Research On Real-time Reconstruction Of 3D Scene Based On UAV Aerial Image Sequences

Nowadays,with the development of the information age,more and more fields are increasing the demand for 3d scene reconstruction such as earth observation,industrial measurement and military applications.However,the traditional three-dimensional reconstruction method requires manual measurement,and then drawing in the computer through the software,which will consume a lot of time costs and labor costs,and it is difficult to ensure the accuracy of outdoor 3D scene model.With the rise of UAV technology,it is possible to obtain the three-dimensional model by means of the low-cost and high-efficiency acquisition of the aerial image sequence by UAV.However,the three-dimensional reconstruction algorithm still has the problem such as reconstruction slowly,poor quality of reconstruction,three-dimensional model quality can not be guaranteed,etc.Aiming at the above problems,this paper presents a method of real-time reconstruction of three-dimensional scene based on UAV aerial image sequences.The main contents of this method are as follows:Firstly,through the UAV to obtain aerial image sequence.In order to avoid the phenomenon of video frame redundancy and video degradation caused by image matching,this paper proposes an image key frame screening algorithm,which can screen the video image sequence and filter the deep layer.This method can guarantee the quality of image feature point extraction and matching,greatly reduce the number of images to be matched,shorten the matching time,and greatly improve the efficiency of3 D reconstruction.Secondly,the image preprocessing is proposed,and an adaptive H-SIFT high-definition video image feature point extraction and matching algorithm is proposed.The algorithm can dynamically adjust the interval of image pixels convoluted by Gaussian function according to the size of the resolution by the video image sequence.Finally,the dense point cloud of the aerial image sequence is generated by structure from motion(SFM)and CMVS-PMVS algorithm.The Poisson Surface Reconstruction algorithm is used to generate the polygonal mesh surface from the dense point cloud.Dense point cloud and surface fitting and texture mapping,complete three-dimensional reconstruction.And with the CPU parallel computing algorithm and GPU CUDA acceleration algorithm to match the hardware to maximize the performance,making 3D reconstruction time significantly shortened.The experimental results show that the proposed method can reconstruct the three-dimensional model through the aerial image sequence,and has the characteristics of short reconstruction time,good reconstruction quality and high reconstruction efficiency,which can meet the real-time demand.