PICTORIAL STRUCTURES MODEL BASED HUMAN INTERACTION RECOGNITION

In this paper, we propose an interaction recognition system using a single camera in the outdoor scene. The proposed system consists of three parts: human detection, pose estimation, and interaction recognition. We build a new human pose model by integrating Pictorial Structures Model (PSM) with the motion masks, which adopts the temporal information to reduce the searching region and result in a better pose estimation. Moreover, we recognize the human interaction by employing the Hierarchical Hidden Markov Model (HHMM), where the lower level describes the interaction pose for a frame and the higher level describes the human interaction for a sequence. We demonstrate that our system has a better performance in different outdoor scenes and can reach 88% recognition rate.