Abstract: In order to effectively handle the tracking of textureless objects, this paper proposes an edge-based adaptive real-time 3D tracking method. While the 3D model of the tracking object is present, the proposed method can robustly detect and track the object edges using historical motion information, and then accurately caculate the extrinsic camera parameters. Contributions of this paper are summarized as follows: 1) Utilizing adaptive threshold and historical information for motion prediction, which can improve the robustness of tracking while movement is fast; 2)Proposing a RANSAC-based edge matching strategy, which can effectively eliminate outliers to make the tracking of complex objects robust;3) Extending the type of model in edge-based tracking from CAD model to general mesh model by extracting the silhouette from the 3D model. Experimental results demonstrate the robustness and efficiency of this method, which can satisfy the demand of augmented reality and virtual assembly.

Key words: edge based tracking, real time 3D tracking, silhouette, Random Sample Consensus （RANSAC） algorithm, adaptive threshold

摘要： 针对缺乏纹理特征的物体，提出了一种基于边的自适应实时三维跟踪方法。在已知物体三维模型的情况下，通过基于历史运动信息的物体边缘检测与跟踪，可以有效准确地求解出摄像机的外参。基于并扩展了现有的基于边的实时跟踪算法，其主要工作体现在以下三个方面： 1）提出自适应阈值和基于历史信息估计当前帧的运动趋势的方法，从而提高边匹配算法在快速运动时的稳定性；2）提出一种基于随机抽样一致性（RANSAC）的边匹配策略，可以有效剔除误匹配的边，从而提高复杂模型的跟踪稳定性；3）利用抽取轮廓边的算法将边跟踪算法从CAD模型扩展到一般的面片模型。实验结果证明了该方法的鲁棒高效，能够满足增强现实、虚拟装配等应用需求。

关键词: 边跟踪, 实时三维跟踪, 轮廓边, 随机抽样一致性算法, 自适应阈值