Online Object Tracking and Learning with Sparse Deformable Template Models

Object tracking is an important and challenging task in the field of computer vision. The objective of object tracking is to associate target objects in consecutive video frames. In this paper, we use SVM trained active basis model as a sparse deformable template for representing objects. Active basis model is a sparse model, which represents each image as a small number of bases selected from an over-complete Gabor dictionary. Given the bounding box of the object in the first frame, the model can be trained on the positive image patch inside the bounding box and negative images outside the bounding box. The tracking is achieved by detection of the object in the subsequent frames in the video by using the learned model. The model will be updated after a few of frames by using new positive and negative images, which are specified by the model. The experiment shows a good performance of the tracking method in some testing videos.