Abstract
A classification method of corresponding points is proposed, which uses a moving monocular camera attached with a 6-axis sensor. It classifies corresponding points between two consecutive frames containing still/moving objects and chooses appropriate corresponding points for 3D measurement. Corresponding point classification experiment with original CG images shows that accuracy, precision, and recall are 0.94, 0.85, and 1.00, respectively. In addition, it is confirmed that erroneous correspondences are removed by outlier detection using Interquartile Range for proposed evaluation function calculated on each corresponding point. The proposed method is planning to be included in 3D measurement method with actual images containing still/moving objects and also to be applied to obstacles avoidance for vehicles or vision system for mobile robots.
| Original language | English |
|---|---|
| Pages | 110-116 |
| Number of pages | 7 |
| Publication status | Published - 2014 |
| Externally published | Yes |
| Event | Joint International Conference of the 10th China-Japan International Workshop on Information Technology and Control Applications and the 6th International Symposium on Computational Intelligence and Industrial Applications, ITCA and ISCIIA 2014 - Changsha, China Duration: 15 Sept 2014 → 20 Sept 2014 |
Conference
| Conference | Joint International Conference of the 10th China-Japan International Workshop on Information Technology and Control Applications and the 6th International Symposium on Computational Intelligence and Industrial Applications, ITCA and ISCIIA 2014 |
|---|---|
| Country/Territory | China |
| City | Changsha |
| Period | 15/09/14 → 20/09/14 |
Keywords
- 3D measurement
- 6-axis sensor
- Corresponding points classification
- Fundamental matrix estimation
- Moving camera