A Visual-Inertial SLAM-Based Localization Method for Intelligent Vehicles in Weak Textured Environments

The accurate extraction of feature points has a significant impact on the pose estimation of visual SLAM system, in weak textured environments such as underground parking lots, robust feature point extraction poses significant challenges. In order to solve the problem, this paper proposes a weak texture environment feature extraction method that combines point and line features to improve the localization performance of visual inertial systems (PIL-VIO). Firstly, to obtain stable visual input features, the traditional line feature detection algorithm is improved by parameter tuning, line segment merging, and length-based filtering strategies, which enhances the efficiency and robustness of line feature detection. Secondly, in order to realize the tight coupling of visual and inertial information, this paper constructs a joint optimization function by combining the line features reprojection error with the point features reprojection error, IMU measurement errors, and marginalized prior constraints in a nonlinear optimization framework based on sliding window and marginalization strategies. Finally, the experimental results on the EUROC dataset show that, compared to traditional visual localization algorithms, PIL-VIO has significant advantages in terms of system positioning accuracy and robustness.