基于级联Adaboost的目标检测融合算法

Object Detection by Combined Model Based on Cascaded Adaboost

单一特征的模型对于颜色纹理变化较大的目标的检测往往存在检测率不高或检测速度慢的缺点.本文提出了一种基于级联Adaboost的"级联-加和"融合算法.融合模型由两个独立训练得到的级联Adaboost分类器组成,分别利用边界片段特征和矩形类Haar小波特征描述整个目标以及目标的一个稳定部件.级联-加和的融合决策以样本在两个分类器中被拒绝或通过的级数信息为依据.在多个数据库上的实验证明这种融合检测算法不仅综合了Haar小波特征检测速度快和边界片段特征鲁棒性好的优点,而且与单一特征的分类器相比,检测性能也有所提高.

Single feature-based model always meets the difficulties of poor detection performance and slow detection speed for object with large variances in color, texture, and shape. A novel cascaded and additive model based on cascaded Adaboost classifier is proposed in this paper. This combined model consists of two cascaded Adaboost classifiers which are independently trained with edge-fragment feature and Haar feature to describe the whole object and one of its stable components, respectively. The final classification decision of the combined model is made according to the stage indexes by which a sample is rejected or accepted in the two cascaded classifiers. Experiments on several test databases show that the combined model can take advantages of the speed merit of Haar feature and the robustness of edge-fragment feature. Compared with single feature-based model, the detection performance of the combined model is greatly improved.