结合图像内容匹配的机器人视觉导航定位与全局地图构建系统

发布时间：2018-07-02 23:00

本文选题：机器人视觉 + 机器人自主定位　；参考：《光学精密工程》2017年08期

【摘要】：为了解决机器人室内定位时的绑架问题和相似物体的干扰,设计了一种具有图像内容匹配功能的视觉系统,从而使机器人能有效提取关键帧序列构建室内全局地图并实现自主定位。考虑影响图像内容匹配的主要干扰是机器人视角和位移造成的图像畸变,本文通过对室内物体的图像畸变建模与特征分析,设计了一种图像内容匹配方法。该方法以图像重叠区提取、基于子块分解匹配的重叠区重建两部分为核心,可将待匹配的两帧图像畸变调整为一致后再进行内容匹配并准确解算它们的相似度。其能有效利用各个房间内不同的景物和布局信息来消除相似物体的影响,从机器人学习环境时采集的视频中提取空间间距大且重叠相连的关键帧序列建立整栋建筑内部的全局导航地图。机器人工作时,实时视觉的图像内容与地图关键帧序列匹配,提取出与每个时刻视觉图像最相似的关键帧对机器人实施定位。在由3个房间和2条走廊组成的实验区进行了实验测试,结果表明:机器人可有效消除相似物体的干扰,绑架发生时仍可通过与全局地图匹配实施准确自主定位,匹配准确率≥93%,定位精度误差(RMSE)0.5m。
[Abstract]:In order to solve the problem of kidnapping and the interference of similar objects, a visual system with image content matching function is designed. Thus the robot can effectively extract the key frame sequence to construct the indoor global map and realize autonomous localization. Considering that the main interference affecting image content matching is image distortion caused by robot visual angle and displacement, this paper designs an image content matching method by modeling and feature analysis of image distortion of indoor objects. This method is based on image overlapping region extraction and reconstruction of two overlapped regions based on sub-block decomposition and matching. The distortion of two frames to be matched can be adjusted to consistency before content matching can be performed and their similarity can be solved accurately. It can effectively use different scene and layout information in each room to eliminate the influence of similar objects. The global navigation map of the whole building is constructed by extracting the sequence of key frames with large space spacing and overlapping connection from the video captured by the robot learning environment. When the robot works, the image content of real-time vision matches with the sequence of map key frames, and the key frame which is the most similar to the visual image at every moment is extracted to locate the robot. Experiments were carried out in the experimental area composed of three rooms and two corridors. The results show that the robot can effectively eliminate the interference of similar objects, and the kidnapping can still be accurately and autonomously located by matching with the global map. The matching accuracy is more than 93 and the positioning accuracy error (RMSE) is 0.5 m.
【作者单位】：中国科学院电子学研究所传感技术联合国家重点实验室;中国科学院大学;
【基金】：国家自然科学基金资助项目(No.61372052,No.61271147)
【分类号】：TP242;TP391.41

【相似文献】