Method and device for searching moving target of unmanned aerial vehicle cluster

Abstract

The invention discloses an unmanned aerial vehicle cluster moving target searching method and device, and the method comprises the steps: determining a probability expectation value and a freshness value of a grid where an unmanned aerial vehicle is located when M unmanned aerial vehicles search N targets in a searching region; determining a local probability expectation matrix and a global probability expectation matrix of the grids according to the probability expectation values, and determining a local freshness matrix and a global freshness matrix of the grids according to the freshness values; determining an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight; and according to the action decision matrix, determining a target grid which maximizes the minimum probability value of the freshness value, and according to the position of the target grid, determining the next motion direction of the unmanned aerial vehicle. According to the scheme, the target position probability information and the prior information collected previously are utilized at the same time, the target with the known position and the target with the unknown position can be searched, and therefore target searching can be more efficient and accurate in a real scene.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method and device for searching for moving targets in a cluster of unmanned aerial vehicles. Background Art

[0002] With the continuous development of research on robot platform technology and sensor technology, the role of UAVs in the field of target search is becoming increasingly important, especially UAVs with airborne vision systems such as high-resolution imaging equipment and image processing devices are more widely used.

[0003] At present, the search methods of drones for targets mainly include the cluster coverage method of drones and the directional search method of drones. When the target location is unknown, the target coverage method is generally used to search the target area once or multiple times until the target location is found. When the target location is known, the directional search of the target is generally used to complete it, specifically through the probability information of the target or the movement information of the target.

[0004] However, actual scenarios are often more complex, and we need to consider both the case where the initial position of the target is known and the case where the target position is unknown. Cluster coverage algorithms can run in most scenarios, but when the target position probability information is known, coverage algorithms cannot use the position prior information, which makes the algorithm inefficient and unable to meet time requirements. Directed search algorithms can use previously collected prior information for path planning to achieve efficient search efficiency, but when the algorithm encounters a situation where the target is unknown, it will fail and cannot search for unknown targets. Summary of the invention

[0005] The present invention aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of the present invention proposes a method for searching for moving targets in a drone cluster, the method comprising:

[0006] When M drones search for N targets in the search area, determine the expected probability value and freshness value of the grid where the drone is located, the expected probability value represents the probability value of each target existing in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N;

[0007] Determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value;

[0008] Based on the local probability expectation matrix and the global probability expectation matrix, a probability expectation decision matrix is ​​calculated, and based on the local freshness matrix and the global freshness matrix, a freshness decision matrix is ​​calculated; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone;

[0009] Determine the current exploration task weight and search task weight according to the search parameters, the linear function and the random noise;

[0010] Determining an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight, and the search task weight;

[0011] The target grid which minimizes the freshness value and maximizes the probability value is determined according to the action decision matrix, and the next movement direction of the UAV is determined according to the position of the target grid.

[0012] Optionally, determining the expected probability value of the grid where the drone is located includes:

[0013] Based on the estimated positions of the respective targets, the probability values ​​of the respective targets existing in the respective grids within the search area are determined, and a probability expectation matrix of the targets composed of the probability values ​​of the respective grids is obtained;

[0014] Obtaining a preset target value matrix, wherein each element in the target value matrix represents the search value of each target;

[0015] The product of the probability expectation matrix and the target value matrix is ​​used as the probability expectation value of the grid.

[0016] Optionally, the calculating a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix includes:

[0017] Obtaining the maximum and minimum values ​​of the elements in the global probability expectation matrix;

[0018] Determine the difference between each element in the global probability expectation matrix and the minimum value to obtain a plurality of first differences;

[0019] Determine a second difference between the maximum value and the minimum value, and determine a sum of the second difference and a preset adjustment parameter to obtain a first sum;

[0020] Determine a second difference between 1 and a preset probability expectation global decision coefficient, and determine a product of the second difference and the local probability expectation matrix;

[0021] A quotient of the first difference and the first sum is determined, and a sum of the quotient and the product is used as a probability expectation decision matrix of the grid.

[0022] Optionally, the calculating a freshness decision matrix based on the local freshness matrix and the global freshness matrix includes:

[0023] Obtaining the maximum and minimum values ​​of the elements in the global freshness matrix;

[0024] Determine the difference between each element in the global freshness matrix and the minimum value to obtain a plurality of first differences;

[0025] Determine a second difference between the maximum value and the minimum value, and determine a sum of the second difference and a preset adjustment parameter to obtain a first sum;

[0026] Determine a second difference between 1 and a preset global decision coefficient of freshness, and determine a product of the second difference and the local freshness matrix;

[0027] A quotient of the first difference and the first sum is determined, and a sum of the quotient and the product is used as a freshness decision matrix for the grid.

