项目名称: 基于视觉的智能机器人场景理解方法研究
项目编号: No.61305114
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 何洪生
作者单位: 东北大学
项目金额: 23万元
中文摘要: 环境场景理解是智能机器人在动态环境下自主决策和行为优化的前提,也是决定智能机器人功能的关键。近来,场景理解逐渐成为智能机器人环境感知和智能信息处理的瓶颈,基于视觉的场景内容和结构理解方面的研究还有待于进一步深入。本项目拟基于机器人视觉系统图像序列的上下文信息,分析场景几何空间结构并构建浮现特征向量,采用视觉注意机制提取场景外观特征,并提出概率图场景理解模型,以研究自然场景结构和内容的信息理解问题,同时阐明场景理解中信息最优化的机理。针对场景结构分析,提出高效的场景几何基元提取算子;针对场景图像特征提取,引入融入智能机器人上层任务反馈的视觉注意机制;针对理解框架,提出结合结构基元和图像特征的场景推理理解模型。本项目的研究成果将进一步发展机器人视觉的理论与方法,并为实际智能机器人的环境感知的研究提供理论依据和新的设计思路,对智能机器人的发展具有重要的理论意义和应用价值。
中文关键词: 场景理解;特征提取;智能机器人;传感器融合;动态分析
英文摘要: Scene understanding plays fundamental roles in autonomous decision-making and behavior optimization while determining the functionalities of intelligent robots in dynamic environments. However, there are many research challenges in understanding of scene structures and contents, which limit the capabilities of intelligent robots for sensing and intelligent information processing. Based on visual context, this project will analyze scene geometrical structure by composing emergent features, extract visual scene features using attention models, and propose probability graph models for scene understanding, in order to perceive the scene structure and contents as well as to illustrate the optimization mechanism in scene understanding. For scene structure analysis, we will propose effective and efficient Geon extractors; for visual scene features, we will introduce an attention model with feedback from the decision-making level of intelligent robots; for scene understanding, we will propose an integrated scene inference model by fusing geometrical and visual features. The expected contributions of theories and techniques in robotic vision will provide theoretical guidelinesand design inspirations for actual intelligent robots in sensing and perception, benefitting the development of intelligent robots both in theory a
英文关键词: Scene Understanding;Feature Extraction;Intelligent Robots;Sensor Fusion;Dynamics Analysis