Real-time High Precision Multi-media Integration System Using Distributed Multi-sensor Network

点击翻译按钮获取中文摘要

The aim of this project is to provide a system to provide users "arbitrary views" of the real-world information depending on the users' demands.To achieve this goal, we have to observe the real world precisely through a lot of sensors and to integrate the sensed data into a meaningful information coincide with the user's request. In this project, we have established a framework of real-time acquisition and integration of information from multiple sensors, especially multiple cameras. The major achievements of this project are summarized as follows.1.Integration of images acquired by multiple camerasWe have established a software framework to integrate, in real-time, images acquired by multiple cameras on a parallel-distributed computers, PC cluster. To achieve the real-time issue, we have also developed a mechanism to make the processing time constant based on variable-resolution strategy, in which the resolution of 3D space is changed depending on the geometrical complexity of an obse … More rved object.2.Real-time execution control of vision systemThere is a trade-off between required confidence and required computational resources of vision processings. For example, when we require precise result, the longer processing time is required. In practical applications, especially real-time ones, we have to make this difficult trade-off depending on the required accuracy and the available resources. Here, we have established a real-time execution control mechanism, called Confidence Driven mechanism, which controls the trade-off dynamically, depending on the user's required confidence, based on prediction of data and accuracies of implemented algorithms.3.Construction of prototypical systems and evaluationBased on the previous results, we have implemented prototypical applications : real-time vision-based human motion capturing and real-time free view-point video on a PC cluster, which has 9 IEEE cameras and 20 or more PCs. The experimental results show the effectiveness of our software framework of real-time vision processing. Less