Search

CN-122028134-A - Unmanned aerial vehicle cluster ad hoc network position topology establishment method based on visual assistance

CN122028134ACN 122028134 ACN122028134 ACN 122028134ACN-122028134-A

Abstract

A visual assistance-based unmanned aerial vehicle cluster self-networking position topology establishment method belongs to the field of unmanned aerial vehicle cluster autonomous cooperation and safety communication. The method comprises the steps of initializing a system, performing self-checking, starting a first camera to lock and track a visual identifier of a target unmanned aerial vehicle, collecting the visual identifier of the unmanned aerial vehicle by using a second camera, extracting geometric feature points, calculating dynamic weights of the feature points, fusing and calculating accurate relative pose of the target unmanned aerial vehicle, generating and broadcasting a local adjacent information table, constructing and weighting and optimizing a global physical topological graph, constructing the global physical topological graph, continuously monitoring node pose change and energy efficiency state, triggering weight calculation when displacement overrun, link interruption or energy efficiency transition is detected, broadcasting an increment update message, carrying out local dynamic correction on the global topological graph, and establishing the unmanned aerial vehicle cluster self-networking position topology based on visual assistance. The invention can obviously improve the performance and reliability of the unmanned aerial vehicle cluster ad hoc network.

Inventors

  • ZHANG YU
  • HUANG XINYI
  • WANG YIMING
  • YANG QIANYU
  • REN YI
  • DIAO WENLAN

Assignees

  • 北京理工大学

Dates

Publication Date
20260512
Application Date
20251218

Claims (9)

