基于视觉注意力机制的机器人感兴趣目标跟踪

项目名称： 基于视觉注意力机制的机器人感兴趣目标跟踪

项目编号： No.61473089

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 于元隆

作者单位： 福州大学

项目金额： 80万元

中文摘要： 与传统的目标跟踪方法需要程序员固化某些目标特征不同，感兴趣目标跟踪是机器人根据当前环境和任务在线自主选取值得注意的目标。同时，跟踪过程自身的动态性也导致在跟踪过程中需要不断的更新关于目标和环境的知识。因此，如何赋予机器人一定的思考能力和自主成长能力，实现具有任务非特异性的感兴趣目标跟踪是一项具有前瞻性的课题。本项目以视觉注意力机制以及源于神经生理学机制的学习理论为基础，研究数据驱动的感兴趣目标检测、任务驱动的感兴趣目标检测、特征在线选取与学习、目标模型在线串行学习等技术，从而构成具有认知能力的机器人感兴趣目标跟踪系统。最后，本系统将在实际机器人平台完成实验验证与性能评估。本项目的研究成果可以广泛应用于民用与国防领域，例如异常行为或突发事件的检测与跟踪、机器人导航、智能交通、辅助医疗、智能监控等。

中文关键词： 目标跟踪；认知机器人；视觉注意力；自主心智成长；特征选取

英文摘要： In contrast to the traditional object tracking methods which require programmers to fix certain features of targets, tracking objects of interest is a process for robots to autonomously select objects to be attended according to current task and environment. Meanwhile, dynamical changes during tracking leads to the online sequential update of knowledge about the targets and environment. Thus it is a promising and challenging issue to endow robots with the ability of autonomous mental development and thinking in order to achieve task-nonspecificity. This research project attempts to build a cognitive system of tracking objects of interest for robots based on visual attention mechanism and biologically-inspired machine learning theories. This project mainly includes data-driven detection of objects of interest, task-driven detection of object of interest, online feature selection and online sequential learning of target models. Finally this tracking system will be tested and evaluated on the real robotic platform. The research can be widely applied in various fields, e.g. detection and tracking of anomaly events and emergencies, robotic navigation, intelligent traffic, auxiliary medical, intelligent surveillance and so on.

英文关键词： Object tracking;Cognitive robots;Visual attention;Autonomous mental development;Feature selection