An image recognition-based method and system for automated cargo delivery by unmanned aerial vehicles (UAVs).

Abstract

This invention relates to an image recognition-based method and system for automatic cargo loading by unmanned aerial vehicles (UAVs), belonging to the field of UAV technology. The method includes the following steps: the UAV starts up and acquires its own parameter data; a visual recognition module acquires real-time images of the target area, obtaining a spatial attitude matrix, cargo information, and vibration characteristic parameters of a red plastic ball; based on the vibration characteristic parameters, turbulence and steady-state wind parameters are calculated, and the center of mass position, center of mass offset, and total mass are calculated; a multi-motor thrust distribution matrix is ​​constructed, and a compensation function integrating center of mass offset and wind field disturbance is used to obtain the safe thrust information of each motor; the UAV is guided to descend along a spatial path; when the distance between the UAV and the upper surface of the cargo is less than a set threshold, it enters a fine-tuning mode; after cargo connection is completed, cargo takeoff is achieved based on the safe thrust information of each motor. This invention achieves autonomous cargo parameter recognition, real-time center of mass calculation, and accurate differentiation and compensation of wind field disturbances, significantly improving flight stability and anti-interference capabilities under complex wind fields.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) technology, specifically relating to an image recognition-based UAV automatic cargo loading method and system. Background Technology

[0002] With the rapid development of drone technology, its applications in logistics, emergency rescue, agricultural and forestry monitoring, and urban delivery are becoming increasingly widespread. Especially in small cargo transportation scenarios, drones, with their flexibility and speed, can effectively improve transportation efficiency and reduce labor costs, demonstrating significant advantages. However, in actual cargo-carrying missions, drones still face many technical challenges, which restrict the further optimization and promotion of their cargo-carrying capabilities.

[0003] Based on the intelligent logistics delivery system and method based on unmanned aerial vehicles (UAVs) disclosed in announcement number CN108520376B, and an analysis of existing UAV cargo systems on the market, current UAV cargo systems suffer from several shortcomings in technical implementation: First, cargo information identification relies on manual intervention and lacks automation capabilities. While existing visual recognition technology can be used to identify cargo information, the amount of encoded information is limited, making it difficult to directly parse and calculate dynamic parameters within the autonomous UAV system. Furthermore, related positioning and identification research largely remains at the stage of ground detection or manual assistance, failing to achieve autonomous identification and docking during flight. Second, the calculation of the center of gravity and load cannot be completed in real time, and the acquisition process is complex. Most UAVs lack real-time center of gravity detection and compensation algorithms for load changes, often relying on experience-based settings or additional external sensor systems. If the overall center of gravity shift of the UAV after loading cargo is not corrected in time, it can lead to unstable flight attitude, increased energy consumption, or even loss of flight control. Third, the accuracy of landing and takeoff processes is low, with deviations or errors easily occurring during landing and uneven thrust distribution during takeoff. In addition, the cargo-carrying process of drones in the existing technology still relies on manual intervention or preset parameters, and wind disturbances (including steady wind and turbulence) will further aggravate flight instability. Traditional wind disturbance detection relies on inertial measurement units or complex meteorological sensors, which are difficult to distinguish between turbulence and steady wind. They have problems such as weak targeting, slow response and low compensation accuracy, and it is difficult to achieve real-time reliable response in complex wind field scenarios.

[0004] Therefore, there is an urgent need for an intelligent unmanned aerial vehicle (UAV) cargo system that is simple in structure, can autonomously identify cargo parameters, distinguish between turbulent and steady-state wind in real time, accurately calculate wind field parameters, and dynamically adjust compensation strategies. Summary of the Invention

[0005] This invention overcomes the shortcomings of existing technologies and provides an image recognition-based method and system for automatic cargo loading by drones. By setting a silicone rod-red plastic ball vibration component on the surface of the cargo, the vibration characteristic parameters of the ball are visually perceived to distinguish between turbulent and steady-state wind. Combined with centroid compensation, dynamic control is achieved, which improves flight stability and anti-interference ability under complex wind fields. Moreover, the vibration component has a simple structure, low cost, and is easy to implement in engineering.

[0006] The technical solution adopted by this invention to solve the problems existing in the prior art is:

[0007] An image recognition-based method for automated cargo delivery by drones includes the following steps:

[0008] Step 1: Start the drone. After startup, the drone control unit obtains the drone's own parameter data, including the drone's weight, through the flight platform module. The geometric dimensions and initial centroid coordinates of the UAV Motor output curves and inertia matrix in the flight control system calibration data;

[0009] Step 2: The drone enters flight mode. The drone uses the camera module in its visual recognition module to acquire real-time images of the target area. The control unit uses a detection algorithm to identify the machine-readable code on the target cargo, obtaining the spatial attitude matrix of the machine-readable code and cargo information. Simultaneously, it identifies the silicone rod-red plastic ball vibration assembly on the cargo surface, extracting the vibration characteristic parameters of the red plastic ball, including its actual vibration frequency. Instantaneous frequency fluctuation range Amplitude variance and peak factor ;

[0010] Step 3: Calculate the center of gravity position using the machine-readable code cargo information. centroid offset , and total mass Based on vibration characteristic parameters, turbulent and steady-state winds are distinguished, and the corresponding wind field parameters are calculated respectively: wind speed of steady-state wind. and the x and y component forces , turbulence intensity and equivalent components in the x and y directions , ;

[0011] Step 4: Construct a multi-motor thrust distribution matrix and obtain the thrust distribution formula for each motor. Construct a compensation function that integrates centroid offset, steady-state wind and turbulence and map it to the motor to obtain the final motor thrust information. Perform amplitude limiting processing to obtain the safe thrust information for each motor.

