F. Niu and M. Abdel-Mottaleb
Human activity recognition, background subtraction, eigen shape, fusion, hidden Markov model, view invariance
Recognizing human activities from image sequences is an active area of research in computer vision. Most of the previous approaches on activity recognition focus on recognition from a single view and ignore the issue of view invariance, and they deal with recognizing a single activity. There are only few published algorithms for segmenting and recognizing complex activities that are composed of more than one activity. In this paper, we present a view invariant human activity recognition approach that uses both motion and shape information. An augmented vector of both optical flow features as well as eigen shape features is used to represent motion and shape of the body in the region of interest in each frame of the sequence. Each activity is represented by a set of hidden Markov models, where each model represents the activity from a different viewing direction, to realize the view invariance. Also, we present a voting-based approach to automatically and effectively segment and recognize complex activities. Experiments on two sets of video clips of different activities show that our method is effective.
Important Links:
Go Back
