Urban scene high-precision three-dimensional reconstruction-oriented unmanned aerial vehicle flight path planning method

Abstract

The invention relates to an unmanned aerial vehicle flight path planning method for urban scene high-precision three-dimensional reconstruction. The method comprises the following steps: generating an initial viewpoint based on a standard method, abstracting a traversal task of the initial viewpoint into a traveling salesman problem model, and then generating a minimum distance circulation path by adopting a heuristic strategy of the traveling salesman problem; updating a viewpoint position in the minimum distance circulation path according to a unit length contribution criterion to obtain an updated viewpoint; calculating the movable radius of the viewpoint by using the contribution degree of the updated viewpoint to scene reconstruction, and constructing a spatial spherical flight corridor based on the movable radius; and under the constraint of the space spherical flight corridor, generating a time optimal flight path by solving an optimal control problem considering the dynamic model of the unmanned aerial vehicle and the security constraint. By adopting the method, the flight path planning efficiency and accuracy of the unmanned aerial vehicle can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of UAV trajectory planning, and in particular to a UAV flight trajectory planning method for high-precision three-dimensional reconstruction of urban scenes. Background Art

[0002] High-precision three-dimensional reconstruction of urban scenes has important application value in urban planning, cultural heritage protection, infrastructure monitoring, and digital twin construction. UAV-based image acquisition technology has become the mainstream data acquisition solution for large-scale three-dimensional reconstruction of urban scenes due to its flexibility, efficiency, and cost-effectiveness. To meet the needs of high-fidelity reconstruction of complex scenes, current mainstream methods generally adopt a two-stage "exploration-exploitation" strategy: first, a rough scene proxy is constructed using sparse images or satellite data from the initial flight, and then additional dense viewpoint sampling is used to enhance the reconstruction details. However, dense sampling relies on a large number of viewpoints, which significantly increases flight time. Limited by the limited battery capacity of drones, how to achieve efficient scene coverage to maximize geometric fidelity has become a core challenge in drone three-dimensional reconstruction trajectory planning.

[0003] In the field of UAV trajectory planning, while existing research has made progress in dynamic modeling and operational efficiency, it still faces challenges in mission adaptation. Classical methods (such as minimizing jerk) primarily improve maneuvering efficiency through polynomial optimization and focus on optimizing flight metrics. Sampling-based planners, while ensuring probabilistic completeness, face scalability bottlenecks in high-dimensional state spaces. Dynamic optimization methods, while capable of generating time-optimal trajectories, lack coordination with photogrammetric targets. Existing methods for 3D reconstruction based on UAV "exploration-exploitation" primarily follow two models. The viewpoint-first model first determines the optimal viewpoint based on geometric heuristics and then generates a path using a traveling salesman strategy. While this model maximizes reconstruction quality for each viewpoint, its rigid adherence to the pre-set viewpoint can lead to lengthy and unsmooth trajectories. Another model employs joint viewpoint-path optimization, such as the "next best viewpoint" planning strategy, which dynamically determines the optimal viewpoint location and flight path based on incremental information. However, this approach is prone to local optima in large-scale urban environments and is computationally complex. Improved methods that update trajectories based on scene agents ignore differences in viewpoint contributions and dynamic constraints, resulting in suboptimal photogrammetric coverage or infeasible trajectories. This reveals the core contradiction of the UAV reconstruction task - strict viewpoint constraints sacrifice flight efficiency, and unconstrained trajectory optimization reduces reconstruction quality. There is an urgent need to establish an optimization framework that coordinates viewpoint flexibility and dynamic feasibility. Summary of the Invention

[0004] Based on this, it is necessary to provide a UAV flight trajectory planning method for high-precision three-dimensional reconstruction of urban scenes, which can improve the efficiency and accuracy of UAV flight trajectory planning and address the above technical problems.

[0005] A method for planning unmanned aerial vehicle (UAV) flight trajectories for high-precision three-dimensional reconstruction of urban scenes, comprising:

[0006] Generate the initial viewpoint based on the standard method, abstract the traversal task of the initial viewpoint into the traveling salesman problem model, and then use the heuristic strategy of the traveling salesman problem to generate the minimum distance cycle path;

[0007] The viewpoint position in the minimum distance cycle path is updated according to the unit length contribution criterion to obtain the updated viewpoint; the movable radius of the viewpoint is calculated using the contribution of the updated viewpoint to scene reconstruction, and a spatial spherical flight corridor is constructed based on the movable radius;

[0008] Under the constraint of a spherical flight corridor in space, the time-optimal flight trajectory is generated by solving the optimal control problem that takes into account the UAV dynamics model and safety constraints.

