CN116182838A

CN116182838A - Plant protection unmanned aerial vehicle global path optimization method based on point cloud map

Info

Publication number: CN116182838A
Application number: CN202310306113.3A
Authority: CN
Inventors: 杨金龙; 王景岩; 潘正阳; 张力璇; 杨坤; 谢小曼
Original assignee: Jiangnan University
Current assignee: Jiangnan University
Priority date: 2023-03-27
Filing date: 2023-03-27
Publication date: 2023-05-30

Abstract

The invention discloses a plant protection unmanned aerial vehicle global path optimization method based on a point cloud map, which relates to the field of intelligent agricultural machinery and comprises the following steps: collecting three-dimensional point cloud data of a terrain to be worked on the basis of multiple sensors, and generating a three-dimensional point cloud map through filtering and data fusion processing; the method comprises the steps of dividing, planning regional paths and finally connecting the regions, and performing global path planning on the delineated region to be worked, so that unmanned aerial vehicle energy consumption is reduced as much as possible, and terrain compatibility and unmanned aerial vehicle operation safety are improved while the operation region can be completely covered. The constructed three-dimensional point cloud map can realize one-time composition warehouse entry of the terrain to be operated, is used for multiple times of unmanned aerial vehicle path planning, realizes foolproof operation of unmanned aerial vehicle agricultural operation, and improves the universality trend of agricultural operation by using the unmanned aerial vehicle.

Description

Plant protection unmanned aerial vehicle global path optimization method based on point cloud map

Technical Field

The invention relates to the field of intelligent agricultural machinery, in particular to a plant protection unmanned aerial vehicle global path optimization method based on a point cloud map.

Background

The existing plant protection unmanned plane mainly has two operation modes: one is simply based on manual control to carry out agricultural operation, and the unmanned aerial vehicle cannot carry out autonomous operation according to the terrain to be operated, so that time and labor are wasted, and the technical requirements on operators are extremely high; another plant protection unmanned plane uses an intelligent chip and a sensor to conduct real-time path planning autonomous operation, and the method reduces manual operation, but has a plurality of defects. Firstly, the problem of energy consumption is solved, the real-time path planning requires that the sensor is opened in real time to acquire data, and the data is processed by a chip in real time, so that the energy consumption is obviously increased compared with the traditional manual control. Secondly, the path problem is that a plurality of uncertain factors exist in real-time path planning, the path is most likely to be limited to an optimal solution by a real-time processing algorithm, so that the unmanned aerial vehicle generates a fluctuation oscillation phenomenon at one point, even if the real-time path planning algorithm is perfect, the path repetition or coverage omission phenomenon is likely to occur, and for crop pesticide spraying, the repeated or omission can cause excessive pesticide spraying or pesticide spraying omission, which is extremely disadvantageous to crops.

Disclosure of Invention

Aiming at the problems and the technical requirements, the inventor provides a plant protection unmanned aerial vehicle global path optimization method based on a point cloud map, and the three-dimensional point cloud map of the terrain to be worked is constructed at one time, so that the three-dimensional point cloud map can be used for multiple times in the path planning process, and the energy consumption is reduced; and presetting an optimal operation path according to the global information by using an intelligent path planning algorithm, wherein the path avoids the phenomena of repeated coverage and coverage omission. The technical scheme of the invention is as follows:

a plant protection unmanned aerial vehicle global path optimization method based on a point cloud map comprises the following steps:

constructing a three-dimensional point cloud map of the terrain to be operated;

the method comprises the steps of defining a to-be-worked area of the plant protection unmanned aerial vehicle in a three-dimensional point cloud map, and carrying out plane fitting on point cloud data in the to-be-worked area to obtain each plane block;

dividing each plane block into subareas, carrying out reciprocating path planning in each subarea, and carrying out backtracking path planning among subareas by utilizing an optimized backtracking method to obtain a full-coverage path of the plant protection unmanned aerial vehicle in the plane block;

and planning paths among the plane blocks by using an optimized A-algorithm to obtain the full-coverage optimal path of the plant protection unmanned aerial vehicle in the area to be worked.

