融合通道互联空间注意力的Siamese网络跟踪算法

基于Siamese网络的跟踪算法在跟踪精度和速度方面展现出巨大的潜力，然而要使离线训练的模型适应在线跟踪仍然面临着挑战。为了提升复杂场景下算法的特征提取以及判别能力，提出了一种融合通道-互联-空间注意力的Siamese网络实时跟踪算法。首先构建以深度卷积网络VGG-Net-16作为主干网络的Siamese跟踪框架，增加特征提取能力；接着设计通道-互联-空间注意力模块，增强模型的适应能力与判别能力；然后加权融合多层响应图，获取更精准的跟踪结果；最后使用大规模数据集对网络进行端到端的训练，在通用数据集OTB-2015上进行跟踪测试。实验结果表明:与当前主流算法相比，所提算法具有较强的稳健性，能更好地适应目标外观变化、相似物干扰、目标遮挡等复杂场景，在NVIDIA RTX 2060 GPU上，跟踪速度平均达到37FPS，满足实时性要求。

Siamese networks tracking algorithm integrating channel-interconnection-spatial attention

The tracking algorithms based on the Siamese networks show great potential in terms of tracking accuracy and speed.However,it is still challenging to adapt the offline trained model to online tracking.In order to improve the feature extraction and discrimination ability of the algorithm in complex scenes,a Siamese network real-time tracking algorithm that combines channel,interconnection and spatial attention mechanisms was proposed.First a Siamese tracking framework with a deep convolutional network VGG-Net-16 as the backbone network was built to increase feature extraction capabilities;then the channel-interconnection-spatial attention module was integrated to enhance the adaptability and discrimination capabilities of the model;then the multi-layer response maps were weighted and fused to obtain more accurate tracking results;and finally the largescale datasets were used to train the end-to-end network,and tracking test on the benchmark OTB-2015 was completed.The experimental results show that compared with the current mainstream algorithms,the proposed algorithm is more robust and better adapt to complex scenes such as target appearance changes,similar distractors,and occlusion.On the NVIDIA RTX 2060 GPU,the average tracking speed reaches 37 FPS,which meets real-time requirements.