An efficient online tracking algorithm to sketch the combined direction and orientation of an object from a video is an arduous task. Most of the discriminative trackers learn from the previous frame output to estimate object in given frame. Many use single stage classifier for classification and the output of the classifier will be the final output in most of the cases. The presence of misclassification by the classifier can ead to failure of the algorithm. To overcome this problem, Two-stage classification is proposed in this paper, where the framework of the algorithm is the novelty. In the proposed algorithm, a Linear SVM classifier is used as a first stage classifier and a Laplacian Regularized Least square learning in Bayesian learning framework is used as a second stage classifier. Using the first stage classification, a set of positive features are selected and by using the second stage classification the tracker location is selected. Hence by doing so, the burden on the classifier is shared and misclassification is reduced considerably. The proposed algorithm is an efficient online tracker, which can handle most of the challenges and performs well in different attributes of selected standard 17 Dataset. The proposed algorithm has improved precision rate (74.1%) and success rate (54.7%) compared with all the available diverse range of trackers.
Discriminative Tracking; Two-Stage Classification; Visual Object Tracking
