Target Tracking Based on Siamese Convolution Neural Networks
Haibo Pang, Qi Xuan, Meiqin Xie, Chengming Liu, Zhanbo Li · 2020
Target tracking is an important research content in the field of computer vision. There is a problem that speed and precision of tracking can’t be balanced. Aim at this problem, this paper proposes a Siamese-SE deep neural network, which is an improvement in the structure of the Siamese-FC Network that add the SE-Network to the network to improve the feature representation ability by refining and extracting the features. And we use Pearson Correlation Coefficient to calculate similarity of two samples by shifting the matching one by one, which makes the tracking speed and accuracy keep balance. The OTB2015 datasets is used to train and test the network, and TRE, SRE and OPE are used as evaluation criteria in this paper. The experimental results demonstrate that the algorithm proposed is superior to other contrast algorithms including Siamese-FC in tracking effect diagram and three evaluation criteria. This paper verifies the feasibility of the proposed network and meets the requirements of real-time target tracking application.