Computing Structure and Motion of General 3D Curves from Monocular Sequences of Perspective Images
نویسندگان
چکیده
This report discusses the problem of recovering 3D motion and structure for rigid curves. For this purpose, we use long monocular sequences of images of the curve and compute some set of derivatives (up to the second order) that are deened on the so-called spatio-temporal surface generated by the curve. For general 3D rigid curves, there is exactly one constraint for each image point that relates these derivatives to the kinematic screw and its rst order time derivative. These equations derive directly from the perspective equations. However, this theory ignores (as does the projective model of cameras) the fact that the viewed object must be entirely in front of the focal plane. In this paper, we provide a complete description (up to second order) of these conditions and of their geometrical consequences for 3D reconstruction. Moreover, the relations between the geometry of the image curve and the kinematic screw are investigated. We show that ignoring these constraints leads to erroneous solutions with a high probability even when using perfect data. Then, an algorithm which takes advantage of these constraints is proposed. The accuracy of the obtained results is demonstrated on both synthetic and real image sequences. Calcul du mouvement et de la structure de courbe rigides 3D ggnnrales partir de ssquences d'images perspectives monoculaires. RRsumm : Ce rapport traite du probllme de la ddtermination du mouvement et de la structure d'une courbe rigide 3D partir d'une longue ssquence d'images monoculaires de celle-ci. Cette ssquence permet de calculer une certain ensemble de grandeurs diiirentielles (jusqu'' l'ordre 2) qui sont dddnies sur la surface spatio-temporelle engendrre par la courbe. Pour des courbe 3D rigides ggnnrales, il existe en chaque point de la courbe image exactement une contrainte reliant ces ddrives au torseur cinnmatique associi au mouvement et la ddrive premiire de ce torseur. Cette relation provient directement des equations de projection perspective. Cepen-dant, cette thhorie ne tient pas compte (tout comme le moddle de cammra utiliss) du fait que l'objet observ doit se trouver entiirement devant le plan focal pour pouvoir tre observable. Dans ce rapport, nous donnons une description compllte (jusqu'' l'ordre 2) des conditions qui doivent tre rralisse pour assurer l'observabilitt de l'objet et des conssquences ggommtriques de ces conditions sur la reconstruction 3D. Une analyse sommaire du nombre de ces contraintes en fonction de la ggommtrie de la courbe image et de la position du torseur cinnmatique est galement …
منابع مشابه
Rigid Motion and Structure from Curves Using Scale Space
This article presents an extension of works by Olivier Faugeras and Th eo Papadopoulo on moving 3D curves. They have developed a theory for the motion elds of 3D curves seen in a monocular sequence of 2D images and described methods for computing the kinematic screw (; V) of the curve and its time derivative (_ ; _ V) in the case of rigid 3D motion. Here the results of implementing one of these...
متن کاملAutomatic detection of liver tumor motion by fluoroscopy images
Background: A method to track liver tumor motion signals from fluoroscopic images without any implanted gold fiducial markers was proposed in this study to overcome the adverse effects on precise tumor irradiation caused by respiratory movement. Materials and Methods: The method was based on the following idea: (i) Before treatment, a series of fluoroscopic images corresponding to different bre...
متن کاملDeformable and articulated 3D reconstruction from monocular video sequences
This thesis addresses the problem of deformable and articulated structure from motion from monocular uncalibrated video sequences. Structure from motion is defined as the problem of recovering information about the 3D structure of scenes imaged by a camera in a video sequence. Our study aims at the challenging problem of non-rigid shapes (e.g. a beating heart or a smiling face). Non-rigid struc...
متن کاملJoint Semantic Segmentation and 3D Reconstruction from Monocular Video
We present an approach for joint inference of 3D scene structure and semantic labeling for monocular video. Starting with monocular image stream, our framework produces a 3D volumetric semantic + occupancy map, which is much more useful than a series of 2D semantic label images or a sparse point cloud produced by traditional semantic segmentation and Structure from Motion(SfM) pipelines respect...
متن کاملIntegrated 3-D Analysis and Analysis-Guided Synthesis of Flight Image Sequences
This paper is concerned with three-dimensional (3D) analysis, and analysis-guided syntheses, of images showing 3-D motion of an observer relative to a scene. There are two objectives of the paper. First, it presents an approach to recovering 3D motion and structure parameters from multiple cues present in a monocular image sequence, such as point features, optical flow, regions, lines, texture ...
متن کامل