Multi-unmanned aerial vehicle monitoring method considering ground safety

Abstract

The application discloses a kind of multi-unmanned aerial vehicle monitoring methods considering ground safety, by modeling and solving multi-objective optimization problem, can maximize the monitoring utility of network while optimizing the safety of ground target. Including the following steps: constantization approximate monitoring utility and safety reward function;Real-time acquisition of the position and orientation of all targets, execute approximate scheme to divide two-dimensional plane into a finite number of sub-regions;In each sub-region, apply candidate monitoring strategy extraction algorithm to obtain a finite set of candidate monitoring strategies;Obtain feasible solution with (1-1 / e-ò) approximation ratio by greedy algorithm;Solve the minimum matching problem to find the new position coordinates that each unmanned aerial vehicle will go to, ensure that the total distance of movement is the shortest, apply trajectory planning algorithm to safely let unmanned aerial vehicle go to the new deployment location, continuously and reliably execute monitoring tasks.The application can solve the problem of insufficient practicality caused by ignoring ground safety in existing unmanned aerial vehicle deployment methods.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of Internet of Things, in particular to a multi-unmanned aerial vehicle monitoring method considering ground safety. BACKGROUND

[0002] Compared with the traditional monitoring network, the multi-unmanned aerial vehicle monitoring system can cover a larger area, has higher monitoring accuracy and flexibility, and has higher robustness. When one unmanned aerial vehicle fails, other unmanned aerial vehicles can quickly take over the monitoring task, thereby ensuring the continuous and reliable operation of the monitoring task. In response to complex monitoring tasks, the multi-unmanned aerial vehicle monitoring system is a very effective monitoring technology.

[0003] The unmanned aerial vehicle monitoring system is also an effective tool for responding to sudden and temporary scenarios, such as in crowd gatherings, traffic accidents, large-scale activities, natural disasters, fire rescue, etc. The multi-unmanned aerial vehicle monitoring system can quickly deploy monitoring and provide an efficient monitoring solution. However, the current monitoring utility optimization methods at home and abroad do not fully consider the impact of unmanned aerial vehicles on ground safety, which may cause the unmanned aerial vehicles used for monitoring to pose a significant safety threat to ground personnel or important property. In addition, due to the relatively low reliability and stability of unmanned aerial vehicles, improper operation or failure of unmanned aerial vehicles may also pose unpredictable risks to ground personnel or facilities. In particular, existing methods usually deploy unmanned aerial vehicles in relatively close proximity to the target to obtain greater monitoring utility, which results in a lack of practicality of these methods in actual applications. SUMMARY

[0004] The purpose of the present application is to provide a multi-unmanned aerial vehicle monitoring method considering ground safety, which solves the problem of lack of practicality caused by ignoring ground safety in existing unmanned aerial vehicle deployment methods.

[0005] Technical scheme: In order to achieve the above-mentioned application purpose, the present application provides a multi-unmanned aerial vehicle monitoring method considering ground safety, comprising the following steps:

[0006] (1) Constant approximation stage: using piecewise constant approximation function to approximate the monitoring quality function and the safety reward function value as a constant in the distance dimension in each distance interval, and discretizing in the angle dimension, the monitoring utility function brought by deploying unmanned aerial vehicles in a single angle interval is approximated as a constant, and the constant approximation scheme is passed to step (2);

[0007] (2) Region discretization stage: real-time acquisition of the positions and orientations of all targets, according to the constant approximation scheme in step (1), the two-dimensional plane is divided into a finite number of sub-regions, and each sub-region obtained by discretization is passed to step (3);

[0008] (3) candidate monitoring strategy extraction stage: apply a candidate monitoring strategy extraction algorithm in each sub-region to find a monitoring strategy equivalent to all monitoring strategies in the region as a candidate monitoring strategy, and pass the obtained candidate monitoring strategy set to step (4);

[0009] (4) greedy selection stage: on the candidate monitoring strategy set, apply a greedy algorithm to iteratively select the element with the maximum current gain to obtain a monitoring strategy set with performance guarantee, until the number of monitoring strategies obtained reaches the upper limit of the UAV number budget, and pass the monitoring strategy set to step (5);

[0010] (5) path selection and trajectory planning stage: apply the Kuhn-Munkres algorithm to find new position coordinates to which each UAV will go, apply a trajectory planning algorithm to safely let the UAV go to the new deployment position, and set the UAV orientation in the corresponding strategy after arrival to realize monitoring of the dynamic scene.