[0028] Optionally, determining the current exploration task weight and search task weight according to the search parameter, the linear function and the random noise includes:

[0029] Add the preset step length to the search parameters of the previous search to obtain the current search parameters;

[0030] Determining a current exploration task weight according to the current search parameters and random noise;

[0031] The difference between 1 and the current exploration task weight is taken as the current search task weight.

[0032] Optionally, determining the action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight includes:

[0033] Determine the product of the probability expectation decision matrix and the exploration task weight to obtain a first product;

[0034] Determine the product of the freshness decision matrix and the search task weight to obtain a second product;

[0035] The difference between the first product and the second product is determined to obtain an action decision matrix.

[0036] Optionally, determining the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix includes:

[0037] The minimum element value in the action decision matrix is ​​obtained, and the grid where the minimum element value is located is determined as the target grid that minimizes the freshness value and maximizes the probability value.

[0038] A second aspect of the present invention provides a drone cluster moving target search device, the device comprising:

[0039] The first determination module is used to determine the probability expectation value and freshness value of the grid where the drone is located when M drones search for N targets in the search area, the probability expectation value represents the probability value of each target existing in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N;

[0040] A second determination module, used to determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and to determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value;

[0041] A decision matrix calculation module, used to calculate a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix, and to calculate a freshness decision matrix based on the local freshness matrix and the global freshness matrix; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone;

[0042] A weight determination module, used to determine the current exploration task weight and search task weight according to the search parameters, linear function and random noise;

[0043] An action decision matrix determination module, used to determine an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight;

[0044] The movement direction determination module is used to determine the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix, and determine the next movement direction of the drone according to the position of the target grid.

[0045] The third aspect of the present invention proposes an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the drone cluster moving target search method as described in the first aspect.

[0046] The fourth aspect of the present invention proposes a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for searching for moving targets in a drone cluster as described in the first aspect is implemented.

[0047] The present invention has the following beneficial effects:

[0048] In an embodiment of the present invention, when M drones search for N targets in a search area, the probability expectation value and freshness value of the grid where the drone is located are determined, the probability expectation value represents the probability value of the existence of each target in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N; according to the probability expectation value, the local probability expectation matrix and the global probability expectation matrix of the grid are determined, and according to the freshness value, the local freshness matrix and the global freshness matrix of the grid are determined; based on the local probability expectation matrix and the global probability expectation matrix, the probability expectation decision matrix is ​​calculated, and based on the The local freshness matrix and the global freshness matrix are used to calculate the freshness decision matrix; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the UAV; the current exploration task weight and the search task weight are determined according to the search parameters, linear functions and random noise; the action decision matrix is ​​determined according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight; the target grid that minimizes the freshness value and maximizes the probability value is determined according to the action decision matrix, and the next movement direction of the UAV is determined according to the location of the target grid. This scheme uses the probability expectation matrix to represent the probability value of the existence of the target in each grid, and uses the freshness matrix to represent the state of the grid being visited by the UAV. At the same time, it uses the target location probability information and the prior information collected before. It can search for targets with known locations as well as targets with unknown locations, which makes the algorithm perform better in real scenarios and the search for targets more efficient and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of the steps of a method for searching for moving targets of a drone cluster provided by an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of a UAV motion space provided by an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the effect of the height of a drone on the field of view provided by an embodiment of the present invention;

[0052] Figure 4 It is a schematic diagram of a sensor field of view range detection method provided by an embodiment of the present invention;

[0053] Figure 5 is a probability expectation graph provided by an embodiment of the present invention;

[0054] Figure 6 It is a structural block diagram of a UAV cluster moving target search device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values ​​may be based on additional conditions or values ​​beyond the described values ​​in practice.

[0057] Figure 1 1 is a flowchart of a method for searching for moving targets in a drone swarm provided by an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0058] Step 101: When M drones search for N targets in a search area, determine the expected probability value and freshness value of the grid where the drones are located, wherein the expected probability value represents the probability value of each target existing in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N.

