Camera Pose Estimation in Unknown Environments using a Sequence of Wide-Baseline Monocular Images

In this work, a feature-based technique is proposed for the camera pose estimation in a sequence of widebaseline images. Camera pose estimation is an important issue in many computer vision and robotics applications such as augmented reality and visual SLAM. The developed method can track captured images taken by a hand-held camera in room-sized workspaces with a maximum scene depth of 3-4 m. This system can be used in unknown environments with no additional information available from the outside world except in the first two images used for initialization. Pose estimation is performed using only natural feature points extracted and matched in successive images. In wide-baseline images, unlike consecutive frames of a video stream, displacement of the feature points in consecutive images is notable, and hence, cannot be traced easily using the patch-based methods. To handle this problem, a hybrid strategy is employed to obtain accurate feature correspondences. In this strategy, first, initial feature correspondences are found using the similarity between their descriptors, and then the outlier matchings are removed by applying the RANSAC algorithm. Further, in order to provide a set of required feature matchings, a mechanism based on the sidelong result of robust estimator is employed. The proposed method is applied on indoor real data with images in VGA quality (640 × 480 pixels), and on average, the translation error of camera pose is less than 2 cm, which indicates the effectiveness and accuracy of the developed approach.