[0011] Further, in step (1), the monitoring quality function is used to measure the monitoring quality obtained by the UAV at a certain angle, which is inversely proportional to the square of the distance and negatively correlated with the angle size of the deviation from the certain angle, the monitoring utility function is the integral of the monitoring quality function in the angle dimension, and the safety reward function is used to measure the safety of the target, which is directly related to the distance of the UAV from the target; two kinds of piecewise constant approximation schemes are used to approximate the safety reward function and the monitoring quality function, respectively.

[0012] Further, in step (2), the target includes a monitoring target without considering safety, a non-monitoring target considering safety, and a monitoring target considering safety, different constant approximation schemes are adopted according to the types of targets to divide the two-dimensional plane into a limited number of sub-regions, at this time, the monitoring utility function and the safety reward function of deploying a UAV at all positions in each sub-region for all monitored targets are approximated as constant values.

[0013] Further, in step (3), for each sub-region, for each target pair that can be monitored by a UAV deployed in the sub-region, first draw a straight line through the coordinates of the target pair, record the intersection of the straight line and the sub-region, deploy monitoring at each intersection, adjust the direction of the monitoring to make the right boundary of the sector overlap with the straight line, and record the corresponding monitoring strategy; second, draw a circular arc with the monitoring angle of the UAV as the central angle through the two coordinates, record the intersection of the circular arc and the sub-region, deploy monitoring at each intersection, adjust the direction of the monitoring to make the left and right boundaries of the sector pass through the two targets, and record the corresponding monitoring strategy; then, randomly select a point on the sub-region, deploy monitoring at the point, and rotate the monitoring direction counterclockwise, when a new target crosses the radius of the sector, record the current monitoring strategy; finally, filter out redundant monitoring strategies to obtain the final monitoring strategy set.

[0014] Further, in step (5), a bipartite graph is constructed using the current positions of all UAVs and the latest deployment positions in the monitoring strategy obtained in step (4), and a Kuhn-Munkres algorithm is applied to solve the minimum matching problem to find new position coordinates to which the UAVs will go.

[0015] Beneficial effects: The present application first proposes a multi-UAV monitoring method considering ground safety, which can effectively solve the problem of insufficient practicality caused by ignoring ground safety in existing UAV deployment methods: first, the present application can maximize the monitoring utility of the network while optimizing the safety of the ground target by solving a multi-objective optimization problem. Second, the present application obtains a feasible solution with performance guarantee under a small time complexity through constant approximation, candidate monitoring strategy extraction and greedy selection stage. Third, the present application generates new deployment positions by changing the target position, and realizes monitoring of dynamic scenes through path selection and trajectory planning. Fourth, the present application verifies the influence of different parameters on the performance of the algorithm through simulation experiment analysis, and the experimental data shows that the monitoring strategy obtained by the method of the present application performs better than the comparative algorithm under various parameter conditions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Fig. 1 is a schematic diagram of the overall flow of the present application;

[0017] Figure 2 Fig. 2 is a schematic diagram of the monitoring utility model of the present application, wherein Figure 2 (a) represents a UAV monitoring model, Figure 2 (b) represents a target model, Figure 2 (c) represents a monitoring model;

[0018] Figure 3 Fig. 3 is a schematic diagram of the distance dimension constant approximation scheme, Figure 3 (a) is a function approximation schematic diagram, Figure 3 (b) represents a schematic diagram of a single target executing the distance constant approximation scheme on a 2D plane;

[0019] Figure 4 Fig. 4 is a schematic diagram of the angle dimension constant approximation scheme;

[0020] Figure 5 Fig. 5 is a schematic diagram of the approximation scheme of three different types of targets, Figure 5 (a), Figure 5 (b) and Figure 5 (c) respectively represent the constant approximation scheme schematic diagram of the monitoring target without considering safety, the non-monitoring target considering safety and the monitoring target considering safety;

[0021] Figure 6Schematic diagram for the region discretization stage;

[0022] Figure 7 Schematic diagram for the candidate monitoring strategy extraction algorithm of Algorithm 1, Figure 7 (a) The steps shown are lines 7-10 of Algorithm 1, Figure 7 (b) The steps shown are lines 11-14 of Algorithm 1, Figure 7 (c) The steps shown are lines 15-16 of Algorithm 1. DETAILED DESCRIPTION

[0023] The present application will be further clarified by the following examples and figures, which should not be taken as limiting the scope of the present application. After reading this application, those skilled in the art will be able to affect various modifications to the present application. Such modifications are intended to be within the scope of the claims appended hereto.