[0059] In the embodiment of the present invention, the drone map construction is mainly divided into grid map, point map, and Thiessen polygon map. Among them, Thiessen polygon has excellent characteristics: each polygon has only one discrete point inside, and each polygon is a convex polygon.

[0060] In the field of drone search, only one drone is used in each polygon, and each drone moves within its own polygon, so that every area can be searched by drones, and drones will not collide with each other, so that the task can be executed more efficiently.

[0061] The present invention adopts the specified plant evolution to solve the problem of drone search. In the model, a large drone scans the whole, and each small drone will explore and scan its own area. When scanning the target, it will increase its own "fitness value". After scanning, it will be sorted according to the "fitness value". The large drone will explore at a position with a higher "fitness value", and other small drones will also fly to these convex polygons with high fitness values, and distribute them again in the search area through normal distribution to search for the target. Through the search of the large drone and the "fitness value" of the small drone, the location of the target is searched. Each convex polygon can be artificially planned or dynamically planned. The Markov method can be used to make the drone cluster search for the target. The key attributes of Markov, such as the state set and the decision set, can be specified. The state is used to decide the next activity of the drone, and the drone decides the activity at the next moment through multiple times. Each grid will also move with the drone, and each grid includes an environmental information map, a target information map, and a digital pheromone map.

[0062] The search area is rasterized into a two-dimensional bounded and regular area. Each grid is the minimum path of the search area. Each grid contains freshness value information and probability information of different targets. The number of drones in the entire mission will be greater than the number of targets. When a target is discovered, it will be discussed whether the target is coordinated through a drone cluster so that the target will be continuously tracked (i.e., ignoring the target's next action), or whether the movement of the drone is continuously tracked only through the weight distribution method. The initial probability information of the target will initialize the map information based on the known prior information. Of course, the target probability will be initialized into different probability heat information depending on the value of the target, so that tasks with higher value will be discovered faster.

[0063] There are many types of drones. According to the flight platform configuration, they can be divided into multi-rotor drones, fixed-wing drones, and unmanned helicopters. Different configurations of drones determine the flight properties of different drones. Among them, multi-rotor drones are more common, and their flight is easier to control, and the next flight position and direction are independent of the previous flight position and direction.

[0064] Figure 2 It is a schematic diagram of a UAV motion space provided by an embodiment of the present invention.

[0065] exist Figure 2 In the figure, the multi-rotor drone can move in any direction, regardless of the direction and speed of the drone, and the drone's motion space is A.

[0066] Drone sensors can detect the motion state of drones, including detecting the rotation and translation of drones; the location state of drones, mainly GPS, which obtains the location information of drones in the world through satellite signals; and the direction and height information of drones. However, this article mainly focuses on the search and coverage tasks of drones and models visual sensors.

[0067] Figure 3 The present invention provides a schematic diagram of the effect of the height of a drone on the field of view.

[0068] exist Figure 3 In the actual drone flight scene, each drone carries a sensor to scan the drone area below the drone. Different drone sensors have different detection ranges and different precisions. The flight altitude of the drone will also affect the range and precision of the sensor. MSAA or Monte Carlo method is used for edge calculation at the edge of FOV.

[0069] Figure 4 It is a schematic diagram of a sensor field of view range detection method provided by an embodiment of the present invention.

[0070] exist Figure 4 In , the sensor field of view (FOV) detection method is adopted. The detection range of the drone is centered on itself, and the sensor detection radius is . However, as the detection distance gets farther and farther away from the drone, the target is in the field of view of the drone, and the regional detection degree is (FOV is the maximum detection range of the drone). It will gradually decrease with the increase of distance, indicating all grids under the detection range of the i-th drone at position L.

[0071]

[0072] Among them, r is the accurate detection radius, that is, when the target is within the range of r, the target can be detected more accurately, and R is the maximum detection radius. The visual sensor may be distorted when it is at the edge of the camera, that is, distortion. Distortion will affect the results, and the closer to the edge, the easier it is to make mistakes, until it exceeds the maximum detection range FOV.

[0073] Define N moving targets in the search area as Target, and for each target j, the position at time t is LoT j (t), and define the target speed to be less than the UAV speed, that is, when the UAV is moving, the target is not necessarily moving.

[0074] Define M drones in the search area as UAVS, forming a drone cluster. The drone cluster jointly maintains the search area information. For each drone i, the position at time t is LoT j(t),Each drone carries a GPS sensor and can autonomously determine its current location information.Each drone has its own FOV.

