| 講演抄録/キーワード |
| 講演名 |
2014-06-20 10:00
A novel preprocessing technique for training Viola-Jones hand detector ○Shuqiong Wu・Hiroshi Nagahashi(Tokyo Inst. of Tech.) PRMU2014-28 |
| 抄録 |
(和) |
(まだ登録されていません) |
| (英) |
Viola-Jones method combines Haar-like features with AdaBoost classifier. It is widely used in real time object detection due to its high performance and rapidity. Viola-Jones method has achieved considerable success in face detection. However, when it is applied to hand detection, the false positive rate is rather high. Compared with face training samples which contain little background area, hand training samples include diverse backgrounds. Therefore it is substantially difficult to find common hand features in the training of hand detectors. To solve this problem, we propose a novel preprocessing technique which can reduce the bad influence of backgrounds included in hand training samples. Experimental results show that our method needs far less training samples than Viola-Jones method and improves the false positive rate significantly. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
Hand detection / Machine learning / Viola-Jones method / Haar-like features / AdaBoost / / / |
| 文献情報 |
信学技報, vol. 114, no. 90, PRMU2014-28, pp. 37-41, 2014年6月. |
| 資料番号 |
PRMU2014-28 |
| 発行日 |
2014-06-12 (PRMU) |
| ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
PRMU2014-28 |