Andrew I. Comport, Éric Marchand, François Chaumette IRISA - INRIA Rennes Campus de Beaulieu, 35042 Rennes, France E-Mail : Firstname.Lastname@irisa.fr
Augmented Reality has now progressed to the point where real-time applications are being considered and needed. At the same time it is important that synthetic elements are rendered and aligned in the scene in an accurate and visually acceptable way. In order to address these issues a real-time, robust and efﬁcient 3D model-based tracking algorithm is proposed for a ’video see through’ monocular vision system. The tracking of objects in the scene amounts to calculating the pose between the camera and the objects. Virtual objects can then be projected into the scene using the pose. Here, non-linear pose computation is formulated by means of a virtual visual servoing approach. In this context, the derivation of point-to-curves interaction matrices are given for different features including lines, circles, cylinders and spheres. A local moving edges tracker is used in order to provide real-time tracking of points normal to the object contours. A method is proposed for combining local position uncertainty and global pose uncertainty in an efﬁcient and accurate way by propagating uncertainty. Robustness is obtained by integrating a M-estimator into the visual control law via an iteratively re-weighted least squares implementation. The method presented in this paper has been validated on several complex image sequences including outdoor environments. Results show the method to be robust to occlusion, changes in illumination and misstracking.
focus on the registration techniques that allow alignment of real and virtual worlds using images acquired in real-time by a moving camera. In such systems AR is mainly a pose (or viewpoint) computation issue. In this paper a markerless model-based algorithm is used for the tracking of 3D objects in monocular image sequences. The main advantage of a model based method is that the knowledge about the scene (the implicit 3D information) allows improvement of robustness and performance by being able to predict hidden movement of the object and acts to reduce the effects of outlier data introduced in the tracking process. Real-time 3D tracking. The most common geometric features used in pose computation which are suitable for AR applications include indoor ﬁducial/marker based [3, 19, 25, 32, 33] and outdoor ﬁducial/marker based , the latter shows how the size of the marker contributes to robustness and ease of use. In the related computer vision literature geometric primitives considered for the estimation are often points [13, 7], segments , lines , contours or points on the contours [21, 24, 10], conics [28, 6], cylindrical objects  or a combination of these different features . Another important issue is the registration problem. Purely geometric (eg, ), or numerical and iterative  approaches may be considered. Linear approaches use a least-squares method to estimate the pose. Full-scale non-linear optimization techniques (e.g., [21, 23, 10]) consists of minimizing the error between the observation and the forward-projection of the model. In this case, minimization is handled using numerical iterative algorithms such as Newton-Raphson or Levenberg-Marquardt. The main advantage of these approaches are their accuracy. The main drawback is that they may be subject to local minima and, worse, divergence. It is important to note that other approaches to on-line augmented reality do not rely on pose estimation but on relative camera motion , planar homography estimation  or optical ﬂow based techniques . These methods have been shown to work in real-time and in outdoor environments, however, they are restricted to planar surfaces which may be problematic in complex environ-
This paper addresses the problem of markerless...