[0009] The above-mentioned drone flight trajectory planning method for high-precision three-dimensional reconstruction of urban scenes, this application drives viewpoint optimization through the unit length contribution criterion, and generates dynamically feasible trajectories based on spherical flight corridors, significantly improving the efficiency of acquiring geometric information of the target scene, achieving a global optimal balance between reconstruction quality and flight efficiency, and can significantly shorten the drone track length and flight time while ensuring high-quality three-dimensional reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 Schematic diagram of a flow chart of a method for planning UAV flight trajectories for high-precision 3D reconstruction of urban scenes in one embodiment;

[0011] Figure 2 A diagram showing a single / dual viewpoint update case in an embodiment; Figure 2 (a) is a single viewpoint update case diagram, Figure 2 (b) Update case diagram for dual viewpoints;

[0012] Figure 3 A mapping relationship diagram of reconstruction contribution and viewpoint proximity distance in one embodiment; Figure 3 (a) is the histogram and distribution diagram of the reconstruction contribution, Figure 3 (b) is the histogram and distribution diagram of the viewpoint adjacent distance, Figure 3 (b) is the mapping relationship diagram between the two;

[0013] Figure 4 A schematic diagram of a spherical flight corridor in another embodiment;

[0014] Figure 5 A schematic diagram of a trajectory crossing a spherical flight corridor in one embodiment;

[0015] Figure 6A schematic diagram of viewpoint adjustment in one embodiment;

[0016] Figure 7 A diagram showing a quantitative comparison of experimental results between virtual scene trajectory planning and 3D reconstruction in one embodiment;

[0017] Figure 8 Graph showing qualitative comparison experimental results of three-dimensional reconstruction of virtual scenes in one embodiment;

[0018] Figure 9 1. A diagram showing qualitative comparison experimental results of 3D reconstruction of a real scene in one embodiment;

[0019] Figure 10 Graph showing trajectory planning comparison experiment results in one embodiment. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0021] In one embodiment, Figure 1 As shown, a UAV flight trajectory planning method for high-precision 3D reconstruction of urban scenes is provided, including the following steps:

[0022] Step 102 : generating an initial viewpoint based on a standard method, abstracting the traversal task of the initial viewpoint into a traveling salesman problem model, and then using the heuristic strategy of the traveling salesman problem to generate a minimum distance cycle path.

[0023] First, the task of traversing the initial viewpoint is abstracted into a traveling salesman problem model. Each viewpoint is modeled as a node in a fully connected graph. The edge weights between nodes are calculated based on visible graph theory to minimize the shortest path in the unobstructed airspace. A heuristic strategy is then used to generate a minimum-distance cyclic path, ensuring that all viewpoints are traversed before returning to the origin. During the initial path generation phase, the initial viewpoint traversal is abstracted into a traveling salesman problem and a heuristic strategy is used to generate a minimum-distance cyclic path, thus reducing inefficient flight distance. In traditional viewpoint-first models, the rigid concatenation of pre-set viewpoints often leads to path redundancy. The core of the traveling salesman problem model is to find the shortest traversal path between multiple nodes, laying the foundation for distance optimization and improving flight efficiency.

[0024] Step 104, updating the viewpoint position in the minimum distance cycle path according to the unit length contribution criterion to obtain the updated viewpoint; calculating the movable radius of the viewpoint using the contribution of the updated viewpoint to scene reconstruction, and constructing a spatial spherical flight corridor based on the movable radius.

