Self-adaptive inertia / vision / height initialization method for unmanned aerial vehicle

Abstract

The invention provides a self-adaptive inertia / vision / height initialization method for an unmanned aerial vehicle, and the method comprises the steps: carrying out the independent initialization of vision and inertia, and obtaining vision measurement information and inertia measurement information; integrating the visual feature matching information, the inertial measurement information and the height information, and distinguishing the current state of the unmanned aerial vehicle; initializing parameters of the unmanned aerial vehicle starting from a ground static state by adopting an inertia / vision joint initialization method to obtain updated vision measurement information and inertia measurement information; a dynamic scene-oriented inertia / vision / height joint initialization method is adopted to initialize parameters of the unmanned aerial vehicle from a motion state, and updated vision measurement information and inertia measurement information are obtained. Initialization of the inertia / vision sensor can be accurately realized in multiple unmanned aerial vehicle application scenes, and high-precision and high-robustness initialization parameter solving is realized.

Description

Technical Field

[0001] This invention belongs to the field of navigation and positioning technology, and relates to an adaptive inertial / visual / altitude initialization method for unmanned aerial vehicles (UAVs). Background Technology

[0002] Modern military warfare is gradually shifting towards intelligent warfare, and unmanned aerial vehicles (UAVs), as an intelligent combat carrier, are playing an increasingly important role on the battlefield. UAV navigation methods are the technologies and methods used to guide aircraft to complete a prescribed flight process along a certain speed and direction. By acquiring high-precision six-degree-of-freedom position and attitude information in real time, and converting it into the UAV's spatial latitude and longitude coordinates, the information is fed back to the aircraft control system, thereby helping the UAV achieve autonomous navigation and landing functions.

[0003] Traditional UAV navigation methods largely rely on satellite information for navigation and positioning, with differential GPS-based methods typically providing relatively accurate spatial positions. However, battlefield situations are unpredictable, and satellite denial is frequent, making autonomous navigation solutions independent of satellites urgently needed. Methods based on combined inertial (INS) and visual sensor navigation and positioning offer complementary advantages and have become a research hotspot in the field of UAV autonomous navigation in recent years. During flight, due to visual scale discrepancies between the scene depth calculated by a monocular camera and the actual scene depth, current methods often utilize altitude information provided by laser rangefinders or barometric altitude measurements to assist UAVs in navigation and positioning.

[0004] To achieve high-precision fusion positioning using inertial / vision sensors, it is necessary to align the inertial and visual data and compensate for sensor errors. This is typically achieved by solving for relevant key parameters through joint inertial / vision initialization. However, inertial initialization generally requires the UAV to be stationary, while visual initialization requires motion excitation, and some observations even require environmental excitations such as acceleration. Therefore, current solutions for initialization parameters have low accuracy and poor robustness. Summary of the Invention

[0005] The purpose of this invention is to at least solve one of the problems existing in the prior art.

[0006] To address this, the present invention provides an adaptive inertial / visual / altitude initialization method for unmanned aerial vehicles (UAVs), which can accurately initialize inertial / visual sensors in multiple UAV application scenarios and achieve high-precision and robust initialization parameter solving.

[0007] The technical solution of the present invention is as follows:

[0008] An adaptive inertial / vision / altitude initialization method for UAVs, comprising the following steps:

[0009] Step 1: Initialize the vision and inertial measurements separately to obtain visual measurement information and inertial measurement information respectively; the visual measurement information is obtained by solving the pose through visual feature matching and confirming the feature point depth through triangulation; the inertial measurement information is obtained through inertial navigation pre-integration.

[0010] Step 2: Integrate visual feature matching information, inertial measurement information, and altitude information to determine the current state of the drone;

[0011] Step 3: Initialize the parameters of the UAV starting from a stationary state on the ground using the inertial / visual joint initialization method to obtain updated visual measurement information and inertial measurement information;

[0012] Step 4: Based on the parameters obtained in Step 3, the UAV parameters are initialized from the start of motion using a joint inertial / visual / altitude initialization method for dynamic scenarios, resulting in updated visual and inertial measurement information.

[0013] Furthermore, in step 1, the process of initializing the vision and obtaining the visual measurement results is as follows:

[0014] First, select image frames: use adjacent preceding and following image frames for pose calculation;

[0015] Then, after acquiring two frames of images to be matched, SURF feature matching can be used to obtain matching feature point pairs between adjacent frames, that is, visual feature matching information is obtained. Then, epipolar geometry is used to calculate the pose of the UAV. Furthermore, based on the obtained feature point pairs, triangulation technology can be used to obtain the depth of three-dimensional map points on the selected two frames of images.

[0016] Finally, pose optimization is performed using cluster adjustment technology. At this point, the optimized UAV pose and initial map points can be obtained. The UAV pose and initial map points are visual measurement information. Obtaining the visual measurement information indicates that visual initialization is complete.

[0017] Furthermore, in step 1, the process of initializing the inertia and obtaining inertial measurement information is as follows:

[0018] By integrating the acceleration and angular velocity information output by the accelerometer and gyroscope, the velocity and attitude of the UAV are solved. The obtained UAV attitude, acceleration, velocity and directly output angular velocity are inertial measurement information. Obtaining the inertial measurement information indicates that the inertial initialization is complete.

[0019] Furthermore, in step 1, the inertial measurement information is synchronized with the visual measurement information.

[0020] Furthermore, in step 2, based on the pose information and visual feature matching information obtained in step 1, the average pixel difference d between matching feature point pairs in two adjacent frames is calculated. k :

[0021]

[0022] In the formula, Let be the pixel coordinates of the matched feature points in the k-th frame image; Let M be the pixel coordinates of the matched feature points in the (k+1)th frame of the image; M is the number of matched feature points, M = 1, 2, ..., m;

[0023] The acceleration, velocity, angular velocity, and rotation angle information of the UAV between two adjacent frames are calculated based on the inertial navigation pre-integration method in step 1.

