近年來，隨著機器學習與電腦視覺快速發展，物件辨識和追蹤是需多應用的主要技術，包括現在很熱門的車用電子輔助駕駛系統Advanced driver assistance systems (ADAS)。在追蹤演算法這方面，光流法因其具有直觀的表達運動模式且不易受到物體外在的干擾等優點而被廣泛採用作為物件追蹤的方法，光流法的演算法主要有HS（Horn and Schunk）和LK（Lucas and Kanade）兩種方法，但是由於計算複雜度相當高，因此雖然已經有軟體程式庫（如Opencv）可支援光流法計算，但是在要求即時處理速度的應用(如前述ADAS)仍有很大的改善空間。為了達到即時之效果，本論文採用了硬體實作的方法來進行加速，而我們選擇效果最好的LK光流法作為研究主軸，在不失其計算精準度下對該演算法設計一套專屬的硬體來加速運算，透過分析硬體運算的時間面積成本與精確度三者相互關係，提出可達即時性且不失精確度的面積成本最小之硬體設計。其中本論文最大的貢獻在於提出簡化版的限制範圍除法器，該除法器相較其它論文可大幅節省面積時間與功耗。此外本論文也提出了將影像切割運算的方法，以解決大範圍運算下金字塔LK光流法需要大量記憶體的問題。

Due to rapid advances of machine learning and computer vision, object track recognition and tracking are fundamental technologies in many applications, including the popular advanced driver assisted systems (ADAS). Optical flow is widely used to compute the motion vectors during object tracking. There are two major optical algorithms: HS (Horn and Schunk) and LK (Lucas and Kanade). Although optical flow has been implemented in software library such as OpenCV, the speed performance is usually not satisfactory in many applications that require real-time processing speed, such as in ADAS. In this thesis, we propose hardware implementations of LK optical flow algorithm considering the trade-off between area cost, speed, and accuracy. A low-cost simplified divider used in the optical flow hardware is presented with reduced computation accuracy. Furthermore, we propose a data partition and computation method to reduce the memory requirement in the pyramid optical flow hardware.

目錄
審定書 i
摘要 ii
英文摘要 iii
圖目錄（List of Figures） vii
表目錄（List of Tables） xi
第1章 緒論（Introduction） 1
1.1 研究動機 1
1.2 本文大綱 3
第2章 研究背景與相關研究 4
2.1 文獻回顧 4
第3章 光流法演算法 7
3.1光流概念(Concept of the Optical Flow) 7
3.2 Lucas-Kanade光流法 9
3.2.1前置處理(Pre. processing) 10
3.2.2梯度計算(Gradient) 11
3.2.3最小平方法(Least Square Matrix) 13
3.2.4求解方程式(Equation Solver) 14
3.2.5應用(Application) 16
3.3 Lucas-Kanade光流法搭配高斯金字塔 17
3.3.1高斯金字塔(Gaussian Pyramid) 17
3.3.2粗略到精細的光流計算(Coarse to Fine Optical Flow) 19
3.3.3演算法總結(Summary of the Pyramidal LK algorithm) 21
第4章 LK光流法演算法-軟體分析 22
4.1 光流評估方法介紹 24
4.2 硬體設計與精確度及計算時間分析比較 26
4.3 金字塔層數與精確度計算時間比較 30
第5章 LK光流法演算法-硬體加速 34
5.1 LK光流法硬體設計(雛形) 35
5.1.1 RGB to Intensity 36
5.1.2 Gaussian Filter 37
5.1.3 Gradient Filter 40
5.1.4 Least Square Matrix 42
5.1.5 Equation Solver 44
5.2 LK光流法硬體設計(低成本) 50
5.2.1 Gaussian Filter 51
5.2.2 Gradient Filter 54
5.2.3 Equation Solver 56
5.3金字塔LK光流法硬體設計(金字塔) 57
5.3.1狀態機(Finite-State Machine)狀態介紹 58
5.3.2狀態機(Finite-State Machine)架構介紹 63
第6章 實驗數據分析與比較 75
6.1 邏輯數據和分析 75
6.2 論文比較 84
第7章 結論與未來希望 89
7.1 結論 89
7.2 未來展望 89
參考文獻（References） 90
附錄A　光流法測試圖像數據介紹 93
附錄B 各種濾波器讀值架構介紹與分析 96

