|
| 基于区域提案孪生网络的优化目标跟踪算法 |
| Optimization of target tracking algorithm based on region proposal Siamese network |
| 投稿时间：2020-07-22 |
| DOI：10.3969/j.issn.1005-5630.2021.01.003 |
| 中文关键词: 孪生网络 区域提案网络 条形池化 通道注意力 |
| 英文关键词:siamese network regional proposal network strip pooling channel attention |
| 基金项目:上海市人工智能计划（2019RGZN01077） |
|
| 摘要点击次数: 2278 |
| 全文下载次数: 1781 |
| 中文摘要: |
| 为了更好地对目标的尺度进行实时估计，避免多尺度测试及提高目标跟踪的速度和精度，提出了一种新的优化目标跟踪算法。通过将跟踪效果比较好的区域提案网络引入普通的孪生网络，并在算法中引进条形池化模块和高效通道注意力模块，应对物体的尺度差异和跟踪过程中较为剧烈的形变。提出的算法在OTB100数据集上取得了0.833的准确度和0.658的成功率，在VOT2016数据集上取得了0.411的EAO指数，在VOT2019数据集上取得了0.275的EAO指数。 |
| 英文摘要: |
| In order to estimate the target scale in real time, avoid multi-scale test and improve the speed and accuracy of target tracking, a new optimized target tracking algorithm is proposed. By introducing the regional proposal network with good tracking effect into the common siamese network, and introducing the strip pooling module and the efficient channel attention module in the algorithm, we can deal with the scale difference of objects and the severe deformation in the tracking process. The proposed algorithm achieves 0.833 accuracy and 0.658 success rate on OTB100 dataset, 0.411 EAO index on VOT2016 dataset, and 0.275 EAO index on VOT2019 dataset. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
