Abstract: Aiming at the shortage of Scale Invariant Feature Transform(SIFT) algorithm in time consumption, this paper proposes a fast SIFT algorithm based on Compute Unified Device Architecture(CUDA), and analyzes its parallelism. It is further optimized according to the detailed analysis on the thread and memory model of the graphic hardware by using the power of Graphics Processing Unit(GPU). The GPU implementation runs 30~50 times faster than the CPU implementation in the experiments. It achieves 24 frames per second processing speed on 640×480 images, and is suitable for the real-time application.

Key words: Scale Invariant Feature Transform(SIFT), feature extraction and match, Graphics Processing Unit(GPU), Compute Unified Device Architecture(CUDA)

摘要： 针对尺度不变特征变换(SIFT)算法耗时多限制其应用范围的缺点，提出一种基于统一计算设备架构(CUDA)的尺度不变特征变换快速算法，分析其并行特性，在图像处理单元(GPU)的线程和内存模型方面对算法进行优化。实验证明，相对于CPU，算法速度提升了30~50倍，对640×480图像的处理速度达到每秒24帧，满足实时应用的需求。

关键词: 尺度不变特征变换, 特征提取与匹配, 图像处理单元, 统一计算设备架构

CLC Number: 

• TP312