基于分裂级联模型的多视角多姿态目标检测(英文)

Multi-view multi-pose object detection using splited cascade model

提出将分裂级联模型用于多视角多姿态的目标检测任务,即通过使用从粗到细的自动集合划分策略,自动生成一个树形结构的分类器,树型分类器的每个节点都是一个具有拒绝功能的级联节点.该分裂级联模型可以使用任意的特征作为输入,并且有很大的适用范围,可用于通常目标而不仅仅是某类特定目标的检测.在INRIA行人库和多视角车辆库上的实验证明了算法的有效性.

A novel boosting-based classifier called splited cascade model was proposed for detecting multiview, multi-pose objects of a known class. By using the coarse-to-fine strategy, a tree-structured classifier in which a rejection cascade was learned at each node could be constructed without predefined intra-class sub-categorization. The proposed method does not limit the types of features that are used and is suitable for detecting various objects. Experiments on INRIA human database and multi-view vehicle data demonstrate the efficacy of the proposed approach.