Dynamic Contextual Recognition of Moving Objects

摘要IIS-9900115Zabih，拉明康奈尔大学，15万-36MOS。运动对象的动态上下文识别这是一个为期三年的标准奖项。这项研究将解决重建从环境中的多个非重叠摄像机暂时可见的所有移动对象所走的路径的问题。有许多有趣的监控应用程序，其中集成来自多个摄像头的信息非常重要。不幸的是，对于人工操作员来说，同时处理多个视频流是非常困难和累人的。PI将采用贝叶斯方法，其中最优解是在给定观测数据的情况下具有最高后验概率的对象路径集。PI计划研究组合优化方法的使用，包括线性规划和图割，以设计这些问题的近似方案。这些方法有望产生高效的算法，并对近似的质量提供有趣的保证。

Abstract IIS-9900115Zabih, RaminCornell University $150,000 - 36 mos.Dynamic Contextual Recognition of Moving ObjectsThis is a standard award of three years. The research will address the problem of reconstructing the paths taken by all moving objects that are temporarily visible from multiple non-overlapping cameras in an environment. There are many interesting surveillance applications where it is important to integrate information from multiple cameras. Unfortunately, it is very difficult and tiring for human operators to simultaneously attend to several video streams. The PI will take a Bayesian approach, where the optimal solution is the set of object paths with the highest posterior probabilitygiven the observed data. The PI plans to investigate the use of combinatorialoptimization methods, including linear programming and graph cuts, to designapproximation schemes for these problems. These methods are expected to result in efficient algorithms with interesting guarantees concerning thequality of the approximation.