张光耀, 倪益华, 吕艳, 等.一种基于改进词袋模型的视觉SLAM算法[J].河南理工大学学报（自然科学版）,2021,40(4):154-161.

ZHANG G Y, NI Y H, LYU Y,et al.A visual SLAM algorithm based on improved word bag model[J].Journal of Henan Polytechnic University(Natural Science) ,2021,40(4):154-161.

一种基于改进词袋模型的视觉SLAM算法

张光耀, 倪益华, 吕艳, 倪忠进, 黄通交

浙江农林大学 工程学院，浙江 杭州310000

摘要:针对室内环境中视觉同时定位与建图（simultaneous localization and mapping， SLAM ）精度不高和实用性较差等问题，采用深度相机作为传感器，提出一种基于改进词袋模型的视觉 SLAM算法。该算法通过增加节点距离的方式，对传统的词袋模型进行改进，采用octree方法转化点云，生成可用于导航的八叉树图，并进行改进前后词袋模型对比实验、数据集精度实验和实验室实测。结果表明，改进后的词袋模型相似度计算能力和区分度更强，SLAM算法在环境有回环和相机运动较慢的情况下，效果较好，可满足室内同时定位与建图及后续导航需求。

关键词:视觉SLAM;回环检测;词袋模型;八叉树地图;TUM数据集

doi:10.16186/j.cnki.1673-9787.2020020053

基金项目:国家自然科学基金资助项目（61175125 ）；浙江省自然科学基金资助项目（LZ15E050003 ，LQ16E050013 ）

收稿日期:2020/02/242

修回日期:2020/07/09

出版日期:2021/07/15

A visual SLAM algorithm based on improved word bag model

ZHANG Guangyao, NI Yihua, LYU Yan, NI Zhongjin, HUANG Tongjiao

School of Engineering， Zhej'iang A&F University， Hangzhou 310000 ， Zhejiang， China

Abstract:Aiming at the problems of low accuracy and poor practicability of visual SLAM in indoor environment ，by using depth camera as the sensor， an algorithm of visual SLAM was proposed based on improved word bag model. The traditional word bag model was improved by increasing the distance between nodes， and octree method was use to transform point cloud to generate octree map which could be used for navigation. Then a comparative experiment between the improved word bag model and the original word bag model， three data set accuracy experiments and a field experiment were carried. The experimental results showed that the improved word bag model was more powerful in similarity calculation and discrimination， the SLAM algorithm worked well when the camera motion was slow and the environment had loop. The requirements of indoor positioning， mapping and subsequent navigation were satisfied.

Key words:visual SLAM;loop closure detection;word bag model;octomap;TUM dataset

一种基于改进词袋模型的视觉SLAM算法_张光耀.pdf