[0012] Step 5: The control unit obtains the UAV's position and attitude adjustment command through machine-readable code; the control unit sends the position and attitude adjustment command and motor thrust distribution information to the UAV's flight control system to guide the UAV to descend along the spatial path;

[0013] Step 6: When the distance between the drone and the top surface of the cargo is less than the set threshold, it enters the fine-tuning mode. After the cargo connection is completed, the drone takes off in cargo-laden state according to the safety thrust information of each motor.

[0014] The machine-readable code for goods includes quality information. Shape type, cargo centroid coordinates Cargo windward area and drag coefficient ;

[0015] In the vibrating assembly consisting of a silicone rod and a red plastic ball, the silicone rod is a composite structure with an inner carbon fiber core and an outer silicone coating. The silicone rod is 8-12cm long and 3-5mm in diameter. The red plastic ball has a diameter of 1.5-2.5cm and a black and white marking ring on its surface. The silicone rod is detachably connected to the surface of the goods via a snap-on fixing seat.

[0016] The specific method for extracting the vibration feature parameters of the red plastic ball is as follows: Images are acquired using a camera module at a frame rate of 30-60 fps; the KCF target tracking algorithm is used to locate the red plastic ball and the surface marking ring; the motion trajectory is fitted; and the actual vibration frequency is extracted using a fast Fourier transform. Calculate the instantaneous frequency fluctuation range within the sliding time window. Amplitude variance and peak factor ,in Peak amplitude, The amplitude is the root mean square (RMS) value.

[0017] Preferably, the centroid position is calculated using cargo information obtained from machine-readable codes. The centroid offset is:

[0018] pass Calculate the position of the centroid Compare with the initial centroid coordinates Get centroid offset , ,Right now ,

[0019] ;

[0020] Calculate the total mass ,in, For the quality of drones, The location of the drone's center of mass. For the quality of goods, The location of the goods.

[0021] Furthermore, the process of distinguishing between turbulent and steady-state wind and calculating wind field parameters in step 3 includes:

[0022] Step 3.1: Set the differentiation threshold and set the threshold for the instantaneous frequency fluctuation range. Amplitude variance threshold and peak factor threshold When satisfied or or If the wind is in a turbulent state, it is considered to be in a steady state; otherwise, it is considered to be in a

[0023] Step 3.2: Calculate steady-state wind parameters:

[0024] Calculate frequency deviation Through the steady-state wind speed formula To calculate the wind speed, where, For frequency-velocity calibration coefficients, Base offset;

[0025] The steady-state wind force formula is used to obtain:

[0026] ,

[0027] ;

[0028] Step 3.3: If turbulence exists, calculate the turbulence parameters:

[0029] Through the turbulence intensity formula Solve for turbulence intensity, where These are turbulence calibration coefficients;

[0030] The following is obtained from the turbulence equivalent force formula:

[0031] ,

[0032] ;

[0033] in, air density; The angle between the main direction of the wind field and the x-axis is calculated by fitting the major axis of the trajectory of the red plastic ball. , These are the effective projected areas of the ball's motion in the x and y directions, respectively; The maximum projected area of ​​the small ball. .

[0034] Preferably, the specific method for constructing the multi-motor thrust allocation matrix and obtaining the thrust allocation formula for each motor is as follows:

[0035] Based on total mass With gravitational acceleration Calculate the total lift ,in To ensure a safety margin of thrust, take 10-15% of the total weight under steady-state wind conditions, and adjust to 15-20% under turbulent conditions;

[0036] The drone is specifically a quadcopter drone, powered by four motors, with the motor distribution matrix as follows: ,in Let be the position coordinates of the i-th motor in the UAV's centroid coordinate system;

[0037] Combined with centroid offset and wind field component , Obtain the thrust balance equations and convert them into matrix form:

[0038] ;

[0039] The distribution of each motor is as follows: Motor 1: Motor 2: Motor 3: Motor 4: Where a and b are the arm lengths in the x and y directions, respectively, and the thrust distribution formula for each motor is as follows:

[0040] ;

[0041] ;

[0042] ;

[0043] .

[0044] Furthermore, the specific method for constructing the compensation function is as follows:

[0045] Based on centroid offset , Steady-state wind force , and turbulent equivalent force , Construct the compensation function:

[0046] ;

[0047] Where i is the motor number. Furthermore, the compensation coefficient matrix is:

[0048] , ,

[0049] , ,

[0050] , ;

[0051] Among them, steady-state wind compensation coefficient , Turbulence compensation coefficient , ; This represents the maximum permissible steady-state wind force.

