A Point-Line Feature based Visual SLAM Method in Dynamic Indoor Scene

Traditional visual SLAM methods employ point features to implement motion estimation and environment map construction. However, in some low-texture indoor scenarios, such as office and corridor, less reliable point features may be found, which could jeopardize the SLAM solution. In addition, scene changes among sequential images are not only caused by the position changes of image acquisition, but also by pedestrians and other moving objects in an indoor dynamic environment. Thus, feature identification for moving objects is needed as an important part for practical application of indoor SLAM. This paper proposes a point-line feature based SLAM method that combines both of points and line segments to enhance the performance of feature extraction in indoor scene, which can extract many line features from walls, furniture and other artificial objects. In this method, added line features help to gain more robust and accurate results. Additionally, a real-time object detection algorithm is introduced to identify the features extracted from pedestrians, so that to eliminate the negative effects caused by moving objects. The experimental results demonstrates that the proposed method can obtain more robust and accurate localization results in dynamic indoor scene.