[0024] The drone's altitude information between two consecutive frames is calculated using a laser rangefinder fixed on the drone.

[0025] Therefore, based on the average pixel difference d obtained above k The drone's current state is distinguished by its acceleration, velocity, angular velocity, rotation angle information, and altitude information between two consecutive frames.

[0026] (1) If the average pixel difference d k If the drone's altitude information between two consecutive frames is less than threshold A, and the acceleration variance is less than threshold B, then the drone is considered to be in a stationary state on the ground.

[0027] (2) If the average pixel difference d k If the drone's altitude information between two consecutive frames is ≥ threshold A or the attitude angle is ≥ threshold D, then the drone is determined to be in an aerial turning / swaying state.

[0028] (3) If the average pixel difference d k If the value of the drone is less than threshold A, the acceleration variance is greater than or equal to threshold B, and the drone's altitude information between two consecutive frames is greater than or equal to threshold C, then the drone is determined to be in a straight-line motion state in the air.

[0029] (4) If the average pixel difference d k If the height information of two consecutive frames is within the threshold A and the change in height information is within the threshold E, then the drone is determined to be in a level flight state.

[0030] Among them, thresholds A, B, C, D, and E are all preset values.

[0031] Furthermore, the specific steps of step 3 are as follows:

[0032] Step 3-1, calculate the initial values ​​of acceleration bias and gyroscope bias for the UAV in a stationary state on the ground: Obtain the initial value of acceleration bias b through coarse alignment of the inertial navigation system on the UAV in a stationary state. a and the initial value of the gyroscope bias;

[0033] Step 3-2: Achieve joint inertial / visual measurement by manipulating the drone's movement, thereby correcting the initial value of the gyroscope bias.

[0034] The pose change of the UAV between images can be calculated based on the continuous image sequence acquired by the monocular camera on the UAV and the visual feature matching in step 1. The inertial navigation system calculates the pose change of the UAV between corresponding time points in the continuous image sequence through pre-integration. Assume that the camera pose calculated by vision at the time point corresponding to the k-th frame and the time point corresponding to the k+1-th frame are q respectively. k and q k+1 The attitude changes of the inertial navigation system at time corresponding to the k-th frame and time corresponding to the (k+1)-th frame are calculated through inertial navigation pre-integration. A fixed monocular camera and an inertial navigation system are used to capture n frames of images. The following cost function is constructed:

[0035]

[0036] In the formula, δb w * This represents the minimum gyroscope bias. This represents the true values ​​of the pose changes of the inertial navigation system at the corresponding time points of frame k and k+1. δb represents the gyroscope bias at the corresponding time point of the k-th frame. w This represents the gyroscope bias. The Jacobian matrix of the inertial navigation pose change relative to the gyroscope bias can be obtained after obtaining the inertial navigation pre-integration results;

[0037] When solving formula (2), the initial gyroscope bias value obtained in step 3-1 is used as the initial time. Substituting into formula (2) and solving it through Gauss-Newton iteration, we obtain the δb corresponding to the minimum error of formula (2). w , and the δb w As a final correction to the gyroscope bias;

[0038] Step 3-3, based on the initial acceleration bias value b from step 3-1 a And the corrected gyroscope bias δb in step 3-2 wThe parameters of the UAV starting from a stationary state on the ground are calculated, including the visual scale s of the monocular camera and the inertial navigation parameters. The visual and inertial solution processes are updated based on these parameters to obtain updated visual measurement information and inertial measurement information, thus completing the initialization starting from a stationary state on the ground.

[0039] Furthermore, the specific process of step 3-3 is as follows:

[0040] First, set the inertial navigation parameter to y. k ={s,R wg ,b,v 0:k}:

[0041] Among them, y k Let be the inertial navigation parameters for the k-th frame;

[0042] R wg Let R be the gravitational rotation from the standard gravity frame to the world coordinate system. wg The gravity vector g = R in the world coordinate system can be obtained. wg (0,0,G) T ;

[0043] b represents the bias, which includes the acceleration bias and the gyroscope bias. The acceleration bias uses the initial value b from step 3-1. a The gyroscope bias uses the corrected gyroscope bias δb from step 3-2. w ;

[0044] v 0:k Let v be the velocity of the UAV in the inertial frame from frame 0 to the k-th frame, where the velocity in the i-th frame is represented as v. ib v ib The conversion can be performed based on the speed of the drone in the camera system of the i-frame, i.e., v ib =R bc v c , where R bc The rotation matrix from the camera frame to the inertial frame is obtained through pre-calibration; v c The initial velocity of the drone under the camera system is obtained by dividing the displacement information of the drone acquired by the monocular camera by the time period.

[0045] Then, the equations of motion are constructed for the inertial navigation system's moving state, and the velocity residuals are obtained based on the inertial navigation system's pre-integration. Translational residual

[0046]

[0047] in, It is the transpose of the rotation matrix from the inertial frame to the world coordinate system at the corresponding moment of frame i, v j Let v be the velocity of the drone in the world coordinate system at the time corresponding to frame j. i Let Δt be the velocity of the drone in the world coordinate system at the corresponding moment of frame i. ij ΔR is the time interval between the corresponding time in frame j and the corresponding time in frame i; bibk This is the pre-integral increment of the rotation amount between the time corresponding to frame k and the time corresponding to frame i. This represents the acceleration information at the corresponding time points in frames k. p represents the acceleration bias at the corresponding time point of frame i, where Δt is the time interval between the corresponding time point of frame i and the corresponding time point of frame k; j Let p be the position of the drone in the world coordinate system at the corresponding moment in frame j. i Let Δv be the position of the UAV in the world coordinate system at the corresponding moment of frame i; ik The pre-integral increment of the velocity at time k compared to time i;

[0048] When solving formula (3), fix b and v. 0:k ,Right now Minimize the iterative solution speed residual Translational residual Obtain the visual scale s and the gravitational rotation R wg ;

[0049] Finally, based on the obtained visual scale s, the visual solution process of the monocular camera is updated to obtain the updated visual measurement information; based on the obtained inertial navigation parameters y... k ={s,R wg ,b,v 0:k The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the gyroscope bias δb modified using step 3-2 w Then, inertial navigation pre-integration calculation is performed to obtain updated inertial measurement information. Thus, the joint initialization process of vision and inertial inertial inertial states under stationary ground conditions is completed.

