and the UAV is accurately landed based on the result of the guided filtering. The landing process of the UAV is optimized from the perspective of AprilTag identification, the edge preservation capability of the guided filtering is improved, and the AprilTag identification is easier to be identified, thereby improving the accuracy of the UAV landing position to a certain extent.

[0059] The present invention also provides a UAV precision landing system based on visual recognition. Figure 4 As shown, the system includes a processor 1 and a memory 3, wherein the memory 3 stores computer program instructions. When the computer program instructions are executed by the processor 1, a method for precise landing of a drone based on visual recognition according to the first aspect of the present invention is implemented.

[0060] The system also includes other components well known to those skilled in the art, such as a communication bus 2 and a communication interface 4. The configuration and functions of these components are known in the art and thus will not be described in detail herein.

[0061] In the present invention, the aforementioned memory 3 may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0062] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0063] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent application. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for accurate landing of a drone based on visual recognition, characterized in that: include: The acceleration and angular velocity of the drone when collecting images are obtained, and the attitude stability of the drone when collecting the current image is calculated. The attitude stability is negatively correlated with the difference in acceleration and angular velocity between the current image collected by the drone and the previous frame of image collected; Calculate the reliability coefficient of the i-th frame image collected by the drone , where the reliability coefficient is negatively correlated with the average and standard deviation of the grayscale values ​​of all pixels in the image, and positively correlated with the stability of the posture; and the image with the largest reliability coefficient among the multiple images before the current image is selected as the guidance map; Based on the guide graph, the filter coefficient of the window corresponding to each pixel in the current image is calculated using the guide filtering algorithm, and the filter coefficient is corrected to obtain the target filter coefficient of pixel m and , the correction formula is: ; In the formula, is the average value of the gradient amplitude of the pixel points in window j corresponding to the pixel point m in the current image, is the pixel m in the current image corresponding to The maximum value among the average values ​​of the gradient amplitudes corresponding to the windows is is the number of filter windows corresponding to pixel m in the current image, and are the filter coefficients of the pixel point m in the current image corresponding to the window j, is an exponential function with the natural constant e as base; Based on the target filter coefficient, the target image after guidance filtering is obtained to guide the UAV to land.

2. The method for accurate landing of a UAV based on visual recognition according to claim 1, characterized in that: Calculate the reliability coefficient of the image collected by the drone, and the calculation formula is: ； In the formula, is the reliability coefficient of the i-th frame image collected by the UAV, is the attitude stability of the UAV when collecting the i-th frame image, is the average grayscale value of all pixels in the i-th frame image collected by the drone, is the standard deviation of the grayscale values ​​of all pixels in the i-th frame image collected by the drone, It is an exponential function with the natural constant e as its base.

3. The method for accurate landing of a UAV based on visual recognition according to claim 1, characterized in that: Calculate the stability of the drone's attitude when collecting the current image, including: Calculate the acceleration value of the drone's attitude stability when collecting the i-th frame image And the value of angular velocity , the acceleration value It is positively correlated with the sum of the differences in acceleration in multiple dimensions when collecting two adjacent frames of images. The angular velocity is It is positively correlated with the sum of the differences in angular velocity in multiple dimensions when two adjacent frames of images are collected, wherein the multiple dimensions are respectively an X-axis dimension, a Y-axis dimension, and a Z-axis dimension; The value of the acceleration And the angular velocity The weighted sum is used to obtain the attitude stability of the drone when collecting the i-th frame image. The attitude stability is related to the performance value , embodied value Both are negatively correlated.

4. The method for accurate landing of a UAV based on visual recognition according to claim 3 is characterized in that: Calculating embodied value , the calculation formula is: ； In the formula, The acceleration of the UAV in the kth dimension when collecting the i-th frame image, The acceleration of the UAV in the kth dimension when collecting the i-1th frame image, where k represents the x-axis dimension, y-axis dimension, and z-axis dimension when it is 1, 2, and 3, respectively. is the standard normalization function.

5. The method for accurate landing of a UAV based on visual recognition according to claim 4, characterized in that: Calculating embodied value , the calculation formula is: ； In the formula, is the angular velocity of the UAV in the kth dimension when collecting the i-th frame image, The angular velocity of the UAV in the kth dimension when collecting the i-1th frame image, is the standard normalization function.

6. The method for accurate landing of a UAV based on visual recognition according to claim 5, characterized in that: The value of the acceleration And the angular velocity Weighted sum, including: ； In the formula, is the attitude stability of the UAV when collecting the i-th frame image, is the acceleration value of the attitude stability when the UAV collects the i-th frame image. is the angular velocity value of the attitude stability of the drone when it collects the i-th frame image. To reflect the value The preset weight size, It is an exponential function with the natural constant e as its base.

7. The method for accurate landing of a UAV based on visual recognition according to claim 1, characterized in that: Get the gradient amplitude of the pixel point in window j corresponding to the pixel point m, including: The Sobel algorithm is used to calculate the gradient amplitude of each pixel in the image.

8. The method for accurate landing of a UAV based on visual recognition according to claim 6, characterized in that: The experience value is 2 / 3.

9. A UAV precision landing system based on visual recognition, comprising a processor and a memory, characterized in that: The memory stores a computer program, and the processor executes the computer program to implement the method for precise landing of a drone based on visual recognition as described in any one of claims 1 to 8.

Images