Continuity-Discrimination Convolutional Neural Network for Visual Object Tracking

This paper proposes a novel model, named Continuity-Discrimination\nConvolutional Neural Network (CD-CNN), for visual object tracking. Existing\nstate-of-the-art tracking methods do not deal with temporal relationship in\nvideo sequences, which leads to imperfect feature representations. To address\nthis problem, CD-CNN models temporal appearance continuity based on the idea of\ntemporal slowness. Mathematically, we prove that, by introducing temporal\nappearance continuity into tracking, the upper bound of target appearance\nrepresentation error can be sufficiently small with high probability. Further,\nin order to alleviate inaccurate target localization and drifting, we propose a\nnovel notion, object-centroid, to characterize not only objectness but also the\nrelative position of the target within a given patch. Both temporal appearance\ncontinuity and object-centroid are jointly learned during offline training and\nthen transferred for online tracking. We evaluate our tracker through extensive\nexperiments on two challenging benchmarks and show its competitive tracking\nperformance compared with state-of-the-art trackers.\n