基于环境自适应孪生网络的低空目标视频跟踪

Low-altitude target visual object tracking based on environment adaptive siamese network

为了实现对低空目标准确、高效的跟踪,通过分析低空目标所处背景可能发生的变化,总结背景变化时跟踪低空目标的技术特点,对基于孪生网络的视频目标跟踪算法进行改进,提出了基于环境自适应孪生网络的低空目标跟踪算法,并利用UAV123数据集对所提算法进行评估。结果显示,加入环境自适应模块的算法比现有的基于孪生网络跟踪算法在精度上提高了5.7%,在成功率上提高了7.3%,对低空目标的跟踪性能有明显的提升。

To achieve accurate and efficient tracking of low-altitude target,by analyzing the possible changes in the background of low-altitude target,the featrues of low-altitude target when the background changes were summarized.The video object tracking algorithm based on siamese network was improved.And a video object tracking algorithm of low-altitude target based on environment adaptive siamese network was proposed.The proposed algorithm was evaluated through UAV123 data set.The results show that compared with fully-convolutional siamese networks for object tracking,the algorithm with environment adaptive module improves the accuracy by 5.7%and the success rate by 7.3%.The tracking performance of low-altitude target has been significantly improved.