为了得到更强的鲁棒性和更好的实时性,针对精确的人眼定位提出了基于随机森林的自适应梯度提升算法(Adaptive Gradient Boosting Random Forest),同时引入了更加多样的训练样本来降低诸如光线强弱,旋转角度等外部环境对定位的影响.在BioID数据库中,此方法的准确率为92%,定位时间少于1ms.实验结果表明,它已经满足在大多数不可控环境对鲁棒性和实时性的要求.
A generative framework for real time object detection and classification.[J].Ian Fasel;Bret Fortenberry;Javier Movellan.Computer Vision and Image Understanding.2004, 1
A generative framework for real time object detection and classification.[J].Ian Fasel;Bret Fortenberry;Javier Movellan.Computer Vision and Image Understanding.2004, 1