面向大尺度三维重建的多核并行SIFT算法

针对大尺度三维重建中图像数据集较大、其特征提取和匹配时间消耗过长等问题,提出了一种针对有序图像数据集的滑动窗口匹配算法.该算法采用并行计算实现了对SIFT特征提取和匹配算法的加速,并在图形处理器上得以实现,减少了匹配计算量.实验结果表明,算法大大提高了特征提取和匹配的运算速度,保证了特征的稳定性和匹配精度,有效地缩短了大尺度三维重建的运算时间,确保了重建精度.

Multicore parallel SIFT algorithm facing large scale three dimension reconstruction

In order to solve the problems that the image dataset is large and the time consumption of feature extraction and matching is too long in the large scale 3D reconstruction, a sliding window matching algorithm for the ordered image dataset was proposed. In the proposed algorithm, the SIFT feature extraction and matching algorithm can be accelerated by using the parallel computing. In addition, the acceleration can be realized on a graphic processing unit, and the matching computation amount reduces. The results show that the proposed algorithm can remarkably enhance the computation speed of feature extraction and matching, ensure the stability and matching accuracy of features, effectively reduce the computation time for the large scale 3D reconstruction, and guarantee the reconstruction accuracy.