基于视觉传感器的半直接SLAM算法  

Semi-direct SLAM algorithm based on vision sensor

作　　者：彭虎 董翔 李腾[1] 樊渊[1] PENG Hu;DONG Xiang;LI Teng;FAN Yuan(School of Electrical Engineering and Automation,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学电气工程与自动化学院,安徽合肥230601

出　　处：《传感器与微系统》2022年第6期114-117,121,共5页Transducer and Microsystem Technologies

基　　金：国家重点研发项目智能机器人专项项目(2018YFB1305804,2019-2021);安徽高校协同创新项目(GXXT-2019-003)。

摘　　要：针对目前基于视觉传感器的同时定位和地图构建(SLAM)系统对特征缺失和无结构等特殊环境比较敏感的缺点,提出了一种基于子图特征增强方案的半直接SLAM算法。首先,使用基于子图的特征增强模块以更稳定的提取图像特征点;并在特征点的基础上,考虑加权融合特征点对应像素的光度信息来估计相机位姿,使系统能够在视觉纹理特征信息比较稀缺的环境下仍然可以稳定工作。其次,为了得到更好的建图效果,运用增量式动态协方差缩放算法来最小化相机位姿估计引起的误差。所提出的算法经过测试,在TUM数据集以及室内环境下具有较好的鲁棒性。Current visual based simultaneous localization and mapping(SLAM)system suffers from feature loss and unstructured scene in hard environments.Addressing this,a fast semi-direct SLAM algorithm is proposed based on subgraph feature enhancement scheme.Firstly,the feature enhancement module based on subgraph is used to extract image feature points more stably.On the basis of feature points,the camera pose is estimated by weighted fusion of the photometric information of of the pixels corresponding to the feature points,so that the system can work stably in the environment where the visual texture feature information is scarce.Secondly,in order to get a better mapping effect,the incremental dynamic covariance scaling algorithm is used to minimize the error caused by the camera pose estimation.The proposed algorithm has been tested extensively on the benchmark TUM dataset and showed superior performance than existing visual based SLAM algorithms in indoor environments.

关 键 词：视觉传感器 同时定位和地图构建 特征增强 位姿估计 增量式动态协方差缩放 

分 类 号：TP399[自动化与计算机技术—计算机应用技术] TP212[自动化与计算机技术—计算机科学与技术]