Transfer AdaBoost learning for action recognition

The universal dataset of human action (such as KTH) includes only simple background, in which the action videos are much different to practical action videos. So the accurate rate of action recognition on practical videos always not so good as on our test videos from the training dataset. However, it will cost lots of human and material resources to establish a labeled video set which includes a large amount of videos with various backgrounds. In this paper, we propose a novel transfer learning framework called TrAdaBoost, which extends boosting-based learning algorithms. By using this algorithm, we can train a action recognition model fitting for most practical situations just relaying on the universal action video dataset and a little set of new action videos with complex background. And by using the TrAdaBoost, the generality of our action recognition model is greatly improved.