面向轻轨的高精度实时视觉定位方法

Real-time Visual Localization Method for Light-rail With High Accuracy

轻轨作为城市公共交通系统的重要组成部分,对其实现智能化的管理势在必行.针对城市轻轨定位系统要求精度高、实时强且易于安装等特点,本文提出一种基于全局-局部场景特征与关键帧检索的定位方法.该方法在语义信息的指导下,从单目相机获取的参考帧中提取区别性高的区域作为关键区域.并结合像素点位置线索利用无监督学习的方式筛选关键区域中描述力强的像素对生成二值化特征提取模式,不仅能够提升匹配精度还显著提高了在线模块场景特征提取与匹配的速度.其次,以场景显著性分数为依据获取的关键帧避免了具有相似外观的场景给定位带来的干扰,并能辅助提高场景在线匹配的精度与效率.本文使用公开测试数据集以及具有挑战性的轻轨数据集进行测试.实验结果表明,本系统在满足实时性要求的同时,其定位准确率均可达到90%以上.

As an important part of the urban public transportation system,it is imperative to realize the intelligent management of light rail.By considering the practical requirements like high accuracy,real-time performance,and easy installation,this paper proposes a visual localization method based on global-local features and keyframe retrieval.Under the guidance of semantic information,the region with high significance in each reference frame obtained by the monocular camera is extracted as the key region.Combined with the location cues of pixels,unsupervised learning is used to filter the pixel pairs with strong description force in the key region to generate the binary pattern,which greatly reduces the computation of feature extraction and matching in the online module while improving the matching accuracy.Secondly,the keyframes obtained based on the discrimination score can effectively avoid the interference caused by the scene with analogous appearance,and assist to improve the accuracy and efficiency of online scene matching.The Nordland dataset and the challenging light rail dataset are used for testing.The experimental results show that the precision of the system can reach more than 90%while meeting real-time requirements.