Unmanned Aerial Vehicle Swarm Ground Tracking And Multi-view Image 3D Reconstruction

As a significant form of intelligent warfare,unmanned aerial vehicle swarm cluster warfare gets rid of the limitation of limited payload and single capability of a single UAV.The effective combination of computer vision technology provides many advantages in battlefield perception,intelligent navigation,target tracking and attack,and three-dimensional battlefield reconstruction,which has significant research significance and application value in both civil and military fields.As a result,the development of high precision and resilience object tracking and 3D reconstruction algorithms is always a popular topic both at home and abroad.In this thesis,the video sequence and multi-view image sequence transmitted by the unmanned aerial vehicle swarm cluster are used to complete the tracking and tracking scene reproduction of the specified target on the ground.Aiming at the problem that the classical discriminative correlation filter tracking algorithm KCF(Kernelized Correlation Filter)has poor anti-target occlusion and anti-scale variation,and the traditional incremental SFM(Structure From Motion)sparse point cloud reconstruction algorithm has low feature matching precision and high point cloud reprojection error.An improved scheme is proposed.1)Aiming at the defects of the KCF algorithm,a multi-feature fusion mechanism is introduced,and a scale filter is added to improve its anti-target occlusion and anti-scale transformation ability.The efficiency of the algorithm is improved by reducing the dimension of the position filter and the difference of the sub-grids of the response map of the scale filter.The experimental results show that the proposed method has good robustness in the case of occlusion,scale transformation,and brief loss of field of view of the tracking target,and the Center Location Erro is smaller than the other trackers participating in the experiment.2)Instead of using the traditional Scale Invariant Feature Transform(SIFT),the Speeded Up Robust Features(SURF)approach is employed to address the issue of low feature extraction efficiency brought on by an excessive description subdimension.Given the drawbacks of the Random Sampling Consensus(RANSAC)algorithm,this paper suggests a false matching point screening algorithm that combines the K-Nearest Neighbor and Progressive Sample Consensus(PROSAC)algorithms to enhance feature point matching and produce the best matching set.The experimental results demonstrate that the improved algorithm can obtain more correct matching pairs in a shorter period and that the precision and recall rate of matching points is better than before when the scale,rotation angle,luminance intensity,and gaussian noise transformation of UAV image are generated.3)The feature point extraction and matching algorithm proposed in this paper is applied to the improved incremental SFM sparse point cloud reconstruction.The PMVS(Patch-based Multiview Stereo)algorithm is selected to further expand the constructed sparse point cloud.After the reconstruction of the dense point cloud is completed,the visual processing is performed,and the 3D model is finally output.Three sets of multi-view image data sets captured by three groups of UAV clusters are used for three-dimensional reconstruction comparison experiments.The experimental results show that the proposed method can restore the target texture,color,contour and depth information after reconstruction in large-view,low-light brightness,high-similarity and fine-grained scenes.The point cloud reprojection error is small,and the reconstruction accuracy and efficiency are better than the traditional three-dimensional reconstruction algorithm.