An Improved RGB-D SLAM Method Based on Manhattan World and Point-Line-Plane Features

At present, most feature-based SLAM methods are based on feature points. But in indoor environments where texture is scarce and light varies greatly, these methods will reduce the accuracy of pose estimation or even lead to pose tracking failure. To tackle the above problem, we propose an improved RGB-D SLAM method which employs point-line-plane feature extraction and matching as well as Manhattan Frames-based rotation estimation. Finally, the proposed method is proven to have good reliability and robustness in open datasets and real indoor scenes.