Citation Context

... motions were presented. Their method used simple gradient descent feature tracking and was therefore unable to match features during high acceleration or close loops after periods of neglect. Nistér =-=[39]-=- presented a real-time system based very much on the standard structure from motion methodology of frame-to-frame matching of large numbers of point features which was able to recover instantaneous mo...

Citation Context

...and Shafer, 1987; Matthies, 1989), with a consistent level of accuracy reported more recently (Olson et al., 2003). Similar work has been reported elsewhere (Zhang et al., 1988; Lacroix et al., 1999; =-=Nistér et al., 2004-=-). Recently, Nistér et al. have reported a successful real-time Visual Odometry implementation (Nistér et al., 2006; Nistér, 2004). This implementation contains two motion estimation schemes: the ster...

Citation Context

...ts that are well localized, because the corners are relatively invariant to change of view. Both these detectors can be implemented very efficiently and have been used in structurefrom-motion systems =-=[2,7,8]-=- because of their accuracy. However, they are not invariant to scale and therefore not very stable across scale changes, which happen constantly with a moving camera. The Harris-Laplace and the Hessia...

Citation Context

...sion communities have developed many techniques for 3D mapping using range scans (e.g. Thrun et al. [53], Triebel et al. [56], May et al. [32], Newman et al. [35]), stereo cameras (e.g. Nister et al. =-=[37]-=-, Akbarzadeh et al. [1], Konolige et al. [27]), monocular cameras (e.g. Clemente et al. [8]), and even unsorted collections of photos (e.g. Snavely et al. [51], Furukawa et al. [14]). Most mapping sys...

Citation Context

...ers [12] and FAST features [33], which are relatively quick to compute and resilient against small viewpoint changes. Methods for robustly matching features across frames include RANSAC-based methods =-=[28, 18, 22]-=- and graph-based consistency algorithms [17]. In the motion estimation process, techniques have ranged from directly minimizing Euclidean distance between matched features [16], to minimizing pixel re...

Citation Context

...ally by updating probability distributions over features and camera pose parameters; and bundle adjustment style optimisation over selected images from the live stream, such as a sliding window [18], =-=[20]-=-, or in particular spatially distributed keyframes [13], [14] which permit drift-free longterm operation. Understanding of the generic character of localisation and reconstruction problems has recentl...

Citation Context

..., [3], [4], some have been extended to use points and lines [5], [6], and some work also for coplanar points [3], [4]. Recent success has also been reported for online structure and motion estimation =-=[7]-=-; where many interest points are extracted, frame-to-frame correspondence is rather easy and no a priori reference to a scene coordinate system is required (for early work in this direction see [8]). ...