Comparison of four visual tracking algorithms based on deep learning
Jin Wei, Haibin Sun, Xin Du, Haojin Li, Linrun Ye, Kun Liu · 2020
To evaluate the visual tracking algorithm proposed by our research team, we compare the algorithm with other three visual tracking algorithms. Firstly, the four visual tracking algorithms are introduced. There are SiamFC, SiamRPN++, ATOM and TDLD, which are all based on deep learning. The first three algorithms are the state-of-the-art trackers of different periods. The last algorithm is proposed by ourselves. And then we do some experiments in seven video sequences from OTB-100 dataset. We qualitatively compare the robustness of the four algorithms on the five tracking challenging factors. The average centre location error (ACLE) and average overlap score (AOC) of the four algorithms are calculated to make a quantitative analysis. The SiamRPN++ algorithm gets the best result of ACLE three times, and the TDLD gets twice. Both the SiamRPN++ and the TDLD get the best result of AOC three times respectively. The analysis results show that performance of the TDLD is very close to the state-of-the-art trackers.