Name
Abstract | ||
|---|---|---|
This paper proposes a face tracking method which uses the concepts of multiple instances and Online Adaboost to successively train and update a face tracking model (FTM) for each tracked face, and then tracks the tracked faces among the image frames. In order to avoid the continuous accumulation of tracking error, this method performs a face detection process around each tracked face. This paper also provides a novel tracking recovery approach when the tracked faces are occluded with other faces. First, it uses an occlusion detection criterion to realize whether a tracked face is occluded by other objects (called shelter) or not. If it is, several candidate regions around each shelter will be constructed, and by using frame difference and skin detection information, the candidate regions will be further filtered out so that the regions with no face clues will be omitted. For the remaining regions, it will apply the FTM preserved just before occlusion to detect whether a tracked face has reappeared or not. If an occluded face indeed has reappeared, it will continue to track this face from its reappeared position. Experiments show that the proposed method provides very good face tracking results on different videos and outperforms both Camshift and Kalman Filters. |
Year | DOI | Venue |
|---|---|---|
2015 | 10.1007/978-3-319-23204-1_25 | GENETIC AND EVOLUTIONARY COMPUTING, VOL I |
Field | DocType | Volume |
Computer vision,Occlusion detection,AdaBoost,Occlusion,Computer science,Frame difference,Kalman filter,Artificial intelligence,Face detection,Machine learning,Facial motion capture,Tracking error | Conference | 387 |
ISSN | Citations | PageRank |
2194-5357 | 0 | 0.34 |
References | Authors | |
2 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Yea-shuan Huang | 1 | 479 | 79.42 |
| Chin-Ie Chang | 2 | 0 | 0.34 |