随着智能手机的飞速发展和普及，移动增强现实应用受到学术界和工业界的广泛关注。姿态跟踪作为移动增强现实中的核心技术成为了迫切需要。然而，传统的移动增强现实跟踪技术通常基于单一的视觉方法或者传感器方法，很难同时满足跟踪精度、鲁棒性和实时性的需求。本课题致力于研究新的融合视觉与惯性传感器的混合跟踪模型以实现智能手机上精确、鲁棒、实时的跟踪。围绕混合跟踪模型中三个具体难点开展深入研究，具体包括： 轻量级在线的特征选择算法；多源传感器融合跟踪方法；基于视觉-传感器数据融合跟踪方法及自适应突变误差校验。本课题将有力推动移动增强现实的理论和应用的发展，为提供一种移动环境下全新的人机交互方式提供核心算法与技术。

With fast development and deployment of smartphones, mobile augmented reality (MAR) on smartphones has attracted increasing interests from academia and industry. Pose tracking is one of the most critical tasks in an MAR system. However, conventional methods which rely on only visual features or inertial sensors are either too computationally expensive on smartphones, or not sufficiently accurate and robust for a satisfactory user experience. This proposal focuses on developing a novel hybrid tracking system and address key challenges in the system, including: lightweight online feature selection for robust feature tracking, multimodel sensor data fusion for inertial tracking, adaptive Kalman filtering with abrupt error verification for vision-inertial hybrid tracking. The research conducted in this proposal will provide robust and efficient pose tracking techniques which can run efficiently on mobile phones. Meanwhile, it will significantly speedup the development of MAR theories and facilitate wide deployment of MAR apps.