Color-based video stabilization for real-time on-board object detection on high-speed trains

Abstract

Abstract

Digital video stabilization through curve warping techniques

“…In order to confirm the validity of the proposed algorithm, a series of experiments have been performed comparing our approach with other digital stabilization techniques based on integral projections [8] and block matching [4].…”

Fast and accurate global motion compensation

“…The vertical and horizontal projections between frames i and iÀ 1 are registered (through cross-correlation or another registration technique), thus producing vertical and horizontal displacement estimations. This method proved very fast, in [20] it was used for real-time obstacle detection on high-speed trains. Image projections were also used for global motion estimation in [22,23].…”

“…Table 2 shows the obtained errors for the Weizmann robust dataset. 'Proj+ XCorr' refers to the use of image projections and cross-correlation to get estimates of global motion, i.e., the method of [20]. Fig.…”

“…The work of [20,21] on video stabilization, for example, uses integral projections of the intensity image to infer a 2D translation. The vertical and horizontal projections between frames i and iÀ 1 are registered (through cross-correlation or another registration technique), thus producing vertical and horizontal displacement estimations.…”

A robust video stabilization using optimized dynamic time warping

“…This block projection is matched in the corresponding block search space in the next frame. In the same field, Piva et al [7] proposed the full frame integral projection (IP) curve extraction for the real time train application. This method uses point to point euclidean distance based matching for motion estimation.…”