Abstract:

Feature point detection is widely used in target recognition, tracking and 3D reconstruction. The computation of the feature point detection algorithm for big data problem is time-consuming and computation-intensive. In this paper, the parallel DoG（Difference-of-Gaussian） feature point detection algorithms are proposed. In the multi-CPU programming model based on OpenMP and the GPU parallel environment based on CUDA and OpenCL architecture, the parallel algorithms of the DoG feature point detection algorithm are designed and implemented. The comparison experiment of hallFeng image set is completed on different platforms, the experimental results show that the multi-CPU feature point detection algorithm based on OpenMP shows good multi-core scalability. The parallel GPU algorithm based on CUDA and OpenCL architecture can achieve the high speedup ratio, up to 96.79, with significant speedup effect, and which has good data and platform scalability.

Key words: Graphics Processing Unit（GPU）, Multi-CPU, Difference-of-Gaussian（DoG）, feature point detection, parallel algorithm

摘要：

特征点检测被广泛应用于目标识别、跟踪及三维重建等领域。针对三维重建算法中特征点检测算法运算量大、耗时多的特点，对高斯差分（Difference-of-Gaussian，DoG）算法进行改进，提出特征点检测DoG并行算法。基于OpenMP的多核CPU、CUDA及OpenCL架构的GPU并行环境，设计实现DoG特征点检测并行算法。对hallFeng图像集在不同实验平台进行对比实验，实验结果表明，基于OpenMP的多核CPU的并行算法表现出良好的多核可扩展性，基于CUDA及OpenCL架构的GPU并行算法可获得较高加速比，最高加速比可达96.79，具有显著的加速效果，且具有良好的数据和平台可扩展性。

关键词: 图形处理器（GPU）, 多核CPU, 高斯差分（DoG）, 特征点检测, 并行算法