No abstract

We present a data-driven approach that synthesizes tree animations from a set of pre-computed motion data. Our approach improves previous motion synthesis algorithms for character animation in several aspects. We first introduce a simple yet effective sampling scheme to generate a rich and reusable motion database for each tree model. We also propose a novel technique to generate a fine set of transitions that are uniformly distributed in the motion database. The transition lengths are adaptively determined according to the similarity of the transiting frame pairs. In the runtime, we employ a greedy searching algorithm to synthesize smooth tree animations under an adjustable wind condition. Experimental results show that our approach achieves comparable quality to physically based methods, while in orders of magnitude faster performance.

Computing world distances of scene features from the captured images is a common task in image analysis and scene understanding. Previous projective geometry based methods focus on measuring distance from one single image. Hence, the scope of measurable scene is limited by the field-of-view (FOV) of one single camera. In this paper, we propose one method of measuring distances of line segments in real world scene using panorama representation. With a full view panorama, the scope of measurable scene is increased and can fully cover the sphere of 360 × 360 FOV. With panorama representation, distance of long-range features, which can not be fully captured by a single image, can be measured from the panoramic image. A prototype system called PanoMeasure is developed for enabling user to interactively measure the distances of line segments. Experiments with simulated data and real measurement results verify that the method offers high accuracy.

No abstract

No abstract