本篇論文提出了一個應用於室內場景的同步定位與地圖建構的原型機。在演算法上，我們參考勞爾設計的系統，使用有方向性的角偵測來進行同步定位與地圖建構（ORB-SLAM）。
首先我們利用慕尼黑工業大學所提供的影像深度資料庫在計算機上模擬同步定位與地圖建構，從模擬的結果中我們指出幀率與定位追蹤間的取捨問題。我們將演算法的各個部分數學模組化，在此我們提出一個的方法來最佳化幀率與定位追蹤的取捨問題，利用控制系統中，比例積分微分控制器的概念，我們設計了動態特徵提取的演算法架構，將當下成功匹配特徵點數量、相機的移動速度作為輸出，來計算與目標值的差值，利用此項誤差即可在下一次的回饋中去補償系統所需要抽取的特徵點數量，這個方法可以隨著影像，適當的調整需要的特徵點數量，而達到穩定的定位能力，也可以兼顧快速的幀率。
接下來，我們修改了決定關鍵幀的規則來降低定位失敗的機率，原本的ORB-SLAM在快速移動或是重新定位後很容易再次定位失敗，我們修改了規則來讓追蹤更穩定，系統在重新定位後也比較不容易再追蹤失敗。
另外，我們使用了進化策略演算法來調整出適合的參數，我們一共處理的三千九百多萬張的影像，以大量的模擬選出了表現較好的參數組合。我們也使用了另一個資料庫來驗證我們所提出的演算法。從幀率、定位結果上來說，我們改善了約15.2%到50.8%的效能，而且我們的系統每秒可以處理20張影像。
在系統整合上，我們建立了一台同步定位與地圖建構用的機器人原型機，由一個影像與深度的感測器、一台筆電、一台掃地機器人組裝而成，加上我們改良的演算法。再機器人的操作上，我們用遙控的方式進行實際場景中的實驗，機器人一邊移動一邊將資料輸入處理器，經由我們的演算法就能得到鏡頭的移動軌跡及重建後的三維模型，我們也會儲存實驗過程中的數據，來分析幀率、定位效果及準確度。

In this thesis, a prototype to implement real-time Simultaneous Localization And Mapping (SLAM) in indoor environment is proposed. In the algorithm, we referred to Oriented FAST (Features from Accelerated Segment Test) and Rotated BRIEF (Binary Robust Independent Elementary Features) SLAM (ORB-SLAM) which is a feature-based SLAM system developed by Raul Mur-Artal and Juan D. Tardos.

At first, we simulated offline SLAM using TUM RGB-D dataset and figured out the issues about frame rate and lost tracking. Then, we assumed some conditions when running real-time SLAM and modeled the feature extraction, feature matching, and motion estimation to an optimized problem. We presented a new method to optimize the problem of lost tracking using the concept of PID control system. A dynamic feature extraction scheme is proposed. We calculate the error between the setpoint and the number of matched features meanwhile considering the velocity of the camera. The feedback would compensate the system in the next feature extraction. The proposed scheme can appropriately extract features from images in order to avoid lost tracking and does not need to slow down the frame rate.

Second, we revised the keyframe decision to reduce the possibility of lost tracking. The original ORB-SLAM is weak when fast motion or relocalization. We modified the rules to make tracking more steady. The system can process stably after relocalization.

Third, we applied evolution strategy to fine-tune the parameters. We simulated numerous offline SLAM. Totally, about 39 million images were processed. Then, we verified our design on another RGB-D benchmark, ICL-NUIM dataset. We improved the performance from 15.2% to 50.8%. The frame rate of our system is about 20 frames per second (FPS).

In the system integration, we built a robot equipped with a sensor, a vehicle, and a processor. Using modified ORB-SLAM, we can remote control the device via Robot Operation System (ROS) and run real-time experiments. When the robot scans the rooms, the process can localize the camera trajectory and display the 3D model meanwhile. The information of SLAM is saved after programming. We can analyze the frame rate, lost tracking, and accuracy.

