为了克服物体表面缺少足够的纹理特征且算法搜索空间太大的缺陷，提出了一种基于模型与自然特征点相融合的三维注册追踪方法.采用保持旋转和尺度不变性的线性并行多模态(LINE-MOD)模板匹配方法快速识别目标物体、获取与当前视角接近的参考视图而完成相机位姿的粗略估计，并缩小算法的搜索空间；采用基于自然特征点的方法完成相机位姿的精确计算；为了避免因特征点较少而引起的位姿抖动或扰动，引入了有效的非迭代透视n点问题(RPnP)算法以提高注册追踪的精度和速度.结果表明，所提出的注册追踪方法能够进行快速三维注册，具有良好的实时性和鲁棒性，其运算速度可达30帧/s.
To tackle the huge searching space and few texture feature points on the object surface problem, we propose a novel method that combines model-based tracking and natural feature tracking for fast and robust camera pose calculation. The proposed method first adapts the state-of-the-art template matching method LINE-MOD into a scale invariant descriptor to recognize the target object and obtain the key-frame that is similar to the current viewpoint. Therefore, it greatly reduces the search space for pose refinement method and improves the searching speed. After that, the keyframe obtained from template matching stage is exploited to refine the tracking accuracy. In order to avoid tracking jitter and improve tracking accuracy and speed, RPnP algorithm is incorporated into our system. The experimental results and performance evaluations demonstrate that our method is fast, accurate and robust for 3D registration and tracking even under partial occlusion, and the tracking speed can reach 30 frame/s.
［1］WESTERFIELD G, MITROVIC A, BILLINGHURST M. Intelligent augmented reality training for motherboard assembly［J］. International Journal of Artificial Intelligence in Education, 2015, 25(1): 157-172.
［2］FIORENTION M, UVA A E, GATTULLO M, et al. Augmented reality on large screen for interactive maintenance instructions［J］. Computers in Industry, 2014, 65(2): 270-278.
［3］CRIVELLARO A, RAD M, VERDIE Y, et al. A novel representation of parts for accurate 3D object detection and tracking in monocular images［C］∥Proceedings of the IEEE International Conference on Computer Vision (ICCV). Santiago, Chile: IEEE, 2015: 4391-4399.
［4］MOTTAGHI R, YU X, SILVIO S. A coarse-to-fine model for 3D pose estimation and sub-category recognition ［C］∥IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, USA: IEEE, 2015, 418-426.
［5］ENGEL J, THOMAS S, DANIEL C. LSD-SLAM: Large-scale direct monocular SLAM［C］∥European Conference on Computer Vision (ECCV). Zurich, Switzerland: Springer, 2014: 834-849.
［6］PAUWELS K, RUBIO L, DIAZ J, et al. Real-time model-based rigid object pose estimation and tracking combining dense and sparse visual cues［C］∥Conference on Computer Vision and Pattern Recognition (CVPR). Portland, USA: IEEE, 2013: 2347-2354.
［7］KYRIAZIS N, ANTONIS A. Scalable 3D tracking of multiple interacting objects［C］∥IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Columbus, USA: IEEE, 2014, 3430-3437.
［8］PAYET N, SINISA T. From contours to 3D object detection and pose estimation［C］∥International Conference on Computer Vision (ICCV). Barcelona, Spain: IEEE, 2011: 983-990.
［9］RADKOWSKI R. Object tracking with a range camera for augmented reality assembly assistance［J］. Journal of Computing and Information Science in Engineering, 2016, 16(1): 011004.
［10］GARRETT T, DEBERNARDIS S, OLIVER J, et al. Poisson mesh reconstruction for accurate object tracking with low-fidelity point clouds［J］. Journal of Computing and Information Science in Engineering, 2017, 17(1): 011003.
［11］WANG G, WANG B, ZHONG F, et al. Global optimal searching for textureless 3D object tracking［J］. The Visual Computer, 2015, 31(6-8): 979-988.
［12］NEWCOMBE R A, IZADI S, HILLIGES O, et al. KinectFusion: Real-time dense surface mapping and tracking［C］∥10th IEEE International Symposium on Mixed and Augmented Reality (ISMAR). Basel, Switzerland: IEEE, 2011: 127-136.
［13］HINTERSTOISSER S, CAGNIART C, ILIC S, et al. Gradient response maps for real-time detection of textureless objects［J］. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2012, 34(5): 876-888.
［14］RUBLEE E, RABAUD V, KONOLIGE K, et al. ORB: An efficient alternative to SIFT or SURF［C］∥International Conference on Computer Vision (ICCV). Barcelona, Spain: IEEE, 2011: 2564-2571.
［15］KALAL Z, MATAS J, MIKOLAJCZYK K. P-N learning: Bootstrapping binary classifiers by structural constraints［C］∥Conference on Computer Vision and Pattern Recognition (CVPR). San Francisco, USA: IEEE, 2010, 49-56.
［16］周凯汀, 郑力新. 基于改进 ORB 特征的多姿态人脸识别［J］. 计算机辅助设计与图形学学报, 2015, 27(2): 287-295.
ZHOU Kaiting, ZHENG Lixin. Multi-pose face recognization based on improved ORB features［J］. Journal of Computer-Aided Design and Computer Graphics, 2015, 27(2): 287-295.
［17］LU C P, HAGER G D, MJOLSNESS E. Fast and globally convergent pose estimation from video images［J］. Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(6): 610-622.
［18］LI S Q, XU C, XIE, M. A robust O(n) solution to the perspective-n-point problem［J］. Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7): 1444-1450.
［19］ABAWI D F, JOACHIM B, RALF D. Accuracy in optical tracking with fiducial markers: An accuracy function for ARToolKit［C］∥Proceedings of the 3rd IEEE/ACM International Symposium on Mixed and Augmented Reality. Arlington, USA: IEEE, 2004: 260-261.
