基于SURF和ORB全局特征的快速闭环检测

发布时间：2018-04-02 14:26

本文选题：同时定位与地图创建　切入点：闭环检测　出处：《机器人》2017年01期

【摘要】：针对移动机器人SLAM(同时定位与地图创建)中的闭环检测问题,提出了一种基于SURF(加速鲁棒特征)和ORB(oriented FAST and rotated BRIEF)全局特征的快速闭环检测算法.首先利用SURF与ORB分别提取查询图像的全局特征,实现对图像的特征表征.在特征提取过程中,对查询图像进行归一化操作,并将归一化的图像中心直接作为SURF与ORB的特征点位置,从而避免了耗时的特征点定位过程.然后将归一化的图像直接作为特征点的邻域区域,把计算的SURF与ORB局部特征描述符作为图像的全局特征.为了融合SURF与ORB全局特征实现查询图像的位置识别,提出了H-KNN(混合K最近邻)的改进算法:WH-KNN(加权混合K最近邻).最后通过跟踪模型实现闭环检测,其核心思想是利用之前闭环检测的结果预测查询图像在地图图像中的位置范围.实验中分别使用采集数据集和牛津大学公开数据集对本文算法进行了验证,同时与传统的BOW(词袋)算法进行了对比.本文算法在两种数据集上分别达到了94.3%和94.5%的准确率,并且查询图像位置识别与全局特征提取的平均时间不到3 ms.其准确性及计算速度都超过了BOW算法,可以准确快速地实现实时闭环检测.
[Abstract]:Aiming at the closed-loop detection problem of mobile robot slam (simultaneous location and map creation), a fast closed-loop detection algorithm based on the global features of surf (accelerated robust feature) and ORB(oriented FAST and rotated (BRIEF) is proposed.Firstly, SURF and ORB are used to extract the global feature of the query image, and the feature representation of the image is realized.In the process of feature extraction, the query image is normalized, and the normalized image center is taken as the feature point position of SURF and ORB directly, thus avoiding the time-consuming feature point localization process.Then the normalized image is directly regarded as the neighborhood region of the feature points, and the computed SURF and ORB local feature descriptors are taken as the global features of the image.In order to combine the global features of SURF and ORB to realize the location recognition of queried images, an improved algorithm of H-KNN (mixed K-nearest neighbor), called weighted mixed K-nearest neighbor (WHH-KNN), is proposed.Finally, the closed-loop detection is realized by tracking model. The core idea is to predict the location range of the query image in the map image by using the results of the previous closed-loop detection.In the experiment, we use the collected data set and Oxford University open data set to verify the algorithm, and compare it with the traditional word bag algorithm.The algorithm achieves the accuracy of 94.3% and 94.5% respectively on the two data sets, and the average time of querying image location recognition and global feature extraction is less than 3 Ms.The accuracy and computing speed of the algorithm are higher than that of BOW algorithm, which can accurately and quickly realize the real-time closed-loop detection.
【作者单位】：河北工业大学电子信息工程学院;武汉理工大学智能交通系统研究中心;
【基金】：国家自然科学基金(51679181,51208168) 河北省普通高等院校青年拔尖人才计划(BJ2014-013) 湖北省技术创新专项重大项目(2016AAA007) 湖北省自然科学基金(2015CFB252)
【分类号】：TP242;TP391.41

【相似文献】