Depth feature fusion target tracking method based on attention mechanism

The invention discloses a depth feature fusion target tracking method based on an attention mechanism. The method comprises the following steps: S1, obtaining a training video set and a test video set; s2, constructing a Siamese network, and sending the video into the constructed network; s3, an ECO-SA module and an ECO-CA module are constructed; s4, extracting a target feature map by adopting an SBT network; and S5, sending the position codes into the target tracking network to complete target classification and positioning. According to the method, the image feature correlation is deeply embedded into the multi-layer feature network, and the features of the two images are widely matched, so that the key features of the images are enhanced, the non-target features are inhibited, and the target tracking precision is improved. 本发明公开了一种基于注意力机制的深度特征融合目标跟踪方法，包括以下步骤：S1：获取训练视频集和测试视频集；S2：构建Siamese网络，将视频送入构建网络中；S3:构建ECO-SA模块和ECO-CA模块；S4：采用SBT网络提取目标特征图；S5：将位置编码送入目标跟踪网络中，完成目标分类定位。本发明通过将图像特征相关性深度嵌入到多层征网络中，广泛匹配两幅图像的特征，从而