시공간 패턴을 이용한 감시 카메라에서의 그룹 행동 분류
우해나, 정순홍, 설상훈 · 대한전자공학회 학술대회 · 2014
In this paper, we propose the method for classifying six group activities, by using several spatiotemporal patterns and their projection. The proposed algorithm first generates spatiotemporal pattern by sampling pixels at a set of horizontal lines in each frame along time axis and projects them into one pattern. Motion trajectories of each group are computed by comparing the several spatiotemporal patterns with projected pattern. Hidden Markov model (HMM) classifier categorizes the motion trajectories by activities of two groups. The experimental results show that the proposed algorithm can estimate motion trajectory precisely and show the better performance than existing approach.
Cited by 0 works.