[0024] The present application provides a multi-unmanned aerial vehicle monitoring method considering ground safety, which solves the problem of insufficient practicability of existing unmanned aerial vehicle deployment methods. As shown in Figure 1 The complete flow of the present application includes a constant approximation stage, a region discretization stage, a candidate monitoring strategy extraction stage, a greedy selection stage, and a path selection and trajectory planning stage. The specific embodiments are described as follows:

[0025] First, the multi-unmanned aerial vehicle monitoring method considering ground safety is formalized, and the technical solutions corresponding to each step are introduced in turn. First, it is assumed that there is a budget of b unmanned aerial vehicles, and these unmanned aerial vehicles are needed to monitor N targets O m ={o1,o2,...,o N} on a two-dimensional plane, considering the safety of K ground objects O s ={o1,o2,...,o K}. The coordinates of the jth target are o j , and the direction is represented by the angle θ oj . The deployment strategy set of the unmanned aerial vehicle is represented as U={<u1,θ u1 >,<u2,θ u2 >,...,<u M ,θ uM >}, and the binary tuple u i and θ ui represent the coordinates and orientation of the unmanned aerial vehicle, respectively.

[0026] The monitoring quality (Quality of Monitoring, QoM) function is formalized as:

[0027]

[0028] where a, b are constants, φ is an arbitrary angle, is θ ui is the unit vector of the direction, is θ oj is the unit vector of the direction, α(·) represents the angle between two vectors, β is the monitoring angle of the UAV, and γ is the angle that the target needs to be monitored.

[0029] The present application defines the monitoring utility of the target o j as the integral of the maximum QoM function Q(u i ,o j ,θ ui ,θ oj ,φ) in each direction, which is specifically as follows:

[0030]

[0031] Figure 2 This monitoring utility model is described figuratively, Figure 2 (a) is the UAV model, Figure 2 (b) is the target model, Figure 2 (c) is the monitoring model, where ω is the angle that the target needs to be captured by the UAV, and ω e is the effective angle of monitoring.

[0032] Intuitively, the safety of the target increases with the increase of the distance, and thus the safety reward function is constructed as follows:

[0033]

[0034] where A, B, and C are all constants, R0 is a constant used to ensure that the value of R(u i ,o j ) is non-negative, D s is the safety distance, and the safety reward function no longer increases with the increase of the distance beyond this distance, and the overall safety reward function of the target is

[0035]

[0036] The first optimization goal of the present application is to maximize the monitoring utility of all the targets to be monitored,

[0037]

[0038] and the second goal is to maximize the safety reward function,

[0039]

[0040] The present application sums both with weights to obtain the overall optimization goal of the problem, and the problem is defined as follows:

[0041] (P1)

[0042] s.t.U={<u i ,θ ui >|θ ui ∈[0,2π)},|U|≤b.

[0043] The constantization approximation stage corresponds to technical solution step (1). The specific implementation is as follows:

[0044] For simplicity of description, the following uses R(d) to represent the safety reward function R(u j ,o i ) of any target o i and any UAV u j when the distance between them is d, and uses to represent the piecewise approximation function of R(d).

[0045] The distance dimension constantization approximation scheme 1 is executed to obtain the following piecewise approximation function

[0046]

[0047] The following introduces the distance dimension constantization approximation scheme 1 used to approximate R(d):

[0048] The distance dimension constantization approximation scheme 1 is specifically: let l(0)=0,l(K1)=D s , wherein The error of

[0049]

[0050] Figure 3 The distance dimension constantization approximation scheme 1 is described, and the space around the target is divided into l(0), l(1),..., l(K1) in total K1 segments according to the distance, forming K1 circular ring regions. In each circular ring region, the value of the approximation function is a constant.

[0051] Similarly, the function Q(d,φ) is used to represent the QoM function on the angle φ, d is the distance between the target and the UAV, and is used to represent the piecewise approximation function of Q(d,φ).

[0052] The distance dimension constantization approximation scheme 2 is executed to obtain the following piecewise approximation function

[0053]

[0054] The following introduces the distance dimension constantization approximation scheme 2 used to approximate Q(d, φ).