The beneficial technical effects of the invention are as follows:

in order to meet the requirement that a plant protection unmanned aerial vehicle can adapt to complex terraces such as a terrace with high and low fluctuation, the application constructs a farmland three-dimensional point cloud map by acquiring farmland topography information and further selecting important data to fuse, and calculates an optimal operation path as soon as possible through a series of path planning algorithms based on the three-dimensional point cloud map, so that the complexity of the path planning algorithm is reduced as much as possible while the path planning accuracy is ensured, the waiting time of a user is reduced, and the use experience and satisfaction degree of the user are improved.

The global path optimization method integrally adopts the dividing and controlling idea, the delineated area to be worked is divided into a plurality of plane blocks, each plane block is divided into a plurality of sub-areas through an optimized decomposition method, and reciprocating path planning is further carried out in each sub-area. Firstly, the method for dividing and controlling can improve the working efficiency of unmanned aerial vehicle agricultural operation, and reduce the energy consumption problem and the safety problem caused by the terrain height difference as much as possible; secondly, the segmented planning in the subareas can effectively complete obstacle avoidance of the unmanned aerial vehicle, the condition that a backtracking path crosses the subareas can be avoided by utilizing an optimized backtracking method, the condition that an adapter subarea is not solved due to incapability of backtracking is avoided, and repetition and omission of unmanned aerial vehicle traversal are effectively prevented; finally, compared with real-time planning, the method for constructing the map and planning the path can avoid the influence caused by network delay, can avoid the fluctuation of the unmanned aerial vehicle at one point caused by the sinking local optimum, can evaluate a plurality of parameters such as working time, energy consumption and the like in advance, and reasonably arrange the source changing points according to the parameters.

According to the three-dimensional point cloud map, one-time composition warehouse entry can be realized, the unmanned aerial vehicle is used for multiple times in path planning, the energy consumption of the unmanned aerial vehicle is reduced, the use difficulty of the unmanned aerial vehicle is reduced, the fool operation of the unmanned aerial vehicle agricultural operation is truly realized, and the universality trend of using the unmanned aerial vehicle for agricultural operation is improved.

Drawings

FIG. 1 is a flow chart of the method of the present application;

FIG. 2 is a voxel filtering front-to-back contrast;

FIG. 3 is a flow chart of data fusion of a point cloud with an image;

FIG. 4 (a) is a diagram of raw point cloud data;

FIG. 4 (b) is a point cloud data fusion effect diagram;

FIG. 5 (a) is a plan block topography;

FIG. 5 (b) is a sub-region exploded effect diagram;

FIG. 6 is an effect diagram of a reciprocating, optimized backtracking path plan;

FIG. 7 is an effect diagram of an optimized A-algorithm for three-dimensional environmental path planning;

FIG. 8 is a schematic diagram of a server architecture for constructing a three-dimensional point cloud map;

fig. 9 is a server operation flow diagram for constructing a three-dimensional point cloud map.

Detailed Description

The following describes the embodiments of the present invention further with reference to the drawings.

The embodiment provides a plant protection unmanned aerial vehicle global path optimization method based on a point cloud map, which comprises the following steps as shown in fig. 1:

step 1: constructing a three-dimensional point cloud map of a terrain to be worked, comprising:

step 1.1: and collecting point cloud data and image data of the terrain to be worked, wherein the point cloud data comprises position coordinates of each point cloud, and the image data comprises RGB information of each pixel point. Optionally, the application acquires point cloud data of the terrain to be operated by using a laser radar, and acquires the terrain coordinates to be assigned to the point cloud by using a dual-mode positioning module; and acquiring image data of the terrain to be worked by utilizing a plurality of cameras.

Step 1.2: voxel filtering is carried out on the point cloud data, dense points are removed on the reserved topographic profile, and the method comprises the following steps:

(1) Creating a three-dimensional voxel grid according to the point cloud data, wherein the side length of each voxel small grid approximately takes the individual length of the plant protection unmanned aerial vehicle;

(2) And calculating the gravity center of each voxel small grid according to the data coordinates of the point cloud, and replacing all points in the voxel small grid with the gravity center to achieve the purpose of filtering.

