Deblurring Of Multi-view Images And Three-dimensional Reconstruction

With the widespread attention from people, the technology ofintelligent three-dimensional industrial inspection in the field of machine vision hasbeen developing rapidly in recent years. And three-dimensional reconstruction based on theimages of targets is the key technology of three-dimensional vision inspection. In actualwork, the target objects to be inspected and the environment of objects are diverse. Thetarget data may be lost for background and noise interference from multi-view imagesobtained by machine vision system. So the reliable information of interest target is difficultto be extracted. The images of object acquired by imaging system can be blurred for thestate of motion of target. In these situations, the obtained three-dimensional informationbased on multi-view images of the targets will always carry a degree of imprecision oreven wrong, which restricts the development of three-dimensional vision inspection andthe other practical application. Aiming at the above difficulties, the technologies of targetextraction of the images, multi-view images deblurring and three-dimensionalreconstruction based on multi-view images are researched in this paper.Firstly, a multi-view imaging and three-dimensional reconstruction experiment systemis designed and implemented to meet the requirements of the three-dimensional inspectionof industrial products. In order to ensure the quality of multi-view images,high-resolution industrial cameras are adopted in our machine vision system. And thecontrol software of multi-view imaging system is developed based on the programminginterfaces of camera system to debug the quality of multi-view images in real time. Theprocessing of these steps has provided a stable environment for the subsequent multi-viewimages processing.Then, a segmentation algorithm based on graph theory is proposed to extract thetargets from the multi-view images. After preprocessing of images, the targets inmulti-view images are extracted with the segmentation method based on graph theory. Thebackground information of images is suppressed and the valuable information of thetargets is preserved at most degree. The initial information is provided accurately and completely for the three-dimensional reconstruction experiments about the target objects.Finally, in order to solve the problem about inspecting actual parameters of theproducts, a method of multi-view images deblurring and three-dimensional reconstructionis proposed to obtain the three-dimensional information. The three-dimensional surfaceinformation of object is achieved through the steps of affine, stereo matching and Poissonsurface reconstruction.A series experiments were performed to test the method of multi-view imagesdeblurring and three-dimensional reconstruction. In the course of the experiments, thefeasibility of imaging system has been verified at first. Next, the target extraction,deblurring of multi-view images and three-dimensional reconstruction had beenimplemented. The experimental results show that the method of multi-view imagesdeblurring and three-dimensional reconstruction can obtain the three-dimensionalinformation accurately and steadily. The proposed algorithm also provides the crucial datafor three-dimensional industrial inspection and other practical applications.