Unmanned aerial vehicle navigation and intelligent pile inspection method for closed stockyard

Abstract

The application discloses a closed yard unmanned aerial vehicle navigation method and an intelligent disc inspection method adopting the navigation method. The navigation method fuses observation values of multiple sensors through a factor graph optimization method to obtain real-time position information of the unmanned aerial vehicle, and performs navigation based on the obtained position information of the unmanned aerial vehicle. The multiple sensors include a laser radar, an IMU carried on the unmanned aerial vehicle, an identification code arranged on the ground, and a holder camera arranged on the unmanned aerial vehicle and used for detecting the identification code. The closed yard unmanned aerial vehicle navigation method disclosed by the application can be used for a long time in a closed yard environment and is not prone to failure.

Description

Technical Field

[0001] This invention relates to a method for unmanned aerial vehicle (UAV) navigation and intelligent inventory inspection in enclosed storage yards, belonging to the field of UAV control. Background Technology

[0002] After mining, resources such as coal and mineral powder are generally stored in stockpiles. To achieve accurate allocation, planning, and timely and rational distribution and use of these resources, regular and precise inventory checks of coal and mineral powder in the storage areas are necessary. Traditional stockpile inventory checks mainly rely on handheld laser instruments for data collection, which results in limited collection angles, large collection errors, and the need for manual operations to climb to heights, posing significant inconvenience and safety risks to workers. Therefore, an unmanned aerial vehicle (UAV) intelligent inventory check system has been proposed.

[0003] In existing technologies, some UAV intelligent inventory inspection systems use lidar to collect data. However, due to the large amount of dust at the bottom of enclosed storage yards, the signal is easily interfered with when using lasers for data collection, resulting in low data collection accuracy.

[0004] In addition, some UAV intelligent inspection systems use visual information for data collection. However, since monocular visual information lacks scale, it is necessary to accurately record the position of each image to facilitate subsequent data processing and analysis such as 3D reconstruction and classification. The accuracy of UAV positioning is related to the flight safety of the UAV system, the modeling accuracy of the 3D reconstruction system, and the overall quality of the entire inspection task. This requires the UAV to have stable, reliable, and accurate navigation functions to provide accurate positioning.

[0005] Existing UAV navigation systems primarily utilize GNSS-INS systems composed of satellite positioning and inertial modules. However, storage yards are often enclosed environments where GNSS is denied, rendering them unusable. Currently, navigation systems in GNSS-denied environments mainly employ two methods: one uses exogenous sensors pre-deployed in the surrounding environment, and the other uses endogenous sensors mounted on the platform. Exogenous navigation systems include UWB and VICON systems. These systems have high pre-installed hardware costs, large blind zones, and are difficult to maintain. During operation, their accuracy is easily affected by dust, metal in the environment, and other factors, and they also suffer from real-time data transmission issues. Endogenous navigation systems mainly use vision or lidar. Vision-based navigation systems, such as ORB-SLAM2 and VINS-Mono, are prone to failure due to strong light variations and unclear features in enclosed storage yards. Furthermore, because UAVs exhibit three-dimensional movement unlike land-based wheeled robots, lidar systems, such as LOAM, LeGO-LOAM, LIO-SAM, FAST-LIO, and LIOM, are more susceptible to failure.

[0006] Therefore, it is necessary to conduct more in-depth research in order to obtain a highly reliable and low-cost intelligent inventory inspection method for closed storage yards using unmanned aerial vehicles (UAVs). Summary of the Invention

[0007] To overcome the above problems, the inventors conducted in-depth research and proposed a closed-yard UAV navigation method. This method uses a factor graph optimization method to fuse observations from multiple sensors to obtain real-time UAV pose information, and then performs navigation based on the obtained UAV pose information.

[0008] Furthermore, the multiple sensors include a lidar and an IMU mounted on the drone, an identification code set on the ground, and a gimbal camera mounted on the drone for detecting the identification code.

[0009] Furthermore, the factor graph includes multiple factors, including:

[0010] IMU prediction factor is used to obtain the change in the UAV state between adjacent time k-1 and k based on the IMU detection value;

[0011] The lidar factor is used to obtain the change in the UAV state between adjacent times k-1 and k based on the lidar detection value.

