融合梯度密度的点线视觉SLAM算法改进  

Improvement of SLAM Algorithm for Point and Line Vision Based on Gradient Density

作　　者：李明好 陈孟元 贡鹏昊[1,2] 龙海燕 LI Minghao;CHEN Mengyuan;GONG Penghao;LONG Haiyan(School of Electrical Engineering,Anhui Polytechnic University,Wuhu 241000,China;Key Laboratory of Advanced Perception and Intelligent Control of High-end Equipment,Anhui Polytechnic University,Wuhu 241000,China;School of Electrical and Electronic Engineering,Anhui Institute of Information Technology,Wuhu 241000,China)

机构地区：[1]安徽工程大学电气工程学院,安徽芜湖241000 [2]安徽工程大学高端装备先进感知与智能控制教育部重点实验室,安徽芜湖241000 [3]安徽信息工程学院电气与电子工程学院,安徽芜湖241000

出　　处：《安徽工程大学学报》2023年第6期55-63,71,共10页Journal of Anhui Polytechnic University

基　　金：国家自然科学基金资助项目(61903002);安徽省高校协同创新项目(GXXT-2021-050);安徽省高校自然科学研究重点项目(KJ2020A0829)。

摘　　要：针对视觉同步定位与地图构建(Simultaneous Localization and Mapping,SLAM)方法在相机快速运动中容易导致图像模糊,以及在稠密场景中线特征提取易造成信息冗余等问题,提出一种融合梯度密度的点线视觉SLAM算法改进。该算法首先利用前后图像帧之间特征点数量信息对模糊图像进行筛选,并使用高斯模糊进行优化处理,得到匹配效果更佳的图像帧。然后利用点特征信息判断是否引入线特征,并引入图像像素梯度密度对LSD(Line Segment Detection)线特征进行多维优化,提取出稳定线特征以提高后续匹配质量。最后结合点线特征误差构建误差函数,最小化投影误差提高位姿估计精度。算法在TUM数据集下进行测试,实验结果表明本算法可以有效提升特征提取的稳健性,进而提高相机位姿估计与建图的精度。Aiming at the problems that the visual synchronous localization and map building(SLAM)method is likely to cause image blur in the rapid camera movement,and that the extraction of centerline features in dense scenes is likely to cause information redundancy,an improved point and line visual SLAM algorithm integrating gradient density is proposed.The algorithm first uses the number of feature points between the front and back image frames to filter the blurred image,and then uses Gaussian blur to optimize the processing to obtain better matching image frames.Then,the point feature information is used to judge whether the line feature is introduced,and the image pixel density gradient is introduced to optimize the LSD(line segment detection)line feature from multi-dimensions,and the stable line feature is extracted to improve the subsequent matching quality.Finally,the error function is constructed based on the point and line characteristic error to minimize the projection error and improve the pose estimation accuracy.The algorithm is tested in TUM data set,and the experimental results show that the algorithm can effectively improve the robustness of feature extraction,thereby improving the accuracy of camera pose estimation and mapping.

关 键 词：同步定位与建图 梯度信息 线特征提取 点线融合 信息熵 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]