Reduced keyframes for fast bundle adjustment using point and line features in monoslam

Visual Simultaneous Localization and Mapping (VSLAM) requires feature detection on visual data. In indoor scenes that include architectures such as plain walls and doors, there are no or less corners detected, in such cases tracking will be lost. To overcome this situation we track different types of features that help in feature extraction even in textureless scenes. Line features are used to get the outline of the structure. This paper explains method of line based localization when no point features are detected and process of mapping the matched lines between two frames. One of the existing state of art monocular SLAM that extracts point features only, is used as the main build and LBD (Line Band Descriptor) is incorporated in it for localization and mapping. Our new system extracts more features compared to the existing system and complexity of the bundle adjustment is decreased as it decreases the keyframes selected for local mapping and optimization when tested using our own dataset and compared with the existing system.