[0075] The entire search area is gridded into a two-dimensional bounded and regular area. Each grid is the minimum path in the search area. Each grid contains freshness value information and probability information of different targets. The existence probability value of each target in the grid constitutes the probability expectation value TD of the grid.

[0076] The freshness value F represents the status of the grid being visited by the drone in history. A high freshness value indicates that the grid has been visited by the drone recently, while a low freshness value means that the drone has not visited the grid recently or has never visited the grid.

[0077] The freshness map maintained by the drone cluster will decrease over time. Some areas will be discovered by the drone cluster. The discovered areas will have their freshness values ​​initialized to 1 and will be defined as newly discovered areas.

[0078] The expected probability value TD of the grid is related to the value matrix of the target.

[0079] F l Indicates the freshness value, F l The update formula is:

[0080]

[0081] Among them, ΔF represents the attenuation value, τ represents the freshness threshold, which is generally set to 0.3. l represents the position of the current grid. When l is in the field of view of the drone, that is, l∈FOV, F l is set to 1; when When F l Attenuation. When F l ≤τ, then take F l .

[0082] Step 102: Determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value.

[0083] The trajectory of the drone can be dynamically planned by the freshness value and the expected value of probability. Specifically, the local freshness values ​​of each adjacent grid of the current grid are determined to obtain a local freshness matrix composed of the local freshness values ​​of the adjacent grids. The global freshness values ​​of each adjacent grid of the grid are determined to obtain a global freshness matrix composed of the global freshness values ​​of the adjacent grids.

[0084] Determine the local probability expectation values ​​of each adjacent grid of the current grid, and obtain a local probability expectation matrix composed of the local probability expectation values ​​of the adjacent grids. Determine the global probability expectation values ​​of each adjacent grid of the grid, and obtain a global probability expectation matrix composed of the global probability expectation values ​​of the adjacent grids.

[0085] Each grid corresponds to a local freshness matrix and a global freshness matrix, and each grid corresponds to a local probability expectation matrix and a global probability expectation matrix.

[0086] Specifically, the local probability expectation matrix TD local The calculation method is as follows:

[0087]

[0088] Each element of the matrix Represents the expected value of probability in motion space A.

[0089]

[0090] Among them, s represents the number of grids in the field of view of the drone at the next moment, FOV t+1 It represents the grid corresponding to the detection range of the current i-th UAV at time t+1. The possible position of the UAV at the next moment can be obtained:

[0091] FOV t+1 =FOV(LoU i (t+1)) (5)

[0092] LoU i (t) represents the current location of the drone, LoU i (t+1) represents the possible position of the drone at the next moment, which is calculated as follows:

[0093]

[0094] In addition, P(FOV i (L) represents the probability that the i-th UAV can accurately identify the target within the detection range of position L, which is calculated as follows:

[0095]

[0096] Among them, r represents the accurate detection radius, that is, when the target is within the range of r, the target can be detected more accurately, R represents the maximum detection radius, and Ri represents the current location of the drone.

[0097] Global probability expectation matrix TD hlobal It represents the expected value of probability calculated through the entire map. The specific calculation formula is as follows:

[0098]

[0099] Among them, i, j represent the location of the UAV, m, n represent the size of the search area, and TD represents the expected probability value of the grid.

[0100] Local freshness matrix F local The calculation method is as follows:

[0101]

[0102] Each element of the matrix Represents the freshness value in motion space A:

[0103]

[0104] Among them, s represents the number of grids in the field of view of the drone at the next moment, FOV t+1 It represents the grid corresponding to the detection range of the current i-th UAV at time t+1, which can be obtained through the possible position of the UAV at the next moment.

[0105] Global freshness matrix F global Represents the freshness value calculated through the entire map. The specific calculation formula is as follows:

[0106]

[0107] Among them, i, j represent the location of the drone, m, n represent the size of the search area, and F represents the expected probability value of the grid.

[0108] Step 103: Calculate a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix, and calculate a freshness decision matrix based on the local freshness matrix and the global freshness matrix; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone.

[0109] Specifically, the probability expectation decision matrix is ​​calculated based on the local probability expectation matrix and the global probability expectation matrix. And the freshness decision matrix is ​​determined based on the local freshness matrix and the global freshness decision matrix. The probability expectation decision matrix and the freshness decision matrix comprehensively represent the possibility of the existence of the target in each grid. Therefore, the probability expectation decision matrix and the freshness decision matrix can determine the next movement direction of the drone.

