基于传感器与高效二阶最小化结合的追踪方法

Object Tracking Solution Based on Sensors and Efficient Second-order Minimization

在室外环境中,由于追踪目标的运动状态可能会随时发生改变,而且手持设备的硬件性能与台式机相比存在局限性,所以可以运用在移动设备上的高效实时追踪技术一直是研究的重点和难点。在移动增强现实(augmented reality,AR)系统中,提出了一种基于手持移动设备的无标记物体追踪方法。该方法能很好地适应目标物体的旋转和运动模糊的情况。首先在特征点匹配阶段,将传感器的旋转参数引入鲁棒独立的基本二进制特征描述子(binary robust independent elementary features,BRIEF)中,使特征描述子具有旋转不变性。在特征匹配成功后采用改进的高效二阶最小化追踪算法(efficient second-order minimization-small grid-motion blur,ESM-SG-M B)来对目标物体进行追踪。该算法引入了摄像头的快门时间,并将模板图像进行网格划分,从中选择高梯度的子网格进行追踪。实验结果表明,与现有的移动增强现实系统中运用的追踪方法相比,该方法具有更好的鲁棒性和实时性。

The movement status of tracking targets may change unpredictably in outdoor environment,and the mobile device has poorer performance than stand-alone device does.Therefore,efficient real-time tracking technology that can be applied to mobile devices has always been the focus and difficulty of research.For this,we propose a mark-less tracking solution based on handheld devices in outdoor augmented reality（ AR）,which can be well adapted to the target＇s rotating and motion blur.At the stage of feature matching,the rotation parameters of the sensors are introduced to BRIEF（ binary robust independent elementary features） descriptor to achieve a rotational-invariant performance.After a successful match is observed,the ESM-SG-MB（ efficient second-order minimization-small grid-motion blur） tracking method starts to serve its role to continue tracking.ESM-SG-MB tracking algorithm introduces camera shutter time and divides the template into sub-grids. The average image gradient in each sub-grid is computed,and only high gradient grids are used in tracking.Experiment shows that the proposed tracking solution has a better robustness and real-time ability than several state-of-the-art tracking method in mobile augmented reality.