[0012] The identifier factor is used to obtain the change in the UAV state between adjacent times k-1 and k based on the identifier detected by the gimbal camera.

[0013] The identification code cloze factor is used to obtain the pose transformation of the UAV between the previous j time and the current k time when the same identification code was detected.

[0014] Preferably, the identification code is an ArUco code.

[0015] Preferably, for any time K, the pose of the UAV at time K under the coordinates of the i-th identifier code and the change in pose of the UAV at time K+1 under the coordinates of the i-th identifier code are used as the identifier code factor from time k to time k+1.

[0016] Preferably, in loop closure detection, when the same identifier code is detected discontinuously, the state with the smallest identifier code factor covariance in each of the previous consecutive detections will be selected, and they will be looped back with the current time K.

[0017] Preferably, in loop closure detection, the interval between when the drone does not detect the same identifier and the distance it travels away from the identifier are used as conditional thresholds to determine whether it is a continuous detection. If both are greater than the threshold, it indicates that it is not a continuous detection; otherwise, it is a continuous detection.

[0018] On the other hand, the present invention also provides an intelligent inventory inspection method for closed stockyards using unmanned aerial vehicles (UAVs). The method uses the above-mentioned closed stockyard UAV navigation method for navigation. During the navigation process, images of the stockpiled materials and the corresponding positions of each image are collected to construct a map. The stockpiled areas and the volume of the stockpiled materials are identified from the model obtained from the map, thereby completing the intelligent inventory inspection.

[0019] Preferably, the mapping is a three-dimensional model mapping or a two-dimensional orthophoto model mapping.

[0020] Preferably, the two-dimensional orthophoto image obtained from the mapping is classified to obtain the specific area of ​​the material pile, and then the specific area of ​​the material pile is mapped to the three-dimensional model to calculate the volume of the material pile.

[0021] The beneficial effects of this invention include:

[0022] (1) It has high reliability and can be used for a long time in a closed storage yard environment. It is not easy to fail, and the navigation and inspection accuracy is high.

[0023] (2) Low cost, simple to use, vision plus lidar fusion, easy to install and maintain. Attached Figure Description

[0024] Figure 1 A schematic diagram of the factor graph structure in a closed-yard unmanned aerial vehicle navigation method according to a preferred embodiment of the present invention is shown.

[0025] Figure 2 This diagram illustrates loop closure detection in a factor graph of a closed-yard UAV navigation method according to a preferred embodiment of the present invention.

[0026] Figure 3 The curves showing the variation of the UAV flight trajectory along the X-axis under different navigation conditions in Experiment Example 1 are shown.

[0027] Figure 4 The curves showing the variation of the UAV flight trajectory along the Y-axis under different navigation conditions in Experiment Example 1 are shown.

[0028] Figure 5 The curves showing the variation of the UAV flight trajectory along the Z-axis under different navigation conditions in Experiment Example 1 are shown. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Through these descriptions, the features and advantages of the present invention will become clearer and more apparent.

[0030] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0031] To address the challenges of large blind spots, high maintenance difficulty, and susceptibility to interference from dust and metals in the environment affecting accuracy in external navigation systems, as well as issues with real-time data transmission and the susceptibility to failure in internal navigation, this invention provides a navigation method for unmanned aerial vehicles (UAVs) operating in enclosed storage yards, along with an intelligent inventory management method based on this navigation approach. This navigation method employs a simple vision-plus-LiDAR fusion approach, utilizing both easily installed and maintained external and internal navigation devices, resulting in a highly reliable and low-cost navigation system.

[0032] Specifically, on the one hand, the present invention provides a navigation method for unmanned aerial vehicles (UAVs) in a closed storage yard, which fuses observations from multiple sensors using a factor graph optimization method to obtain real-time pose information of the UAV, and performs navigation based on the obtained UAV pose information.

[0033] In this invention, any navigation system can be used for navigation, such as the navigation system carried by a drone.

[0034] The multiple sensors include a lidar and an IMU mounted on the drone, a ground-based identification code, and a gimbal camera mounted on the drone for detecting the identification code.

