摘 要近两年来,旋翼无人机风靡全球,在各个领域中都能够见到它的身影,其中,以 无人机为载体的视觉跟踪系统成为了近期国内外研究的热点课题之一。以此同时,计 算机视觉也是当下人工智能发展的重要技术之一。本论文以旋翼无人机为平台,在其 上面搭载三轴无刷云台、视觉传感器和微型计算机——Jetson TK1,以此搭建一套用 于无人机平台上的视觉跟踪系统。该系统不但能够隔离无人机飞行时候带来的扰动, 还可以保证跟踪目标始终处在视觉传感器的视野中,以及将跟踪目标方位信息告知飞 行控制器,实现无人机对目标的跟随。针对视觉跟踪,本文介绍了一种基于核函数相 关分类器的跟踪器。该跟踪器是利用核函数技巧对多通道的方向梯度直方图特征进行 融合,最后通过循环位移的方式对视频序列图像进行稠密采样并结合快速傅立叶变换 对分类器进行快速的训练,使得训练的分类器对目标检测的能力更强。以此,在嵌入 式平台编写基于 OpenCV 库的相对应的跟踪程序。80790
毕业论文关键词:视觉跟踪;核函数相关分类器;旋翼无人机;OpenCV;Jetson TK1
Abstract In recent years, rotor-wing UAV has swept all over the world。 It has been used in the walks of life。 And UAV as the carrier of visual tracking has become one of hot projects in the domestic and overseas。 At the same time, computer vision is one of the important technologies in the development of artificial intelligence。 This thesis takes rotor-wing UAV platform, equipped with three axis brushless head thereon, visual sensor and mini-computer Jetson TK1, so that we can build a set of visual tracking system。 This system can not only provide the isolation of UAV fight disturbance, but also make sure the tracking object still being in the sight of vision sensor, so that the UAV can adjust flight control unit to continue tracking the object by processing the position information of tracking。 This paper is to introduce a kind of tracker, based on kernelized correlation filters, for visual tracking。 This tracker uses kernel to amalgamate multi channel directional HOG, finally it takes the dense samples of video sequence image by cyclic shift and combines these samples with fast fourier transform to trains categorizer quickly, which makes categorizer trained more sensitive。 Based on this tracking algorithm, we programme the corresponding tracking code with the library of OPENCV。
Keywords: visual tracking; kernelized correlation filters; rotor-wing UAV; OpenCV; Jetson TK1
目 录
第一章 绪论 1
1。1 研究背景及意义 1
1。2 旋翼无人机跟踪系统研究的现状和发展 1
1。3 旋翼无人机跟踪系统的难点及性能要求 3
1。4 本文的主要内容 3
第二章 无人机跟踪系统平台 4
2。1 硬件模块 4
2。2 系统环境 5
2。3 CUDA 与 OpenCV 6
2。4 小结 7
第三章 核函数相关分类器的快速跟踪方式 9
3。1 引言 9
3。2 基于核函数相关分类器的跟踪器 9
3。2。1 理论基础 9
3。2。2 核函数理论及应用 12
3。2。3 相关核函数