[0055] The distance dimension constantization approximation scheme 2 is specifically: fixing φ, setting l(0) = 0, l(K2) = D, wherein The error of Q(d, φ) is limited to a constant ò2:

[0056]

[0057] After the above distance dimension constantization approximation scheme 2, the function Q has been successfully segmented and approximated, and then the function is used to represent the monitoring utility function U at this time, is used to represent the approximation function of , and Figure 4 sets forth the key idea that by discretizing the angle into multiple angle intervals of size ΔA, only the angle intervals that are completely covered are used to calculate the approximate monitoring utility function.

[0058] The angle dimension constantization approximation scheme is executed to obtain the following approximation function

[0059]

[0060] wherein, represents the approximate monitoring utility of the kth angle interval of the unmanned aerial vehicle u i , which is formally represented as

[0061]

[0062] The angle dimension constantization approximation scheme is specifically: fixing the distance, setting ΔA = cd(γ, ω) cd(·,·) represents a function of finding the common factor of two numbers, and at this time The error of Q(d, φ) is limited to a constant ò Δ :

[0063]

[0064] The region discretization stage corresponds to technical scheme step (2). The specific implementation is:

[0065] Firstly, the positions and orientations of all targets are obtained in real time. According to different target types, the two-dimensional plane is divided into a finite number of sub-regions using the piecewise constantization method described in step (1). Specifically, the targets are divided into three types: the first type is a monitoring target without considering safety, the second type is a non-monitoring target considering safety, and the third type is a monitoring target considering safety. Figure 5 The different constantization approximation schemes for the three types of targets are shown by examples. For the first type of target, the distance dimension constantization approximation scheme 2 and the angle dimension constantization approximation scheme are implemented, as shown in Figure 5 (a); for the second type of target, the distance dimension constantization approximation scheme 1 is implemented, as shown in Figure 5 (b); and for the third type of target, the distance dimension constantization approximation scheme 1 and the distance dimension constantization approximation scheme 2 are implemented simultaneously, as well as the angle dimension constantization approximation scheme, as shown in Figure 5 (c).

[0066] Figure 6 The examples show how to divide the two-dimensional plane for three third-type targets. The sub-regions can be divided into three categories according to the number of effectively monitored targets: (I) the number of effectively monitored targets is 0; (II) the number of effectively monitored targets is 1; and (III) the number of effectively monitored targets is greater than or equal to 2. Figure 6 The midpoints p1, p2, and p3 are located in the sub-regions of type (I), type (II), and type (III), respectively.

[0067] The candidate monitoring strategy extraction stage corresponds to step (3) of the technical solution. The specific implementation is as follows:

[0068] After step (2), the space is divided into a finite number of sub-regions. In each sub-region, the functions R and U have been approximated. Therefore, only the representative monitoring strategies in each sub-region need to be found as candidate monitoring strategies, which have the characteristic that the set of targets they monitor is not a subset of the set of targets monitored by any other strategy. Algorithm 1 records how to extract candidate monitoring strategies, Figure 7 The main process of algorithm 1 is shown in Figure 7 (a) shows steps 7-10 of algorithm 1, Figure 7 (b) shows steps 11-14 of algorithm 1, Figure 7 (c) shows steps 15-16 of algorithm 1.

[0069]

[0070]

[0071] For the set of sub-regions A = {a1, a2,..., an}, the candidate monitoring strategy set is denoted as C = {c1, c2,..., cm}, where m is the number of candidate monitoring strategies. |A|Algorithm 1 is applied to all sub-regions in the dataset to obtain the candidate monitoring set Γ for that sub-region. k ,make The problem is now transformed into selecting a set S of monitoring strategies S that satisfy the quantity budget b from Γ. The problem is redefined as:

[0072] (P2)

[0073]

[0074] The greedy selection phase corresponds to step (4) of the technical solution. The specific implementation method is as follows:

[0075] After step (3), the problem is redefined as P2, which can be proven to be a monotonic submodulus function maximization problem satisfying uniform matroid constraints. A greedy algorithm can be used to iteratively select the element with the largest current gain to add to the final monitoring policy set, let ∈1=∈2+∈2∈ Δ +∈ Δ It can obtain The feasible solution for the approximate ratio, where The specific steps of the algorithm are as follows:

[0076]

[0077]

