Research of single object tracking method based on Siamese Network and Level Set
Tianbo Liu, Li Ping Su, Shuai Yuan, Gong Cheng, Feng Zhang · 2020
In this paper, we propose a single object tracking method based on a Siamese network and level set to increase the accuracy of object tracking and segmentation while maintaining real-time tracking. First of all, feature extraction and similarity measurement are performed on the object template image and the search image by using the Siamese backbone network based on ResNet. Then the top-down object tracking contour refinement is performed on the feature map using the level set method. Finally, the loss function of the algorithm is defined, and the proposed algorithm is evaluated and compared on the VOT and DAVIS dataset. The experimental results illustrate that the proposed algorithm maintains the real-time performance, and at the same time the accuracy has been significantly improved.