[0110] Step 104: Determine the current exploration task weight and search task weight according to the search parameters, the linear function and the random noise.

[0111] Each drone needs to perform a search mission or an exploration mission. The search mission is to search for areas with a higher probability of target existence when the probability of target existence is known. The exploration mission is to explore locations with an unknown probability of target existence to obtain information about the target.

[0112] In order to balance the exploration task and the search task, the exploration task weight and the search task weight are set. When the map information is initially unknown or the target information is less and the drone needs to explore, the exploration task weight should be greater, and when the map information is more, the search task weight should be greater.

[0113] Step 105: Determine an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight, and the search task weight.

[0114] The final motion decision of the drone is determined by the probability expectation decision matrix, the freshness decision matrix, the exploration task weight, and the search task weight.

[0115] Step 106: Determine the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix, and determine the next movement direction of the drone according to the position of the target grid.

[0116] The area with the smallest freshness value and the largest probability value indicates that the time since the last drone search in this area is long and the probability of the target existing is high. The possibility of finding the target in this area is the highest. Therefore, the target grid with the smallest freshness value and the largest probability value is determined according to the action decision matrix.

[0117] Specifically, the direction of the line between the current grid and the target grid is used as the next movement direction of the drone. When all targets have been searched, or the drone's movement step has reached the upper limit, the task ends, and the drone's search step and the number of searched targets are counted.

[0118] This solution comprehensively considers multiple key issues in target search and tracking, improves search efficiency and accuracy, and is suitable for UAV missions in complex unknown environments. It has the following advantages:

[0119] (1) Adaptability to complex scenarios: The algorithm can flexibly cope with various practical search challenges in view of the complex motion and diverse initial probabilities of the target;

[0120] (2) Global and local balance: The FTD algorithm balances coverage and search in the search, taking into account both global and local perspectives, making it perform well in real-world scenarios;

[0121] (3) Collaborative search: Although the algorithm maintains collaborative nature when the UAVs are flying, future optimization plans include introducing more rigorous collaborative algorithms to improve search efficiency and path planning.

[0122] In summary, in an embodiment of the present invention, when M drones search for N targets in a search area, the probability expectation value and freshness value of the grid where the drone is located are determined, the probability expectation value represents the probability value of the existence of each target in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N; according to the probability expectation value, the local probability expectation matrix and the global probability expectation matrix of the grid are determined, and according to the freshness value, the local freshness matrix and the global freshness matrix of the grid are determined; based on the local probability expectation matrix and the global probability expectation matrix, the probability expectation decision matrix is ​​calculated, and based on Based on the local freshness matrix and the global freshness matrix, the freshness decision matrix is ​​calculated; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone; the current exploration task weight and search task weight are determined according to the search parameters, linear functions and random noise; the action decision matrix is ​​determined according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight; the target grid that minimizes the freshness value and maximizes the probability value is determined according to the action decision matrix, and the next movement direction of the drone is determined according to the location of the target grid. This scheme uses the probability expectation matrix to represent the probability value of the existence of the target in each grid, and uses the freshness matrix to represent the state of the grid being visited by the drone. At the same time, it uses the target location probability information and the prior information collected before, and can search for targets with known locations as well as targets with unknown locations. This allows the algorithm to perform better in real scenarios and search for targets more efficiently and accurately.

[0123] In a possible implementation manner, determining the expected probability value of the grid where the drone is located in step 101 includes:

[0124] Step 1011: Based on the estimated positions of the respective targets, the probability values ​​of the presence of the respective targets in the respective grids within the search area are determined, and a probability expectation matrix of the targets composed of the probability values ​​of the respective grids is obtained.

[0125] Specifically, the probability expectation matrix of the target is calculated as follows:

[0126]

[0127] Among them, (i, j) represents the estimated position of the target, δ represents the variance, and L i , L j Respectively represent the length and width of the search area.

[0128] Step 1012: Obtain a preset target value matrix, wherein each element in the target value matrix represents the search value of each target.

[0129] Large drones fly higher and carry optoelectronic instruments with a wider detection range. They have greater search capabilities, but the error may also be greater. There may be areas where there are no targets but targets are found, and areas where there are targets but targets are not found. At this time, the small drones in this solution are needed to search these areas in detail so that the areas can be searched accurately. However, since the search value of each target is different, high-value targets should be searched first. Therefore, the search value of the target is used as one of the decision factors to determine the direction of the drone's movement.