The calculation formula of the voxel filtering method is as follows:

in (x) _i 、y _i 、z _i ) Representing three-dimensional coordinates of an ith point cloud; x is x _max 、y _max Representing the maximum value, x, on two coordinate axes in a point cloud data set _min 、y _min 、z _min Representing minimum values on three coordinate axes in the point cloud data set; r is the side length of the voxel small grid; h is a _i And (3) for indexes of the ith point cloud in the point cloud data set in the voxel small grids, calculating the gravity center of each voxel small grid by sequencing all indexes in the voxel small grids from small to large.

The comparison of the three-dimensional point cloud diagrams before and after filtering is shown in fig. 2, and the filtering result reduces the number of points and saves the shape characteristics of the point cloud. If the selected length of the side length of the voxel small grid is too small, the processing speed and the processing difficulty of the subsequent path planning are increased; the selected length of the side length of the voxel small grid is too large, so that the traversing effect of the unmanned aerial vehicle can be influenced, and the potential safety hazard of the unmanned aerial vehicle is increased, so that the individual length of the unmanned aerial vehicle is approximately selected to be the most suitable side length of the voxel small grid.

Step 1.3: performing data fusion on the filtered point cloud data and the image data to obtain a three-dimensional point cloud map with RGB, wherein the method comprises the following steps:

as shown in fig. 3, each point in the filtered point cloud data set is subjected to a coordinate transformation formula to find a corresponding point in the image data coordinate system, R, G, B pixel values of the corresponding point are obtained and assigned to corresponding points in the point cloud data set, and a three-dimensional point cloud map with RGB is obtained, as shown in fig. 4. The fused point cloud has RGB color attributes, can provide better visual effect, and is convenient for ground object identification and application.

The coordinate transformation formula from the point cloud coordinates to the pixel coordinates is as follows:

in the method, in the process of the invention,

an internal reference matrix of each camera for collecting image data; />

A correction matrix from each camera to the reference camera; />

Rotating the translation matrix for coordinate transformation; x (X, y, z) is the three-dimensional coordinates of the point cloud X; y (u, v) is the pixel point Y determined by the number of rows u and the number of columns v in the pixel coordinate.

Step 2: and defining a to-be-worked area of the plant protection unmanned aerial vehicle in the three-dimensional point cloud map, and performing plane fitting on point cloud data in the to-be-worked area to obtain each plane block.

The method for carrying out plane fitting on the point cloud data in the area to be worked comprises the following steps:

fitting the point cloud data in the area to be worked to the plane equation z=a using the least square method ₀ x+a ₁ y+a ₂ To

As a fitting condition, a fitting condition satisfying the fitting condition is obtained by a plurality of times of cyclic fittingCoefficient a of plane equation ₀ 、a ₁ 、a ₂ Where (x, y, z) represents the three-dimensional coordinates of the point cloud, z _i Representing the z-coordinate of the ith point cloud, min represents the minimum.

Specifically, the point cloud data is imported

The normalized covariance matrix is calculated as: />

Wherein p is _i ∈P；/>

The expression is +.>

And then calculating eigenvalues and eigenvectors of the covariance matrix according to a singular value decomposition formula lambdav=mv, wherein lambdav is the eigenvalue of M, and V is the eigenvector corresponding to lambdav. Finally searching the minimum eigenvalue to obtain the coefficient a of the plane equation ₀ 、a ₁ 、a ₂ . Multiple fitting planes can be obtained by cycling the steps of the section, and finally the to-be-worked area is decomposed into various plane block areas.

Taking terraced fields as an example, the least squares plane fitting is aimed at obtaining coefficients of the plane equation

And as a fitting condition, through repeated circular fitting, each ladder face can be finally independently fit into a plane block, the labels are assigned in sequence, and the corresponding plane equation coefficients are recorded in the storage areas of the corresponding labels. The least square method is simple and effective in implementation, and can find the optimal plane function of the data by minimizing the square sum of errors. />

Step 3: and carrying out sub-region segmentation on each plane block.

