项目名称: 动态环境下基于概率图模型的机器人地点识别及实时语义地图构建新方法
项目编号: No.61201362
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 电子学与信息系统
项目作者: 杨金福
作者单位: 北京工业大学
项目金额: 26万元
中文摘要: 动态场景下定位与地图构建是当前机器人研究领域的核心问题。由于真实动态场景充满复杂性和不确定性,此类问题的研究具有挑战性和非常重要的研究价值。本课题以机器人地图构建为研究对象,以图像为信息载体,以机器视觉为主要技术手段,研究机器人地点和物体识别以及实时语义地图构建方法。首先,在分析动态场景复杂变化特性的基础上,提出一种基于BOW技术的快速局部不变性特征的提取和选择方法;然后针对场景的复杂性和不确定性,提出一种基于概率主题模型的机器人地点和物体识别方法,以及一种基于稀疏在线核学习判别的运动物体实时检测方法;进而将图论、三维计算机视觉理论以及人工智能理论融合,创新性地建立一个适用于动态场景的实时语义地图构建计算框架;最后构建一套基于机器人地点和物体识别的实时语义地图构建软件原型系统。
中文关键词: 概率图模型;地点识别;机器视觉;语义地图;
英文摘要: Visual-based localization and mapping is the main research issue in robotics community.It is very difficult to sovle the problems of complexity and uncertainty in dynamic enviroment.In this proposal,we plan to study on the methods of place and object recogniton based on probabilistic graphical model(PGM) and building a novel semantic knowledge-based map for mobile robots. First,an algorithm of local feature extraction and selection based on bag of words technique is proposed to cope with scale and affine changes of images. Second,an efficent method for place and object recognition is described to enchance recogniton rate and reduce computational complexity.Then,in order to identify effectively moving object in dynamic enviroment, we present a new detecting method based on sparse on-line leaning with kernels.After that, a novel framework of real-time mapping with semantic knowledge using graphical theory,3D visual computing theory and artificial intelligence theory is proposed. At last,a demo system is designed to perform experiments on building a semantic map for mobile robots.
英文关键词: probabilistic garphical model;place recognition;robot vision;semantic map;