[0035] Among them, lidar, IMU, and gimbal camera are internal devices, which are common equipment carried by drones. The identification code is an external device that can be set on the ground in any way, such as by spraying.

[0036] In a preferred embodiment, the identifier is ArUco, which is a synthetic square marker commonly used in machine vision inspection. Preferably, the specific detection principle of the ArUco code is as described in the literature Garrido-Jurado S, Mu~noz-Salinas R, Madrid-Cuevas FJ, et al. Automatic generation and detection of highly reliable fiducial markers under occlusion[J]. Pattern Recognition, 2014, 47(6):2280-2292., and will not be repeated in this invention.

[0037] The overall process of the factor graph optimization method can be found in the literature Frank D, Michael K. Factor Graphs for Robot Perception[J]. Foundations & Trends in Robotics, 2017, 6(1-2): 1-139. It will not be repeated in this invention. The following only describes the modifications to the factor graph optimization method.

[0038] According to the present invention, in the factor graph, the state of the UAV at time k is set as: X k =(p k v k R k b ak b gk t ak ), where p k v is the translation vector of the UAV in the geodetic coordinate system. k R is the velocity vector of the UAV in the geodetic coordinate system. k Let b be the rotation matrix of the UAV in the geodetic coordinate system. ak b represents the accelerometer bias in the IMU. gk t represents the gyroscope bias. ak It is a matrix consisting of the position vectors of all identifiers in the geodetic coordinate system.

[0039] Furthermore, the factor graph contains multiple factors, including not only commonly used IMU prediction factors and lidar factors, but also identifier code factors and identifier code loopback factors, such as... Figure 1 As shown,

[0040] The identification code factor is used to obtain the change in the state of the UAV between adjacent times k-1 and k based on the identification code detected by the gimbal camera.

[0041] IMU prediction factor is used to obtain the change in the UAV state between adjacent time k-1 and k based on the IMU detection value;

[0042] The lidar factor is used to obtain the change in the UAV state between adjacent times k-1 and k based on the lidar detection value.

[0043] The identification code cloze factor is used to obtain the pose transformation of the UAV between the previous j time and the current k time when the same identification code was detected.

[0044] The IMU predicted factor is set in the same way as in the traditional factor graph, and will not be described in detail in this invention. Preferably, the method described in the literature Forster C, Carone L, Dellalt F, et al. On-Manifold Preintegration for Real-Time Visual--Inertial Odometry[J]. ieee transactions on robotics, 2017, 33(1):1-21 is adopted. The IMU raw data is processed and the IMU predicted factor is constructed. According to the kinematic equation, the attitude, position, velocity, accelerometer deviation, and gyroscope deviation are calculated so that the result is only related to the previous state.

[0045] The lidar factors, including lidar keyframe selection and lidar factor construction, can be implemented in any manner. Preferably, the method disclosed in the literature Shan, T., Englot, B., Meyers, D., Wang, W., Ratti, C., & Rus, D. LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping [C]. 2020 IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS).

[0046] More preferably, motion distortion of the lidar is also removed, and the removal method can be the one described in the literature Bai W, Li G, Han L. Correction Algorithm of LiDAR Data for Mobile Robots [C]. International Conference on Intelligent Robotics and Applications. Springer, Cham, 2017: 101-110.

[0047] In this invention, the identification code factor can greatly assist the navigation system in maintaining stability and enhancing robustness. This is because the scene in the industrial environment is complex, and traditional LiDAR motion calculation is prone to degradation, leading to navigation system failure.

[0048] In a preferred embodiment, after the camera recognizes the identification code, it can calculate the relative position of the identification code with respect to the camera coordinate system, which is expressed as:

[0049]

[0050] in, This indicates the relative position of the identifier code with respect to the camera coordinate system. This represents the rotation matrix of the identifier relative to the camera coordinate system. This represents the translation vector of the identifier code relative to the camera coordinate system.

[0051] Furthermore, by using the gimbal rotation angle at that moment, the relative relationship between the camera and the drone body can be obtained:

[0052]

[0053] in, This indicates the relative relationship between the camera and the drone's body. This represents the rotation matrix between the camera and the drone's fuselage;

[0054] The relative pose of the drone body with respect to the identification code can be obtained.