[0050] Furthermore, in step 4, during the drone's level flight phase:

[0051] During the level flight phase, visual triangulation is used to determine the depth of the 3D map points corresponding to the first 10 frames. For the i-th frame, the average depth d is calculated. i The height information h measured by the laser rangefinder at the corresponding time is obtained by aligning with the timestamp. i ;

[0052] Height information h measured by laser rangefinder iThe depth value obtained by a monocular camera in the laser ranging direction differs from the depth value obtained by a visual scale factor s. i The visual scale factor s of the i-th frame i for: Then, the visual scale factor s was calculated for the first 10 frames. i The average is calculated to obtain the optimized result of visual scale s in step 3, i.e., the average visual scale.

[0053] Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the gyroscope bias δb modified using step 3-2 w The inertial navigation pre-integration calculation is performed to obtain the updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the level flight phase is completed.

[0054] Furthermore, in step 4, during the drone's turning / swaying phase in the air:

[0055] During the turning / swaying phase in the air, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera is fixed, and the amount of gravitational rotation R is... wg Acceleration bias Gyroscope bias δb w and speed v 0:k The solution is obtained through iterative optimization, using the optimized average visual scale during the level flight phase. Instead of the visual scale s, the final iteratively optimized gravity rotation amount R is obtained. wg ', Acceleration bias b a 'Gyroscope bias δb' w 'and speed v 0:k ';

[0056] Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... Bias b' includes the iteratively optimized acceleration bias b a 'and gyroscope bias δb w Then, perform inertial navigation pre-integration calculation to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the turning / swaying phase in the air is completed.

[0057] Furthermore, during the straight-line movement phase of the drone in the air:

[0058] During the linear motion phase in the air, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera and the gyroscope bias are fixed, and the gravitational rotation R is... wg Acceleration bias and speed v 0:k The solution is obtained through iterative optimization, using the optimized average visual scale during the level flight phase. Instead of the visual scale s, the gyroscope bias uses the gyroscope bias δb during the turning / shaking phase in the air. w Alternatively, the modified gyroscope bias δb from step 3-2 can be used. w Finally, the gravity rotation amount R after iterative optimization is obtained. wg "Acceleration bias b" a "and speed v 0:k ”;

[0059] Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... "Bias b" includes the iteratively optimized acceleration bias b a "and the gyroscope bias δb during the turning / shaking phase in the air" w 'or gyroscope bias δb' w Then, perform inertial navigation pre-integration calculation to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the straight-line motion phase in the air is completed.

[0060] In the phase where the UAV transitions directly from level flight to linear motion, the gyroscope bias uses the corrected gyroscope bias δb obtained in step 3-2. w When the UAV enters a level flight phase, followed by a turning / swaying phase and then a straight-line motion phase, the gyroscope bias is the same as the bias δb used during the turning / swaying phase. w '.

[0061] By applying the above technical solution, the present invention has the following beneficial effects:

[0062] (1) This invention proposes an adaptive inertial / visual / altitude initialization method for UAVs, which combines the advantages and requirements of different sensors (including visual sensors, inertial sensors and laser rangefinders) for various scenarios such as static and dynamic UAV startup. The motion state is determined by the joint action of the three sensors. Then, two types of inertial / visual / altitude initialization methods are proposed for static startup scenarios and dynamic startup scenarios of UAVs, respectively, based on the motion characteristics in the scenario. The sensor parameters are solved in a refined manner, including inertial acceleration bias, gyroscope bias, visual scale, gravity rotation, etc., which improves the accuracy and robustness of the application scenario initialization method.

[0063] (2) Based on the motion characteristics of sensors, this invention proposes a method for estimating the motion state of UAVs by comprehensively utilizing inertial, visual, and altitude observation information, thereby facilitating the differentiation of motion excitation situations.

[0064] (3) This invention proposes a joint initialization method for inertial / visual / altitude for dynamic and static initialization scenarios, based on the differences in initialization scenarios. On the one hand, it adds parameter solving strategies for different scenarios to the traditional initialization method, and on the other hand, it introduces altitude channel information observation to obtain more accurate joint initialization results.

[0065] In summary, the method proposed in this invention comprehensively utilizes the flight state of the UAV to simulate the profile process from takeoff from the ground to flight. It can effectively handle the joint initialization of sensors in the stationary state of the UAV and the joint initialization of sensors in the moving state after in-flight restart or repositioning. The whole process has strong adaptability and effectively improves the accuracy of parameter solving after initialization. Attached Figure Description

[0066] The accompanying drawings, which form part of this specification, are provided to further illustrate embodiments of the invention and, together with the textual description, explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0067] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0068] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0070] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0071] Example 1:

[0072] This embodiment provides an adaptive inertial / visual / altitude initialization method for UAVs. See Appendix. Figure 1 The steps of this method are as follows:

[0073] Step 1: Initialize the vision and inertial measurements separately to obtain visual measurement information and inertial measurement information respectively; the visual measurement information is obtained by solving the pose through visual feature matching and confirming the feature point depth through triangulation; the inertial measurement information is obtained through inertial navigation pre-integration.