  1. 1. A method for establishing an unmanned aerial vehicle cluster self-networking position topology based on visual assistance is characterized by comprising the following steps: Step 1, system initialization and self-checking, namely loading a core algorithm module, activating a communication and visual module based on an unmanned aerial vehicle, and realizing hardware self-checking and parameter calibration; Step 2, ID recognition and relative pose calculation, namely starting a first camera to lock and track a visual identifier of a target unmanned aerial vehicle to obtain a unique ID; acquiring a visual identifier of the unmanned aerial vehicle by using a second camera, extracting geometric feature points, calculating dynamic weights of the feature points, and fusing and solving the accurate relative pose of the target unmanned aerial vehicle; step 3, the local adjacent information table is generated and broadcasted, namely the residual electric quantity information is read, the local adjacent information table is generated by combining the target ID and the relative pose, and the local adjacent information table is broadcasted in a trunking channel through a WLAN communication module; Step 4, global physical topological graph construction and weighted optimization, namely receiving and analyzing broadcast data, and carrying out data validity verification and initial topological graph construction; calculating pose weights and energy efficiency weights, fusing the pose weights and the energy efficiency weights to generate final edge weights, and constructing a global physical topological graph; and 5, monitoring node position change and energy efficiency state continuously in real time, triggering weight calculation when displacement overrun, link interruption or energy efficiency transition is detected, broadcasting increment update message, and carrying out local dynamic correction on the global topological graph to realize the establishment of the unmanned aerial vehicle cluster self-networking position topology based on visual assistance.
  2. 2. The method of claim 1, wherein the system initialization and self-test in step 1 are implemented by, Step 1.1, core System Loading and initialization After the unmanned aerial vehicle is electrified, the core processing unit automatically guides and loads an embedded operating system, and based on the operating system, a core algorithm module in the solidified storage is transferred into an operation memory; Step 1.2, communication subsystem activation and function verification The core processing unit sends an instruction to the WLAN communication module through the high-speed PCIe interface, activates the hardware function of the WLAN communication module, and tests the receiving-transmitting link function through the self-checking program; step 1.3, function self-checking and precision confirmation of double-camera vision module Yun Taidui the first camera performs a full-range motion test, the self-checking program performs a full-reciprocating zooming test on the first camera, and the self-checking program checks whether hardware synchronous signals between the two cameras are aligned or not; step 1.4, calibration and confirmation of the measurement reference of the second fixed wide-angle camera The system loads an internal reference, and a radial distortion coefficient k 1 、k 2 and a tangential distortion coefficient p 1 、p 2 for correcting the optical error of the lens; The relation of the internal reference matrix is specifically as follows: (1) Wherein, the Is the pixel coordinates of the target point in the image; is the three-dimensional coordinates of the target point in the camera coordinate system, wherein Representing depth; is an internal reference matrix of the camera; Is the equivalent focal length of the camera in the x-axis and y-axis directions, in pixels; Is the projection of the camera's optical center onto the imaging plane, i.e., the principal point coordinates.
  3. 3. The method of claim 1, wherein the step 2 of obtaining the unique ID of the target unmanned aerial vehicle is implemented by, After the suspected target is detected preliminarily, the target is locked in the center of the visual field through the first camera, continuous tracking is kept, a visual identifier on the target body is collected, and the visual identifier is decoded to obtain the unique identity ID of the target unmanned aerial vehicle.
  4. 4. The method of claim 1, wherein the accurate relative pose solution of the target unmanned aerial vehicle in step 2 is implemented by, 2.2.1, Analyzing the recognized visual identifier in the visual field by the second camera; Step 2.1.1.1, visual identification Fu Jiaodian distortion refinement First extracting 4 corner original pixel coordinates of visual identifier K=1, 2,3,4, and corresponds to four vertexes of the identifier, and correcting the visual identifier based on camera internal parameters and distortion coefficients according to the following correction formula: (2) Wherein, the Is the original pixel coordinates of the corner point, The coordinates are normalized for the corner points, The radius square is normalized for the corner point, As the radial distortion coefficient of the lens, As a tangential distortion coefficient, In order to normalize the coordinates of the corrected corner points, The pixel coordinates of the corner points after correction; step 2.1.1.2, multi-feature ranging value calculation Based on 2.1.1.1 corrected coordinates Respectively calculating distance measurement values: (3) (4) (5) Wherein, the Ranging values of "side length feature", "diagonal feature", "center distance feature" respectively, For the actual side length of the visual identifier, To correct the post-identifier center pixel coordinates, , ; Step 2.1.1.3 dynamic weight Allocation Calculating consecutive 5 frames Statistical characteristics: (6) (7) (8) Wherein, the Is a feature index, i=1 corresponds to D1, i=2 corresponds to D2, i=3 corresponds to D3, Is to count the number of frames of the window, For the ranging values of the ith class of features of the t-th frame, Is the 5-frame mean of the i-th class of features, A standard deviation of 5 frames for class i features; the entropy of information for the i-th class of features, Is a minimum value; Is the dynamic weight of the i-th type characteristic, meets the following conditions ; Step 2.1.1.4, fusion ranging Based on dynamic weights (9) Where n is the current frame index, The distance measurement value is the fusion distance measurement value of the nth frame 3 type characteristics, namely the final optimized distance; Step 2.2.2 relative orientation resolution Based on the coordinates after 2.1.1.1 correction The relative orientation is calculated by the formula (10) (11): (10) (11) Wherein, the Representing a yaw angle of the target drone relative to the host; representing the pitch angle of the target unmanned aerial vehicle relative to the host; step 2.2.3 data smoothing Using Kalman filters for successive multi-frame images And carrying out data fusion and filtering to output smoother and more stable relative pose data.
  5. 5. The method of claim 1, wherein the step 3 of generating and broadcasting the local adjacency information table is implemented by, The local node reads the current residual electric quantity through the self power management module ; Each unmanned aerial vehicle uses the self ID as the source address identification of the information table, combines the target unmanned aerial vehicle ID and the relative pose obtained in the step 2 to integrate into a structured local adjacent information table { information source ID, neighbor ID, relative distance, relative azimuth, time stamp, }; The WLAN communication module broadcasts a local adjacent information table in a trunking public channel on a trunking preset public channel.
