项目名称: 基于稀疏描述的非结构化环境地形识别研究
项目编号: No.61272220
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 赵春霞
作者单位: 南京理工大学
项目金额: 80万元
中文摘要: 复杂的非结构化环境下地形识别研究是面向移动机器人环境感知和识别的一个重要的问题。环境受时间、光照、干扰、遮挡等因素的影响,给非结构化环境地形识别带来了许多亟待解决的问题,如:数据的多模态性、无适用性观念等。如何缓解这些问题,以提高地形识别的能力是目前面临的一个重大挑战。面对非结构化环境地形识别的众多实际问题,仅限的成果无法有效地适应具体应用。针对以上问题,本项目将以在非结构环境下地形图像的稀疏描述、基于稀疏描述的特征融合、以及组合模型的一位密度估计为攻关内容对非结构环境地形识别问题进行基础研究探讨。本项目的创新之处是拟提出鲁棒的稀疏描述算法,通过用联合1、2范数表征传统的稀疏描述模型,并提出一个特征选择机制;基于稀疏描述,拟提出融合稀疏判别特征、环境要素特征等方法;拟提出基于稀疏多投影支持向量机并建立了一维密度估计的组合模型。本课题将有利推动模式识别、机器学习和智能机器人理论、技术和应用。
中文关键词: 地形识别;稀疏描述;非结构化环境;机器学习;
英文摘要: Terrain recogntion in complex and unstructured environments is an extremely important problem in the field of robot intelligence. The evironment is always affected by time、illumination、occlusion and corruption,leading to many diffucult but unsolved problems in terrain recognition, such as data mulimodality,no concept of applicablity and so on. How to resolve these issues to provide the further performance improvement of terrain recogntion is currently a challenge.The previous works cannot be effectivley applied to terrain recogntion due to the existing above problems. Aiming at overcoming the above challenge, this project mainly discusses the saprse representation、the sparse representation based feature fusion as well as the one-dimensional density estimatation models. The innovation of the project is: (1) to propose a robustly sparse representation algorithm, which includes the constrcutions of a sparse model via joint 1、2 norms and a feature selection approach; (2)to develop a sparsity-based feature fusion system incorporating sparse discriminant features and environment factors; (3) to present a multi-weight projection SVM and construct a combination model based the one-dimensional density estimation. This study will promote the development of theroey, technology and application in pattern recogntion, machine
英文关键词: Terrain Recognition;Sparse Representation;Unstructured Enviroments;Machine Learning;