• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Journal of visual communication & image representation >Visual tracking via dynamic weighting with pyramid-redetection based Siamese networks
【24h】

Visual tracking via dynamic weighting with pyramid-redetection based Siamese networks

机译：通过动态加权与基于金字塔检测的暹罗网络进行视觉跟踪

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Siamese network based similarity-learning algorithm is currently a significant branch of visual tracking. However, most of existing deep Siamese networks depend much on the offline-trained knowledge and always assume the same importance for different prediction views. In this paper, we first introduce a dynamic weighting module in Siamese framework, which could make the offline-trained network adapt to the current circumstance well and weight predictive response maps discriminatively. The thought stems from the basis that different maps have different predictive preference, which should not be treated equally. Secondly, in order to focus more on the accurate preference, we then introduce the residual structure to form the residual dynamic weighting module. Thirdly, we construct a simple online pyramidredetection module to avoid local search and also consider the global viewpoint. Extensive experiments on both short-term and long-term tracking demonstrate that the proposed tracker possesses the competitive tracking performance over many mainstream state-of-the-art trackers. (C) 2019 Elsevier Inc. All rights reserved.
机译：基于连体网络的相似性学习算法目前是视觉跟踪的重要分支。但是，大多数现有的深度暹罗网络在很大程度上取决于离线训练的知识，并且对于不同的预测视图始终具有相同的重要性。在本文中，我们首先在暹罗框架中引入了动态加权模块，该模块可以使脱机训练的网络很好地适应当前情况，并能区分判别加权预测响应图。这种想法源于不同地图具有不同的预测偏好的基础，因此不应同等对待。其次，为了更专注于准确偏好，我们接着引入残差结构以形成残差动态加权模块。第三，我们构建了一个简单的在线金字塔重检测模块，以避免局部搜索，同时考虑全局视角。短期和长期跟踪的大量实验表明，与许多主流的最新跟踪器相比，该跟踪器具有竞争性的跟踪性能。 （C）2019 Elsevier Inc.保留所有权利。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号