并行正向反馈ICP算法研究

ICP算法用于点云对齐，对点云质量有很高的要求，Kinect设备采集的原始点云较为粗糙，给ICP计算带来很大不便。文中设计了一种并行正向反馈ICP算法，将新旧点云带权叠加，以优化新点云的质量，并提供给ICP的下一次循环，同时又采用一种基于GPU的并行二叉树约减算法，提高计算速度。该算法用于增强现实的注册环节，运行结果流畅稳定。这种改进算法增加了深度数据的可用性和稳定性。

Research of Positive Feedback ICP Algorithm in Parallel

The ICP algorithm,which was utilized to align points clouds,had to use high quality points clouds. The points clouds,captured by Kincct,was too rough to be utilized. A positive feedback ICP algorithm in parallel was introduced in this paper. It added current point clouds with the last point clouds, to optimize the quality of point clouds. Tben the point clouds could bc provided into the next ICP turn. And used GPU to calculate matrixes, so that the time was saved. This algorithm was fast and steady when used into augmented reality, which augmented the liability and stability of depth data.