[0055]

[0056] Furthermore, since the identifier is fixed to the ground, the pose of the UAV at time K, under the coordinates of the i-th identifier acquired by the gimbal camera, can be obtained as follows:

[0057]

[0058] And obtain the pose of the UAV at time K+1 under the coordinates of the i-th identifier code acquired by the gimbal camera.

[0059]

[0060] The pose transformation is used as the change in the UAV state between time k and time k+1, which is used as the identification code factor from time k to time k+1.

[0061] Existing LiDAR navigation typically uses horizontal position determination to generate loop closures and then uses LiDAR point cloud matching for loop closure detection. This method is acceptable for two-dimensional motion wheeled robots and unmanned surface vessels, but its application in three-dimensional motion drones has certain limitations. When drones operate at different altitudes, the overlap of LiDAR fields of view is not high, and in industrial scenarios, there are many similar scenarios. Traditional loop closure methods are prone to mismatches, leading to increased errors.

[0062] According to the present invention, by employing an identification code loop closure factor, the identification code is used to determine the generation of loop closures in loop closure detection, which greatly expands the scope of loop closure detection.

[0063] Specifically, when the same identifier code is detected discontinuously, the state with the smallest identifier code factor covariance in each of the previous consecutive detections will be selected, and this state will be looped back with the current time K, such as... Figure 2 As shown.

[0064] For example, during flight, the drone has passed the i-th identifier twice. Let time h and time j be the times when the identifier factor covariance is minimized in these two consecutive detections, respectively. Then, at the current time k, the result will be... and Calculations were performed separately to obtain the pose transformations of the UAV at times h and k. pose transformation of the UAV at times j and k The pose transformations described above are used as identification closure factors to participate in factor graph optimization.

[0065] In a preferred embodiment, the interval between when the drone does not detect the same identifier and the distance it travels away from the identifier are used as conditional thresholds to determine whether it is a continuous detection. If both are greater than the threshold, it is not a continuous detection; otherwise, it is a continuous detection.

[0066] The specific threshold values ​​can be set by those skilled in the art according to actual needs, and are not particularly limited in this invention. For example, the interval time threshold can be set to 30 seconds and the distance threshold can be set to 10 meters.

[0067] On the other hand, the present invention also provides an intelligent inventory inspection method for closed stockpile unmanned aerial vehicles (UAVs). During the navigation process of the UAV, images of the stockpile and the corresponding positions of each image are collected to build a map. The stockpile area and the volume of the stockpile are identified from the model obtained from the map to complete the intelligent inventory inspection.

[0068] Preferably, Photoscan software is used for 3D model building and 2D orthophoto model building. Using images for 3D reconstruction can minimize the impact of dust in the enclosed storage yard on the modeling accuracy.

[0069] Furthermore, the two-dimensional orthophotos obtained from the mapping are classified to obtain specific areas of the stockpile, and then the specific areas of the stockpile are mapped to the three-dimensional model to calculate the stockpile volume.

[0070] Preferably, an optimized RF method is used to classify the two-dimensional orthophoto image to obtain the specific area of ​​the material pile. The optimized RF method can be found in the literature Zhang T, Su J, Xu Z, Luo Y, Li J. Sentinel-2 Satellite Imagery for Urban Land Cover Classification by Optimized Random Forest Classifier. Applied Sciences. 2021; 11(2):543., which will not be elaborated in this invention.

[0071] Example

[0072] Example 1

[0073] The navigation system was tested inside a dry coal shed. A coal shed coordinate system was set up, fixed to the geodetic coordinate system, with the XYZ axes pointing north-west-sky, and the origin set as the UAV's takeoff point. The origin of the navigation system coordinate system was set as the position when the navigation system was started. The XYZ directions corresponded to the front-left-up direction of the lidar. The experimental platform used a DJI M210RTK UAV, equipped with a Velodyne-16 lidar, an MTI-300 inertial navigation element, and an Intel NUC8. A Zenmuse X5S gimbal camera was used for image acquisition. Ground ARUCO codes were marked with DICT_6×6_50, with a side length of 0.8m. Two codes were placed around the flight point, and one code was placed every 10m on the pedestrian walkway at the center of the coal shed (y = -18.5m, z = 1.5m). The layout of these codes in the coal shed coordinate system is shown in Table 1.

