Abstract | ||
---|---|---|
This paper presents an improved method for Fast Visual Odometry estimation from a Kinect-style RGB-D camera. In order to improve the accuracy and robustness of Fast Visual Odometry, we propose a modified ICP algorithm called Semi-probabilistic Trimmed ICP and a transform strategy between frame-to-model approach and frame-to-frame approach. An overlap parameter is computed to reject outlier before the registration. And if it comes to a occasional large camera motion, we skip the current frame and compute a coarse initial guess by the RANSAC algorithm between the next frame and the previous frame, finally refine the pose of camera by the original ICP. The evaluation on TUM RGB-D benchmark shows that Our Visual Odometry outperforms state-of-the-art in certain scenarios like a small-scale camera motion and it's capable of dealing with an occasional large camera motion. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1145/3234664.3234666 | PROCEEDINGS OF THE 2018 2ND HIGH PERFORMANCE COMPUTING AND CLUSTER TECHNOLOGIES CONFERENCE (HPCCT 2018) |
Keywords | DocType | Citations |
Visual odometry, semi-probabilistic trimmed-ICP, frame-to-model, frame-to-frame | Conference | 0 |
PageRank | References | Authors |
0.34 | 10 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zhang Hong | 1 | 18 | 3.74 |
Shiqiang Hu | 2 | 53 | 10.05 |