Cloud-shift: A novel algorithm for visual object tracking
Bo Jiang, Wenyan Gan, Hailong Zhang, Xin Wang · 2015
Building effective and efficient appearance models is a challenging task for robust object tracking. Online learned strategy is widely used to solve appearance changes for its adaptive ability. However, it will introduce potential drifting problems for the accumulation of errors during the self-updating. As a result, this paper attempts to create the visual object appearance model using cloud model for the first time to solve the research gap of target object appearance changes in the tracking. Then we present a novel online cloud-shift algorithm for object tracking based on the cloud model representation. Based on a set of comprehensive experiments, our algorithm has demonstrated efficiency in appearance changes and other challenging factors and performs favorably against state-of-the-art algorithms on challenging sequences in terms of efficiency, accuracy and robustness.