Research On 3D Reconstruction Algorithms Of Large Outdoor Scenes Based On Video Images

With the continuous development of computer technology and the continuous updating of digital equipment,more and more fields have improved the demand for the accuracy of three-dimensional model reconstruction.However,it is a time-consuming and labor-consuming task to manually measure and construct three-dimensional models by using modeling software.It is a hotspot to study how to obtain the three-dimensional model directly and quickly through the algorithms.However,the three-dimensional model reconstructed by the algorithms still has some problems,such as slow reconstruction speed and insufficient quality of the three-dimensional model.In order to solve the above problems,this paper studies and improves the three-dimensional reconstruction algorithm of large outdoor scenes based on video images.The main contents are as follows:Firstly,in the stage of video image preprocessing,a hierarchical key frame extraction technology is proposed,which combines shallow extraction with deep filtering.Shallow extraction uses improved key frame sampling extraction algorithm,which takes into account the factors of image sharpness evaluation to extract a group of clear and partly redundant video image sequences.Deep filtering finds regions of interest by graying and segmenting images and eliminates redundant images in image sequences by combining image similarity measurement algorithm based on histogram Euclidean distance.Experiments show that the hierarchical key frame extraction technology can eliminate more than 90%of the video image frames,and the reconstruction efficiency is 6-15 times higher than the traditional method.Secondly,in the stage of three-dimensional reconstruction of video image sequence,a method of image feature point matching based on spatiotemporal sequence characteristics is proposed.The number of images participating in feature point matching is limited by the"correlation"between images.And then,the matched feature points are restored to three-dimensional space by using the structure from motion algorithm,and the position and attitude of the camera are calibrated.Point clouds are clustered and expanded into dense three-dimensional point clouds by multi-view stereo algorithms.Finally,Poisson surface reconstruction algorithm is used to mesh dense point clouds,and texture mapping is carried out to complete three-dimensional reconstruction.Finally,the parallel acceleration method based on GPU is used to make full use of the advantages of GPU in parallel processing and large-scale computing.The parallel analysis and optimization of the structure from motion algorithm and multi-view stereo algorithm are carried out.The time of three-dimensional reconstruction is shortened from the two levels of algorithm and hardware.Compared with serial three-dimensional reconstruction algorithm,the reconstruction efficiency is improved by nearly 9 times.Combined with the practical application scene of three-dimensional reconstruction,the significance of three-dimensional reconstruction technology in cultural relic's protection,digital research and display is discussed.