PLDS-SLAM: Point and Line Features SLAM in Dynamic Environment

Visual simultaneous localization and mapping (SLAM), based on point features, achieves high accuracy map construction. They primarily perform static features. Despite their efficiency precision, they are prone to instability even failure in complex environments. In a dynamic environment, it is easy keep track of failures work. The object elimination method, semantic segmentation, often recognizes objects without distinction. If there many segmentation or the distribution uneven camera view, this may result feature offset deficiency for matching motion tracking, which will lead problems, such as reduced system accuracy, tracking failure, loss. To address these issues, we propose novel point-line SLAM method obtains prior region features by detecting segmenting region. It realizes separation proposing geometric constraint line segments, combined with epipolar points. Additionally, Bayesian theory proposed eliminate noise points lines improve robustness system. We have performed extensive experiments KITTI HPatches datasets verify claims. experimental results show that our has excellent performance scenes.