摘要
针对Adaboost行人检测训练非常耗时的问题,在梯度方向直方图(Histograns of Oriented Gradiant,HOG)特征计算上引入积分向量图,同时对于作为AdaBoost学习过程中的分类器---线性SVM,应用序列最小优化(Seguential Minimal Optimigation,SMO)来解决其二次规划(QP)问题。实验结果表明,通过这两个方面的改进,不仅行人检测训练检测速度得到了提升,而且取得了令人满意的检测效果。
Aiming at the problem that training time of Adaboost Pedestrian detection is extremely long, Integral Vector Image is presented to compute HOG features, and applyied linear SVM to AdaBoost as a classifier. The experiment results show that not only the training time is improved, but also the result of detection is satisfactory through improvement of the above two sides.
出处
《科学技术与工程》
2009年第13期3646-3651,共6页
Science Technology and Engineering