基于CPU—GPU协同处理的DSM粗差探测方法研究

DSM Gross Error Detection Algorithm Based on CPU- GPU Collaborative Parallel Processing

DSM粗差探测是高精度DSM数据生产中的关键技术环节，本文针对影像匹配自动生成的DSM数据，设计了一种基于虚拟格网的DSM数据粗差探测CPU—GPU协同并行处理算法。该算法首先在CPU上统计高程频数，其次在GPU上剔除高程频数低的点，再利用未剔除点在CPU上统计每个虚拟格网所包含的点数以及这些点的高程值之和，最后在GPU上进行阈值的计算和剔除粗差点。实验结果表明：该算法粗差探测准确率高，且与CPU算法相比，执行效率显著提升，可满足实时处理的要求。

Gross error detection of Digital Surface Model （ DSM ） is a key technical step in high - precision DSM data pro- duction. Dealing with automatically generated DSM data by image matching, this paper designs a CPU - GPU collaborative paral- lel processing algorithm based on pseudo - grid for DSM gross error detection. The algorithm firstly counts the height frequency on CPU, then eliminates low points of height frequency, calculates the number and the sum of elevation values of points in each pseudo- grid on CPU using points that have not been eliminated, and finally calculates the threshold value and eliminates gross error points. The experimental results show that this algorithm has high accuracy in gross error detection, and the efficiency is improved significantly compared with that of CPU algorithm, it also can meet the requirement of real - time processing.