基于光流和自注意力的视觉里程计方法  

Visual Odometry Method Based on Optical Flow and Self-attention

作　　者：高静 任明武[1] GAO Jing;REN Mingwu(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2023年第11期2666-2670,共5页Computer & Digital Engineering

摘　　要：视觉里程计作为基于视觉的同时定位与地图构建的重要组成部分,在无人驾驶、移动机器人等领域都发挥了重要的作用。为弥补现有方法的不足,论文提出了一种端到端的基于光流和自注意力的视觉里程计方法。首先,该方法引入先进的光流估计RAFT模型提取富含运动信息的光流。其次,结合双向长短期记忆网络学习图像序列之间的联系。最后,借鉴传统方法中关键帧的思想,引入了自注意力机制,通过连续多帧的特征得到最后两帧图像之间的相对位姿。在KIT⁃TI数据集上的实验结果表明该方法显著降低了旋转、平移的均方根误差。As an important part of visual Simultaneous Localization and Mapping(SLAM),visual odometry is widely used in fields such as unmanned driving and mobile robots.In order to make up for the shortcomings of the existing methods,this paper pro⁃poses an end-to-end visual odometry method based on optical flow and self-attention.First,the method introduces the advanced op⁃tical flow estimation RAFT model to extract optical flow rich in motion information.Secondly,the relationship between image se⁃quences is learned by combining Bi-LSTM.Finally,drawing on the idea of key frames in traditional methods,the paper introduces self-attention mechanism and the relative pose between the last two frames of images is obtained through the characteristics of con⁃secutive multiple frames.Experimental results show that the method significantly reduces the root mean square error of rotation and translation.

关 键 词：视觉里程计 光流 双向长短期记忆网络 自注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]