[0130] The target value is TV k =Value(i,j),TV k represents the value of the kth target, and Value(i,j) represents the value of the target at the current position in the i-th row and j-th column.

[0131] Step 1013: The product of the probability expectation matrix and the target value matrix is ​​used as the probability expectation value of the grid.

[0132] The probability expectation matrix TD(i,j) of the grid is the product of the probability expectation matrix and the target value matrix, as follows:

[0133] TD(i,j)=Value(i,j)×P(i,j) (13)

[0134] Among them, Value(i,j) represents the target value matrix, and P(i,j) represents the probability expectation matrix.

[0135] In a possible implementation manner, calculating a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix in step 103 includes:

[0136] Step 1031, obtaining the maximum value and the minimum value of the elements in the global probability expectation matrix;

[0137] Step 1032: determine the difference between each element in the global probability expectation matrix and the minimum value to obtain a plurality of first differences;

[0138] Step 1033: determine a second difference between the maximum value and the minimum value, and determine the sum of the second difference and a preset adjustment parameter to obtain a first sum;

[0139] Step 1034, determining a second difference between 1 and a preset probability expectation global decision coefficient, and determining a product of the second difference and the local probability expectation matrix;

[0140] Step 1035: determine the quotient of the first difference and the first sum, and use the sum of the quotient and the product as the probability expectation decision matrix of the grid.

[0141] In step 1031-step 1035, the probability expectation decision matrix is ​​calculated as follows:

[0142]

[0143] Among them, max(TD global ) represents the maximum value of the elements in the global freshness matrix, min(TD global ) represents the minimum value of the elements in the global freshness matrix. ε represents the preset adjustment parameter, which is an infinitesimal value. β TD Represents the probability expectation global decision coefficient, through β TD To balance the expected value of global probability and the expected value of local probability.

[0144] In a possible implementation, calculating the freshness decision matrix based on the local freshness matrix and the global freshness matrix in step 103 includes:

[0145] Step 1036: Obtain the maximum and minimum values ​​of the elements in the global freshness matrix;

[0146] Step 1037, determining the difference between each element in the global freshness matrix and the minimum value, and obtaining a plurality of first differences;

[0147] Step 1038: determining a second difference between the maximum value and the minimum value, and determining a sum of the second difference and a preset adjustment parameter to obtain a first sum;

[0148] Step 1039, determining a second difference between 1 and a preset global decision coefficient of freshness, and determining a product of the second difference and the local freshness matrix;

[0149] Step 1040: determine the quotient of the first difference and the first sum, and use the sum of the quotient and the product as the freshness decision matrix of the grid.

[0150] In step 1036-step 1040, the calculation method of the freshness decision matrix is ​​as follows:

[0151]

[0152] Among them, max(F global ) represents the maximum value of the elements in the global freshness matrix, min(F global ) represents the minimum value of the elements in the global freshness matrix. ε represents the preset adjustment parameter, which is an infinitesimal value. α FRepresents the global decision coefficient of freshness, through α F To balance the global freshness value and the local freshness value.

[0153] In a possible implementation, step 104 includes:

[0154] Step 1041, adding a preset step length to the search parameters of the previous search to obtain the current search parameters;

[0155] Step 1042: Determine the current exploration task weight according to the current search parameters and random noise;

[0156] Step 1043: take the difference between 1 and the current exploration task weight as the current search task weight.

[0157] In step 1041 to step 1043, the calculation method of the exploration task weight α and the search task weight β is as follows:

[0158]

[0159] β = 1-α (16)

[0160] Where x represents the current search parameter, x∈[-1,1], and x increases with the number of searches according to the preset step size. Subtract the preset step size from the search parameter of the previous search to obtain the current search parameter x. For example, if the preset step size is 0.1, in the first search, x is -1, then in the second search, x is -1+0.1=-0.9, and so on.

[0161] In addition, ε represents random noise.

[0162] In a possible implementation, step 105 includes:

[0163] Step 1051: determine the product of the probability expectation decision matrix and the exploration task weight to obtain a first product;

[0164] Step 1052: determine the product of the freshness decision matrix and the search task weight to obtain a second product;

[0165] Step 1053: determine the difference between the first product and the second product to obtain an action decision matrix.

