Research On 3D Reconstruction Of Motion Blurred Visual Features

In aerospace and some other fields,it is necessary to accurately measure the 3D shape of high-speed moving targets.Compared with the traditional contact measurement method,the machine vision measurement technology with non-contact,fast speed and good flexibility is more suitable for measurement of the moving targets.However,most of the existing methods of machine vision measurement are based on the image with the clear edges and shapes,and it is difficult to apply to the 3D reconstruction of the images with motion blur effect shoot from moving objects.For this reason,this paper sought an effective and accurate 3D vision measurement method for measuring fast moving objects with certain motion blur effect.The main contents and innovations of this paper are as follows.1.On the basis of brief introduction of the structure of active visual feature points,more specially a kind of coded targets,the automatic segmentation,automatic decoding and localization of code on clear images were discussed,and some binocular stereo vision techniques were studied,which include the camera model,camera calibration method,binocular stereo vision model,epipolar constraint and 3D reconstruction method.2.For the purpose of finding a suitable motion blur model for 3D reconstruction of motion objects,a new model based on spatial motion path(SMP)was proposed.Since the large amount of real motion blurred images needed for the networks training are difficult to obtain,the GMBC model driven by six parameters for generating simulated motion blurred images of the coded targets was built according to the SMP motion blur model.3.For the purpose of recognizing motion blurred coded targets,a novel method based on the convolutional neural networks was proposed.A carefully designed convolutional neural networks(MBCNet)was constructed and analyzed,whose training data set including a lot of coded targets images with different blur degree was generated by GMBC model.The recognition performance of MBCNet and several traditional pattern recognition algorithms on motion blurred coded targets was compared and analyzed from some different angles.4.A 3D reconstruction algorithm based on the multiple view synchronized temporal images array was proposed for reconstructing spatial motion path of feature points laid on moving objects.This algorithm combines motion blur effect modeling,3D geometric reconstruction and motion path estimation in nested coupling of 3D reconstruction of moving objects,and expresses it through an integral optimization objective function.The methods for estimating initial parameters and iteration optimization of this function were given.Finally,a detailed experimental scheme was designed to reconstruct the 3D images with motion blur effect.The experimental results show that the reconstruction precision is improved by one order of magnitude compared with the conventional 3D reconstruction algorithm without considering motion blur effect.