摘要
闭环检测是同步定位与地图构建(SLAM)的关键组成部分之一。传统的闭环检测多是基于手工设计特征的图像描述方法,很容易受环境的影响。而基于深度学习图像描述方法的闭环检测通常没有考虑到图像的局部空间特性。基于这些因素,提出一种CNN与VLAD融合的图像描述方法,应用于闭环检测。实验结果表明,与传统的基于手工设计特征的图像描述相比,基于CNN与VLAD融合的图像描述方法在实现100%的准确率下,召回率最高提高59.71%,与基于深度学习的图像描述方法相比召回率提高28.33%。
Loop closure detection is one of the most essential components of Simultaneous Localization And Mapping (SLAM).The traditional loop closure detection is mostly based on the image description method of hand-crafted features,which is easily affected by the environment.Loop closure detection,based on deep learning image description method,usually doesn't consider the local spatial characteristics of images. Considering these factors,proposes an image description method that CNN and VLAD fusion for Loop closure detection.The results of the experiment show that the proposed method,compared with the traditional image description based on hand-crafted features,can increase 59.71% of recall rate when achieves a 100% precision.Compared with the traditional image description based on deep learning,the recall rate increased by 28.33%.
作者
林辉
LIN Hui(School of Automation,Guangdong University of Technology,Guangzhou 510006)
出处
《现代计算机》
2018年第24期17-21,25,共6页
Modern Computer