Abstract: To enhance the tracking performance of textureless parts in the augmented reality assembly system,a region-based 3D tracking method was proposed.A new smooth step function was used to optimize the image segmentation method based on level-set function,which improved the segmentation effect of contour edges.Then,a pixel foreground and background color posterior probability statistical model was designed to enhance the time consistency between continuous frames and improve the robustness and accuracy against motion blur.Finally,the Gauss-Newton method was used for pose optimization,with its fast convergence and numerical stability ensuring real-time and stability of the algorithm.Experimental results demonstrated that the proposed 3D tracking method could accurately track textureless parts.Meanwhile,the image segmentation and pose estimation exhibit stronger robustness against disturbances such as motion blur or cluttered backgrounds,meeting the requirements of tracking textureless parts in industrial scenarios.

Key words: augmented assembly, foreground and background region, image segmentation, pose estimation, 3D object tracking

摘要： 为提升增强现实装配系统中对于弱纹理零件的跟踪效果,提出一种基于区域的三维跟踪方法。首先,采用全新的平滑阶跃函数优化基于水平集函数的图像分割方法,提高了轮廓边缘的分割效果;然后,设计了像素前背景颜色后验概率统计模型,增强了连续帧之间的时间一致性,提高了对运动模糊的鲁棒性和准确性;最后,采用高斯牛顿方法进行姿态优化,利用其快速收敛和数值稳定的性质,保证算法的实时性和稳定性。实验结果表明,所提出的三维跟踪方法能对弱纹理零件进行精确的跟踪。同时,在面对运动模糊或背景杂乱等干扰时,图像分割与位姿估计表现出更强的鲁棒性,满足了工业场景中对弱纹理零件跟踪的要求。

关键词: 增强装配, 前背景区域, 图像分割, 姿态估计, 三维目标跟踪