Simultaneous localization of a monocular camera and mapping of the environment in real time

Abstract

In this work a Visual SLAM system (Simultaneous Localization and Mapping) that performs in real time, building feature-based maps and estimating the camera trajectory is presented. The camera is carried by a person that moves it smoothly with six degrees of freedom in indoor environments. The features correspond to high quality corners parametrized with inverse depth representation. They are detected inside regions of interest and an occupancy criterion is applied in order to avoid feature agglomeration. The association process is developed using active search. The final representation is made in a three-dimensional environment.