[0052] Furthermore, the process of obtaining the final motor thrust information is as follows:

[0053] First, calculate the basic thrust: ;

[0054] Secondly, calculate the compensation amount in each direction:

[0055] , ,

[0056] , ,

[0057] , ;

[0058] The final motor thrust information is obtained by combining the base thrust and the compensation amounts in each direction:

[0059] ( ).

[0060] The limiting process involves inputting the motor thrust information into a limiting function for calculation to obtain the safe thrust information for each motor. The limiting function is as follows:

[0061] ,

[0062] in, and These are the minimum and maximum thrust limits for the motor, respectively.

[0063] Preferably, the control unit obtains the positional deviation of the UAV from the cargo docking point through machine-readable code, calculates the UAV position and attitude adjustment value, and obtains attitude adjustment commands; in turbulent environments, the thrust adjustment cycle is shortened, and the attitude control response speed is improved.

[0064] When the drone is less than a set threshold away from the top of the cargo, it enters fine-tuning mode. Based on the vibration component images and machine-readable code images collected in real time by the camera module, the position deviation and wind field compensation parameters are corrected simultaneously. After the docking mechanism makes contact, it automatically closes. After the cargo connection is completed, the flight control system realizes the drone take-off in cargo-laden state based on the safety thrust information of each motor.

[0065] An image recognition-based unmanned aerial vehicle (UAV) automated cargo delivery system includes a flight platform module, a vision recognition module, a control unit, a docking mechanism module, and a vibration component.

[0066] The flight platform module includes a quadcopter UAV airframe, flight control system, motor assembly, propeller, optical flow module, inertial measurement unit, GPS module, and battery system;

[0067] The visual recognition module includes a camera module, which is located on the bottom of the drone and has a frame rate of no less than 30fps.

[0068] The docking mechanism module is located at the bottom of the UAV, and is specifically a servo-controlled pin structure.

[0069] The vibration assembly includes a thin silicone rod, a small red plastic ball, and a fixed base. The thin silicone rod has a composite structure of "carbon fiber core + silicone outer layer". The surface of the small red plastic ball is marked with a black and white identification ring. The fixed base is detachably connected to the surface of the goods.

[0070] The control unit integrates a wind field sensing module, a compensation and correction module, and a wind field adaptation module. The wind field sensing module is used to extract vibration feature parameters from the vibration component image, distinguish between turbulent and steady-state wind, and calculate wind field parameters. The compensation and correction module is used to fuse multi-dimensional compensation amount correction compensation function and motor thrust distribution command. The wind field adaptation module is used to adjust the safety margin thrust and thrust adjustment cycle in turbulent environment.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] First, this invention uses a silicone rod with a carbon fiber core and a silicone outer layer composite structure and a red plastic ball with a marking ring to form a vibration component. The structure is simple, low-cost, and has a sensitive vibration response, ensuring both bending resistance and accurate response to wind field disturbances. By extracting multi-dimensional vibration characteristic parameters, including frequency, fluctuation range, amplitude variance, and peak factor, it can accurately distinguish between turbulent and steady-state winds, solving the problem that traditional technologies have difficulty distinguishing complex wind field types.

[0073] Secondly, this invention constructs a multi-dimensional compensation function that integrates centroid offset, steady-state wind, and turbulence. It dynamically adjusts the safety margin thrust and thrust adjustment cycle for different wind field types, achieving real-time matching of "centroid-wind field type-power". This effectively offsets the dual effects of centroid offset and complex wind fields, significantly improving the stability and anti-interference capability of UAV cargo flight, and reducing energy consumption and runaway risk.

[0074] Finally, this invention achieves integrated acquisition of cargo information through machine-readable codes, and combines the wind field sensing of vibration components with an automatic docking device to form a closed-loop collaborative system that requires no manual intervention throughout the process and has a high degree of automation. Attached Figure Description

[0075] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0076] Figure 1 This is a schematic diagram illustrating the control principle of an image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method and system according to the present invention.

[0077] Figure 2 This is a flowchart of an image recognition-based unmanned aerial vehicle (UAV) automatic cargo loading method according to the present invention. Detailed Implementation

[0078] The following is in conjunction with the appendix Figures 1-2 The present invention provides a more detailed description of an image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method and system, but this is not intended to limit the scope of the invention.

[0079] like Figure 1 As shown, an image recognition-based unmanned aerial vehicle (UAV) automated cargo-carrying system includes a flight platform module, a vision recognition module, a control unit, a docking mechanism module, and a vibration component.

[0080] The flight platform module includes a quadcopter drone fuselage, flight control system, motor assembly, propellers, optical flow module, inertial measurement unit, GPS module, and battery system. The quadcopter drone fuselage, motor assembly, propellers, and battery system are essential structures for drone flight; the inertial measurement unit, optical flow module, and GPS module are used to acquire the current flight status and position information of the drone. In this embodiment, the flight control system is specifically the Pixhawk flight control system. This flight platform module performs attitude control, position control, and power output, providing stable flight support for the entire system.

