Reducing False Positive Rate for Viola-Jones Method in Hand Detection

摘要

The Viola-Jones object detector is a machine learning based method which combines Haar-like features with AdaBoost classifier. It is known that this method is the mainstream in face detections due to its high performance of both detection speed and precision. However, the false positive rate increases dramatically when this approach is applied to detect hands in a sign language video which includes other skin color parts such as faces or arms. In this paper, we analyze the reason for mis-detecting of these points and propose a new method to reduce the false positive rate. Experimental results have shown that our method is more robust to hand motion and decreases the false positive rate significantly.