Human action detection and classification using optimal bag-of-words representation

Based on interest keypoints extracted as salient space-time volumes, human-action can be described as a “bag of visual words”. This representation has been frequently used in the classification of image and video data. The representation choices regarding the dimension, selection, and weighting of visual words are crucial to the classification performance. In this paper, we address the problem of efficient human-action classification by selecting an optimal bag-of-words representing an action. We introduce genetic algorithm based SVM classifier which performs dimension reduction, features subset selection, Instance selection, visual words weighting and SVM parameter selection. The impact of this optimization to human-action classification is studied through extensive experiments on the TRECVID and CMU video collections.