In the present embodiment, the baseDividing each plane block into subareas according to an optimized Morse decomposition method, wherein the Morse decomposition method is one of unit decomposition methods, and a Morse function f is utilized ₁ (x, y) =x, and the binary map in the two-dimensional plane is divided into areas. Compared with other area division algorithms, the Morse decomposition method is more accurate in describing the internal barriers of the environment and reduces the complexity of the search process. The method is further optimized, and the dividing times can be reduced through the two additional planning strategies, so that the number of subareas is reduced. The optimized planning strategy comprises the following steps: the first divides the subareas along the walking direction of the unmanned aerial vehicle, and the second divides the boundaries of the obstacles in the plane block into the boundaries of the subareas. The basic steps of the optimized Morse decomposition method are as follows:

(1) Firstly, map environment data of a plane block are read, and binarization processing is carried out on the map environment data;

(2) The direction of the gradient at each pixel is calculated using the Sobel operator. Using this information, the direction that best fits the cell, i.e. the direction that makes the cell longer, can be calculated and the map rotated in this way, while the direction that the cell longer is the direction in which the drone walks;

(3) One slice (Morse function) is scanned through a given map and the connectivity of this line, i.e. how many connected segments are checked. If the connection increases, i.e. more segments occur, an IN event will occur that generates a new individual cell, and if the connection decreases, i.e. segments merge, an OUT event will occur that merges the two cells together. When any event occurs, the algorithm checks the critical points along the current line, i.e. the point of the trigger event, starts to move leftwards/rightwards from the critical points until the black pixel of the obstacle is hit, and then starts to draw the boundary of the cell from the position of the black pixel of the obstacle, so that the boundary of the obstacle is divided into the boundaries of the subareas;

(4) After the scan slice determines all the cells, the algorithm finds the cells using contour detection, and provides a set of pixel points for each cell, and creates a generic polygon for each cell using the pixel points;

(5) And finally, storing the division file and displaying the division result.

Step 4: a reciprocating path plan is performed within each sub-area.

The reciprocating method, i.e. reciprocating traversing walking, is preferred to walk along the long side to reduce the number of turns of the drone. The round-trip traversal walking is one of the basic ways to implement full coverage traversal, and is not described here again.

Step 5: and performing backtracking path planning among the sub-areas by using an optimized backtracking method to obtain the full coverage path of the plant protection unmanned aerial vehicle on the plane block.

For the path planning between subareas, the backtracking points and the backtracking paths are required to be designed by using an optimized backtracking method. The selection of the backtracking points is based on a backtracking list, the backtracking list is a data structure, vertex information generated by the intersection point of paths in the subareas and the borders of the subareas is stored, each backtracking considers the nearest backtracking point between the current point and the starting point of the next execution subarea, and the backtracking paths are generated based on the positions of the backtracking points.

The optimized backtracking method meets the following two priority selection principles: the middle region priority, that is, the first priority is to take the region closer to the middle in the region adjacent to the currently executed region as the next executed region; the adjacent region priority is employed when the first priority is not satisfied, i.e., the second priority is to take the adjacent sub-region closest to the currently executing sub-region as the next executing sub-region.

The specific workflow of the optimized backtracking method is as follows:

(1) Updating a backtracking list according to the current position of the plant protection unmanned aerial vehicle;

(2) Selecting a backtracking point in the backtracking list according to a priority selection principle;

(3) Judging whether the backtracking list is empty, ending planning and outputting a full coverage path when the backtracking list is empty, otherwise, performing (4);

(4) And (3) taking the backtracking point as a target point connected between the subareas, and planning to the target position and then carrying out (1) again.

The optimized backtracking method is used for the adapter region, so that the situation that the backtracking path spans the sub-region and leaks the sub-region can be avoided, and repeated coverage and missing coverage are reduced.

Step 6: and planning paths among the plane blocks by using an optimized A-algorithm to obtain the full-coverage optimal path of the plant protection unmanned aerial vehicle in the area to be worked.

