Research On Path Planning Based On Jumping Robots And 3D Reconstruction Of Complex Military Scenes

In recent years,military robots have gradually become an independent system.Among them,jumping robots have occupied a considerable position,and the research in the field of vision related to them has gradually become the focus of current research and has become one of the active areas of artificial intelligence.Among them,the path planning of jumping robots in complex military scenes and the three-dimensional reconstruction of scenes during the search task execution based on the exploration path are the hot topics of related research.How to make a robot efficiently plan a safe optimal path,and perform a three-dimensional reconstruction of the surrounding environment during the task after searching the path to allow those people who cannot get involved in the battlefield environment to interact with the remote environment and obtain more data to support various subsequent information surveys,target positioning,scene analysis and other related work are core computer vision technologies in the decision and control of jumping robots.The following research work has been done on this thesis:First,by analyzing the environmental modeling method and we select the grid method for modeling,in view of the jumping movement characteristics of the jumping robot,considering the possibility of movement in the z-axis direction,improvements are made to the traditional low-attitude two-dimensional grid space.This thesis proposes a three-dimensional raster map model,which is extended to three dimensions,and a height matrix is added to describe the height information of the z-axis,so that the robot can perform the jumping obstacle avoidance possibility in this environment model.Secondly,the basic principle and search process of A * algorithm path planning are discussed.Based on the 3D grid map model,a new path cost function is constructed to evaluate the grid node cost,and an improved A * path search algorithm based on path intermediate points is proposed.By implementing the path intermediate point detection and reduced neighborhood search strategy for the A * algorithm,the search for redundant grid nodes is reduced and the direction of the path is constrained;Matlab policy experiments were performed on the improved algorithm and the comparison algorithm from the three aspects of the optimal path length,the number of extended grid nodes,and the algorithm running time.Through comparative analysis,the improved algorithm in this thesis has better planning effect and higher path quality.Then,for the robot's exploration and search process,the problem that the reconstructed 3D scene cannot capture and highlight the hidden and disguised heat source targets,this thesis proposes a three-dimensional reconstruction framework based on multiple information fusion based on visible infrared image sequence matching.The framework can capture the thermal radiation information of targets that cannot be obtained by pure visible light,while preserving the authenticity of the scene,and highlight them in the scene for subsequent target detection and spatial positioning.Finally,for the order matching problem of infrared and visible images in 3D reconstruction,this thesis proposes a non-rigid image matching algorithm based on mixed features.We extract edge and corner mixed features based on structural features for image pairs,introduce a multi-feature Gaussian mixture model,constrain it under the same non-rigid transformation to perform probability modeling and suppress cross-error matching,and then use the maximum expected value method EM)solve the problem.The improved matching algorithm and comparison algorithm are simulated on different data sets.Through comparative analysis,the improved algorithm in this paper can obtain more correct matching feature point pairs and a higher accurate matching rate,which is convenient for subsequent point cloud 3D coordinate calculation and infrared targets preservation and rebuilding authenticity.Finally,experiments are performed on a three-dimensional reconstruction framework that can integrate light and infrared multiple information,and the reconstruction results are analyzed.The framework proposed in this thesis has better reconstruction results and solves related problems we face.