一种基于点线特征融合的视觉惯性SLAM算法

A Visual Inertial SLAM Algorithm Based on Point and Line Feature Fusion

针对室内弱纹理环境下特征点导向的SLAM算法难以高效提取特征,甚至无法追踪到足够特征导致运动估计失败的问题,提出了一种融合点线特征的视觉惯性SLAM算法。该算法通过长度抑制策略及对输入图像直方图均衡化来改进传统LSD算法,提高线特征提取速度和质量,然后利用线特征辅助点特征对场景结构提供额外的约束,同时联合IMU数据得到更准确的位姿估计,减少运动速度过快、光照条件不佳的情况对系统稳定性的影响。对比多组公开数据集的实验结果表明:该方法提高了弱纹理环境下系统的定位精度和鲁棒性。

Aiming at the problem that feature point-oriented SLAM algorithms in indoor weak texture environments are difficult to extract features efficiently,and even fail to track enough features to cause motion estimation failure,a visual inertial SLAM algorithm that fuses point and line features is proposed.The algorithm improves the traditional LSD algorithm through a length suppression strategy and histogram equalization of the input image to improve the speed and quality of line feature extraction,and then uses the line features to assist the point features to provide additional constraints on the scene structure,and at the same time unites the IMU data to obtain a more accurate position estimation and reduce the impact of excessive motion speed and poor lighting conditions on the stability of the system.Experimental results comparing multiple publicly available datasets show that the method improves the positioning accuracy and robustness of the system in weak texture environments.