• 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 >3DSportNet: 3D sport reconstruction by quality-aware deep multi-video summation
【24h】

3DSportNet: 3D sport reconstruction by quality-aware deep multi-video summation

机译：3DSportNet：通过质量意识的深度多视频求和重建3D运动

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

Automatically reconstructing 3D sceneries from video sequences is an indispensable technique in computer 3D games, urban planning, and intelligent navigation. Many previous work relies on complicated and expensive equipment to fulfill 3D reconstruction under constrained environments. Nevertheless, such schemes are not readily to be applied for reconstructing 3D sport sceneries, such as basketball and mountain climbing. In this work, we propose a novel deep architecture: 3DSportNet, which reconstructs 3D sport sceneries by making use of multiple handheld videos captured by smart phones. In particular, given a rich of mobile videos captured by users, we extract multiple deep/shallow visual features from each sport video frame by leveraging the weakly-supervised semantic encoding. Afterward, a geometry-aware quality model is designed to summarize the multiple videos into multiple key frames from each single video, wherein the objective is that the selected key frames can maximally reconstruct the multiple sport videos. Based on this, we employ the key frames to reconstruct sport videos by utilizing the PMVS2 software. Comprehensive experimental comparisons and visualization results have shown that our method can produce very real 3D sport sceneries and athletes. Besides, the 3D reconstruction time consumption is reduced by 95% compared to conventional methods. (C) 2019 Published by Elsevier Inc.
机译：从视频序列自动重建3D场景是计算机3D游戏，城市规划和智能导航中必不可少的技术。以前的许多工作都依赖复杂且昂贵的设备来在受限环境下完成3D重建。然而，这样的方案并不容易应用于重建3D运动场景，例如篮球和登山。在这项工作中，我们提出了一种新颖的深度架构：3DSportNet，该架构通过利用智能手机捕获的多个手持视频来重建3D体育场景。特别是，鉴于用户捕获了丰富的移动视频，我们利用弱监督的语义编码从每个运动视频帧中提取了多个深/浅视觉特征。之后，设计几何感知质量模型以将多个视频从每个单个视频中总结为多个关键帧，其中目的是所选择的关键帧可以最大程度地重构多个运动视频。基于此，我们利用关键帧通过PMVS2软件重建体育视频。全面的实验比较和可视化结果表明，我们的方法可以生成非常真实的3D运动场景和运动员。此外，与传统方法相比，3D重建时间消耗减少了95％。 （C）2019由Elsevier Inc.发布

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