[0081] The visual recognition module includes a camera module, specifically a USB high-definition camera with a frame rate of 60fps, mounted on the bottom of the drone. It is used to detect machine-readable codes and vibrating components attached to the surface of the cargo in real time, extracting encoded information, spatial coordinates, and the ball motion characteristics of the vibrating components, providing a foundation for precise drone landing, wind field perception, and cargo docking. In this embodiment, the machine-readable code is specifically an AprilTag code.

[0082] The control unit, specifically a Raspberry Pi 4B, connects to the camera module and flight control system. It is responsible for running core algorithms, including wind field perception, compensation and correction, and wind field adaptation. The wind field perception function extracts vibration characteristic parameters from images of vibrating components, distinguishes between turbulent and steady-state wind, and calculates wind field parameters. The compensation and correction function integrates a multi-dimensional compensation amount correction function with motor thrust distribution commands. The wind field adaptation function is used to adjust the safety margin thrust and thrust adjustment cycle in turbulent environments. The control unit also interacts with the flight control system via the Mavlink protocol to achieve real-time transmission of attitude parameters, thrust commands, and visual feedback information.

[0083] The docking mechanism module is located at the bottom of the UAV and is used to connect cargo during the landing phase. Specifically, it is a servo-controlled pin structure. In this embodiment, the servo control voltage is 5V and the pin pop-out time is ≤0.5s.

[0084] The vibration assembly includes a thin silicone rod, a small red plastic ball, and a snap-on mounting bracket adapted to the silicone rod. The silicone rod passes through the center of the small red plastic ball and is fixed to the surface of the goods via the snap-on bracket, making it detachably connected to the goods surface. The reference vibration frequency of the small red plastic ball in a windless environment is [insert frequency here]. The silicone rod is a composite structure with an inner carbon fiber core and an outer silicone coating. This structure ensures both bending resistance and flexibility to respond to wind disturbances. The silicone rod is 8-12cm long and 3-5mm in diameter. The red plastic ball has a diameter of 1.5-2.5cm and features 1-2 black and white marking rings on its surface, with a ring width of 0.3-0.5cm, enhancing visual tracking accuracy. This vibration component requires no external power supply and vibrates naturally due to wind disturbances. The reference vibration frequency is [not specified] in windless environments. The frequency is 2-4Hz, and this component is detachably connected to the surface of the goods via a snap-fit ​​mechanism, making it simple in structure and easy to install. In this embodiment, a 10cm long, 4mm diameter silicone rod and a 2cm diameter red plastic ball with a black and white marking ring on its surface are specifically selected.

[0085] like Figure 2 As shown, an image recognition-based method for automated cargo loading by unmanned aerial vehicles includes the following steps:

[0086] Step 1: Start the drone. After startup, the drone control unit obtains the drone's own parameter data, including the drone's weight, through the flight platform module. The geometric dimensions and initial centroid coordinates of the UAV Motor output curves and inertia matrix in the flight control system calibration data;

[0087] Step 2: The drone enters flight mode. The drone uses the camera module in its visual recognition module to acquire real-time images of the target area. The control unit uses a detection algorithm to identify the machine-readable code on the target cargo, obtaining the spatial attitude matrix of the machine-readable code and cargo information. Simultaneously, it identifies the silicone rod-red plastic ball vibration assembly on the cargo surface, extracting the vibration characteristic parameters of the red plastic ball, including its actual vibration frequency. Instantaneous frequency fluctuation range Amplitude variance and peak factor ;

[0088] Step 3: Calculate the center of gravity position using the machine-readable code cargo information. centroid offset , and total mass Based on vibration characteristic parameters, turbulent and steady-state winds are distinguished, and the corresponding wind field parameters are calculated respectively: wind speed of steady-state wind. and the x and y component forces , turbulence intensity and equivalent components in the x and y directions , ;

[0089] Step 4: Construct a multi-motor thrust distribution matrix and obtain the thrust distribution formula for each motor. Construct a compensation function that integrates centroid offset, steady-state wind and turbulence and map it to the motor to obtain the final motor thrust information. Perform amplitude limiting processing to obtain the safe thrust information for each motor.

[0090] Step 5: The control unit obtains the UAV's position and attitude adjustment command through machine-readable code; the control unit sends the position and attitude adjustment command and motor thrust distribution information to the UAV's flight control system to guide the UAV to descend along the spatial path;

[0091] Step 6: When the distance between the drone and the top surface of the cargo is less than the set threshold, it enters the fine-tuning mode. After the cargo connection is completed, the drone takes off in cargo-laden state according to the safety thrust information of each motor.

[0092] In this embodiment, after the drone starts flying, it takes off and uses the camera module to photograph the cargo, obtaining image information of the AprilTag code affixed to the cargo, as well as image and time information of the vibration component, and then transmits the information to the Raspberry Pi.