[0025] High-quality 3D reconstruction requires image acquisition at specific locations, which imposes a constraint on the UAV's flight trajectory, requiring it to traverse viewpoints sequentially. This requirement creates a key contradiction: while adjusting viewpoint pose can improve trajectory continuity and flight efficiency, it can also reduce reconstruction quality. To address this, this application proposes the concept of a movable radius. By establishing a quantitative mapping between inter-viewpoint distance and reconstruction contribution, this approach ensures that the loss in reconstruction quality caused by viewpoint pose adjustment remains within an acceptable range. First, viewpoint positions are updated based on a unit-length contribution criterion. This criterion optimizes the geometric information acquired per unit flight distance and dynamically updates the viewpoints in the initial path. For viewpoints with low contribution, flight cost can be reduced by shortening their distance to surrounding high-contributing viewpoints. For high-contributing viewpoints, their positions are ensured to maximize information acquisition. This avoids redundant coverage of invalid viewpoints and improves the accuracy of geometric information acquisition by optimizing the spatial distribution of viewpoints. This ensures that each flight distance corresponds to a higher reconstruction value, fundamentally improving planning accuracy and concentrating resources on areas that are more important for reconstruction. Traditional joint optimization methods either sacrifice trajectory smoothness due to fixed viewpoints or reduce reconstruction quality due to unconstrained optimization. However, the spherical flight corridor sets a movable range for each viewpoint. High-contribution viewpoints have a small movable radius, ensuring accuracy in the core area; low-contribution viewpoints have a large movable radius, providing more flexibility in trajectory optimization. This solves the dilemma between the inefficiency of strict constraints and the loss of focus due to unconstrained optimization.

[0026] Step 106 , under the constraints of the spatial spherical flight corridor, generate a time-optimal flight trajectory by solving an optimal control problem that takes into account the UAV dynamics model and safety constraints.

[0027] Under the corridor constraint, by solving the optimal control problem that takes into account the UAV dynamics model and safety constraints, a feasible time-optimal flight trajectory is generated, thereby converting discrete viewpoints into efficient trajectories. After completing the flight trajectory planning, the position of the original viewpoint needs to be adjusted to ensure that the captured image fully covers the scene. Figure 6 As shown in the figure, each original viewpoint is repositioned to the nearest point on the trajectory. At the same time, the ray intersection point of each adjusted viewpoint with the target scene is consistent with its original viewpoint, ensuring equivalent scene coverage without changing the number of viewpoints. This ensures that the final 3D reconstruction quality is not degraded, achieving optimal image capture and high-fidelity 3D reconstruction of the target scene.

[0028] The above-mentioned drone flight trajectory planning method for high-precision three-dimensional reconstruction of urban scenes, this application drives viewpoint optimization through the unit length contribution criterion, and generates dynamically feasible trajectories based on spherical flight corridors, significantly improving the efficiency of acquiring geometric information of the target scene, achieving a global optimal balance between reconstruction quality and flight efficiency, and can significantly shorten the drone track length and flight time while ensuring high-quality three-dimensional reconstruction.

[0029] In one embodiment, the unit length contribution criterion includes:

[0030] The unit length contribution of a path segment is defined as the weighted sum of the contributions of all viewpoints in the segment divided by its path length. For a path consisting of n viewpoints (v1, v2, ..., v n The calculation formula of the unit length contribution of ) is

[0031]

[0032] Among them, d i Represents the viewpoint v i and v i+1 The distance between c i represents the contribution of the i-th viewpoint to scene reconstruction;

[0033] Take the adjacent viewpoint v n-1 and v n+1 For fixed endpoints, the calculation includes v n With and bypass v n The unit length contribution ratio of the path segment, when bypassing v n When the unit length contribution of is greater, it is determined that the viewpoint v needs to be updated n .

[0034] In a specific embodiment, after obtaining the traveling salesman path, the unit length contribution criterion is used to optimize the path efficiency. This criterion defines the unit length contribution of a path segment as the weighted sum of the contributions of all viewpoints in the segment divided by its path length, where the weight of the endpoint is set to 0.5 and the weight of the internal point is set to 1 (the endpoint is a shared connection point between adjacent path segments, and its contribution needs to be calculated in half). For a path consisting of n viewpoints (v1, v2, ..., v n ), the calculation formula for unit length contribution is

[0035]

[0036] in

[0037]

[0038] d i Represents the viewpoint v i and v i+1The distance between them.

[0039] Contribution of viewpoint to scene reconstruction c n The marginal benefit of local geometric reconstruction at visible sampling points is quantified and solved by the following formula:

[0040]

[0041] Among them, {s s} represents the viewpoint v n The set of visible sampling points; {v v} represents the viewpoint set; and Represents all viewpoints and removes viewpoint v n The rest of the viewpoints for point s j The reconstruction is achieved through a pre-trained Transformer network.