The algorithm a is a direct search method in a static road network for solving the shortest path most effectively, and is a common heuristic algorithm for many other problems. The formula is: f (n) =g (n) +h (n), where f (n) is the distance estimate from the initial node to the target node via node n, g (n) is the actual distance from the initial node to node n in the node space, and h (n) is the estimated distance of the best path from node n to the target node. The essence of the algorithm a is the fusion of the greedy algorithm and the Dijkstra algorithm, which not only can search the shortest path, but also can guide itself by using a heuristic function, and has the advantages of the two algorithms. The application further optimizes the algorithm A, improves the algorithm A from three aspects of heuristic functions, weight coefficients and search neighborhoods respectively, and specifically comprises the following working flows:

(1) Creating a START node START, a TARGET node TARGET, an OPEN list and a CLOSE list, wherein the CLOSE list is initially empty;

(2) Adding START to OPEN list;

(3) Checking nodes in the OPEN list, and if the list is empty, no feasible path exists; otherwise, selecting a node k with the minimum value of the adjacent point cost function f (x, y, z);

(4) Removing the node k from the OPEN list, adding the node k to the CLOSE list, and outputting a full-coverage optimal path if the node k is a TARGET node TARGET; otherwise, continuing to expand the five neighborhoods of the node k, generating a sub-node set q of the node k, calculating corresponding f (x, y, z) values for all the sub-nodes, selecting the node with the smallest f (x, y, z) value, and placing the node in a CLOSE list;

(5) Re-executing (3) until full coverage optimal path is obtained or no solution exits.

In the above flow, the method optimizes the extended eight neighborhood to five neighborhood to reduce the number of search nodes. The prox cost function f (x, y, z) =g (x, y, z) +w (n) ×h (x, y, z); wherein g (x, y, z) is an actual cost function, h (x, y, z) is a heuristic function, the optimized a algorithm sets the heuristic function as a manhattan distance function, and a weight coefficient w (n) is further added, and the weight coefficient dynamically changes along with the numerical value fed back by the heuristic function, and is expressed as follows: when the heuristic value is larger, the weight coefficient is larger, so that the A algorithm can be expanded to the end point as much as possible and used for quickly reaching the destination when searching starts; when the heuristic is small, the weight coefficient is chosen to be small, so the a algorithm will tend to search for the optimal path for the best path to reach the destination at the end of the search. The number of nodes searched by using the optimized A-algorithm is smaller, the searching speed is faster, and the planned path is smoother.

Step 7: the plant protection unmanned aerial vehicle calls the dual-mode positioning module during actual operation, so that the traversing task of the area to be worked is completed according to a preset full-coverage optimal path.

The three-dimensional point cloud map result obtained by executing the step 1 is verified:

the experimental results of first verifying voxel filtering using the kitti dataset as experimental verification data are shown in fig. 2. The experimental result shows that the number of the point clouds becomes sparse, but the shape structure of the point clouds can be shown, and the test finds that the individual length of the unmanned aerial vehicle is the most suitable for the side length of the voxel small grid, so that when the unmanned aerial vehicle path planning is carried out later, the time calculated by the algorithm is reduced to the greatest extent on the premise that the global coverage of the unmanned aerial vehicle path can be realized.

And the kitti data set is also used as experimental verification data to verify point cloud data fusion. The whole algorithm flow is as shown in fig. 3, and the data reading, transformation matrix preparation, coordinate transformation and point cloud display are sequentially executed. The final effect diagram obtained through the four steps is shown in fig. 4, wherein (a) is original point cloud data, and (b) is point cloud data obtained by adding a photo to take color and fuse. The fused point cloud has actual colors, and has important significance for subsequent regional division and obstacle recognition.

The global path planning result obtained by executing the steps 2-6 is verified: the area after a block least square plane fit shown in fig. 5 (a) was selected, wherein the black area represents an obstacle. First, it is sub-region decomposed using an optimized Morse decomposition method, and finally divided into 7 sub-regions as shown in fig. 5 (b). Then, a reciprocating method and an optimized backtracking method are used for planning the paths of the subareas, as shown in fig. 6, the plus sign in the figure represents the start point of the path in each subarea, the triangle represents the end point of the path in each subarea, the solid line represents the reciprocating path, and the dotted arrow represents the backtracking path of the adapter subarea. And finally, the planned regional effect diagram shows complete coverage, reasonably avoids obstacles, and backtracking the condition that the path does not cross the subarea.

