针对维修诱导系统中采用基于3D模型的注册算法跟踪复杂机械设备时要求手动初始化位姿、容易产生误差偏移以及不能自动更新3D模型等诸多问题,提出一种基于边缘特征的3D模型混合跟踪注册算法.首先选取边缘像素数量作为测度标准,采用粒子滤波算法对摄像机位姿进行初始化;然后在持续位姿估计中利用非线性最小平方误差估计算法对位姿进行持续更新,并且通过对模型控制点数量的实时监控消除误差偏移;最后利用关键点匹配实现进程的识别和3D模型实时更新.在原型系统进行实验的结果表明,该算法具有较强的稳定性和较高的效率,能够有效地克服误差偏移.%In guided maintenance systems, the 3D model based registration algorithm requires user assisted initial pose, is prone to drift error, and cannot automatically update the 3D model, when tracking complex mechanical objects. To address these issues, a hybrid edge-based 3D registration and tracking is proposed. it uses the particle filter algorithm to set the number of edge pixels as a criterion and initialize the pose of camera. In the later continuous pose estimation, it adopts a nonlinear least square method to update the poses, and eliminates error drifting by real-time monitoring the number of control points on the model. The pro-posed algorithm achieves the process recognition by key-point matching to finally get an auto-updating 3D model. The experimental results on the prototype system show that the proposed algorithm is robust and ef-ficient, and effectively eliminates error drifting at the same time.
展开▼
