【作者】 李辉；

【导师】 朱奎宝； 张乐；

【作者基本信息】 河北科技大学 ， 电子信息（专业学位）， 2025， 硕士

【摘要】 在自动驾驶及机器人等领域,同时定位与建图(Simultaneous Localization And Mapping,SLAM)技术应用愈加广泛。在此背景下,仅依靠单一传感器在现实场景中已愈发难以达到定位精度要求。因此,多传感器融合成为SLAM技术的研究热点,但仅依靠点特征在弱纹理、光照变换等环境下定位效果不理想。针对上述问题,本文引入线特征,结合IMU,提出了一种基于点线特征融合的视觉惯性SLAM算法。主要研究内容如下:(1)在图像信息处理部分,针对ORB特征点提取算法存在特征点堆积问题,引入四叉树算法实现特征点均匀化提取。针对LSD算法提取的线段存在大量短线问题,提出基于条件筛选的线特征提取算法,改进后的算法保留了高质量线段,有效克服了原始算法的缺点。并使用RANSAC算法剔除点线特征误匹配,提升匹配精度。(2)设计了视觉点线特征与IMU紧耦合非线性优化模型。首先,构建点、线及IMU残差模型。其次,采取基于滑动窗口的边缘化策略,减少系统计算量。构建包含残差模型与边缘化的先验信息的目标函数,通过非线性优化对目标函数进行迭代优化,得到位姿最优估计。最后建立点线特征视觉词袋模型用于回环检测,消除累计误差。(3)对本文算法进行实验验证。在EuRoC数据集上,将主流SLAM方案和本文算法进行对比,结果表明本文算法的平均误差降低了15.2%。搭建实物验证平台,进行多传感器联合标定,在现实场景中进行实验验证。实验结果表明,本文算法具有良好的定位精度以及实用性。更多还原

【Abstract】 Simultaneous Localization And Mapping(SLAM)technology is widely used in the fields of autonomous driving and robotics.In this context,relying on a single sensor in the real world has become increasingly difficult to meet the positioning accuracy requirements.Therefore,SLAM technology based on multi-sensor fusion has become a research hotspot.However,it is not ideal to rely only on point features in weak texture,light transformation and other environments.Therefore,a visual inertial SLAM algorithm based on point-line feature fusion is proposed by introducing line feature and IMU.The main research contents are as follows:(1)In the image information processing part,aiming at the problem of accumulation of feature points in ORB feature point extraction algorithm,quadtree algorithm is introduced to realize the homogenization of feature points extraction.Aiming at a lot of short lines in line segments extracted by LSD algorithm,a line feature extraction algorithm based on conditional screening is proposed.The improved algorithm retains high quality line segments and overcomes the shortcomings of the original algorithm.RANSAC algorithm is used to eliminate the mismatching of point and line features and improve the matching accuracy.(2)A nonlinear optimization model with tight coupling of visual point-line features and IMU is designed.Firstly,the point,line and IMU residual models are constructed.Secondly,the marginalization strategy based on sliding window is adopted to reduce the calculation amount of the system.The objective function containing residual model and marginal prior information is constructed,and the objective function is iteratively optimized by nonlinear optimization to obtain the pose optimal estimation.Finally,the dot and line feature visual word bag model is established for loop detection to eliminate the accumulated error.(3)The algorithm in this paper is experimentally verified.On the Eu Ro C dataset,the mainstream SLAM schemes were compared with the algorithm in this paper.The results show that the average error of the algorithm in this paper is reduced by 15.2%.Build a physical verification platform,conduct joint calibration of multiple sensors,and carry out experimental verification in real scenarios.The experimental results show that the algorithm in this paper has good positioning accuracy and practicability.更多还原