首页> 外文期刊>Fortschritt-Berichte VDI, Reihe 8. Mess-, Steuerungs- und Regelungstechnik >Prediction Strategies for Dynamic Objects in Visual Scenes
【24h】

Prediction Strategies for Dynamic Objects in Visual Scenes

机译:视觉场景中动态对象的预测策略

获取原文
           

摘要

The novel contributions from this dissertation to the tracking research community can be summarized under the following aspects which have been integrated into a single robust tracking framework for arbitrary objects and whose qualitative and quantitative benefits have been demonstrated in a series of comparative evaluations. On the sensory side, it relies on a multiple-cue based, illumination invariant measurement method which can cope with a large variety of appearance transformation. On the modeling side, this work introduced, on one hand, a probabilistic way to fuse the cues. On the other hand, it focused on the hierarchical modeling of prediction models and proposed a scheme for probabilistic exchange of top-down and bottom-up influences between the modules inside a prediction model hierarchy. Base on this scheme, a modularized inference mechanism in a hierarchical Bayesian modeling structure is enabled. In addition, this work modeled and solved Depth-from-Size as a Bayesian inference problem inside three coupled HAAM chains.
机译:本文从以下几个方面总结了本论文对跟踪研究界的新贡献:这些方面已被整合到一个针对任意物体的鲁棒跟踪框架中,并且在一系列比较评估中证明了其定性和定量收益。在感觉方面,它依赖于基于多提示的照明不变性测量方法,该方法可以应对各种各样的外观变换。在建模方面,一方面,这项工作引入了一种融合提示的概率方式。另一方面,它着重于预测模型的层次建模,并提出了一种在预测模型层次结构内的模块之间概率自上而下和自下而上的影响进行概率交换的方案。基于该方案,启用了分层贝叶斯建模结构中的模块化推理机制。此外,这项工作将三个深度耦合的HAAM链中的深度深度作为贝叶斯推理问题建模并求解。

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