2D/3D视觉信息融合仿生SLAM关键问题研究

项目名称： 2D/3D视觉信息融合仿生SLAM关键问题研究

项目编号： No.61503401

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 肖军浩

作者单位： 中国人民解放军国防科技大学

项目金额： 21万元

中文摘要： 长时间大范围可靠的同步定位与建图（SLAM）能力不足是制约移动机器人走向实际应用的瓶颈问题之一。相比较基于概率计算和图优化的方法，受大脑导航定位机制启发的2D视觉仿生SLAM已经显示了其处理长时间大范围SLAM问题的优势，但还存在精度低、缺少几何信息、对光照条件鲁棒性不高等缺陷，尚未研究利用对动物导航定位而言重要的3D深度信息。项目研究融合2D/3D视觉信息的仿生SLAM，以克服上述缺陷：1、提出基于深度图的与特征无关的6自由度视觉里程计，满足实时性和精度要求；2、提出一种新的拓扑米制混合地图表示方法，基于2D/3D视觉信息增量式地构建全局拓扑一致，局部几何信息精确的混合地图；3、提出基于多模态信息的闭环检测方法，最大限度地融合两个视觉模态的信息量，提高闭环检测的鲁棒性和准确率。研究成果对提高机器人SLAM能力具有重要意义，能为移动机器人早日实用于复杂环境奠定一定的理论基础和提供技术支撑。

中文关键词： 多模态视觉信息融合；仿生SLAM

英文摘要： A bottle-neck problem which restricts mobile robots moving from controlled and structured environments to practical applications is that the long-term Simultaneous Localization and Mapping (SLAM) ability is insufficient. Inspired from the brain navigation mechanism and based on its computational model, pure 2D visual bio-inspired SLAM has already overstepped fliters and graph optimization based SLAM approaches. However, it still suffers from limitations such as low accuracy, lack of geometric information, relies on sufficient illumination. Therefore, we propose to fuse multi-modal visual information in the bio-inspired SLAM system to overcome the above-mentioned limitations: .1. A featureless 3D vision based visual odometry is proposed, which can fulfill real-time and accuracy requirements with 6 degree of freedom pose estimation..2. A novel topological-metric hybrid map representation is proposed; The hybrid map is incremental constructed based on 2D/3D visual information, where the topological map has global consistency and local metric maps are accurate according to geometry..3. A multi-modal visual information based loop closure detection method is presented, which makes use of as much information in both modalities to improve the robustness of accuracy of loop closure. .The research result is significant for improving the SLAM ability for mobile robots, and can provide certain theory foundation and technical support to make mobile robots really applicable in complex environments.

英文关键词： multi-modal visual information fusion;bio-inspired SLAM