Citation Context

...ule can run on the GPU with more than 30 Hz for images of size 1024 × 768 tracking 1000 salient features (Sinha et al. 2007). We also propose an extension of the KLT Fig. 2 (Color online) Processing modules and data flow of the 3D reconstruction system tracker that estimates the gain of the camera (Kim et al. 2007). 3. 3D tracking/geo-location estimates the camera pose. A Kalman filter uses the 2D correspondences and GPS/INS data to estimate geo-located camera projection matrices. Alternatively, in the absence of INS/GPS data, we perform pose estimation using structure from motion techniques (Nistér 2004). 4. Sparse scene analysis examines the 3D feature points to determine three orthogonal sweeping directions (one for the ground and two for the facades). In addition, it examines the distribution of feature points in 3D to determine the most likely planes of the scene. This information enhances the effectiveness and efficiency of our multi-way plane sweeping stereo module. 5. Stereo depth estimation computes depth maps from the (geo-located) camera poses and the images using a fast GPU implementation of an advanced multi-view plane sweeping stereo algorithm. Our algorithm has provisions to dea...

Citation Context

... sets are similar to stereo SLAM in that sparse features are extracted from images to solve the correspondence problem. Projective geometry is used to define the spatial relationship between features =-=[21, 5, 29]-=-, a much harder problem to solve than correspondence in ICP. For the loop closure problem, most recent approaches to 3D mapping rely on fast image matching techniques [29, 5, 16, 20]. Once a loop clos...

Citation Context

...ethod that relies on the five-point algorithm [7] combined with bundleadjustment over the last few frames has been shown to yeld good results if one is willing to tolerate a short delay in processing =-=[8]-=-. However, most of the methods of this nature derive the camera position by concatenating transformations between adjacent frames. Over short sequences, the tracking may be accurate and jitter-free bu...

Citation Context

...into account the 3D point triangulation certainty (i.e. the shape of the covariance matrices) [18]. Once an initial pair of images is found, their relative pose is determined by the five-point method =-=[19]-=-. The remaining views of the current component are added using perspective pose estimation from 2D-3D correspondences. The order of insertion is determined from the edges of the underlying maximum spa...

Citation Context

...rmance for two hand-held video sequences (V1,V2) with respect to different 3D document strategies. Our evaluation includes also a comparison to a pure image based method, with the Five-Point algorithm=-=[14]-=- used for pose verification (relevant parameters are adjusted to get comparable timings to the 3-point method). Note, V1 was taken at nearby positions as the views from model reconstruction, therefore...

Citation Context

...n 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 stereo scheme, which is an iterative pose refinement scheme, and the monocular scheme, which was based on the framework of 5-point al...

Citation Context

...zation [12], [19] and SLAM navigation [6]. Given two image views of the same scene, five point of correspondences are selected for use in an implementation of the “five point algorithm ” described in =-=[18]-=- which, notably, is capable of dealing with coplanar correspondence points. Given two views of the same scene, five or more point of correspondences (this whole procedure should run inside a RANSAC ro...