CN-121994246-A - Unmanned plane cooperation-based forest blind spot inspection path planning method and system

CN121994246ACN 121994246 ACN121994246 ACN 121994246ACN-121994246-A

Abstract

The application discloses a forest blind spot inspection path planning method and system based on unmanned plane cooperation, and relates to the technical field of path planning. And then, mapping the visual information into a three-dimensional space by utilizing characteristic back projection, quantitatively evaluating the visual confidence of the region, and constructing an illumination confidence model for characterizing the perceived reliability. Based on the method, a visual degradation area which is easy to cause positioning failure is identified, anchor point mapping is established by combining visibility searching, and a navigation cost field which fuses illumination risks is generated. And finally, solving a multi-machine collaborative path under the constraint of a cost field to generate a chain track which avoids dead zones and keeps perception communication. Therefore, perceived risks caused by illumination drafts can be converted into hard constraints of path planning, the problem of visual positioning drifting under complex illumination of a forest zone is effectively solved, and high-reliability collaborative inspection of the unmanned aerial vehicle group is realized.

Inventors

WANG YUMING
LIAO SHENGHUA
ZHANG JIA
Fei Gangqiang
CHEN MI
LI WEI
WANG QI
HE JUN

Assignees

湖北长林生态科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260224

Claims (9)

