摘要行人检测是目标检测中的一个热点研究领域,并且成为了越来越多研究者的研究课题。稀疏表示理论的快速发展,使得其在模式识别领域得到了越来越多的应用,也应用到了行人检测问题上。然而,在稀疏表示中,字典的学习是一个非常重要的步骤。通常,为了适应不同的输入信号的需求,训练学习到的字典都是过完备的。本文使用 K-SVD 字典学习算法来训练过完备的行人字典,采用稀疏表示实现行人检测,得到在行人检测中最优的字典尺寸。在此实验之后,提出了采用虚拟场景行人为训练集,来进行现实世界的行人检测。 31429 毕业论文关键词 K-SVD算法 字典训练 稀疏表示 多重稀疏行人检测 虚拟行人
Title Dictionary Training based on Sparse classification pedestrian detection
Abstract Pedestrian detection is a hot research field of target detection, and becoming the research topic of more and more researchers. The rapid development of the sparse representation theory makes it more and more used in the field of pattern recognition, also used in the problem of pedestrian detection. However, in the sparse representation, the dictionary learning is a very important step. In order to meet the requirement of different input signal, the trained learning dictionary is overcomplete. In this paper, we use the K-SVD dictionary learning algorithm to train a overcomplete pedestrian dictionary, then use sparse representation to achieve pedestrian detection, finally get the optimal size of dictionary in pedestrian detection. After the experiment, we put forward using virtual pedestrians as the training set to work on the real-world pedestrian detection.
Keywords K-SVD algorithm dictionary learning sparse representation multiple sparse pedestrian detection virtual pedestrian
目次
1引言1
1.1国内外对行人检测的研究2
2图像稀疏表示原理6
2.1过完备稀疏表示6
2.2稀疏表示优化方法8
2.3字典学习11
2.4稀疏表示的应用14
3基于多重稀疏的行人检测方法15
3.1字典训练16
3.2多重稀疏字典直方图21
3.3多重稀疏字典特征提取23
3.4实验24
4虚拟行人训练集的生成28
4.1虚拟行人的生成28
4.2后期研究工作29
结论31
致谢32
参考文献33