  6. 6. The method of claim 1, wherein the data validity verification and initial topology construction in step 4 is implemented by, Step 4.1 data validity verification Checking and filtering the received information source ID, neighbor ID, relative distance, relative azimuth and time stamp in the local adjacent information table, identifying and removing abnormal data which does not accord with a preset structural format, has the time stamp expired or has the electric quantity field lower than a normal threshold value, and screening to obtain an effective data packet for subsequent topology construction; Step 4.2, maintaining a global physical topological graph in a local memory by a core processing unit of each unmanned aerial vehicle, and performing iterative updating based on the parsed effective data to complete an initial global topological graph Is constructed according to the following steps; step 4.2.1 set of vertices Updating Vertex definition, namely, each vertex uniquely maps one unmanned aerial vehicle in a cluster, takes the global unique ID of the unmanned aerial vehicle as a vertex identifier, and under an initial state Only contains the vertex corresponding to the local ID; Update logic, checking information source ID and neighbor ID in effective data packet, if some ID does not exist in the top point set The ID is added as a new vertex At the same time, the initial electric quantity information of the vertex is correlated to ensure the vertex set Completely covering all known nodes of the cluster; Step 4.2.2 edge set Initial update Edge definition, in which each edge represents a visually-confirmed direct physical link between two vertices, initial edge weights Is generated by the quantization of the relative pose information between two vertexes, and is in an initial state Is an empty set; update logic, based on information in data packet, edge set Performing an add or update operation: a. newly added edge, if there is no existing edge between two vertexes, calculating initial weight based on relative pose information And adds a new edge with the weight, Normalized by the following formula: (12) Wherein, the For connecting nodes And node Is a side of (2); Is a node And (3) with Is a relative distance of (2); Respectively a horizontal azimuth angle and a pitching azimuth angle between the two nodes; The method comprises the steps of b, updating edges, namely, if edges exist between two vertexes, comparing time stamps of new and old adjacent relations, and replacing old edges with relative pose information updated by the time stamps And meanwhile, the electric quantity information of the vertex corresponding to the edge is synchronously updated, so that the link information is ensured to reflect the network state and the node energy consumption state in real time.
  7. 7. The method of claim 1, wherein the calculating the energy efficiency weight in step 4 is implemented by, Step 4.3.1 fixed parameter Loading Loading preset hardware and task parameters from a local nonvolatile storage: Power consumption parameter communication module power consumption Visual module power consumption Core processor power consumption ; Power threshold, minimum working power Full capacity ; Task parameters Single Cluster task duration ; Energy efficiency coefficient, energy efficiency sensitivity coefficient ; Step 4.3.2 node energy consumption Rate calculation Based on the loaded power consumption parameters, calculating the total energy consumption rate of all vertexes: (13) Wherein, the For vertex sets Any vertex of (a); Is a node Reflecting the energy consumption level per unit time of the node; Step 4.3.3 residual energy efficiency calculation Based on vertex set The associated power threshold and task parameters, and calculating the remaining energy efficiency of each vertex: (14) Wherein, the Is a node Is the current residual quantity of electricity; Is a node The remaining time of the task can be maintained in the current electric quantity state; if a certain vertex does not acquire the validity Then preset default electric quantity value is adopted Substituting calculation, and updating in real time after receiving effective data packet broadcast by the node ; Based on energy efficiency coefficient and residual energy efficiency Energy efficiency priority of quantization node: (15) Wherein, the Is a node The energy efficiency weight of (2) is as follows ; The larger the coefficient is, the more remarkable the influence of the residual energy efficiency on the weight is; When the energy efficiency is remained Equal to the duration of the task When, energy efficiency weight And when the residual energy efficiency of the node approaches 0, the energy efficiency weight maintains the lowest threshold value, and the node is prevented from completely separating from the route selection.
  8. 8. The method of claim 1, wherein the fusing of the location weights and the energy efficiency weights to generate final edge weights in step 4 is implemented by constructing a global physical topology map, Step 4.4 edge weight fusion optimization Weighting the initial edge Node energy efficiency weight Fusion is carried out to generate final edge weight considering pose precision and node energy efficiency : (16) Wherein, the The pose weight duty ratio coefficient is used for ensuring that the relative pose is still a core influencing factor of the link weight; 、 Respectively is a side Two-end node 、 Energy efficiency weights of (2); Comprehensively reflecting pose advantages and node energy efficiency advantages of the links for the fused final edge weights; Collect edges Initial weights for all sides in (1) Replaced by after fusion Completing topology map optimization updating; Recalculating affected nodes every time 1 valid data packet is received thereafter or a preset update period is reached And And the real-time matching of the edge weight with the node energy consumption state and the link pose state is ensured.
  9. 9. The method of claim 1, wherein the step 5 is implemented by triggering weight calculation and topology map update in real time based on the monitored pose change and node energy efficiency change to complete topology maintenance, Step 5.1, the unmanned aerial vehicle enters a high-frequency monitoring mode based on the double-camera vision module in the step two to calculate the current relative pose of the neighbor node in real time, and meanwhile, a core processing unit reads the data of the power management module in real time to monitor the residual energy efficiency of the node ; Step 5.2, setting a topology update triggering threshold, and triggering a local update flow when the system detects that any one of the following conditions is met: (a) The visual pose mutation is that the deviation between the current calculated relative distance or azimuth and the recorded numerical value in the local adjacent information table exceeds a preset pose tolerance threshold; (b) The link state changes, namely the visual system continuously multi-frame can not detect the visual identifier of the neighbor node, and the link is judged to be blocked or interrupted; (c) Energy efficiency level transition: remaining energy efficiency of node The change results in an energy efficiency weight The fluctuation range of (2) exceeds a preset sensitivity threshold; after the step 5.3 triggers the update, the core processing unit performs the following operations: for the changed nodes or links, re-invoking equation (12) to calculate real-time pose weights Equation (15) calculates the real-time energy efficiency weight And regenerating the final edge weight according to the formula (16) And after other nodes in the cluster receive the message, the global physical topological graph Gp maintained locally is subjected to targeted correction, the weight value of the corresponding edge is updated or the invalid edge is deleted, and the real-time consistency of the topological graph and the physical cluster state is maintained.