1. The method for planning the blind spot inspection path of the forest based on unmanned aerial vehicle cooperation is characterized by comprising the following steps: S1, performing luminosity preprocessing on an acquired cluster original sensor data stream to obtain a visual feature set and a luminosity statistics list, wherein the cluster original sensor data stream comprises camera image streams, inertial navigation data and illumination sensor readings of each unmanned aerial vehicle; S2, based on the visual feature set and the luminosity statistics list, performing feature back projection and visual confidence score calculation on a preset three-dimensional voxel grid to obtain an illumination confidence grid chart; s3, carrying out connected domain segmentation and visibility search on the illumination confidence grid graph by using a preset threshold value to obtain a vision degradation region set and an anchor point mapping table corresponding to the vision degradation region set; S4, generating a navigation cost field based on the illumination confidence grid graph and the vision degradation area set; S5, carrying out multi-machine system collaborative accompanying path solving on the navigation cost field and the anchor point mapping table to obtain a visual chain track set; And S6, controlling the unmanned aerial vehicle group to cooperatively fly according to the visual chain track set.
2. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S1 comprises: Carrying out data flow analysis and self-adaptive histogram equalization on the data flow of the original sensor of the cluster to obtain an enhanced gray image flow; Extracting primary characteristic points based on gradient from the enhanced gray image stream to obtain a candidate characteristic coordinate set; And determining a local neighborhood window in the enhanced gray image stream according to the candidate feature coordinate set, and carrying out local variance statistics and descriptor extraction on pixel data in the local neighborhood window to obtain a visual feature set and a luminosity statistics list.
3. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S2 comprises: Based on boundary constraint of prior geometric map and preset resolution parameter, discretizing projection and index coding are carried out on three-dimensional coordinates in the visual feature set so as to obtain a voxel index mapping table; traversing the voxel index mapping table, and carrying out voxel-level photometric statistical feature aggregation on the feature points in the non-empty voxels to obtain a voxel aggregation data packet containing stable feature point quantity and average photometric variance data; And carrying out multidimensional visual confidence calculation on the voxel aggregation data packet to obtain an illumination confidence grid chart.
4. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S3 comprises: performing connected domain segmentation on the light confidence grid graph based on a confidence threshold through a preset risk threshold to obtain a vision degradation region set; Sampling surrounding type high-confidence candidate points of each region in the visual degradation region set to obtain a region candidate anchor point cluster associated with a potential observation position; And carrying out visibility detection and collaborative observation utility evaluation based on a ray casting technology on candidate points in the regional candidate anchor point cluster based on the illumination confidence grid chart, and recording the candidate point coordinate with the highest utility value into an anchor point mapping table.
5. The unmanned aerial vehicle collaboration-based forest blind spot routing inspection path planning method according to claim 4, wherein performing visibility detection and collaborative observation utility evaluation based on a ray casting technology on candidate points in a regional candidate anchor cluster based on an illumination confidence grid chart, and recording candidate point coordinates with highest utility values in an anchor mapping table comprises: Carrying out ray stepping sampling on the connecting lines between the candidate anchor points and target voxels in the vision degradation region set, calculating the integral of the occupation probability density of the passing voxels on a ray path to obtain accumulated optical thickness, and processing the accumulated optical thickness by utilizing an exponential decay function to obtain a probability view factor; Acquiring local luminosity variance of a target voxel, and carrying out mapping calculation on the local luminosity variance through a nonlinear activation function to obtain characteristic information weight; based on the self confidence of the anchor point, the feature information weight and the probability view factor, calculating to obtain the probability visual information gain fraction, and taking the probability visual information gain fraction as the collaborative observation utility value.
6. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S4 comprises: Performing nearest barrier distance calculation on occupied voxels in the illumination confidence grid chart and performing geometric rejection cost calculation by combining a preset safety radius to obtain a geometric distance field; judging the spatial attribute of the voxels in the illumination confidence grid graph by using the vision degradation region set to determine a region self-adaptive penalty weight, and performing inverse proportion risk mapping on the illumination confidence score of the voxels to obtain a vision risk field; and carrying out multi-source cost weighted fusion on the geometric distance field and the visual risk field to obtain a navigation cost field.
7. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S5 comprises: Searching a main task path based on a cost field for the navigation cost field to obtain a segmentation mark original path; Traversing the segmented marked original path, matching corresponding cooperative anchor points for the identified blind zone traversing segments according to an anchor point mapping table, and constructing a main assistant machine task queue containing relative time sequence constraint so as to obtain a cooperative navigation point pair sequence; And performing multi-machine space-time track joint optimization on the coordinated waypoint pair sequence to obtain a visual chain track set.
8. The method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation according to claim 1, wherein step S6 comprises: Analyzing and tracking a bottom instruction of the visual chain track set, and carrying out extended Kalman filtering collaborative correction based on inter-machine relative observation data when the state estimation covariance is beyond the limit so as to obtain a real-time observation data stream; Carrying out instantaneous illumination confidence reevaluation on the real-time observation data stream to obtain an instantaneous confidence measurement packet reflecting the real illumination condition of the current position; Based on the instantaneous confidence measurement package, online numerical correction is carried out on the illumination confidence grid graph by utilizing a recursive Bayesian updating strategy and an observation uncertainty coefficient so as to obtain an updated local confidence graph.
9. Unmanned aerial vehicle cooperation-based forest blind spot inspection path planning system, which is characterized by comprising: the luminosity preprocessing module is used for carrying out luminosity preprocessing on the acquired data flow of the original sensor of the cluster to obtain a visual feature set and a luminosity statistics list, wherein the data flow of the original sensor of the cluster comprises camera image flows, inertial navigation data and illumination sensor readings of each unmanned aerial vehicle; the confidence level map building module is used for carrying out characteristic back projection and visual confidence level score calculation on a preset three-dimensional voxel grid based on the visual characteristic set and the luminosity statistics list so as to obtain an illumination confidence level grid map; the illumination visibility analysis module is used for carrying out connected domain segmentation and visibility search on the illumination confidence grid graph by utilizing a preset threshold value so as to obtain a vision degradation region set and an anchor point mapping table corresponding to the vision degradation region set; The navigation cost field construction module is used for generating a navigation cost field based on the illumination confidence grid graph and the vision degradation area set; the collaborative satellite flight path solving module is used for carrying out multi-machine system collaborative satellite flight path solving on the navigation cost field and the anchor point mapping table so as to obtain a visual chain track set; and the cluster cooperative flight control module is used for controlling the unmanned aerial vehicle clusters to cooperatively fly according to the visual chain track set.