[0078] The corresponding technical solution step (5) for the candidate monitoring strategy extraction stage is as follows: After obtaining the set of monitoring strategies with performance guarantees through step (4), we use the coordinate set V0 = {u1, u2, ..., u b} represents the current location of all drones, using V1={u1',u2',...,u b '} represents the set of the latest positions obtained in step (4). A bipartite graph G = (V, E) is constructed, where the vertex set V consists of V0 and V1. All vertices in V0 and V1 are connected by edges with weights equal to the distance between the coordinates of the two points. The Kuhn-Munkres algorithm is applied to solve the minimum matching problem to find the edge set with the minimum weight, so as to find the new position coordinates that each UAV will go to. Such matching can guarantee that the total distance traveled is the shortest. Finally, the existing trajectory planning algorithm is applied to safely guide the UAV to the new deployment position. After arriving, the UAV orientation in the corresponding strategy is set, thus realizing the monitoring of the dynamic scene.

Claims

1.A multi-UAV surveillance method considering ground safety, comprising the following steps: (1) Constant approximation stage: using piecewise constant approximation function to approximate the surveillance quality function and the safety reward function as constants in each distance interval in the distance dimension, and discretizing in the angle dimension to approximate the surveillance utility function brought by deploying a UAV in a single angle interval as a constant, and passing the constant approximation scheme into step (2); (2) Region discretization stage: obtaining the positions and orientations of all targets in real time, dividing the two-dimensional plane into a finite number of sub-regions according to the constant approximation scheme in step (1), and passing the discretized sub-regions into step (3); (3) Candidate surveillance strategy extraction stage: applying a candidate surveillance strategy extraction algorithm in each sub-region to find a surveillance strategy equivalent to all surveillance strategies in the region as a candidate surveillance strategy, and passing the obtained candidate surveillance strategy set into step (4); (4) Greedy selection stage: applying a greedy algorithm to iteratively select the element with the maximum current gain on the candidate surveillance strategy set to obtain a performance- guaranteed surveillance strategy set, until the number of obtained surveillance strategies reaches the upper limit of the UAV number budget, and passing the surveillance strategy set into step (5); (5) Path selection and trajectory planning stage: applying the Kuhn-Munkres algorithm to find new position coordinates for each UAV to go to, applying a trajectory planning algorithm to safely let the UAVs go to the new deployment positions, and setting the UAV orientations in the corresponding strategies after arrival to realize surveillance of the dynamic scene; in step (1), the surveillance quality function is used to measure the surveillance quality obtained by the UAV for a target at a specific angle, which is inversely proportional to the square of the distance and negatively correlated with the angle size of the deviation from the target orientation, the surveillance utility function is the integral of the surveillance quality function in the angle dimension, and the safety reward function is used to measure the safety of the target, which is directly related to the distance of the UAV from the target; two piecewise constant approximation schemes are used to approximate the safety reward function and the surveillance quality function, respectively; in step (2), the targets include surveillance targets without considering safety, non-surveillance targets considering safety, and surveillance targets considering safety, different constant approximation schemes are adopted according to the types of targets to divide the two-dimensional plane into a finite number of sub-regions, and at this time the surveillance utility function and the safety reward function of deploying a UAV at all positions in each sub-region for all monitored targets are approximated as constant values; in step (3), for each sub-region, for each target pair that can be monitored by a UAV deployed in the sub-region, first draw a straight line through the coordinates of the target pair, record the intersection points of the straight line and the sub-region, deploy monitoring at each intersection point, adjust the direction of the monitoring to make the right boundary of the fan-shaped region overlap with the straight line, and record the corresponding monitoring strategy; Secondly, draw a circular arc with the monitoring angle of the UAV as the center angle, record the intersection of the circular arc and the sub-region, deploy monitoring on each intersection, adjust the direction of the monitoring so that the left and right boundaries of the sector just pass through the two targets, and record the corresponding monitoring strategy; then randomly select a point on the sub-region, deploy monitoring on the point, and rotate the monitoring direction counterclockwise; when a new target crosses the radius of the sector, record the current monitoring strategy; finally, filter out redundant monitoring strategies to obtain the final monitoring strategy set; in step (5), a bipartite graph is constructed using the current positions of all UAVs and the latest deployment positions in the monitoring strategies obtained in step (4), and the Kuhn-Munkres algorithm is applied to solve the minimum matching problem to find the new position coordinates of each UAV.

Images