Pattern Recognition and understanding for Visual Information Media

模式识别与理解是实现智能视觉信息媒介的基础技术，它可以根据每个人的需求对现实世界中的视觉信息进行选择、总结和扩充。本项目的目的是将PRU技术改造为多媒体领域的关键技术，期望信息媒体支持人类的活动，增强人类的能力。这两种情况的关键区别在于，在多媒体系统中，人类消耗PRU的输出，而在自动化系统中，机器消耗PRU的输出来替代人类。作为项目的共同研究对象，我们重点研究了PRU在信息媒体中的三个典型和基本问题。L)处理人类行动和行为的机制。2)处理人类模糊需求的机制。3)处理简单建模不适用的真实世界的机制。充分利用以下三个假设，成功解决了这些经典难题。I)识别结果的最终消费者是人类。2)计算机与人性化的实时交互成为可能。3)可获得长周期、多摄像头、宽视野的视觉数据。我们开发了具有智能摄像头控制能力的远程学习系统、具有目光接触恢复能力的混合现实社区空间、具有视觉信息汇总机制的个人情节记录系统和采用多个全方位图像传感器的视频监控系统等几个视觉信息系统。

Pattern recognition and understanding, PRU, is a fundamental technology to realize intelligent visual information media, which can select, summarize, and augment the visual information in the real world according to the need of each individual. The purpose of this project is to restructure the PRU technology, which has been developed for making automatic machines to replace humans, into a key technology in the multimedia fields where the information media are expected to support the human activity and to augment the human ability. The critical difference between the two cases is that a human consumes the output of PRU in multimedia systems, while a machine consumes it in automated systems for substituting humans.As common research targets of the project, we have focused on three problems that are typical and fundamental in the PRU for information media. L) The mechanism to handle the action and behavior of humans. 2) The mechanism to handle the ambiguous demand of humans. 3) The mechanism to handle the real world for which simple modeling is not applicable. These classic and difficult problems have been solved successfully by fully utilizing the following three assumptions. I) The final consumer of recognition results is a human. II) The real time interaction between computers and humane is possible. III) Visual data with long period, multiple cameras, wide field-of-vision are available.We have developed several visual information systems, such as a distance learning system with intelligent camera control capability, a mixed-reality community space with eye-contact recovery capability, a personal episode recording system with visual information summary mechanism, and a video surveillance system using multiple omni-directional image sensors.