Fast and Robust Object Tracking Using Image Foresting Transform

We propose a fast and robust approach to track deformable and rigid objects from a stationary/moving camera. The method combines seed estimation by the Kanade-Lucas-Tomasi (KLT) algorithm with object delineation by the image foresting transform to segment the object along the frames. The segmentation does not need to be perfect and it considerably improves efficiency and robustness in the KLT algorithm. The tests in real videos demonstrate good results, by outperforming two other KLT-based methods, in the tracking of objects, such as people, cars, motorcycles, under partial occlusion and rotation.