A Fast Vision-inertial Odometer Based on Line Midpoint Descriptor  

作　　者：Wen-Kuan Li Hao-Yuan Cai Sheng-Lin Zhao Ya-Qian Liu Chun-Xiu Liu 

机构地区：[1]State Key Laboratory of Transducer Technology,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100190,China [2]University of Chinese Academy of Sciences,Beijing 100190,China

出　　处：《International Journal of Automation and computing》2021年第4期667-679,共13页国际自动化与计算杂志（英文版）

基　　金：by National Natural Science Foundation of China(Nos.61774157 and 81771388);Beijing National Natural Science Foundation of China(No.4182075);National Key Research and Development.Plan(Nos.2020YFC2004501,2020YFC2004503 and 2017YFF0107704).

摘　　要：Visual simultaneous localization and mapping(VSLAM) are essential technologies to realize the autonomous movement of vehicles. Visual-inertial odometry(VIO) is often used as the front-end of VSLAM because of its rich information, lightweight, and robustness. This article proposes the FPL-VIO, an optimization-based fast vision-inertial odometer with points and lines. Traditional VIO mostly uses points as landmarks;meanwhile, most of the geometrical structure information is ignored. Therefore, the accuracy will be jeopardized under motion blur and texture-less area. Some researchers improve accuracy by adding lines as landmarks in the system.However, almost all of them use line segment detector(LSD) and line band descriptor(LBD) in line processing, which is very time-consuming. This article first proposes a fast line feature description and matching method based on the midpoint and compares the three line detection algorithms of LSD, fast line detector(FLD), and edge drawing lines(EDLines). Then, the measurement model of the line is introduced in detail. Finally, FPL-VIO is proposed by adding the above method to monocular visual-inertial state estimator(VINSMono), an optimization-based fast vision-inertial odometer with lines described by midpoint and points. Compared with VIO using points and lines(PL-VIO), the line processing efficiency of FPL-VIO is increased by 3-4 times while ensuring the same accuracy.

关 键 词：High efficiency visual-inertial odometry(VIO) non-linear optimization points and lines sliding window 

分 类 号：TN96[电子电信—信号与信息处理] TP391.41[电子电信—信息与通信工程]