Description

Unmanned plane cooperation-based forest blind spot inspection path planning method and system Technical Field The application relates to the technical field of path planning, in particular to a forest blind spot inspection path planning method and system based on unmanned plane cooperation. Background Forest resources are taken as the main body of the land ecological system and are important to maintain global ecological balance and biodiversity. In order to ensure the safety of forestry resources, the normalized inspection aiming at tasks such as fire early warning, pest and disease damage monitoring, illegal theft and the like is indispensable. Traditional forest inspection mainly relies on manual hiking or driving vehicles, and the mode is limited by complicated and rugged topography, low in efficiency and various potential safety hazards, and full coverage monitoring of a large-area forest area is difficult to realize. In recent years, with the rapid development of unmanned aerial vehicle technology, multi-rotor unmanned aerial vehicle gradually replaces traditional manual operation by virtue of the advantages of flexibility, vertical take-off and landing and low-altitude operation, and becomes core equipment for modern forestry management. Particularly, the cluster operation mode based on multi-machine cooperation can remarkably improve the inspection efficiency and coverage rate of a large-scale forest region through task distribution and cooperation, and becomes an important development direction of current intelligent forestry construction. However, most existing unmanned aerial vehicle forest zone inspection path planning schemes focus on obstacle avoidance and shortest path solution based on geometric information, and default environment awareness information is continuous and reliable. However, in an actual under-forest flight scene, the inside illumination condition of a forest zone is extremely complex and presents a bright-dark alternation characteristic with a high dynamic range under the influence of crown shading degree difference and sun altitude angle change. The strong illumination rapid change can cause overexposure or underexposure of a visual sensor carried by the unmanned aerial vehicle during imaging, and seriously damages the luminosity consistency assumption relied on by a visual synchronous positioning and map building (SLAM) algorithm. When the light and shadow mottled woodland gap is passed or the light and shadow area is entered, the stable characteristic points are difficult to be extracted from the vision front end, so that characteristic tracking failure and positioning drift are caused, and finally, the planned path is invalid due to positioning loss. The current path planning strategy lacks the ability of prejudging the vision degradation risk, can not identify and avoid geometrically passable but visually unreliable illumination blind areas, is difficult to realize effective coverage of the blind points of the forest area on the premise of ensuring the perception stability, and limits the deep application of unmanned aerial vehicle clusters in complex forestry scenes. Therefore, an optimized forest blind spot routing path planning method and system based on unmanned plane cooperation are needed. Disclosure of Invention The present application has been made to solve the above-mentioned technical problems. According to one aspect of the application, a method for planning a blind spot inspection path of a forest based on unmanned aerial vehicle cooperation is provided, which comprises the following steps: S1, performing luminosity preprocessing on an acquired cluster original sensor data stream to obtain a visual feature set and a luminosity statistics list, wherein the cluster original sensor data stream comprises camera image streams, inertial navigation data and illumination sensor readings of each unmanned aerial vehicle; S2, based on the visual feature set and the luminosity statistics list, performing feature back projection and visual confidence score calculation on a preset three-dimensional voxel grid to obtain an illumination confidence grid chart; s3, carrying out connected domain segmentation and visibility search on the illumination confidence grid graph by using a preset threshold value to obtain a vision degradation region set and an anchor point mapping table corresponding to the vision degradation region set; S4, generating a navigation cost field based on the illumination confidence grid graph and the vision degradation area set; S5, carrying out multi-machine system collaborative accompanying path solving on the navigation cost field and the anchor point mapping table to obtain a visual chain track set; And S6, controlling the unmanned aerial vehicle group to cooperatively fly according to the visual chain track set. According to another aspect of the present application, there is provided a system for planning a blind spot inspection path in a forest bas