[0074] Step 2: Integrate visual feature matching information, inertial measurement information, and altitude information to determine the current state of the drone;

[0075] Step 3: Use the inertial / vision joint initialization method to perform general initialization of the parameters of the UAV starting from a stationary state on the ground, and obtain updated visual measurement information and inertial measurement information;

[0076] Step 4: Based on the parameters obtained in Step 3, the UAV parameters are initialized from the start of motion using a joint inertial / visual / altitude initialization method for dynamic scenarios, resulting in updated visual and inertial measurement information.

[0077] Example 2:

[0078] This embodiment, based on embodiment 1, provides a detailed process for each step:

[0079] In this embodiment, the visual measurement of the UAV is achieved through the visual sensor of a monocular camera, the inertial measurement of the UAV is achieved through an inertial sensor (including an accelerometer and a gyroscope), and the altitude measurement is achieved through a laser rangefinder; the monocular camera, inertial sensor, and laser rangefinder are all fixed on the UAV, and their positions are fixed.

[0080] In step 1, the process of initializing vision and obtaining visual measurement results is as follows:

[0081] A monocular camera cannot obtain the true depth information of a scene. Pure vision achieves pose calculation and visual initialization by matching and locating adjacent preceding and following image frames or key image frames. The visual initialization process is as follows:

[0082] First, select image frames: If key image frames (hereinafter referred to as key frames) are used to complete pose calculation, key frames need to be selected in the image sequence. The following conditions should be met: there are enough matching feature point pairs between two key frames (a feature point pair includes two mutually matching feature points, and the two feature points are in two key frames respectively); the two key frames need to reach a certain viewpoint difference, that is, the displacement of feature points in the two frames needs to reach a certain threshold. Therefore, key frames are generally not adjacent in the image sequence; if adjacent preceding and following image frames (hereinafter referred to as adjacent frames) are used for pose calculation, the pose result of each frame image can be obtained; this embodiment uses adjacent frames for visual initialization, and the initialization process of key frames is consistent with that of adjacent frames in this embodiment.

[0083] Then, after acquiring two frames of images to be matched, SURF feature matching can be used to obtain matching feature point pairs between adjacent frames (i.e., visual feature matching information is obtained). Then, epipolar geometry is used to calculate the pose of the monocular camera, which includes the rotation matrix R and translation vector t of the monocular camera. Based on the obtained feature point pairs, triangulation techniques can be used to obtain the depth of 3D map points on the selected two frames of images.

[0084] Finally, pose optimization is performed using bundle adjustment (BA) technology. At this point, the optimized pose of the monocular camera and the initial map points can be obtained. The pose of the monocular camera and the initial map points are visual measurement information. Obtaining the visual measurement information indicates that visual initialization is complete. Among them, the rotation in the pose of the monocular camera does not involve the visual scale, while the translation and map points still need to determine the visual scale.

[0085] In step 1, the process of initializing the inertia and obtaining inertial measurement information is as follows:

[0086] The inertial sensor (hereinafter referred to as the inertial navigation system) is initialized using the inertial navigation pre-integration method proposed by Forster et al.: by integrating the acceleration and angular velocity information output by the accelerometer and gyroscope, the velocity and pose (including position and attitude) of the inertial navigation system are solved. The attitude can be represented by rotation matrices, rotation quaternions, etc. The obtained pose, acceleration, velocity, and directly output angular velocity of the inertial navigation system constitute the inertial measurement information. Obtaining the inertial measurement information indicates that the inertial initialization is complete.

[0087] Since the monocular camera, inertial sensor, and laser rangefinder are all fixed on the UAV, the pose of the monocular camera and the pose of the inertial sensor refer to the pose of the UAV, and the acceleration, velocity, angular velocity, and rotation angle of the inertial sensor refer to the acceleration, velocity, angular velocity, and rotation angle of the UAV.

[0088] In this step, the inertial navigation data (i.e., inertial measurement information) measured by the inertial sensor is synchronized with the visual data (i.e., visual measurement information) measured by the monocular camera.

[0089] In step 2, based on the pose information and visual feature matching information (i.e., feature point pair information between two adjacent frames) obtained in step 1, the average pixel difference d between the matching feature point pairs in the two adjacent frames is calculated. k :

[0090]

[0091] In the formula, Let be the pixel coordinates of the matched feature points in the k-th frame image; Let M be the pixel coordinates of the matched feature points in the (k+1)th frame of the image; M is the number of matched feature points, M = 1, 2, ..., m.

[0092] The acceleration, velocity, angular velocity, and rotation angle information of the UAV between two adjacent frames are calculated based on the inertial navigation pre-integration method in step 1.

[0093] The drone's altitude information between two consecutive frames is calculated using a laser rangefinder fixed on the drone.

[0094] Therefore, based on the average pixel difference d obtained above k The drone's current state is distinguished by its acceleration, velocity, angular velocity, rotation angle information, and altitude information between two consecutive frames.

[0095] (5) If the average pixel difference d k If the drone's altitude information between two consecutive frames is less than threshold A, and the acceleration variance is less than threshold B, then the drone is considered to be in a stationary state on the ground.

[0096] (6) If the average pixel difference d k If the drone's altitude information between two consecutive frames is ≥ threshold A or the attitude angle is ≥ threshold D, then the drone is determined to be in an aerial turning / swaying state.

[0097] (7) If the average pixel difference d k If the value of the drone is less than threshold A, the acceleration variance is greater than or equal to threshold B, and the drone's altitude information between two consecutive frames is greater than or equal to threshold C, then the drone is determined to be in a straight-line motion state in the air.

[0098] (8) If the average pixel difference d k If the height information of two consecutive frames is within the threshold A and the change in height information is within the threshold E, then the drone is determined to be in a level flight state.

