项目名称: 融合全景与深度信息的机器人室内复杂环境目标搜索与观测规划方法研究
项目编号: No.61473090
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 自动化技术、计算机技术
项目作者: 何炳蔚
作者单位: 福州大学
项目金额: 80万元
中文摘要: 随着视觉技术在机器人领域的广泛应用,使机器人得以向智能化和复杂化方向更好的发展。机器人在室内复杂场景进行目标搜索的研究近年来已成为研究的热点和重点,此研究对于扩展机器人的应用范围有着重大意义。 针对以往研究中机器人配以普通视觉传感器进行目标搜索中的不足(如:视场范围有限,需多个视角获取场景图像,目标检测的效率低)。本项目拟提出将全向视觉与RGB-D系统搭建在同一台机器人上,成为机器人获取外界信息的主要来源。对室内复杂场景中目标检测、搜索路径规划等关键问题进行了研究,其中:通过融合全景与深度信息,提升机器人在复杂环境中的感知能力;将深度信息与视觉注意机制模型融合实现室内复杂场景显著区域与显著目标的检测,提高复杂场景目标检测效率;基于混合视觉系统的多帧数据在线融合算法,大大提升数据融合效率和表面数据精度。研究成果将提升机器人在复杂场景中自主搜寻、获取目标的能力,提高机器人环境适应性和智能水平。
中文关键词: 全向视觉;深度信息;信息融合;目标搜索;室内环境
英文摘要: With the wide application of visual sensors, robots have been developed with a lot of intelligent functions. Visual target search is becoming more and more important for robots in the real-world domestic scenes in recent years, and has a big impact on the expansion of the application area of robots. In previous literatures, many robots utilize classic vision sensors to perceive the environment. However, these sensors with the limited field of view require extensive image processing, thus increase the operation cost and reduce the efficiency of information acquisition. To overcome this disadvantage, a novel hybrid vision system consisting of an omnidirectional camera and an RGB-D system is proposed and implemented on a mobile robot. The corresponding technologies, which include the object detection and search path planning, are studied. In this proposal, the perception cability of the robot will be enhanced by fusing the omnidirectional visual and depth information, and the efficiency of target detection will be improved by fusing the depth information model and visual attention mechanism to detect significant area and targets. An online multi-frame data fusion algorithm based on the hybrid vision system will greatly enhance the efficiency of data fusion and surface data accuracy. This research will enhance the ability of robot on automatic target search in complex scenes and improve its adaptability in different environments.
英文关键词: omnidirectional vision;depth information;information fusion;target search;domestic environments