[0042] To determine the viewpoint v n Whether it needs to be updated, this application uses adjacent viewpoints v n-1 and v n+1 For fixed endpoints, the calculation includes v n The path segment and bypass v n The path segment (i.e. direct v n-1 and v n+1 The unit length contribution ratio of the new path segment is formed. n When the unit length contribution of is greater, it is determined that the viewpoint v needs to be updated n The mathematical expression of this criterion is:

[0043]

[0044] The update rate η quantifies the viewpoint v n The marginal benefit of the contribution per unit length, in this application, η=1 is set as the baseline.

[0045] When multiple consecutive viewpoints need to be updated, this application adopts a parallel processing mechanism. n-1 and v′ n+N′ ), so based on the marginal contribution of a single viewpoint, the update criterion formula can be generalized as:

[0046]

[0047] Among them, N' is the number of viewpoints to be updated. In the specific implementation, first, candidate positions are sampled in the triangular area with the initial viewpoint and the boundary viewpoint as vertices; then the feasibility of each candidate position replacing the original viewpoint is evaluated according to the judgment formula; finally, the feasible candidate viewpoint with the smallest Euclidean distance to the initial viewpoint is selected as the replacement position. If there is no feasible candidate viewpoint, the viewpoint is deleted. The initial path planning scheme after the unit length contribution update can generate a more efficient coarse-grained waypoint sequence, providing a topological basis for subsequent further trajectory optimization. Figure 2 Shows single / dual viewpoint update cases, Figure 2 (a) is a single viewpoint update case diagram, Figure 2 (b) is a case diagram of dual-viewpoint update.

[0048] In one embodiment, calculating the movable radius of the viewpoint using the contribution of the updated viewpoint to scene reconstruction includes:

[0049] Calculate the viewpoint v using the contribution of the updated viewpoint to the scene reconstruction n The movable radius is:

[0050]

[0051]

[0052] Among them, ∈=0.01m sets the lower threshold of the negative mapping result, c n Represents the viewpoint v n Contribution to scene reconstruction, {s s} represents the viewpoint v n The set of visible sampling points; {v v} represents the viewpoint set; and Represents all viewpoints and removes viewpoint v n The rest of the viewpoints for point s j Reconstructibility.

[0053] In a specific embodiment, the movable radius is used to quantify the maximum offset distance of the viewpoint position without affecting the quality of 3D reconstruction. This application first analyzes the viewpoint reconstruction contribution c in the initial path. n Distance d from adjacent viewpoints i The distribution of (histogram and distribution of viewpoint reconstruction contribution and adjacent viewpoint distances as shown in Figure 3 (a) Figure 3 (b)). The Shapiro-Wilk normality test verifies that both obey normal distribution; therefore, a mapping relationship can be established through normal quantiles (see Figure 3 (c)), namely:

[0054] Φ1(μ1+σ1zα )=1-Φ2(μ2+σ2z 1-α ),

[0055] Among them, c m ~N(μ1,σ1), r m ~N(μ2,σ2), Φ1 and Φ2 are cumulative distribution functions, z α and z 1-α are the α and 1-α quantiles of the standard normal distribution. To adapt the scale, let the movable radius be distributed identically to half of the distance between the viewpoints. After simplification, we get the viewpoint v n The formula for calculating the movable radius is:

[0056]

[0057] Here, ∈ = 0.01m sets the lower bound threshold for negative mapping results. This formula adaptively assigns a smaller radius to viewpoints with higher reconstructability contributions to ensure acquisition quality, while assigning a larger radius to viewpoints with lower importance to improve trajectory flexibility. This allows for more efficient flight trajectories while maintaining reconstruction accuracy.

[0058] In one embodiment, constructing a spatial spherical flight corridor based on a movable radius includes:

[0059] With viewpoint position q n Construct movable radius r for the center n A spherical region is constructed based on the initial path sequence, and the constraints of the spherical flight corridor are set as follows:

[0060]

[0061] Among them, p(t n ) indicates that the drone is at t n The position of the sphere n at any moment, q n Represents the viewpoint v n location.

[0062] In a specific embodiment, the movable radius of each viewpoint significantly improves the optimizability of the trajectory by expanding the viewpoint access space. n Construct radius r for the center n A semi-open navigation corridor can be established based on the initial path sequence (such as Figure 4 This corridor constraint requires that the trajectory must intersect all spheres in sequence, ensuring that the viewpoint movement is restricted to its movable radius:

[0063]

[0064] Among them, p(t n) indicates that the drone is at t n The position of the sphere n at any moment, q n Represents the viewpoint v n The spherical flight corridor provides free space for trajectory optimization while ensuring viewpoint accessibility, thus achieving efficient data acquisition and high-quality 3D reconstruction.