[0093] The Raspberry Pi uses the AprilTag detection algorithm, specifically the Tag36h11 encoding family recognition algorithm based on OpenCV, to parse the AprilTag code information on the target goods, obtain the tag's ID number, and retrieve the corresponding goods information. The spatial pose matrix of the AprilTag code is calculated using the Zhang's calibration algorithm. The goods information from the AprilTag code includes mass... Shape type, cargo centroid coordinates Cargo windward area and drag coefficient .

[0094] On the other hand, the Raspberry Pi will use the information collected by the camera module, employ the KCF target tracking algorithm to lock onto the red plastic ball and the surface marking ring, fit the motion trajectory, and extract the actual vibration frequency through Fast Fourier Transform (FFT). Calculate the instantaneous frequency fluctuation range within the sliding time window. Amplitude variance and peak factor ,in The peak amplitude is the maximum amplitude taken within the calculation time. The root mean square amplitude is the root mean square of the amplitude within the calculation time, which is usually 1 second.

[0095] The centroid position is calculated using the cargo information obtained through machine-readable codes. The centroid offset is:

[0096] pass Calculate the position of the centroid Compare with the initial centroid coordinates Get centroid offset , ,Right now , ;

[0097] Calculate the total mass ,in, For the quality of drones, The location of the drone's center of mass. For the quality of goods, The location of the goods.

[0098] The process of distinguishing between turbulent and steady-state wind and calculating wind field parameters includes:

[0099] Step 3.1: Set the differentiation threshold and set the threshold for the instantaneous frequency fluctuation range. Amplitude variance threshold Peak factor threshold When satisfied or or If the wind is strong, it is considered to be turbulent; otherwise, it is considered to be steady-state wind.

[0100] Step 3.2, Steady-state wind parameter calculation:

[0101] Calculate frequency deviation Through the steady-state wind speed formula Calculate the wind speed, where This is the frequency-velocity calibration coefficient, specifically measured in m / (s·Hz). Through experimental calibration, its value ranges from 0.3 to 0.5. This invention provides an embodiment where, when the silicone rod is 10cm long... . This is the base offset, which defaults to 0.

[0102] The steady-state wind force formula is used to obtain:

[0103] ,

[0104] ;

[0105] Step 3.3: If turbulence exists, calculate the turbulence parameters:

[0106] Through the turbulence intensity formula Solve for turbulence intensity, where This is the turbulence calibration factor, with a value ranging from 0.8 to 1.2, representing the turbulence intensity. The larger the value, the more intense the turbulence;

[0107] The following is obtained from the turbulence equivalent force formula:

[0108] ,

[0109] ;

[0110] In the above formula, The density of air is typically taken as 1.225 kg / m³. The angle between the main direction of the wind field and the x-axis is calculated by fitting the angle between the major axis of the trajectory of the red plastic ball and the x-axis. , These are the effective projected areas of the ball's motion in the x and y directions, respectively, calculated from the ball's motion amplitude and diameter. The maximum projected area of ​​the small ball. .

[0111] The and The specific calculation method is as follows: calculate the effective projected area of ​​the ball's motion in the x and y directions at time t. and ,

[0112]

[0113]

[0114] in, It is the maximum cross-sectional area of ​​the ball when it is vertical. Let be the angle of the ball's swing in the yz plane. Let be the angle of the ball's swing in the xz plane. , For calculation time , The mean value is typically calculated in 1 second.

[0115] The specific method for constructing the multi-motor thrust allocation matrix and obtaining the thrust allocation formula for each motor is as follows:

[0116] Based on total mass With gravitational acceleration (Default value is 9.8 m / s²), calculate the total lift. ,in To ensure a safety margin of thrust, 10-15% of the total weight is used in steady-state wind conditions, and adjusted to 15-20% in turbulent conditions to improve the disturbance rejection margin;

[0117] The drone is specifically a quadcopter drone, powered by four motors, with the motor distribution matrix as follows: ,in Let be the position coordinates of the i-th motor in the UAV's centroid coordinate system;

[0118] Combined with centroid offset , and wind field component , We obtain the thrust balance equations and further convert them into matrix form:

[0119] ;

[0120] Due to the symmetrical structure of the quadcopter drone, the distribution of the motors is as follows: Motor 1: Motor 2: Motor 3: Motor 4: Where a and b are the arm lengths in the x and y directions, respectively, determined by the UAV's geometry. The thrust distribution formula for each motor is as follows:

[0121] ;

[0122] ;

[0123] ;

[0124] .

[0125] The specific method for constructing the compensation function is as follows:

[0126] Based on centroid offset , Steady-state wind force , and turbulent equivalent force , Construct the compensation function:

[0127] ;

[0128] The compensation coefficient matrix is ​​as follows:

[0129] , ,

[0130] , ,

[0131] , ;

[0132] Steady-state wind compensation coefficient , Turbulence compensation coefficient , The turbulence compensation coefficient is slightly higher than that for steady-state wind to improve turbulence resistance. The maximum permissible steady-state wind force is 4-8N, calibrated according to the payload capacity of the UAV.

