A feature-based approach for monocular camera tracking in unknown environments

Camera tracking is an important issue in many computer vision and robotics applications, such as, augmented reality and Simultaneous Localization And Mapping (SLAM). In this paper, a feature-based technique for monocular camera tracking is proposed. The proposed approach is based on tracking a set of sparse features, which are successively tracked in a stream of video frames. In the developed system, camera initially views a chessboard with known cell size for few frames to be enabled to construct initial map of the environment. Thereafter, Camera pose estimation for each new incoming frame is carried out in a framework that is merely working with a set of visible natural landmarks. Estimation of 6-DOF camera pose parameters is performed using a particle filter. Moreover, recovering depth of newly detected landmarks, a linear triangulation method is used. The proposed method is applied on real world videos and positioning error of the camera pose is less than 3 cm in average that indicates effectiveness and accuracy of the proposed method.