And selecting a mountain three-dimensional point cloud map as data of test optimization A-algorithm three-dimensional path planning, wherein the finally planned obstacle avoidance path is shown in fig. 7. Test results show that the optimized A-algorithm can well realize point-to-point obstacle avoidance path planning in a three-dimensional environment, plays a key role in the engagement of all planes fitted by the least square method, and is also a key element for overcoming complex terrains such as terrains.

The present application also provides two sets of control experiments to demonstrate the advantages of the methods of the present application:

comparative example 1: referring to step 2-6, the comparison analysis is performed on the backtracking paths under two conditions without using the priority selection principle in step 5 and without changing the others. The result shows that: the backtracking path planned by the backtracking method without using the priority principle is longer and can cross the traversed area, so that repeated coverage is caused; the backtracking method using the priority principle not only does not traverse across the partition but also reduces the risk of collision with obstacles.

Comparative example 2: referring to step 2-6, the algorithm A in step 6 is respectively changed into a bee colony algorithm, a genetic algorithm and an annealing algorithm, and the experimental results under the four conditions are compared and analyzed without change. The result shows that: compared with other algorithms, the optimized A-type algorithm has simple and clear thought, is convenient for practical application, and simultaneously has more visual obstacle avoidance traversal and shortest path finding effect between two points, and the algorithm runs faster.

The method of the step 1 is carried on a system, and the system comprises a data acquisition module, a data processing module and a data uploading and downloading module. The data acquisition module comprises a dual-mode positioning module, a laser radar module and a camera module and is used for acquiring point cloud data and image data of the terrain to be worked; the data processing module processes the received data in the step 1.2 and the step 1.3 through the construction server; and the data uploading module is used for uploading the point cloud file processed by the data processing module to the cloud platform.

The system configuration comprises the requirements of server construction, data acquisition of a data acquisition module and sending and receiving of point cloud files.

1. Collecting topographic data:

the dual-mode positioning module can acquire position information through the self-contained RF radio frequency chip and the baseband chip, and transmit acquired data to the STM32 server end in a wired serial port transparent transmission mode, so that data transmission in the dual-mode positioning module is realized.

The dual-mode positioning module is connected with the STM32, and the raspberry group server is connected with the STM32 through an on-board serial port, so that UART pins of the raspberry group are needed to be used. As shown in fig. 8, the UART pin of the raspberry group has a pair of serial ports, the GPIO numbers of which are 14 and 15, but the serial ports are allocated to the newly added bluetooth module by default in the raspberry group 4B, so that the serial ports cannot be directly used as common serial ports. Therefore, the Bluetooth is firstly closed to use the hardware serial port, and the hardware serial port is restored to be a common UART serial port; after the UART serial port is restored to a common UART serial port, data is read from a ttyAMA0 port mapped by the on-board serial port by utilizing a WiringSerial library in the raspberry Pi, and the baud rate is set to 115200. After receiving the collected position information, STM32 performs verification on the data, and when the verification is successful, sends a 4B receiving signal to the raspberry.

2. Processing topographic data:

and (3) constructing a server: the raspberry pie 4B based on ARM Cortex-A72 architecture is adopted, support of Linux kernel version and STM32F103ZET6 are provided, and a small server is built. And a data processing tool MATLAB supporting a general programming language is adopted to realize analysis of txt format data files obtained by raspberry party. The server can automatically run the appointed script when being started, and the initialization is realized, so that the functions of raspberry group and STM32 data receiving and transmitting are realized. The method comprises the following specific steps:

adding a command under the raspberry pie/rc/local directory, su pi-c 'python/home/pi/Desktop/running. Py', wherein su pi-c refers to executing an instruction with the identity of pi user, because the command in this file runs with the root identity by default, thus enabling self-launching python script with pi identity;

and adding a 10-second delay in the running.py, and after waiting for the completion of the starting of the raspberry group related service, adding a control command of a system port and a server execution script, wherein the running flow of the server execution script is shown in fig. 9:

(1) Starting a server, waiting for a receiving instruction sent after STM32 is checked, and receiving a response by a raspberry group;

(2) Initializing a serial port transmission mode of STM32 and GPS, and receiving data from a ttyAMA0 serial port;

(3) Reading a point cloud file received under a system directory, initializing a coordinate tag and controlling variables (coordinate information obtained by last measurement);

(4) Calling a MATLAB process through a DBUS, and creating a MATLAB process control object;

(5) The server contains a raspberry group, an STM32 and a GPS module, a raspberry group algorithm program runs, the contents of the step 1.2 and the step 1.3 are specifically executed, the data fusion processing is realized based on an OpenCV visual library, and the rest of the contents are not repeated here.

