基于CUDA的红外小目标检测算法的快速实现-计算数学专业论文.docxVIP

II II Abstract With the rapid development of computer vision technology, infrared target detection and tracking has found wide application in such areas as military, aerospace, remote sensing and so on. In the pass the research of infrared target detection method focused on improving the target detection accuracy, but this paper studied real-time target detection. In order to accelerate the speed of small infrared target detection, this paper introduces GPU and CUDA. This paper first introduced the LCM-based small infrared target detection algorithm and presented the basic flow chart of the algorithm, then parallelized the method, mapped the traditional serial LCM algorithm to the CUDA parallel programming model and proposed a fast CUDA-based LCM method. According to the basic flow the LCM parallel algorithm, on GPU the first step is calculating multi-scale LCM, then computing the threshold by using shared memory and reduction operations, next using a bit array to mark the target position, finally calculating the target position on CPU, and marking target on the image. At last, the LCM parallel algorithm was programmed under the platform of CUDA5.0 and the feasibility of the CUDA-based LCM method is verified by the experiments. The experimental results also showed that: under the premise of ensuring high accuracy, the algorithm proposed in this paper is more efficient than the traditional LCM algorithm, and could achieved an approximately 10 times speed-up with the CPU implementation. Key words: CUDA, Target detection, LCM, GPU, Derived kernel III III 目录 摘 要 I HYPERLINK \l _bookmark0 Abstract II HYPERLINK \l _bookmark1 1 引言 HYPERLINK \l _bookmark2 1.1 研究背景及研究意义 (1) HYPERLINK \l _bookmark4 1.2 目标检测方法概述 (2) HYPERLINK \l _bookmark5 1.3 GPU 和 CUDA 概述 (4) HYPERLINK \l _bookmark8 1.4 本文内容简介 (6) HYPERLINK \l _bookmark9 2 导出核模型和 LCM 算法 HYPERLINK \l _bookmark10 2.1 导出核模型 (8) HYPERLINK \l _bookmark11 2.2 LCM 算法 (10) HYPERLINK \l _bookmark12 基于 CUDA 的 LCM 算法实现