[0065] In one embodiment, under the constraints of a spherical flight corridor in space, a time-optimal flight trajectory is generated by solving an optimal control problem that takes into account the UAV dynamics model and safety constraints, including:

[0066] The UAV is modeled as a nonlinear system with differential flatness. Based on the differential flatness, the full state variables of the system are determined by planning the position trajectory of the UAV and the yaw angle trajectory specified according to the viewpoint pose. The dynamic constraints of the UAV during flight are set based on the full state variables of the system.

[0067] Given a spherical flight corridor containing multiple spheres, the goal of trajectory optimization is to generate a smooth trajectory that sequentially passes through all the spheres within time T. The smooth trajectory is decomposed into multiple trajectory segments, each of which is parameterized using a quintic polynomial. The coefficients of the quintic polynomial are solved through boundary conditions, and the trajectory optimization problem of the UAV is transformed into an optimal control problem. The IPOPT is used to solve the optimal control problem under the dynamic and safety constraints of the UAV to generate a time-optimal flight trajectory.

[0068] In a specific embodiment, the flight corridor defines the free navigation area of ​​the drone. To ensure the dynamic feasibility of the drone's trajectory, the present application models the drone as a nonlinear system with differential flatness, whose flat output is σ = [x, y, z, ψ]T, where p = [x, y, z]T represents the position of the drone and ψ represents the yaw angle. Based on the differential flatness, the full state variables of the system can be fully determined by planning the drone's position trajectory p(t) and the yaw angle trajectory ψ specified according to the viewpoint pose. The dynamic constraints during flight are described by the following ordinary differential equations:

[0069]

[0070] Where s represents the state variable, and u represents the control variable. p, v, a, and j represent the position, velocity, acceleration, and jerk of the point on the trajectory, respectively. By solving the optimal control problem under these constraints, a dynamically feasible UAV flight trajectory can be generated.

[0071] like Figure 5 As shown in Figure 2, given a flight corridor sequence containing M spheres, the goal of trajectory optimization is to generate a smooth trajectory p(t) that passes through all spheres in sequence within time T: The trajectory is decomposed into M segments p k (t), corresponding to the time period t∈[0,T k ]. The total flight time is the sum of the flight times of each segment:

[0072]

[0073] To ensure acceleration continuity (thrust continuity) and jerk boundedness (angular velocity continuity), each trajectory segment p k (t),t∈[0,T k ] parameterized by a quintic polynomial:

[0074]

[0075] where p k (0),v k (0),a k (0) is the starting state of the kth trajectory, are polynomial coefficients. The trajectory must satisfy the viewpoint access constraints:

[0076]

[0077] Adjacent segment trajectory state s k (T k ) and s k+1 (0) must be equal to ensure the continuity of state transition:

[0078] s k (T k )=s k+1 (0),

[0079] Among them, s k (0) and s k (T k ) are the start and end states of the kth trajectory. In addition, to achieve the periodicity of the closed-loop trajectory, the following constraints must be met:

[0080] s M (T M )=s1(0)=d0,

[0081] Where d0 is the state value at the start and end of the trajectory, and the velocity and acceleration are both zero. Due to dynamic constraints, the velocity, acceleration, and jerk need to meet the maximum limit:

[0082]

[0083] Among them, v max ,a max ,j maxThey are the maximum speed, acceleration, and jerk of the UAV during flight. In addition, flight safety is also a key requirement for UAV trajectory planning:

[0084]

[0085] Among them, A ij ,B ij ,C ij ,D ij are the coefficients defining the jth half-space of the ith polyhedron.

[0086] For each trajectory, the coefficients of the quintic polynomial are solved by the boundary conditions:

[0087]

[0088] Where Δp k =p k (T k )-p k (0),Δv k =v k (T k )-v k (0),Δa k =a k (T k )-a k (0). Therefore, the trajectory planning problem of the UAV can be transformed into an optimal control problem, where the optimization variables are:

[0089] X={T k},{p k (0)},{v k (0)},{a k (0)},

[0090] The objective function is:

[0091]

[0092] In order to obtain the optimal solution, this application uses CasADi for symbolic modeling and IPOPT for nonlinear solution. Each trajectory is confined to the corresponding space, and the feasibility of the dynamics is ensured by path constraints. In a specific embodiment, according to the actual situation, v max Set it to 5.0m / s, and set a max Set to 7.0 m / s 2 , ensuring that the drone can fly stably.

