Human Activity Tracking using Star Skeleton and Activity Recognition using HMM ’ s and Neural Network

Human motion detection and analysis is currently an important area of research, motion analysis help us to solve many problems. An Automated Video Surveillance Model is presented in this paper. The model which I proposed is capable of detecting and monitoring people in both environments Indoor and outdoor. The Model is capable to find out the Suspicious and NonSuspicious activities. It also detects multiple peoples in video and monitoring their activities. Moving targets are detected and their boundaries extracted, we use star skeletonization technique with the adaptive centroid point to create human skeletons. In this paper we use HMM-based methodology for action recognition. In our proposed method, a series of star skeletons is generated according to action over time. Then, time-sequential images frame is converted into a feature vector sequence and the feature vector is translated into sequence of symbols after that we use Neural Network for action recognition which represents the particular action in Suspicious and Non-Suspicious category. We design a codebook of posture, which contains representative star skeletons of various human action types and define a star distance to find out the similarity between feature vectors. Each and every sequence of feature vector is matched against the codebook and Neural Network after that Symbol is assigned to the most similar. Then the time-sequential images or frames are converted to a sequence of symbol posture by this, we easily categorize the actions.