基于LK光流法的单目视觉里程计  

Monocular visual odometry based on LK optical flow

作　　者：陈伟杰 孙作雷[1] Chen Weijie;Sun Zuo lei(Machine Perception and Interaction Group （ MPIG）, Shanghai Maritime University, Shanghai 201306, China)

机构地区：[1]上海海事大学机器辨识与交互研究组,上海201306

出　　处：《微型机与应用》2017年第21期45-47,50,共4页Microcomputer & Its Applications

摘　　要：在移动机器人快速发展的今天,视觉里程计是通过视觉传感器进行导航定位的主要方式之一。文章介绍了基于单个摄像头的单目视觉里程计的一种实现方法,使用SURF算法提取特征点,用LK光流法进行特征追踪,相比于传统特征匹配在运算效率上有极大的提高,再对特征点计算位移与旋转量,以里程计的方式达到定位目的。详细介绍了视觉里程计的数学原理以及SURF算法和LK算法的原理及其数学推导、单目视觉里程计的尺度不确定原理。最后通过对比确定LK算法的优势以及最优参数的选择。Nowadays,with the rapid development of mobile robots,visual odometer is one of the main ways of navigating through visual sensors.In this paper,we introduce a method of monocular visual odometer based on single camera. We use SURF to extract feature points and then use LK optical flow to track features. The operational efficiency of LK is higher than usual matching. Finally,we use matching points to get translation and rotation to achieve the purpose of positioning. We introduce how SURF and LK work and what＇s projective ambiguity. At last,we show the result that LK is really faster than usual matching,and the optimal parameters are determined by comparing the selection.

关 键 词：视觉里程计 LK光流法 图像处理 特征跟踪 多视图几何 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置]