[0093] In one embodiment, the dynamic constraints of the UAV during flight are set based on the full state variables of the system as follows:

[0094]

[0095] Among them, s represents the state variable, u represents the control variable, p, v, a, j represent the position, velocity, acceleration and jerk of the point on the trajectory respectively.

[0096] In one embodiment, a fifth-order polynomial parameterization is applied to each trajectory segment, including:

[0097] For each trajectory segment p k (t),t∈[0,T k ] parameterized by a quintic polynomial:

[0098]

[0099] Where t∈[0,T k ] represents the time period, p k (0),v k (0),a k (0) is the starting state of the kth trajectory, are the polynomial coefficients.

[0100] In one embodiment, the trajectory optimization problem of the UAV is transformed into an optimal control problem by solving the coefficients of the quintic polynomial by boundary conditions, including:

[0101] For each trajectory, the coefficients of the quintic polynomial are solved by the boundary conditions:

[0102]

[0103] Where Δp k =p k (T k )-p k (0),Δv k =v k (T k )-v k (0),Δa k =a k (T k )-a k (0), p k (0),v k (0),a k (0) is the starting state of the kth trajectory, are the polynomial coefficients;

[0104] The trajectory planning problem of the UAV is then transformed into an optimal control problem, where the optimization variables are:

[0105] X={T k},{p k(0)},{v k (0)},{a k (0)},

[0106] The objective function is:

[0107]

[0108] Where M represents the number of spheres, T k represents the flight time of the kth trajectory.

[0109] In one embodiment, due to dynamic constraints, velocity, acceleration, and jerk need to meet maximum limits:

[0110]

[0111] Among them, v max ,a max ,j max are the maximum speed, acceleration and jerk of the drone during flight, respectively.

[0112] In one embodiment, the security constraint is

[0113]

[0114] Among them, A ij ,B ij ,C ij ,D ij are the coefficients defining the jth half-space of the ith polyhedron.

[0115] In the specific embodiment, in order to illustrate the advantages of the method of the present application, Figure 7 Comparative experimental results of trajectory planning and 3D reconstruction using this method and previous methods are presented. These experiments were conducted using rough geometric proxies of the Castle, Bridge, and School virtual scenes from the UrbanScene 3D dataset. These comparative results demonstrate that this method significantly reduces drone track length and flight time while maintaining comparable or even improved reconstruction quality compared to other methods.

[0116] Figure 8 The qualitative comparison experimental results of the three-dimensional reconstruction of the present application method and previous methods in Castle, Bridge and School scenes are shown. Figure 9 The qualitative comparison experimental results of the three-dimensional reconstruction of the proposed method, the latest algorithm and the oblique photography method in the Museum and Concert Hall scenes are demonstrated. Figure 10Comparative results of trajectory planning between this proposed method and previous methods in the Castle, Bridge, and School scenarios are presented. These experimental results demonstrate that this proposed method can achieve smooth trajectory generation and high-fidelity reconstruction across scenarios, maximizing flight efficiency and trajectory feasibility. Regardless of the viewpoint planning method used, all viewpoints can be optimally connected. Furthermore, these experimental results demonstrate that although this proposed method is trained in virtual scenarios, it can achieve comparable results in real-world scenarios, demonstrating its potential for practical application.

[0117] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0118] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and such modifications and improvements are intended to fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A UAV flight trajectory planning method for high-precision 3D reconstruction of urban scenes, characterized by: The method comprises: Generate the initial viewpoint based on the standard method, abstract the traversal task of the initial viewpoint into the traveling salesman problem model, and then use the heuristic strategy of the traveling salesman problem to generate the minimum distance cycle path; The viewpoint position in the minimum distance cycle path is updated according to the unit length contribution criterion to obtain an updated viewpoint; the movable radius of the viewpoint is calculated using the contribution of the updated viewpoint to scene reconstruction, and a spatial spherical flight corridor is constructed based on the movable radius; Under the constraints of the spatial spherical flight corridor, the time-optimal flight trajectory is generated by solving the optimal control problem that takes into account the UAV dynamics model and safety constraints.

