Object tracking algorithm of Siamese network based on feature fusion and attention mechanism

Abstract

With the development of computer vision technology, object tracking is widely used in various fields of social life. The traditional object tracking method mainly characterizes the object by considering the designed features, and then realizes the object tracking. With the proposal of the deep learning algorithm, the object tracking field has ushered in a new development. The object tracking algorithm based on the Siamese neural network calculates the target position of the current frame by calculating the response between the template image and the feature map of the image to be searched. It is concise and efficient, but it is prone to tracking drift when facing complex backgrounds. The visual attention mechanism is beneficial to highlight the background and suppress the background. Therefore, this paper introduces an attention mechanism based on the Siamese network to improve the performance of object tracking. In addition, a key point for object tracking based on Siamese network is object and background classification. Compared with high-level features, low-level features have stronger discriminative ability. Therefore, a feature fusion mechanism is also introduced and combined with the attention mechanism to optimize the object tracking algorithm based on Siamese network. Compared with the OTB2013 benchmark evaluation, the algorithm in this paper has better tracking performance than the original algorithm.