Target Tracking Methods Based on Deep Learning

Target tracking has also received a great deal of attention from society. The basic problem of a target tracking algorithm is to select a target of interest in a video or image sequence, and in the next consecutive frames, find the exact position of that target and form its motion trajectory. In the early days of dealing with target tracking the optical flow method (Lucas-Kanade), Kalman filtering, kernel methods, etc. were usually used to achieve target tracking. But since 2016, deep learning has also made breakthroughs in target tracking as well as target detection. In this paper, we introduce the traditional target tracking algorithms and then describe and discuss in detail the target tracking algorithms around deep learning, to explore the excellence of deep learning in target tracking compared with the earlier tracking algorithms as well as the broad prospects for development in the future.