[0074] Table 1

[0075]

[0076]

[0077] The observations from multiple sensors are fused using a factor graph optimization method to obtain the real-time pose information of the UAV. Navigation is then performed based on the obtained UAV pose information. The factor graph is set with IMU prediction factor, LiDAR factor, identifier factor, and identifier loop closure factor. In loop closure detection, when the same identifier is detected discontinuously, the state with the smallest identifier factor covariance in each of the previous consecutive detections will be selected, and they will be looped back with the current time K.

[0078] The interval between when the drone does not detect the same identifier and the distance it travels away from the identifier are used as threshold conditions to determine whether it is a continuous detection. The interval threshold is set to 30 seconds and the distance threshold is set to 10 meters.

[0079] During the experiment, the drone was manually controlled to fly, and the collected lidar point cloud data, IMU data, image data, pose data calculated in real time, and data calculated by the drone platform's own navigation system were recorded.

[0080] Experimental Example 1

[0081] The data collected in Example 1 were substituted into the DJI, LOAM, Lego-LOAM, LIO-SAM, and F-LOAM navigation systems for calculation, and compared with the navigation results in Example 1.

[0082] DJI used sensors such as barometers, magnetometers, binocular vision, and downward-looking optical flow to estimate the position of the aircraft platform.

[0083] Loam is a lidar odometry based on surface and angular features. For a detailed introduction, please refer to the literature Zhang J, Singh S. LOAM: Lidar Odometry and Mapping in Real-time[C]. Robotics: Science and Systems Conference. 2014.

[0084] Lego-LOAM separates ground points based on LOAM, optimizing it for applications in wheeled robots adapted to two-dimensional motion. For a detailed introduction, please refer to the literature Shan T, Englot B. LeGO-LOAM: Lightweight and Ground-Optimized Lidar Odometry and Mapping on Variable Terrain[C]. 2018 IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS).

[0085] LIO-SAM uses factor maps to optimize navigation systems by combining LIOAM and the predicted scores commonly used in visual odometry. For a detailed introduction, please refer to the paper Shan, T., Englot, B., Meyers, D., Wang, W., Ratti, C., & Rus, D. LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping [C]. 2020 IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS).

[0086] Floam improves upon LOAM by optimizing the navigation system in aspects such as motion distortion removal from LiDAR point cloud acquisition, feature point weighting, and point cloud matching cost function. For a detailed introduction, please refer to the literature H.Wang, C.Wang, C.-L.Chen and L.Xie, F-LOAM: Fast LiDAR Odometry and Mapping[C]. 2021 IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS), 2021, pp.4390-4396.

[0087] The changes of the X, Y, and Z axes of each navigation system over time in the navigation coordinate system and the failure status of each navigation system are as follows: Figure 3-5 As shown.

[0088] from Figure 3-5 It can be seen that the DJI, LOAM, Lego-LOAM, LIO-SAM, and F-LOAM navigation systems all experienced failures at different stages, while Example 1 did not show any failures. Specifically,

[0089] DJI experienced a failure during the cruise phase. The high altitude stability during the cruise phase was due to the use of altitude hold mode in the experiment, with the altitude hold control data sourced from DJI. Therefore, the altitude did not appear to change significantly in the graph. However, based on the camera pose calculated from various navigation systems and offline visual 3D mapping, the drone was not flying at the same altitude. DJI mistakenly believed that the aircraft was always at the same altitude. Analysis revealed that the barometer altitude was inaccurate due to the influence of temperature and gusts of wind inside the coal shed during the test.

[0090] The failure of the LOAM navigation system during the aircraft's climb phase is mainly manifested in the altitude direction, resulting in a large number of mismatches in the horizontal direction as a result.

[0091] Because Lego-loam is adapted to navigation in two-dimensional motion systems, it did not show any advantage during the takeoff waiting phase due to uneven ground and the low laser reflectivity of coal, and it quickly became ineffective after the aircraft took off.

[0092] F-LOAM failed to function properly after takeoff without human assistance and became completely inoperable shortly after the cruise phase.