[1] G. Klein and D. Murray, “Parallel Tracking and Mapping for Small AR Workspaces,” in 2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality, Nov 2007, pp. 225–234.
[2] F. Endres, J. Hess, J. Sturm, D. Cremers, and W. Burgard, “3-D Mapping With an RGBD Camera,” IEEE Transactions on Robotics, vol. 30, no. 1, pp. 177–187, Feb 2014.
[3] J. Engel, T. Sch¨ops, and D. Cremers, LSD-SLAM: Large-Scale Direct Monocular SLAM. Cham: Springer International Publishing, 2014, pp. 834–849. [Online]. Available: http://dx.doi.org/10.1007/978-3-319-10605-2 54
[4] R. Mur-Artal, J. M. M. Montiel, and J. D. Tards, “ORB-SLAM: A Versatile and Accurate Monocular SLAM System,” IEEE Transactions on Robotics, vol. 31, no. 5, pp. 1147–1163, Oct 2015.
[5] J. Engel, V. Koltun, and D. Cremers, “Direct Sparse Odometry,” in arXiv:1607.02565, July 2016.
[6] W. Hess, D. Kohler, H. Rapp, and D. Andor, “Real-time loop closure in 2D LIDAR SLAM,” in 2016 IEEE International Conference on Robotics and Automation (ICRA), May 2016, pp. 1271–1278.
[7] E. Rosten and T. Drummond, “Machine Learning for High-speed Corner Detection,” in Proceedings of the 9th European Conference on Computer Vision - Volume Part I, ser. ECCV’06. Berlin, Heidelberg: Springer-Verlag, 2006, pp. 430–443. [Online]. Available: http://dx.doi.org/10.1007/11744023 34
[8] D. G. Lowe, “Distinctive Image Features from Scale-Invariant Keypoints,” Int. J. Comput. Vision, vol. 60, no. 2, pp. 91–110, Nov. 2004. [Online]. Available: http://dx.doi.org/10.1023/B:VISI.0000029664.99615.94
[9] E. Rublee, V. Rabaud, K. Konolige, and G. Bradski, “ORB: An efficient alternative to SIFT or SURF,” in 2011 International Conference on Computer Vision, Nov 2011, pp. 2564–2571.
[10] R. B. Rusu and S. Cousins, “3D is here: Point Cloud Library (PCL),” in IEEE International Conference on Robotics and Automation (ICRA), Shanghai, China, May 9-13 2011.
[11] M. Calonder, V. Lepetit, C. Strecha, and P. Fua, BRIEF: Binary Robust Independent Elementary Features. Berlin, Heidelberg: Springer Berlin Heidelberg, 2010, pp. 778–792.
[12] M. A. Fischler and R. C. Bolles, “Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography,” Commun. ACM, vol. 24, no. 6, pp. 381–395, Jun. 1981. [Online]. Available: http://doi.acm.org/10.1145/358669.358692
[13] A. Handa, T. Whelan, J. McDonald, and A. Davison, “A Benchmark for RGB-D Visual Odometry, 3D Reconstruction and SLAM,” in IEEE Intl. Conf. on Robotics and Automation, ICRA, Hong Kong, China, May 2014.
[14] K. H. Ang, G. Chong, and Y. Li, “PID control system analysis, design, and technology,” IEEE Transactions on Control Systems Technology, vol. 13, no. 4, pp. 559–576, July 2005.
[15] T. Salimans, J. Ho, X. Chen, and I. Sutskever, “Evolution Strategies as a Scalable Alternative to Reinforcement Learning,” ArXiv e-prints, Mar. 2017.
[16] A. P. Bradley, “The use of the area under the ROC curve in the evaluation of machine learning algorithms,” Pattern Recognition, vol. 30, no. 7, pp. 1145 – 1159, 1997. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S0031320396001422