2. The method according to claim 1, characterized in that The unit length contribution criteria include: The unit length contribution of a path segment is defined as the weighted sum of the contributions of all viewpoints in the segment divided by its path length. For a path consisting of n viewpoints (v1, v2, ..., v n The calculation formula of the unit length contribution of ) is Among them, d i Represents the viewpoint v i and v i+1 The distance between c i represents the contribution of the i-th viewpoint to scene reconstruction; Take the adjacent viewpoint v n-1 and v n+1 For fixed endpoints, the calculation includes v n With and bypass v n The unit length contribution ratio of the path segment, when bypassing v n When the unit length contribution of is greater, it is determined that the viewpoint v needs to be updated n .

3. The method according to claim 1, characterized in that The movable radius of the viewpoint is calculated using the contribution of the updated viewpoint to the scene reconstruction, including: Calculate the viewpoint v using the contribution of the updated viewpoint to the scene reconstruction n The movable radius is: Among them, ∈=0.01m sets the lower threshold of the negative mapping result, c n Represents the viewpoint v n Contribution to scene reconstruction, {s s } represents the viewpoint v n The set of visible sampling points; {v v } represents the viewpoint set; and Represents all viewpoints and removes viewpoint v n The rest of the viewpoints for point s j Reconstructibility.

4. The method according to claim 1, wherein Constructing a spatial spherical flight corridor based on the movable radius includes: With viewpoint position q n Construct movable radius r for the center n A spherical region is constructed based on the initial path sequence, and the constraints of the spherical flight corridor are set as follows: Among them, p(t n ) indicates that the drone is at t n The position of the sphere n at any moment, q n Represents the viewpoint v n location.

5. The method according to claim 1, wherein Under the constraints of the spatial spherical flight corridor, the optimal control problem considering the UAV dynamics model and safety constraints is solved to generate a time-optimal flight trajectory, including: The UAV is modeled as a nonlinear system with differential flatness. Based on the differential flatness, the full state variables of the system are determined by planning the position trajectory of the UAV and the yaw angle trajectory specified according to the viewpoint pose. The dynamic constraints of the UAV during flight are set based on the full state variables of the system. Given a spherical flight corridor containing multiple spheres, the goal of trajectory optimization is to generate a smooth trajectory that sequentially passes through all the spheres within time T. The smooth trajectory is decomposed into multiple trajectory segments, each of which is parameterized using a quintic polynomial. The coefficients of the quintic polynomial are solved using boundary conditions, and the trajectory optimization problem of the UAV is transformed into an optimal control problem. The IPOPT is used to solve the optimal control problem under the dynamic and safety constraints of the UAV to generate a time-optimal flight trajectory.

6. The method according to claim 5, characterized in that The dynamic constraints of the UAV during flight are set based on the full state variables of the system as follows: Among them, s represents the state variable, u represents the control variable, p, v, a, j represent the position, velocity, acceleration and jerk of the point on the trajectory respectively.

7. The method according to claim 6, characterized in that Each trajectory segment is parameterized using a fifth-order polynomial, including: For each trajectory segment p k (t),t∈[0,T k ] parameterized by a quintic polynomial: Among them, t∈[0,T k ] represents the time period, p k (0),v k (0),a k (0) is the starting state of the kth trajectory, are the polynomial coefficients.

8. The method according to claim 5, characterized in that By solving the coefficients of the quintic polynomial under boundary conditions, the trajectory optimization problem of the UAV is transformed into an optimal control problem, including: For each trajectory, the coefficients of the quintic polynomial are solved by the boundary conditions: Where Δp k =p k (T k )-p k (0),Δv k =v k (T k )-v k (0),Δa k =a k (T k )-a k (0), p k (0),v k (0),a k (0) is the starting state of the kth trajectory, are the polynomial coefficients; The trajectory planning problem of the UAV is then transformed into an optimal control problem, where the optimization variables are: X={T k },{p k (0)},{v k (0)},{a k (0)} The objective function is: Where M represents the number of spheres, T k represents the flight time of the kth trajectory.

9. The method according to claim 7, characterized in that The method further comprises: Due to dynamic constraints, velocity, acceleration, and jerk must meet maximum limits: Among them, v max ,a max ,j max are the maximum speed, acceleration and jerk of the drone during flight, respectively.

10. The method according to claim 5, characterized in that The safety constraints are Among them, A ij ,B ij ,C ij ,D ij are the coefficients defining the jth half-space of the ith polyhedron.

Images