[0099] In this embodiment, (1) is determined to be the stationary state of the drone, and (2), (3) and (4) are determined to be the motion state of the drone. Other motion states of the drone are not considered in this embodiment. Among them, threshold A, threshold B, threshold C, threshold D and threshold E are all preset values.

[0100] The specific steps for step 3 are as follows:

[0101] Step 3-1: Calculate the initial values ​​of the inertial acceleration bias and gyroscope bias for the UAV when it is stationary on the ground.

[0102] In the average pixel difference d kIn a stationary ground state where the ground velocity is less than threshold A, the acceleration variance is less than threshold B, and the drone's altitude information between two consecutive frames is less than threshold C, the inertial sensor detects the Earth's rotation angular velocity and the triaxial accelerometer information as the component of gravitational acceleration. Through coarse alignment of the inertial sensor in a stationary state, the initial acceleration bias value b can be obtained. a and the initial value of the gyroscope bias;

[0103] Step 3-2: Achieve joint inertial / visual measurement by slightly moving (i.e., shaking) the drone, thereby correcting the initial value of the gyroscope bias.

[0104] Since the visual data does not drift when the drone moves slightly, the pose change of the drone calculated by vision is considered to be more accurate. The initial gyroscope bias value obtained in step 3-1 is corrected by aligning the pose calculated by vision with the pose obtained by inertial navigation pre-integration (i.e., joint inertial / visual measurement). The specific process is as follows:

[0105] Based on the continuous image sequence acquired by the monocular camera and the visual feature matching in step 1, the pose change of the camera between images (i.e., the pose change of the UAV) can be calculated. The inertial sensor calculates the pose change of the inertial navigation system (i.e., the pose change of the UAV) between corresponding moments in the continuous image sequence through pre-integration. Assume that the camera pose calculated by vision at the corresponding moment of the k-th frame and the corresponding moment of the k+1-th frame are q respectively. k and q k+1 The attitude changes of the inertial navigation system at time corresponding to the k-th frame and time corresponding to the (k+1)-th frame are calculated through inertial navigation pre-integration. A fixed monocular camera and an inertial navigation system are used to capture n frames of images. The following cost function is constructed:

[0106]

[0107] In the formula, δb w * This represents the minimum gyroscope bias. This represents the true values ​​of the pose changes of the inertial navigation system at the corresponding time points of frame k and k+1. δb represents the gyroscope bias at the corresponding time point of the k-th frame. w This represents the gyroscope bias. The Jacobian matrix of the inertial navigation pose change relative to the gyroscope bias can be obtained after obtaining the inertial navigation pre-integration results;

[0108] When solving formula (2), the initial gyroscope bias value obtained in step 3-1 is used as the initial time. Substituting into formula (2) and solving it through Gauss-Newton iteration, we obtain the δb corresponding to the minimum error of formula (2). w, and the δb w As a final correction to the gyroscope bias;

[0109] Step 3-3, based on the initial acceleration bias value b from step 3-1 a And the corrected gyroscope bias δb in step 3-2 w The parameters of the UAV starting from a stationary state on the ground are calculated. These parameters include the visual scale *s* of the monocular camera and the inertial navigation parameters. Based on these parameters, the visual and inertial measurement processes are updated to obtain updated visual and inertial measurement information, thus completing the initialization from a stationary state on the ground. The specific process is as follows:

[0110] When using the monocular camera for visual navigation, the position and movement information of the UAV in the camera frame is obtained. However, since the monocular camera lacks the visual scale s of real space, it is impossible to obtain the proportional relationship between the position of the UAV in the camera frame and the actual position of the UAV calculated in the inertial frame. Therefore, it is necessary to calculate the visual scale s first. The calculation process is as follows:

[0111] First, set the inertial navigation parameter to y. k ={s,R wg ,b,v 0:k}:

[0112] Among them, y k Let be the inertial navigation parameters for the k-th frame;

[0113] R wg Let R be the gravitational rotation from the standard gravity frame to the world coordinate system. wg The gravity vector g = R in the world coordinate system can be obtained. wg (0,0,G) T ;

[0114] b represents the bias, which includes the acceleration bias and the gyroscope bias. The acceleration bias uses the initial value b from step 3-1. a The gyroscope bias uses the corrected gyroscope bias δb from step 3-2. w ;

[0115] v 0:k Let v be the velocity of the UAV in the inertial frame from frame 0 to the k-th frame, where the velocity in the i-th frame is represented as v. ib v ib The conversion can be performed based on the speed of the drone in the camera system of the i-frame, i.e., v ib =R bc v c , where R bc The rotation matrix from the camera frame to the inertial frame is usually obtained through pre-calibration; vc The initial velocity of the drone under the camera system is obtained by dividing the displacement information of the drone acquired by the monocular camera by the time period.

[0116] Then, the equations of motion are constructed for the inertial navigation system's moving state, and the velocity residuals are obtained based on the inertial navigation system's pre-integration. Translational residual

[0117]

[0118] in, It is the transpose of the rotation matrix from the inertial frame to the world coordinate system at the corresponding moment of frame i, v j Let v be the velocity of the drone in the world coordinate system at the time corresponding to frame j. i Let Δt be the velocity of the drone in the world coordinate system at the corresponding moment of frame i. ij ΔR is the time interval between the corresponding time in frame j and the corresponding time in frame i; bibk This is the pre-integral increment of the rotation amount between the time corresponding to frame k and the time corresponding to frame i. This represents the acceleration information at the corresponding time points in frames k. p represents the acceleration bias at the corresponding time point of frame i, where Δt is the time interval between the corresponding time point of frame i and the corresponding time point of frame k; j Let p be the position of the drone in the world coordinate system at the corresponding moment in frame j. i Let Δv be the position of the UAV in the world coordinate system at the corresponding moment of frame i; ik The pre-integral increment of the velocity at time k compared to time i is given.

