Author(s)

Yibo Min, Jianwei Ma, Shaofei Zang, Hongyun Zheng

Corresponding Author

Jianwei Ma

ABSTRACT

Deep learning methods have been widely used in target tracking with superior performance, but most target trackers still require a large amount of video training data to ensure the robustness of the network, resulting in low real-time performance. In order to address this problem, we propose a deep feedforward network that does not require online training to learn the general relationship between target motion and appearance. Then, target position can be output through multi-layer regression network. Finally, the tracker shows excellent results on the benchmark data sets.

KEYWORDS

Object tracking, Deep learning, Convolutional neural network