[0093] The LIO-SAM began to drift as soon as it was placed on the ground, and completely diverged during the landing phase due to incorrect loop detection matching.

[0094] Experiment Example 2

[0095] Based on the data obtained in Example 1, 3D modeling and 2D orthophoto modeling were performed using Photoscan software. Since the actual volume of the stockpile and the true value of the 3D model were unknown, the accuracy of the 3D modeling was inferred only from the position of the ArUco markers during modeling. Because the relative position of the ArUco markers was known, we measured the error corresponding to the ArUco markers on the pedestrian walkway in the coal shed coordinate system. The accuracy of the model after inventory was inferred from the ArUco marker position error.

[0096] The errors of ARUCO on the pedestrian walkway in the obtained 3D model in the coal shed coordinate system are shown in Table 2.

[0097] Table 2

[0098]

[0099]

[0100] ArUco 0 and ArUco 1 are inconvenient to compare in the take-off and landing area and nearby, and the camera did not capture images of ArUco2 during flight, so there is no corresponding data in Table 2.

[0101] As can be seen from Table 2, the error between the ARUCO coordinates obtained from the model and the actual ARUCO coordinates is very low. It can be inferred that, in conjunction with classification and volume measurement, the task of accurately measuring the stockpile volume can be completed.

[0102] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship in the working state of this invention, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0103] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0104] The present invention has been described above with reference to preferred embodiments; however, these embodiments are merely exemplary and illustrative. Various substitutions and modifications can be made to the present invention based on these embodiments, all of which fall within the scope of protection of the present invention.

Claims

1. A navigation method for unmanned aerial vehicles (UAVs) in a closed storage yard, characterized in that, The observations from multiple sensors are fused using a factor graph optimization method to obtain the real-time pose information of the UAV, and navigation is performed based on the obtained UAV pose information. The multiple sensors include a lidar and an IMU mounted on the drone, an identification code set on the ground, and a gimbal camera set on the drone for detecting the identification code. The factor graph contains multiple factors, including: IMU prediction factor is used to obtain the change in the UAV state between adjacent time k-1 and k based on the IMU detection value; The lidar factor is used to obtain the change in the UAV state between adjacent times k-1 and k based on the lidar detection value. The identifier factor is used to obtain the change in the UAV state between adjacent times k-1 and k based on the identifier detected by the gimbal camera. The identification code cloze factor is used to obtain the pose transformation of the UAV between the previous j time and the current k time when the same identification code was detected. For any time K, the pose of the UAV at time K under the coordinates of the i-th identifier code and the change in pose of the UAV at time K+1 under the coordinates of the i-th identifier code are used as the identifier code factor from time k to time k+1. In loop closure detection, when the same identifier code is detected discontinuously, the state with the smallest identifier code factor covariance in each of the previous consecutive detections will be selected, and they will be looped back with the current time K. In loop closure detection, the interval between when the drone does not detect the same identifier and the distance it travels away from the identifier are used as conditional thresholds to determine whether it is a continuous detection. If both are greater than the threshold, it is not a continuous detection; otherwise, it is a continuous detection.

2. The closed-yard UAV navigation method according to claim 1, characterized in that, The identification code is the ArUco code.

3. A method for intelligent inventory inspection using unmanned aerial vehicles (UAVs) in a closed storage yard, characterized in that, Navigation is performed using the closed-yard UAV navigation method described in claim 1 or 2. During the navigation process, images of the stockpiled material and the corresponding locations of each image are collected to create a map. The stockpiled area and the volume of the stockpiled material are identified from the model obtained from the map, and intelligent inventory checks are completed.

4. The intelligent inventory check method for closed storage yards using unmanned aerial vehicles according to claim 3, characterized in that, The mapping refers to the creation of three-dimensional models and two-dimensional orthophoto models.

5. The intelligent inventory check method for closed storage yards using unmanned aerial vehicles according to claim 3, characterized in that, The two-dimensional orthophotos obtained from the mapping are classified to obtain specific areas of the material stockpile. Then, the specific areas of the material stockpile are mapped to the three-dimensional model to calculate the volume of the material stockpile.

Images