[0166] In step 1051-step 1053, the action decision matrix Action t+1 The calculation method is as follows:

[0167] Action t+1 =argmin(α×DF-β×DTD) (17)

[0168] Among them, α represents the exploration task weight, DF represents the probability expectation decision matrix, β represents the search task weight, and DTD represents the freshness decision matrix.

[0169] In a possible implementation manner, determining the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix in step 106 includes:

[0170] The minimum element value in the action decision matrix is ​​obtained, and the grid where the minimum element value is located is determined as the target grid that minimizes the freshness value and maximizes the probability value.

[0171] In the embodiment of the present invention, the grid with the smallest freshness value and the largest probability value indicates that the grid has been searched by the drone for a long time and the probability of the target existing is large. The possibility of finding the target in this area is the greatest, and the smallest element value in the action decision matrix has the characteristics of the smallest freshness value and the largest probability value. Therefore, the next movement direction of the drone is determined according to the grid where the smallest element value in the action decision matrix is ​​located.

[0172] Simulation experiment:

[0173] The simulation environment takes an area of ​​40,000x40,000, with each grid of 100m. Therefore, the search area is a 400x400 grid, with M drones and N targets. The field of view of each drone is r=2, R=3. When a drone observes a target, the target is automatically tracked by other drones, and the search drone will no longer search for the target. The freshness value is initialized to 0 before the search, and TD will be given the first approximate probability by the large drone.

[0174] Figure 5 It is a probability expectation graph provided by an embodiment of the present invention.

[0175] like Figure 5 , there are 5 targets. From the probability expectation graph, we can see that targets 1-4 are identified by large drones, target 5 is not scanned by large drones, and there are two misjudgment areas, which will interfere with the search of small drones. In the experiment, there is always one target that is not identified, and there are always two misjudgment areas. And the drone moves one grid each time, and the target moves 0.1 grid each time, that is, the drone moves ten times, and the target moves once, and the maximum number of flight steps is 1000 steps.

[0176] The calculation criteria are the discovery rate, that is, the average step length when all targets are discovered, and the number of targets discovered, that is, the number of targets discovered under a limited step length. Suppose the average search rate is:

[0177]

[0178] N find is the number of targets found, T is the time step when all targets are found, and if not all targets are found, it is the maximum step.

[0179] The present invention proposes a scenario that is closer to reality for the problem of searching for moving targets. The motion characteristics of moving targets are more complex, and the initial probabilities of targets are more diverse, including known target positions, unknown target positions, and erroneous target positions. At the same time, the scene modeling is also closer to the real search scene and has a limited field of view. In this scenario, the FTD algorithm is proposed, which can search for targets with known positions as well as targets with unknown positions. This allows the algorithm to be flexibly used in different scenarios and achieve good results. At the same time, since the algorithm takes into account coverage and search, and has both global and local fields of view, the algorithm performs better in more realistic scenarios. However, since this article adopts a global field of view, it may have a larger time complexity in global calculations. Perhaps the operating efficiency can be improved by narrowing the global field of view. At the same time, when the algorithm is flying, the coordination between drones is relatively loose, and a more rigorous algorithm can be used for collaborative search, which greatly improves the search efficiency and path planning efficiency.

[0180] Figure 6 200 is a structural block diagram of a drone cluster moving target search device provided by an embodiment of the present invention. The device 200 comprises:

[0181] The first determination module 201 is used to determine the probability expectation value and freshness value of the grid where the M drones are located when searching for N targets in the search area, wherein the probability expectation value indicates the probability value of each target existing in the grid, and the freshness value indicates the state of the grid being visited by the M drones; M<N;

[0182] A second determination module 202, configured to determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and to determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value;

[0183] A decision matrix calculation module 203 is used to calculate a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix, and to calculate a freshness decision matrix based on the local freshness matrix and the global freshness matrix; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone;

[0184] A weight determination module 204, for determining a current exploration task weight and a search task weight according to a search parameter, a linear function and random noise;

[0185] An action decision matrix determination module 205 is used to determine an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight;

[0186] The movement direction determination module 206 is used to determine the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix, and determine the next movement direction of the drone according to the location of the target grid.

[0187] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0188] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0189] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0190] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for searching moving targets of drone clusters, characterized in that: The method comprises: When M drones search for N targets in the search area, determine the expected probability value and freshness value of the grid where the drone is located, the expected probability value represents the probability value of each target existing in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N; Determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value; Based on the local probability expectation matrix and the global probability expectation matrix, a probability expectation decision matrix is ​​calculated, and based on the local freshness matrix and the global freshness matrix, a freshness decision matrix is ​​calculated; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone; Determine the current exploration task weight and search task weight according to the search parameter, the linear function and the random noise; Determining an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight, and the search task weight; The target grid which minimizes the freshness value and maximizes the probability value is determined according to the action decision matrix, and the next movement direction of the UAV is determined according to the position of the target grid.

