Activity recognition in thermal infrared video


In this paper, we investigate the tracking and recognition of limited activity in thermal infrared video. We have improved the pose segmentation from the background using a universal segmentation technique. Gait energy images (GEI) have been developed for collected repetitive and non-repetitive activities. Seven invariant moments features are extracted from the sequences of GEI of each activity and concatenated to a feature vector. Naïve Bayesians classifier is used for classification of feature vectors. Experimental result on limited activity shows the effectiveness of our proposed activity recognition algorithm.

Publication Title

Conference Proceedings - IEEE SOUTHEASTCON