[0133] The final motor thrust information is obtained as follows:

[0134] First, calculate the basic thrust: The total lift is evenly distributed among the four motors.

[0135] Secondly, calculate the compensation amount in each direction:

[0136] , ,

[0137] , ,

[0138] , .

[0139] The final motor thrust information is obtained by combining the base thrust and the compensation amounts in each direction: ( ).

[0140] To ensure system safety, the thrust of each motor is limited. The motor thrust information is input into the limiting function for calculation to obtain the safe thrust information for each motor. The limiting function is as follows:

[0141] ,

[0142] in, and These are the minimum and maximum thrust limits for the motor, with values ​​ranging from 5-8N and 25-35N respectively.

[0143] The control unit obtains the positional deviation of the UAV from the cargo docking point through machine-readable codes, calculates the UAV's position and attitude adjustment values, and obtains attitude adjustment commands. In turbulent environments, the control unit shortens the thrust adjustment cycle to 45-60ms, improving the attitude control response speed and enhancing anti-turbulence capabilities. The control unit sends the attitude adjustment commands and motor thrust distribution information to the UAV's flight control system, which guides the UAV to descend along the spatial path, performing a precise landing.

[0144] In this embodiment, the Raspberry Pi uses the AprilTag coordinate information to obtain the positional deviation of the UAV from the cargo docking point and acquires adjustment commands. The Raspberry Pi then sends the adjustment commands and the motor thrust distribution information calculated and acquired in the Raspberry Pi to the UAV's Pixhawk flight control system via MAVLink, thereby enabling the UAV to descend.

[0145] When the drone is less than a set threshold away from the top surface of the cargo, it enters fine-tuning mode. Based on real-time images of the vibration components captured by the camera module and machine-readable code images, it synchronously corrects positional deviations and wind field compensation parameters. After contact, the docking mechanism automatically closes under the action of the servo motor, completing the cargo connection. Upon receiving feedback that docking is complete, the flight control system controls the drone to take off smoothly and execute the transportation mission based on the safety thrust information of each motor.

[0146] The present invention also provides another embodiment:

[0147] After the drone starts up, the drone control unit Raspberry Pi 4B obtains the drone's own parameter data, including the drone's weight, through the flight platform module. Initial centroid coordinates x-axis arm length y-axis arm length Minimum thrust of motor Maximum thrust Gravitational acceleration The flight control system is calibrated so that the motor output curve is a linear response, meaning the thrust is proportional to the control signal, and the inertial matrix... .

[0148] The drone initiates flight mode, hovers 10 meters above the target area, and captures images of the cargo at 60fps using its bottom camera module, transmitting the images to a Raspberry Pi 4B. The control unit then uses OpenCV's Tag36h11 encoding family recognition algorithm to detect the AprilTag code and determine the cargo's weight. centroid coordinates The cargo is a regular cuboid with dimensions of 0.4m, 0.3m, and 0.2m (length, width, and height), and a windward area of ​​[missing information]. drag coefficient With spatial attitude matrix .

[0149] Simultaneously, the KCF target tracking algorithm was used to lock onto the red plastic ball and the marker ring, fit their motion trajectories, and extract vibration characteristic parameters. The specific parameters of the vibration component were: a 10cm long, 4mm diameter silicone rod; a 2cm diameter red plastic ball; and a windless reference vibration frequency of [missing information]. Frequency-speed calibration coefficient Turbulence calibration coefficient The Raspberry Pi vibration detection component calculates the actual vibration frequency. Instantaneous frequency fluctuation range Amplitude variance Peak factor .

[0150] The centroid position, centroid offset, and total mass are calculated using machine-readable code information about the cargo, while also distinguishing between turbulent and steady-state winds and estimating wind field parameters.

[0151] Total mass: ;

[0152] The coordinates of the centroid of the combined system:

[0153] ;

[0154] .

[0155] Centroid offset:

[0156] ;

[0157] .

[0158] Set a threshold to differentiate between instantaneous frequency fluctuation ranges. Amplitude variance threshold Peak factor threshold .

[0159] Steady-state wind parameter estimation: Based on , , The current environment is determined to be turbulent.

[0160] Steady-state wind parameters are calculated, including air density. The maximum permissible steady-state wind component is set to The angle between the wind direction and the x-axis , , , .

[0161] The specific steady-state wind calculation process is as follows:

[0162] ;

[0163] ;

[0164] ;

[0165] .

[0166] The specific calculation process for turbulence parameters is as follows:

[0167] ;

[0168] ;

[0169] .

[0170] Construct a multi-motor thrust distribution matrix and obtain the thrust distribution formula for each motor; construct a compensation function and perform amplitude limiting processing:

[0171] Calculate total gravity: .

[0172] The safety margin thrust is taken as 12% under steady-state wind conditions and 18% under turbulent conditions.

[0173] Safety margin thrust: ,

[0174] Total lift: ;

[0175] Compensation coefficient calibration: due to the maximum allowable steady-state wind component force Therefore, the compensation coefficient is:

[0176] ,

[0177] ,

[0178] ,

[0179] .

