This paper studies the challenging activity recognition task and a novel discriminative action pattern is proposed. We construct action cube pool and propose a novel scheme to represent action cubes as action patterns...
详细信息
ISBN:
(纸本)9781538609361
This paper studies the challenging activity recognition task and a novel discriminative action pattern is proposed. We construct action cube pool and propose a novel scheme to represent action cubes as action patterns. Candidates of discriminative patterns are selected among action patterns according to motion energy of corresponding action cube. spatial constraint clustering algorithm clusters these candidates into several disjoint clusters. Discriminative patterns are learned with the help of negative patterns and multi-class SVM. The presented approach is validated on the MSRAction3D dataset and the MSRDailyActivity3D dataset. The experimental results prove the effectiveness of our approach.
暂无评论