Research on Object Tracking Based on Data Enhancement of Sequence Inversion

• 1、School of Computer Science,Beijing University of Posts and Telecomunications,Beijing 100876

Abstract：Recntly, many object tracking algorithms based on deep learning encounter the problem of insufficient training data, which leads to the model can not learn an effective representation of moving objects. However, most deep learning algorithms by fine-tuning the neural network online to a certain extent make up for the lack of training data.Fine-tuningonline leads to increased computational complexity making them can not be applied to real-time target tracking scenarios. This paper proposes a new training data enhancement methodto solve this problem.By reversing the original video sequences,we can obtain twice the size of original training data.Utilizing all the datathe model can learn highly robust motion representation even only trained off-line.This can save online fine-tuning steps. The experimental results show that the model\'s performance trained in this way improves on the VOT2014 test set .?

Keywords： Deep Learning Object Tracking Data Enhancement VOT2014