[0180] Calculate the thrust of each motor:

[0181] ;

[0182] ;

[0183] ;

[0184] ;

[0185] Limit the thrust of each motor: Set F max =30N, F min =5N, the above thrust is limited to obtain the final safe thrust as follows:

[0186]

[0187]

[0188]

[0189] .

[0190] Raspberry Pi 4B uses AprilTag coordinate information to calculate the positional deviation at the docking point between the drone and the cargo. , , The Raspberry Pi 4B generates attitude adjustment commands, i.e., roll angle correction. Pitch angle correction and yaw angle correction The attitude adjustment commands and motor thrust distribution information are sent to the Pixhawk flight control system via Mavlink. In turbulent conditions, the thrust adjustment cycle is set to 50ms to guide the UAV to descend along a preset path.

[0191] When the drone is 0.4m away from the top surface of the cargo (less than the threshold of 0.5m), it enters fine-tuning mode, acquiring images in real time through the camera module and simultaneously correcting positional deviations and wind field compensation parameters. After the docking mechanism contacts the cargo, the servo motor drives the pin to close, completing the cargo connection. Upon receiving feedback that docking is complete, the flight control system starts the motors according to the safety thrust command, and the drone takes off smoothly at a speed of 0.5m / s to perform the transportation mission.

[0192] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for automated cargo loading by unmanned aerial vehicles based on image recognition, characterized in that, Includes the following steps: Step 1: Start the drone. After startup, the drone control unit obtains the drone's own parameter data, including the drone's weight, through the flight platform module. The geometric dimensions and initial centroid coordinates of the UAV Motor output curves and inertia matrix in the flight control system calibration data; Step 2: The drone enters flight mode. The drone uses the camera module in its visual recognition module to acquire real-time images of the target area. The control unit uses a detection algorithm to identify the machine-readable code on the target cargo, obtaining the spatial attitude matrix of the machine-readable code and cargo information. Simultaneously, it identifies the silicone rod-red plastic ball vibration assembly on the cargo surface, extracting the vibration characteristic parameters of the red plastic ball, including its actual vibration frequency. Instantaneous frequency fluctuation range Amplitude variance and peak factor ; Step 3: Calculate the centroid position using the machine-readable code cargo information. centroid offset , and total mass Based on vibration characteristic parameters, turbulent and steady-state winds are distinguished, and the corresponding wind field parameters are calculated respectively: wind speed of steady-state wind. and the x and y component forces , turbulence intensity and equivalent components in the x and y directions , ; Step 4: Construct a multi-motor thrust distribution matrix and obtain the thrust distribution formula for each motor. Construct a compensation function that integrates centroid offset, steady-state wind and turbulence and map it to the motor to obtain the final motor thrust information. Perform amplitude limiting processing to obtain the safe thrust information for each motor. Step 5: The control unit obtains the UAV's position and attitude adjustment command through machine-readable code; the control unit sends the position and attitude adjustment command and motor thrust distribution information to the UAV's flight control system to guide the UAV to descend along the spatial path; Step 6: When the distance between the drone and the top surface of the cargo is less than the set threshold, it enters the fine-tuning mode. After the cargo connection is completed, the drone takes off in cargo-laden state according to the safety thrust information of each motor.

2. The image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method according to claim 1, characterized in that, The machine-readable code for goods includes quality. Shape type, cargo centroid coordinates Cargo frontal area and drag coefficient ; In the vibrating assembly consisting of a silicone rod and a red plastic ball, the silicone rod is a composite structure with an inner carbon fiber core and an outer silicone coating. The silicone rod is 8-12cm long and 3-5mm in diameter. The red plastic ball has a diameter of 1.5-2.5cm and a black and white marking ring on its surface. The silicone rod is detachably connected to the surface of the goods via a snap-on fixing seat.

3. The image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method according to claim 2, characterized in that, The specific method for extracting the vibration characteristic parameters of the red plastic ball is as follows: Images are acquired using a camera module at a frame rate of 30-60 fps; the KCF target tracking algorithm is used to locate the red plastic ball and the surface marking ring; the motion trajectory is fitted; and the actual vibration frequency is extracted using a fast Fourier transform. Calculate the instantaneous frequency fluctuation range within the sliding time window. Amplitude variance and peak factor ,in Peak amplitude, The amplitude is the root mean square (RMS) value.

4. The image recognition-based unmanned aerial vehicle (UAV) automatic cargo loading method according to claim 2, characterized in that, The centroid position is calculated using the cargo information obtained through machine-readable codes. and centroid offset , for: pass Calculate the position of the centroid Compare with the initial centroid coordinates Get centroid offset , ,Right now , ； Calculate the total mass ,in, For the quality of drones, The location of the drone's center of mass. For the quality of goods, The location of the goods.