3. Uploading and downloading of topographic data:

the Bluetooth socket of the server is configured, signal broadcasting and channel monitoring of the server are started according to UUID (universal unique identification code) of the Bluetooth serial port service of the mobile client, and the server can respond in a related manner according to instructions issued by a user of the mobile client; the responding processing request comprises uploading and downloading of three-dimensional point cloud map data, full-coverage path planning of terrains in a selected range, control of unmanned aerial vehicle operation progress and the like.

The method of the steps 2-6 is also carried on a system for realizing, and the system configuration comprises the requirements of two aspects of construction of a server, real-time positioning of a positioning module and the like. The construction of the server is the same as that described in the topographic data processing in step 1, and will not be described here again.

And (3) constructing a positioning module: the main body is composed of a dual-mode positioning module and an STM32, and the STMF103ZET6 is the upper computer responsible for receiving the position information transmitted by the positioning module. The dual-mode positioning module (GPS+Beidou) adopts an S1216F8-BD module, is provided with an IPX interface, can be provided with various active antennas, supports hot start, and can also store position information in a FLASH in the module. The serial port baud rate between the dual-mode positioning module and the STM32 upper computer defaults to 38400,8-bit data bits, 1 stop bit, no parity check and 5HZ updating frequency. The output protocol of the position information adopts NMEA-0183 protocol output by default, and the module can be configured through SkyTraq protocol.

The path optimization algorithm provided in step 2-6 is run in the raspberry group after the construction is completed, and will not be described in detail here.

What has been described above is only a preferred embodiment of the present application, and the present invention is not limited to the above examples. It is to be understood that other modifications and variations which may be directly derived or contemplated by those skilled in the art without departing from the spirit and concepts of the present invention are deemed to be included within the scope of the present invention.

Claims

1. A plant protection unmanned aerial vehicle global path optimization method based on a point cloud map is characterized by comprising the following steps:

constructing a three-dimensional point cloud map of the terrain to be operated;

defining a to-be-operated area of the plant protection unmanned aerial vehicle in the three-dimensional point cloud map, and performing plane fitting on point cloud data in the to-be-operated area to obtain each plane block;

dividing each plane block into subareas, carrying out reciprocating path planning in each subarea, and carrying out backtracking path planning among subareas by using an optimized backtracking method to obtain a full coverage path of the plant protection unmanned aerial vehicle in the plane block;

2. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 1, wherein the method for constructing the three-dimensional point cloud map of the terrain to be worked comprises the following steps:

collecting point cloud data and image data of a terrain to be worked, wherein the point cloud data comprises position coordinates of each point cloud, and the image data comprises RGB information of each pixel point;

and carrying out voxel filtering on the point cloud data, and carrying out data fusion on the filtered point cloud data and the image data to obtain the three-dimensional point cloud map with RGB.

3. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 2, wherein the method for voxel filtering the point cloud data comprises the following steps:

creating a three-dimensional voxel grid according to the point cloud data, wherein the side length of each voxel small grid approximately takes the individual length of the plant protection unmanned aerial vehicle;

calculating the gravity center of each voxel small grid according to the data coordinates of the point cloud, and replacing all points in the voxel small grids with the gravity centers so as to achieve the purpose of filtering;

the calculation formula of the voxel filtering method is as follows:

in (x) _i 、y _i 、z _i ) Representing three-dimensional coordinates of an ith point cloud; x is x _max 、y _max Representing the maximum value, x, on two coordinate axes in a point cloud data set _min 、y _min 、z _min Representing minimum values on three coordinate axes in the point cloud data set; r is the voxel small gridSide length; h is a _i And (3) indexing the ith point cloud in the point cloud data set in the voxel small grids, and calculating the gravity center of each voxel small grid.

4. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 2, wherein the method for carrying out data fusion on the filtered point cloud data and the image data to obtain the three-dimensional point cloud map with RGB comprises the following steps:

finding out corresponding points in an image data coordinate system through a coordinate transformation formula from each point in the filtered point cloud data set, acquiring R, G, B pixel values of the corresponding points, and assigning the R, G, B pixel values to the corresponding points in the point cloud data set to obtain a three-dimensional point cloud map with RGB (red, green and blue);

in the method, in the process of the invention,

an internal reference matrix of each camera for collecting image data; />

A correction matrix from each camera to the reference camera; />

Rotating the translation matrix for coordinate transformation; x (X, y, z) is the three-dimensional coordinates of the point cloud X; y (u, v) is the pixel point Y determined by the number of rows u and the number of columns v in the pixel coordinate. />

5. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 1, wherein the method for performing plane fitting on the point cloud data in the area to be worked comprises the following steps:

fitting a plane equation z=a by using a least square method to the point cloud data in the region to be worked ₀ x+a ₁ y+a ₂ To

As a fitting condition, coefficients a of a plane equation satisfying the fitting condition are obtained by a plurality of cyclic fitting ₀ 、a ₁ 、a ₂ Where (x, y, z) represents the three-dimensional coordinates of the point cloud, z _i Representing the z-coordinate of the ith point cloud, min represents the minimum.

6. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 1, wherein the method for dividing each plane block into subareas comprises the following steps:

and carrying out sub-region segmentation on each plane block based on an optimized Morse decomposition method, wherein an optimized planning strategy comprises the following steps: dividing the subareas along the walking direction of the unmanned aerial vehicle, and dividing the boundaries of the obstacles in the plane block into the boundaries of the subareas.

7. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 1, wherein the method for performing backtracking path planning between sub-areas by using an optimized backtracking method comprises the following steps:

updating a backtracking list according to the current position of the plant protection unmanned aerial vehicle, wherein the backtracking list stores vertex information generated by the intersection point of a path in the subarea and the subarea boundary;

selecting a backtracking point in the backtracking list according to a priority selection principle, wherein the priority selection principle is as follows: the first priority is to take a sub-region closer to the middle among regions adjacent to the currently executed sub-region as a next executed sub-region; the second priority is to take the adjacent subarea closest to the currently executed subarea as the next executed subarea;

and outputting a full-coverage path when the backtracking list is empty, otherwise, taking the backtracking point as a target point for connection between the subareas, planning to a target position, and updating the backtracking list according to the current position of the plant protection unmanned aerial vehicle.

8. The method for optimizing a global path of a plant protection unmanned aerial vehicle based on a point cloud map according to claim 1, wherein the method for planning the path between the plane blocks by using the optimized a-algorithm comprises the following steps:

creating a START node START, a TARGET node TARGET, an OPEN list and a CLOSE list, wherein the CLOSE list is initially empty, and adding the START to the OPEN list;

checking nodes in the OPEN list, and if the list is empty, no feasible path exists; otherwise, selecting a node k with the minimum value of the adjacent point cost function f (x, y, z);

removing the node k from the OPEN list, adding the node k to the CLOSE list, and outputting a full-coverage optimal path if the node k is the TARGET node TARGET; otherwise, continuing to expand the five neighborhoods of the node k, generating a sub-node set q of the node k, calculating corresponding f (x, y, z) values for all the sub-nodes, selecting the node with the smallest f (x, y, z) value, and placing the node into the CLOSE list;

and re-executing the nodes in the check OPEN list until the full-coverage optimal path is obtained or no solution exits.

9. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to claim 8, wherein the proximity point cost function is expressed as:

f(x，y，z)＝g(x，y，z)+w(n)×h(x，y，z)；

wherein g (x, y, z) is an actual cost function, h (x, y, z) is a heuristic function, and the heuristic function is a Manhattan distance function; w (n) is a weight coefficient, and the weight coefficient dynamically changes along with the numerical value fed back by the heuristic function.

10. The plant protection unmanned aerial vehicle global path optimization method based on the point cloud map according to any one of claims 1 to 9, wherein the method further comprises:

and the plant protection unmanned aerial vehicle calls the dual-mode positioning module during actual operation, so that the traversing task of the area to be worked is completed according to a preset full-coverage optimal path.