混合3D模型跟踪注册算法

发布时间：2019-03-15 06:35

【摘要】：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.
[Abstract]:In the maintenance guidance system, the registration algorithm based on 3D model requires manual initialization of position and pose, easy to generate error offset, and can not automatically update 3D model, and many other problems, such as manual initialization of pose, error offset and automatic updating of 3D model are needed in maintenance guidance system. A hybrid registration algorithm for 3D model tracking based on edge features is proposed. Firstly, the number of edge pixels is selected as the measure standard, and the particle filter algorithm is used to initialize the camera position and pose. Then the nonlinear least square error estimation algorithm is used to continuously update the pose estimation, and the error offset is eliminated by real-time monitoring of the number of control points in the model. Finally, key point matching is used to realize process recognition and real-time updating of 3D model. The experimental results on the prototype system show that the algorithm has strong stability and high efficiency, and can effectively overcome the error migration.
【作者单位】：空军工程大学防空反导学院;西安交通大学机械制造系统工程国家重点实验室;
【基金】：国家自然科学基金(51405505)
【分类号】：TP391.41

【相似文献】