Feature selection and pose estimation from known planar objects using monocular vision

In this paper, we develop a way to accurately and precisely estimate the pose of a calibrated camera with a single picture which includes a known planar object. For the proposed algorithm, we first use SURF detector for feature extraction and matching. Then, we use the information from known reference image to retrieve 3D point coordinates. Based on resulting 2D-2D correspondences and 3D coordinates, multiple-view geometry constraints are adopted to calculate the camera pose. Comparing with previous work, the proposed algorithm introduces an advanced feature selection algorithm, which eliminates pose ambiguity and improve the pose estimation result. The feature selection algorithm is based on the assumption that most 3D feature points should be coplanar. We conduct tests on traffic sign and evaluated the test results. The test results show that pose estimation with standard RANSAC turns out to be ambiguous occasionally. Conversely, the estimation with the proposed feature selection strategy leads to high robustness and accuracy.