Although, offline object detectors have shown a tremendous success. One major
drawback of offline techniques is that a complete set of training data has to be collected
beforehand. In addition, once learned, an offline detector cannot make use of
newly arriving data. In order to alleviate these shortcomings, online learning has been
adopted with following objectives:- the technique should be computational and storage
efficient; and the updated classifier must maintain its high classification accuracy.
Improving upon GSLDA classifier, an effective and efficient framework for learning
an online GSLDA model is proposed. Unlike existing online object detection algorithms,
e.g., Grabner and Bischof or Pham and Cham, our online approach
makes use of LDA's learning criterion which has been shown in our previous experiment
to outperform the AdaBoost's learning criterion for offline object detection task.
Our updating algorithm is very efficient since we neither replace weak learners nor
throw away any weak learners during updating phase. Finally, we adopt a learning
technique similar to semi-supervised learning where the classifier makes use of the
unlabeled data in conjunction with a small amount of labeled data.