PL-ALF: A Novel Point-Line Feature Autonomous Localization and Flight Framework Based on Multisensor Fusion and Optimization

Unmanned aerial vehicles (UAVs) are extensively utilized across diverse domains due to their flexibility. However, for UAVs relying solely on visual sensors, achieving autonomous flight in low-textured environments presents a significant challenge. The scarcity of visual features in such environments makes it difficult for UAVs to accurately estimate their position and perform robust autonomous path planning. To address this issue, this article proposes a novel autonomous flight framework called point-line feature autonomous localization and flight framework (PL-ALF). PL-ALF consists of a Stereo/Stereo-Inertial simultaneous localization and mapping (SLAM) system and a path-planning module. By capturing point-line features in each image frame, the framework enables feature tracking, attitude estimation, local bundle adjustment (BA), and global BA. Additionally, it integrates a global multimap system and loop closure detection functionality. The path-planning module utilizes the positioning results from the visual SLAM and depth camera information to update trajectories, allowing the UAV to circumvent local obstacles. It further optimizes obstacle-avoidance trajectories by considering multiple cost factors. Experiments conducted on public datasets and simulation platforms demonstrate that PL-ALF (Stereo-SLAM) achieves an average accuracy improvement of over 30% in indoor environments. Furthermore, it achieves leading results in simulated flight experiments when combined with inertial measurement units (IMUs). Successful real-world flight experiments confirm that the proposed PL-ALF framework enables autonomous flight in indoor low-textured environments.