多视频多图像实时配准

MULTI-VIDEO AND MULTI-IMAGE REAL-TIME REGISTRATION

提出三个加速方法用于视频配准:①利用阶段划分来避免重复计算,减少计算量。②只有重叠区域的兴趣点才有提取价值,引进区域选择算法选取兴趣区域,摒弃原则不重叠的区域。③利用Canny算子对SURF算法进行改进,只针对边缘周边区域进行特征点的提取。实验结果显示结合Canny算子改进的SURF算法提取兴趣点的速度更快,同时也继承了SURF算法的强鲁棒性。三个加速方法减少了计算量,在不影响匹配精度的情况下,有效缩减了视频配准时间。

In this paper three acceleration methods are proposed for video registration： The first is to use stage division to prevent repeated calculation and to reduce computation amount; The second is to introduce regional selection algorithm to select interested region and to get rid of those regions without overlapping principles because that only the interest points in overlapping regions have the value of extraction; The third is to use Canny operator to improve SURF algorithm and to extract feature points only aiming at the perimeter areas of the edge. Experimental result shows that the SURF algorithm improved by Canny operator has faster speed in extracting the interest points. These three acceleration methods reduce the computation amount greatly and cut down the time of video registration effectively under the condition of the matching accuracy unaffected.