[0119] When solving formula (3), fix b and v. 0:k ,Right now Minimize the iterative solution speed residual Translational residual Obtain the visual scale s and the gravitational rotation R wg ;

[0120] Finally, based on the obtained visual scale s, the visual solution process of the monocular camera is updated (this process is existing technology), and the updated visual measurement information is obtained; based on the obtained inertial navigation parameters y k ={s,R wg ,b,v 0:k The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the gyroscope bias δb modified using step 3-2 w Then, perform subsequent inertial navigation pre-integration calculations to obtain updated inertial measurement information. At this point, the joint initialization process of vision and inertial measurement in the ground stationary state is completed.

[0121] In step 4, when the UAV is in motion in the air and under motion excitation, the observability of the inertial / visual sensor can be increased, and the accuracy of parameter solving can be improved. At the same time, the altitude information of the UAV should be integrated into the initialization process of the inertial / visual sensor to further improve the accuracy of parameter solving.

[0122] The movement of a drone in the air includes the level flight phase, the straight-line motion phase, and the turning / swaying phase. Since the level flight phase is the basis for the straight-line motion phase and the turning / swaying phase, it is necessary to calculate the parameters of the drone when it is in the level flight phase first.

[0123] (1) During the level flight phase, visual triangulation is used to solve the depth of the three-dimensional map points corresponding to the first 10 frames. For the i-th frame image, the average depth d is calculated. i The height information h measured by the laser rangefinder at the corresponding time is obtained by aligning with the timestamp. i ;

[0124] Height information h measured by laser rangefinder i The depth value obtained by a monocular camera in the laser ranging direction differs from the depth value obtained by a visual scale factor s. i Since laser ranging has a pointing direction, the 3D map points need to be selected from the area of ​​the pointing direction. Here, a matrix area of ​​15*15 pixels is defined in the center area of ​​the monocular camera, and only the 3D map points corresponding to these pixels are selected for calculation.

[0125] Then the visual scale factor s of the i-th frame i for: Then, the visual scale factor s was calculated for the first 10 frames. i The average is calculated to obtain the optimized result of visual scale s in step 3, i.e., the average visual scale.

[0126] Finally, based on the obtained average visual scale The visual solution process for the updated monocular camera (this process is existing technology) is used to obtain updated visual measurement information; based on the obtained inertial navigation parameters... The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the gyroscope bias δb modified using step 3-2 w Then, the inertial navigation pre-integration calculation is performed to obtain the updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the level flight phase is completed.

[0127] (2) During the turning / swaying phase in the air, the observability of errors in the inertial sensor and the visual sensor of the monocular camera is improved due to the positive excitation generated by the violent movement of the UAV; therefore, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera is fixed, and the gravity rotation R is... wg Acceleration bias Gyroscope bias δb w and speed v 0:k The solution is obtained through iterative optimization, using the optimized average visual scale during the level flight phase. Instead of the visual scale s, the final iteratively optimized gravity rotation amount R is obtained. wg ', Acceleration bias b a 'Gyroscope bias δb' w 'and speed v 0:k ';

[0128] Finally, based on the obtained average visual scale The visual solution process for the updated monocular camera (this process is existing technology) is used to obtain updated visual measurement information; based on the obtained inertial navigation parameters... Bias b' includes the iteratively optimized acceleration bias b a 'and gyroscope bias δb w Then, perform subsequent inertial navigation pre-integration calculations to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the turning / swaying phase in the air is completed.

[0129] (3) During the linear motion phase in the air, the accelerometer of the inertial sensor is excited, while the gyroscope does not receive significant motion excitation; therefore, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera and the gyroscope bias are fixed, and the gravitational rotation R is... wg Acceleration bias and speed v 0:k The solution is obtained through iterative optimization, using the optimized average visual scale during the level flight phase. Instead of the visual scale s, the gyroscope bias can be the gyroscope bias δb during the turning / shaking phase in the air. w Alternatively, the modified gyroscope bias δb from step 3-2 can be used. w Finally, the gravity rotation amount R after iterative optimization is obtained. wg "Acceleration bias b" a "and speed v 0:k ”;

[0130] Finally, based on the obtained average visual scale The visual solution process for the updated monocular camera (this process is existing technology) is used to obtain updated visual measurement information; based on the obtained inertial navigation parameters... "Bias b" includes the iteratively optimized acceleration bias b a "and the gyroscope bias δb during the turning / shaking phase in the air" w 'or gyroscope bias δb' w Then, perform subsequent inertial navigation pre-integration calculations to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the straight-line motion phase in the air is completed.

[0131] In the phase where the UAV transitions directly from level flight to linear motion, the gyroscope bias uses the corrected gyroscope bias δb obtained in step 3-2. w When the UAV enters a level flight phase, followed by a turning / swaying phase and then a straight-line motion phase, the gyroscope bias is the same as the bias δb used during the turning / swaying phase. w '.

[0132] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein will be interpreted accordingly.

[0133] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, the above terms have no special meaning and therefore should not be construed as limiting the scope of protection of this invention.

[0134] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive inertial / vision / altitude initialization method for unmanned aerial vehicles (UAVs), characterized in that, The steps of this method are as follows: Step 1: Initialize the vision and inertial measurements separately to obtain visual measurement information and inertial measurement information respectively; the visual measurement information is obtained by solving the pose through visual feature matching and confirming the feature point depth through triangulation; the inertial measurement information is obtained through inertial navigation pre-integration. Step 2: Integrate visual feature matching information, inertial measurement information, and altitude information to determine the current state of the drone; Step 3: Initialize the parameters of the UAV starting from a stationary state on the ground using the inertial / visual joint initialization method to obtain updated visual measurement information and inertial measurement information; Step 4: Based on the parameters obtained in Step 3, the UAV parameters are initialized from the start of motion using a joint inertial / visual / altitude initialization method for dynamic scenarios, resulting in updated visual and inertial measurement information.

2. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 1, characterized in that, In step 1, the process of initializing vision and obtaining visual measurement results is as follows: First, select image frames: use adjacent preceding and following image frames for pose calculation; Then, after acquiring two frames of images to be matched, SURF feature matching can be used to obtain matching feature point pairs between adjacent frames, that is, visual feature matching information is obtained. Then, epipolar geometry is used to calculate the pose of the UAV. Furthermore, based on the obtained feature point pairs, triangulation technology can be used to obtain the depth of three-dimensional map points on the selected two frames of images. Finally, pose optimization is performed using cluster adjustment technology. At this point, the optimized UAV pose and initial map points can be obtained. The UAV pose and initial map points are visual measurement information. Obtaining the visual measurement information indicates that visual initialization is complete.

3. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 1, characterized in that, In step 1, the process of initializing the inertia and obtaining inertial measurement information is as follows: By integrating the acceleration and angular velocity information output by the accelerometer and gyroscope, the velocity and attitude of the UAV are solved. The obtained UAV attitude, acceleration, velocity and directly output angular velocity are inertial measurement information. Obtaining the inertial measurement information indicates that the inertial initialization is complete.

4. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 1, characterized in that, In step 1, the inertial measurement information is synchronized with the visual measurement information.

5. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 2, characterized in that, In step 2, based on the pose information and visual feature matching information obtained in step 1, the average pixel difference d between matching feature point pairs in two adjacent frames is calculated. k : In the formula, Let be the pixel coordinates of the matched feature points in the k-th frame image; Let M be the pixel coordinates of the matched feature points in the (k+1)th frame of the image; M is the number of matched feature points, M = 1, 2, ..., m; The acceleration, velocity, angular velocity, and rotation angle information of the UAV between two adjacent frames are calculated based on the inertial navigation pre-integration method in step 1. The drone's altitude information between two consecutive frames is calculated using a laser rangefinder fixed on the drone. Therefore, based on the average pixel difference d obtained above k The drone's current state is distinguished by its acceleration, velocity, angular velocity, rotation angle information, and altitude information between two consecutive frames. (1) If the average pixel difference d k If the drone's altitude information between two consecutive frames is less than threshold A, and the acceleration variance is less than threshold B, then the drone is considered to be in a stationary state on the ground. (2) If the average pixel difference d k If the drone's altitude information between two consecutive frames is ≥ threshold A or the attitude angle is ≥ threshold D, then the drone is determined to be in an aerial turning / swaying state. (3) If the average pixel difference d k If the value of the drone is less than threshold A, the acceleration variance is greater than or equal to threshold B, and the drone's altitude information between two consecutive frames is greater than or equal to threshold C, then the drone is determined to be in a straight-line motion state in the air. (4) If the average pixel difference d k If the height information of two consecutive frames is within the threshold A and the change in height information is within the threshold E, then the drone is determined to be in a level flight state. Among them, thresholds A, B, C, D, and E are all preset values.

6. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 1, characterized in that, The specific steps for step 3 are as follows: Step 3-1, calculate the initial values ​​of acceleration bias and gyroscope bias for the UAV in a stationary state on the ground: Obtain the initial value of acceleration bias b through coarse alignment of the inertial navigation system on the UAV in a stationary state. a and the initial value of the gyroscope bias; Step 3-2: Achieve joint inertial / visual measurement by manipulating the drone's movement, thereby correcting the initial value of the gyroscope bias. The pose change of the UAV between images can be calculated based on the continuous image sequence acquired by the monocular camera on the UAV and the visual feature matching in step 1. The inertial navigation system calculates the pose change of the UAV between corresponding time points in the continuous image sequence through pre-integration. Assume that the camera pose calculated by vision at the time point corresponding to the k-th frame and the time point corresponding to the k+1-th frame are q respectively. k and q k+1 The attitude changes of the inertial navigation system at time corresponding to the k-th frame and time corresponding to the (k+1)-th frame are calculated through inertial navigation pre-integration. A fixed monocular camera and an inertial navigation system are used to capture n frames of images. The following cost function is constructed: In the formula, This represents the minimum gyroscope bias. This represents the true values ​​of the pose changes of the inertial navigation system at the corresponding time points of frame k and k+1. This represents the gyroscope bias at the k-th frame, in dB. w This represents the gyroscope bias. The Jacobian matrix of the inertial navigation pose change relative to the gyroscope bias can be obtained after obtaining the inertial navigation pre-integration results; When solving formula (2), the initial gyroscope bias value obtained in step 3-1 is used as the initial time. Substituting into formula (2) and solving it using Gauss-Newton iteration, we obtain the db corresponding to the minimum error of formula (2). w , the db w As a final correction to the gyroscope bias; Step 3-3, based on the initial acceleration bias value b from step 3-1 a And the corrected gyroscope bias dB in step 3-2 w Calculate the parameters of the UAV starting from a stationary state on the ground, including: the visual scale s of the monocular camera and the inertial navigation parameters; The visual and inertial solution processes are updated based on this parameter to obtain updated visual and inertial measurement information, thus completing the initialization from the ground stationary state.

7. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 6, characterized in that, The specific process of step 3-3 is as follows: First, set the inertial navigation parameter to y. k ={s,R wg ,b,v 0:k }: Among them, y k Let be the inertial navigation parameters for the k-th frame; R wg Let R be the gravitational rotation from the standard gravity frame to the world coordinate system. wg The gravity vector g = R in the world coordinate system can be obtained. wg (0,0,G) T ; b represents the bias, which includes the acceleration bias and the gyroscope bias. The acceleration bias uses the initial value b from step 3-1. a The gyroscope bias uses the corrected gyroscope bias dB obtained in step 3-2. w ; v 0:k Let v be the velocity of the UAV in the inertial frame from frame 0 to the k-th frame, where the velocity in the i-th frame is represented as v. ib v ib The conversion can be performed based on the speed of the drone in the camera system of the i-frame, i.e., v ib =R bc v c , where R bc The rotation matrix from the camera frame to the inertial frame is obtained through pre-calibration; v c The initial velocity of the drone under the camera system is obtained by dividing the displacement information of the drone acquired by the monocular camera by the time period. Then, the equations of motion are constructed for the inertial navigation system's moving state, and the velocity residuals are obtained based on the inertial navigation system's pre-integration. Translational residual in, It is the transpose of the rotation matrix from the inertial frame to the world coordinate system at the corresponding moment of frame i, v j Let v be the velocity of the drone in the world coordinate system at the time corresponding to frame j. i Let Dt be the velocity of the drone in the world coordinate system at the corresponding moment of frame i. ij DR is the time interval between the corresponding time of frame j and the corresponding time of frame i. bibk This is the pre-integral increment of the rotation amount between the time corresponding to frame k and the time corresponding to frame i. This represents the acceleration information at the corresponding time points in frames k. p represents the acceleration bias at the corresponding time point of frame i, where Dt is the time interval between the corresponding time point of frame i and the corresponding time point of frame k; j Let p be the position of the drone in the world coordinate system at the corresponding moment in frame j. i Dv represents the position of the drone in the world coordinate system at the corresponding moment in frame i; ik The pre-integral increment of the velocity at time k compared to time i; When solving formula (3), fix b and v. 0:k ,Right now Minimize the iterative solution speed residual Translational residual Obtain the visual scale s and the gravitational rotation R wg ; Finally, based on the obtained visual scale s, the visual solution process of the monocular camera is updated to obtain the updated visual measurement information; based on the obtained inertial navigation parameters y... k ={s,R wg ,b,v 0:k The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the corrected gyroscope bias dB using step 3-2 w Then, inertial navigation pre-integration calculation is performed to obtain updated inertial measurement information. Thus, the joint initialization process of vision and inertial inertial inertial states under stationary ground conditions is completed.

8. The adaptive inertial / vision / altitude initialization method for UAVs as described in claim 7, characterized in that, In step 4, during the level flight phase of the drone: During the level flight phase, visual triangulation is used to determine the depth of the 3D map points corresponding to the first 10 frames. For the i-th frame, the average depth d is calculated. i The height information h measured by the laser rangefinder at the corresponding time is obtained by aligning with the timestamp. i ; Height information h measured by laser rangefinder i The depth value obtained by a monocular camera in the laser ranging direction differs from the depth value obtained by a visual scale factor s. i The visual scale factor s of the i-th frame i for: Then, the visual scale factor s was calculated for the first 10 frames. i The average is calculated to obtain the optimized result of visual scale s in step 3, i.e., the average visual scale. Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... The bias b includes the initial acceleration bias value b obtained in step 3-1. a and the corrected gyroscope bias dB using step 3-2 w The inertial navigation pre-integration calculation is performed to obtain the updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the level flight phase is completed.

9. The adaptive inertial / vision / altitude initialization method for unmanned aerial vehicles as described in claim 8, characterized in that, In step 4, during the drone's turning / swaying phase in the air: During the turning / swaying phase in the air, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera is fixed, and the amount of gravitational rotation R is... wg Acceleration bias Gyroscope bias dB w and speed v 0:k The solution is obtained through iterative optimization, using the optimized average visual scale during the level flight phase. Instead of the visual scale s, the final iteratively optimized gravity rotation amount R is obtained. wg ', Acceleration bias b a 'Gyroscope bias dB' w 'and speed v 0:k '; Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... Bias b' includes the iteratively optimized acceleration bias b a 'and gyroscope bias dB w Then, perform inertial navigation pre-integration calculation to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the turning / swaying phase in the air is completed.

10. The adaptive inertial / vision / altitude initialization method for unmanned aerial vehicles as described in claim 9, characterized in that, During the straight-line motion phase of the drone in the air: During the linear motion phase in the air, for formulas (2) and (3) in step 3, only the visual scale s of the monocular camera and the gyroscope bias are fixed, and the gravitational rotation R is... wg Acceleration bias and speed v 0:k The solution is obtained through iterative optimization. During the solution process, the optimized average visual scale s during the level flight phase is used. i Instead of the visual scale s, the gyroscope bias uses the gyroscope bias db during the turning / shaking phase in the air. w Alternatively, the corrected gyroscope bias (db) from step 3-2 can be used. w Finally, the gravity rotation amount R after iterative optimization is obtained. wg "Acceleration bias b" a "and speed v 0:k ”; Finally, based on the obtained average visual scale The visual solution process for the monocular camera is updated to obtain updated visual measurement information; based on the obtained inertial navigation parameters... "Bias b" includes the iteratively optimized acceleration bias b a "and gyroscope bias dB during mid-air turning / shaking phases" w 'or gyroscope bias dB' w Then, perform inertial navigation pre-integration calculation to obtain updated inertial measurement information. At this point, the joint initialization process of vision, inertial and altitude during the straight-line motion phase in the air is completed. Specifically, during the direct linear motion phase after the UAV's level flight phase, the gyroscope bias uses the corrected gyroscope bias (db) from step 3-2. w When the UAV enters a level flight phase, followed by a turning / swaying phase and then a straight-line motion phase, the gyroscope bias should be the same as the bias used during the turning / swaying phase (db). w '.

Images