5. The image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method according to claim 1, characterized in that, Step 3, the process of distinguishing between turbulent and steady-state wind and calculating wind field parameters, includes: Step 3.1: Set the differentiation threshold and set the threshold for the instantaneous frequency fluctuation range. Amplitude variance threshold and peak factor threshold When satisfied or or If the wind is in a turbulent state, it is considered to be in a steady state; otherwise, it is considered to be in a steady state. Step 3.2: Calculate steady-state wind parameters: Calculate frequency deviation Through the steady-state wind speed formula To calculate the wind speed, where, For frequency-velocity calibration coefficients, Base offset; The steady-state wind force formula yields the following: ， ； Step 3.3: If turbulence exists, calculate the turbulence parameters: Through the turbulence intensity formula Solve for turbulence intensity, where These are turbulence calibration coefficients; The following is obtained from the turbulence equivalent force formula: ， ； in, air density; The angle between the main direction of the wind field and the x-axis is calculated by fitting the major axis of the trajectory of the red plastic ball. , These are the effective projected areas of the ball's motion in the x and y directions, respectively; The maximum projected area of ​​the small ball. .

6. The image recognition-based unmanned aerial vehicle (UAV) automated cargo loading method according to claim 5, characterized in that, The specific method for constructing the multi-motor thrust allocation matrix and obtaining the thrust allocation formula for each motor is as follows: Based on total mass With gravitational acceleration Calculate the total lift ,in To ensure a safety margin of thrust, take 10-15% of the total weight under steady-state wind conditions, and adjust to 15-20% under turbulent conditions; The drone is specifically a quadcopter drone, powered by four motors, with the motor distribution matrix as follows: ,in Let be the position coordinates of the i-th motor in the UAV's centroid coordinate system; Combined with centroid offset and wind field component , Obtain the thrust balance equations and convert them into matrix form: ； The distribution of each motor is as follows: Motor 1: Motor 2: Motor 3: Motor 4: Where a and b are the arm lengths in the x and y directions, respectively, and the thrust distribution formula for each motor is as follows: ； ； ； 。 7. The image recognition-based unmanned aerial vehicle (UAV) automatic cargo loading method according to claim 6, characterized in that, The method for constructing the compensation function is as follows: Based on centroid offset , Steady-state wind force , and turbulent equivalent force , Construct the compensation function: ； The compensation coefficient matrix is ​​as follows: ， ， ， ， ， ； Among them, steady-state wind compensation coefficient , Turbulence compensation coefficient , ; This represents the maximum permissible steady-state wind force.

8. The image recognition-based unmanned aerial vehicle (UAV) automatic cargo loading method according to claim 7, characterized in that, The final motor thrust information obtained in step 4 is as follows: First, calculate the basic thrust: ; Secondly, calculate the compensation amount in each direction: ， ， ， ， ， ； The final motor thrust information is obtained by combining the base thrust and the compensation amounts in each direction: （ ）； The limiting process involves inputting the motor thrust information into a limiting function for calculation to obtain the safe thrust information for each motor. The limiting function is as follows: ， in, and These are the minimum and maximum thrust limits for the motor, respectively.

9. The image recognition-based unmanned aerial vehicle (UAV) automatic cargo loading method according to claim 1, characterized in that, The control unit obtains the positional deviation of the UAV from the cargo docking point through machine-readable codes, calculates the UAV's position and attitude adjustment values, and obtains attitude adjustment commands; in turbulent environments, it shortens the thrust adjustment cycle and improves the attitude control response speed. When the drone is less than a set threshold away from the top of the cargo, it enters fine-tuning mode. Based on the vibration component images and machine-readable code images collected in real time by the camera module, the position deviation and wind field compensation parameters are corrected simultaneously. After the docking mechanism makes contact, it automatically closes. After the cargo connection is completed, the flight control system realizes the drone take-off in cargo-laden state based on the safety thrust information of each motor.

10. An image recognition-based unmanned aerial vehicle (UAV) automated cargo loading system, comprising an image recognition-based UAV automated cargo loading method according to any one of claims 1 to 9, characterized in that, This includes a flight platform module, a vision recognition module, a control unit, a docking mechanism module, and vibration components; The flight platform module includes a quadcopter UAV airframe, flight control system, motor assembly, propeller, optical flow module, inertial measurement unit, GPS module, and battery system; The visual recognition module includes a camera module, which is located on the bottom of the drone and has a frame rate of no less than 30fps. The docking mechanism module is located at the bottom of the UAV, and is specifically a servo-controlled pin structure. The vibration assembly includes a silicone rod, a red plastic ball, and a fixing base. The silicone rod has a composite structure of "carbon fiber core + silicone outer layer". The surface of the red plastic ball is marked with a black and white identification ring. The fixing base is detachably connected to the surface of the goods. The control unit integrates a wind field sensing module, a compensation and correction module, and a wind field adaptation module. The wind field sensing module is used to extract vibration feature parameters from the vibration component image, distinguish between turbulent and steady-state wind, and calculate wind field parameters. The compensation and correction module is used to fuse multi-dimensional compensation amount correction compensation function and motor thrust distribution command. The wind field adaptation module is used to adjust the safety margin thrust and thrust adjustment cycle in turbulent environment.

Images