2. The method according to claim 1, characterized in that The determining of the expected probability value of the grid where the drone is located includes: Based on the estimated positions of the respective targets, the probability values ​​of the respective targets existing in the respective grids within the search area are determined, and a probability expectation matrix of the targets composed of the probability values ​​of the respective grids is obtained; Obtaining a preset target value matrix, wherein each element in the target value matrix represents the search value of each target; The product of the probability expectation matrix and the target value matrix is ​​used as the probability expectation value of the grid.

3. The method according to claim 1, characterized in that The calculating of a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix comprises: Obtaining the maximum and minimum values ​​of the elements in the global probability expectation matrix; Determine the difference between each element in the global probability expectation matrix and the minimum value to obtain a plurality of first differences; Determine a second difference between the maximum value and the minimum value, and determine a sum of the second difference and a preset adjustment parameter to obtain a first sum; Determine a second difference between 1 and a preset probability expectation global decision coefficient, and determine a product of the second difference and the local probability expectation matrix; A quotient of the first difference and the first sum is determined, and a sum of the quotient and the product is used as a probability expectation decision matrix of the grid.

4. The method according to claim 1, characterized in that: The calculating of the freshness decision matrix based on the local freshness matrix and the global freshness matrix comprises: Obtaining the maximum and minimum values ​​of the elements in the global freshness matrix; Determine the difference between each element in the global freshness matrix and the minimum value to obtain a plurality of first differences; Determine a second difference between the maximum value and the minimum value, and determine a sum of the second difference and a preset adjustment parameter to obtain a first sum; Determine a second difference between 1 and a preset global decision coefficient of freshness, and determine a product of the second difference and the local freshness matrix; A quotient of the first difference and the first sum is determined, and a sum of the quotient and the product is used as a freshness decision matrix for the grid.

5. The method according to claim 1, characterized in that The determining of the current exploration task weight and the search task weight according to the search parameter, the linear function and the random noise includes: Add the preset step length to the search parameters of the previous search to obtain the current search parameters; Determining a current exploration task weight according to the current search parameters and random noise; The difference between 1 and the current exploration task weight is taken as the current search task weight.

6. The method according to claim 1, characterized in that Determining the action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight includes: Determine the product of the probability expectation decision matrix and the exploration task weight to obtain a first product; Determine the product of the freshness decision matrix and the search task weight to obtain a second product; The difference between the first product and the second product is determined to obtain an action decision matrix.

7. The method according to claim 6, characterized in that The step of determining the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix includes: The minimum element value in the action decision matrix is ​​obtained, and the grid where the minimum element value is located is determined as the target grid that minimizes the freshness value and maximizes the probability value.

8. A drone cluster moving target search device, characterized in that: The device comprises: The first determination module is used to determine the probability expectation value and freshness value of the grid where the drone is located when M drones search for N targets in the search area, the probability expectation value represents the probability value of each target existing in the grid, and the freshness value represents the state of the grid being visited by the M drones; M<N; A second determination module, used to determine a local probability expectation matrix and a global probability expectation matrix of the grid according to the probability expectation value, and to determine a local freshness matrix and a global freshness matrix of the grid according to the freshness value; A decision matrix calculation module, used to calculate a probability expectation decision matrix based on the local probability expectation matrix and the global probability expectation matrix, and to calculate a freshness decision matrix based on the local freshness matrix and the global freshness matrix; the probability expectation decision matrix and the freshness decision matrix are used to determine the next movement direction of the drone; A weight determination module, used to determine the current exploration task weight and search task weight according to the search parameters, linear function and random noise; An action decision matrix determination module, used to determine an action decision matrix according to the probability expectation decision matrix, the freshness decision matrix, the exploration task weight and the search task weight; The movement direction determination module is used to determine the target grid that minimizes the freshness value and maximizes the probability value according to the action decision matrix, and determine the next movement direction of the drone according to the position of the target grid.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the drone cluster moving target search method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method for searching for moving targets in a cluster of unmanned aerial vehicles as described in any one of claims 1 to 7 is implemented.

Images