基于CUDA的动态摄影测量系统关键算法并行设计优化

Design and Optimization of Key Algorithms for Dynamic Photogrammetric System Based on CUDA

在动态摄影测量系统中,应用CUDA技术对特征点中心提取的现有并行算法进行分析和优化。针对特征点中心提取的图像预处理算法,利用GPU存储器特性进行优化,使算法的处理时间减少20%-30%。针对ROI区域提取,利用CPU主机端分配锁页内存方法进行优化,在点中心数达到20000时,并行算法比串行算法加速了近6倍。将CUDA并行技术应用于特征点匹配算法,对畸变校正、极线约束和距离判断算法进行了并行设计。试验表明,在特征点数达到20000时,并行算法相对于串行算法的速度提高了4.5倍,有效提高了动态摄影测量系统的处理速度。

The existing parallel target center location algorithms are analyzed and optimized by using the CUDA technology in dynamic photogrammetric system. By applying the GPU memory to the image processing algorithm of target center location, the time consumption is reduced by 20% - 30%. And by applying the pinned memory in CPU to the ROI extraction algorithm, the processing speed of 20000 targets is almost 6 times faster. The CUDA technology is also applied to the algorithm of targets matching for distortion correction, epipolar constraint and distance judgment, and the experimental results show that, compared with the serial algorithm, the processing speed of 20000 targets is 4.5 times faster, which effectively improves the processing speed of the system.