Description

Unmanned aerial vehicle cluster ad hoc network position topology establishment method based on visual assistance Technical Field The invention relates to a visual navigation, ad hoc network communication and energy efficiency optimization method, in particular to a visual assistance-based unmanned aerial vehicle cluster ad hoc network position topology establishment method, which realizes high-precision positioning, stable topology construction and dynamic maintenance of an unmanned aerial vehicle cluster through double-camera multi-feature pose resolving and energy efficiency perception route optimization, is suitable for a collaborative operation scene of a large-scale unmanned aerial vehicle cluster, and belongs to the technical field of unmanned aerial vehicle cluster autonomous collaboration and safety communication. Background Along with the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicle cluster formation shows huge application potential in fields such as military reconnaissance, disaster relief, logistics distribution, environmental monitoring and the like. The unmanned aerial vehicle cluster establishes a dynamic communication network through an ad hoc network communication mode to realize the interaction and collaborative operation of the position information among the nodes, and the stability, communication efficiency and positioning accuracy of the topological structure directly determine the execution effect of the flight task of the cluster. However, the existing unmanned aerial vehicle cluster formation flight still faces significant technical bottlenecks in positioning accuracy, communication resource utilization, topology dynamic adaptability, energy efficiency matching and the like, and the reliable application of the unmanned aerial vehicle cluster ad hoc network in a large-scale scene is restricted. The method is characterized in that (1) the visual positioning robustness is insufficient, the precision is difficult to ensure, the existing unmanned aerial vehicle cluster positioning is dependent on Beidou (GPS) navigation, positioning signals are easy to be shielded or interfered in complex scenes such as urban streets, dense jungle and the like, so that the positioning error is obviously increased (up to several meters or even tens of meters), and the safety distance requirement of dense formation cannot be met. Although the vision-aided technology is applied, the vision-aided technology is limited to distance measurement or simple target recognition of single features (such as a mark Fu Bianchang), is easily affected by imaging distortion, local shielding and illumination interference, is difficult to realize accurate unmanned aerial vehicle ID recognition and high-precision relative pose measurement meeting the topology construction requirement at the same time, and cannot provide stable and reliable data support for global topology establishment. (2) And the communication link resources are tense, the topology construction efficiency is low, along with the expansion of the cluster scale of the unmanned aerial vehicle, the channel bearing capacity of the existing frequency band (such as WLAN communication of IEEE 802.11 standard) is severely limited, and the channel competition during simultaneous communication of multiple nodes can cause the increase of the data packet loss rate and the communication delay, so that the communication efficiency of the ad hoc network is obviously reduced. The traditional ad hoc network topology is constructed to be highly dependent on a communication link to perform node discovery and information interaction, so that channel load is further increased, route reconstruction time is long (high-dynamic scenes of fast moving of an unmanned aerial vehicle and frequent joining/exiting of nodes cannot be adapted), and network performance is easy to be rapidly deteriorated. (3) The topology construction does not consider the energy efficiency of the nodes, and has poor survivability, namely the topology construction of the existing unmanned aerial vehicle cluster ad hoc network only focuses on the physical link connection (such as relative pose) among the nodes, and does not consider the energy consumption state difference of the nodes. When the low-power node is preferentially selected as the routing node, the low-power node is easy to be suddenly offline due to power exhaustion, so that the local topology is broken, and the collaborative operation continuity of the whole cluster is further affected. Meanwhile, the existing topology maintenance mechanism only passively responds to node displacement or link interruption, and potential failure risks are predicted by the unbonded node energy efficiency, so that the long-term stability of the topology is further reduced. (4) The existing vision auxiliary technology cannot be in communication with an ad hoc network and is in topology optimization deep fusion, when an ad hoc network t