Robust and efficient pose estimation from line correspondences
ACCV'12 Proceedings of the 11th Asian conference on Computer Vision - Volume Part III
| Hi-index | 0.00 |
This paper presents an approach of visual SLAM using line segments as primitives. We choose Pl 篓ucker lines to represent spatial lines and derive a novel fast line projection function based on the relationship between Pl篓ucker line and Pl 篓ucker matrix. During normal SLAM procedure, the Near by Line Tracking (NLT) method is adopted to track lines and an Extended Kalman Filter (EKF) is used to predict and update the state of the camera and line landmarks. After each update step, a robust spatial line reconstruction algorithm is used to find new line landmarks and to add them into the map. The SLAM procedure is under supervision and when failure is detected, a recovery method based on the angle histogram of key frames is adopted to relocalize the camera and re-start the SLAM procedure. We demonstrate that our monocular SLAM system is robust to illumination change, partial occlusion and fast camera motion.