[1] P. Viswanath, et al., “A Diverse Low Cost High Performance Platform for Advanced Driver Assistance System (ADAS) Applications,” IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2016.
[2] Z. Kalal, K. Mikolajczyk and J. Matas, “Tracking-learning-detection. Pattern Analysis and Machine Intelligence,” IEEE Transactions on, vol. 34, no. 7, pp. 1409-1422, 2012.
[3] S. Hare, A. Saffari and P. H Torr, “Struck: Structured output tracking with kernels,” In Computer Vision (ICCV), 2011 IEEE International Conference, pp. 263-270, Nov.2011
[4] R. E. Kalman, “A new approach to linear filtering and prediction problems,” Transactions of the ASME–Journal of Basic Engineering, vol. 82, no. Series D, pp. 35–45, 1960.
[5] M. S. Arulampalam, et al., “A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking,” IEEE Trans. Signal Process, vol. 50, no. 2, pp. 174–188, Feb. 2002.
[6] Barron J, Fleet D and Beauchemin S, “Performance of optical flow techniques,” International Journal of Computer Vision, 1994, 12 (1): 42-77.
[7] B. K. P. Horn, and B. G. Schunck, “Determining Optical Flow,” Artificial Intelligence, Vol. 17, pp.185-203, 1981
[8] B.D. Lucas and T. Kanade, “An Iterative Image Registration Technique with an Application to Stereo Vision,” DARPA Image Understanding Workshop, pp. 121-130, 1981
[9] M. V. Correia and A. C. Campilho, “Real-time implementation of an optical flow algorithm,” in Proc. Int. Conf. Pattern Recognition, 2002, pp. 247–250..
[10] J. Dìaz, et al., “FPGA based real-time optical-flow system,” IEEE Trans. Circuits Syst. Video Technol, vol. 16, no. 2, pp. 274–279, Feb. 2006.
[11] J. Diaz, E. Ros, R. Agis, and J. Bernier, “Superpipelined high-performance optical-flow computation architecture,” Comput. Vision Image Understand., vol. 112, no. 3, pp. 262–273, 2008.
[12] V. Mahalingam, et al., “A VLSI architecture and algorithm for Lucas–Kanade-based optical flow computation,” IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 18, no. 1, pp. 29–38, Jan. 2010.
[13] F. Barranco, et al., “Parallel architecture for hierarchical optical flow estimation based on FPGA,” IEEE Trans. VLSI Syst. vol. 20, no. 6, pp.1058–1067, 2012
[14] H.-S. Seong, and H.-J. Lee, “A VLSI design of real-time and scalable Lucas-Kanade optical flow,” ICEIC 2014Kota Kinabalu, Malaysia, pp. 305–306.Jan.2014.
[15] H.-S. Seong, C.- E. Rhee and H.-J. Lee, “A Novel Hardware Architecture of the Lucas–Kanade Optical Flow for Reduced Frame Memory Access,” IEEE Circuits and Systems Society,Volume: 26 Issue: 6, May 2015
[16] J. J. Gibson, “The Perception of the Visual Wrld,” Oxford, U.K.: Houghton Mifflin, 1950
[17] H. Greenspan, C.H. Anderson, and S. Akber,“Image enhancement by nonlinear extrapolation in frequency space, ”IEEE Trans. Image Process., 2000, pp. 1035–1048.
[18] Jean-Yves Bouguet, “Pyramidal implementation of the Lucas Kanade feature tracker description of the algorithm,” Intel Corporation Microprocessor Research Labs, 2001
[19] S.Baker, D. Scharstein, and J.P. Lewis, “A Database and Evaluation Methodology for Optical Flow,” Eleventh IEEE International Conference on Computer Vision (ICCV 2007), Rio de Janeiro, Brazil, Oct. 2007
[20] M. H. Seyed Javadia, and H. R. Mahdianib, “Efficient Utilization of Imprecise Blocks for Hardware Implementation of a Gaussian Filter,” IEEE Computer Society Annual Symposium on VLSI, 2015
[21]H. Ishihara, M. Miyama, and Y. Mastsuda, “A VGA 30-fps Optical-Flow Processor Core Based on Pyramidal Lucas and Kanade Algorithm,” IEEE Asian Solid-State Circuits Conference, Korea, Nov. 2007
[22]Y. Murachi, Y.Fukuyama, and R. Yamamoto, “A VGA 30-fps Optical-Flow Processor Core Based for Moving Picture Recognition,” IEICE TRANS. ELECTION, April. 2008
[23] Q. Zhu, N. Garg, Y. Tsai, and K. Pulli, “An energy efficient timesharing pyramid pipeline for multi-resolution computer vision,” IEEE 21st International Conference on Very Large Scale Integration. IEEE, 2013, pp. 278–281.
[24] Tomasi, M, et al., “High-Performance Optical-Flow Architecture Based on a Multi-Scale, Multi-Orientation Phase-Based Model,” IEEE Transactions on circuits and systems for video technology, Dec. 2010