工業4.0為現今工廠自動化的重要趨勢，自動倉儲系統(Automated Storage and Retrieval System, ASRS)更是其中較重要的議題，它被廣泛地應用在各種類型的工廠中去處理各種不同的儲存應用。然而建造一套完整的自動倉儲系統的花費過於高昂，普遍的中小企業並沒有辦法負擔得起。而AGV(Automatically Guided Vehicle)則是一個低成本並且可以替代部分自動倉儲系統的方案，通過在堆高機上安裝AGV系統，便能夠達到自動化運輸的目的，在其中關鍵則是必須知道棧板的位置。
    因此本論文提出一個基於自適應結構特徵(Adaptive Structure Feature, AS)與方向加權重疊率(Direction Weighted Overlapping Ratio, DWO)的棧板辨識系統，用以輔助堆高機的取放貨。透過所提出的自適應結構特徵與方向加權重疊率結合棧板檢測，可以移除大多數非棧板的背景，並且增加處理效率。在我們的方法中應用了Haar-like特徵，結合Adaboost與階層式分類器的架構來檢測棧板，並使用變異數特徵與自適應結構特徵來優化檢測結果。最後計算各候選區與追蹤紀錄的方向加權重疊率來進行非最大抑制，以選出最適合的結果。實驗結果證明我們的方法，可以在光線較少的環境下檢測出棧板，同時這樣的混合式檢測系統，對於棧板的檢測率可以達到平均95%。

Nowadays, Industry 4.0 is an important trend in factory automation (FA), which automated-storage-and-retrieval-system (ASRS) is one of the most important issues in the industry. It has been widely used in a variety of industries to handle a variety of storage applications in factories, warehouses, etc. However, the cost to construct an ASRS is too expensive such that most of the small/medium enterprises cannot afford it. A forklift system is another alternative solution to replace the complicated ASRS due to its low cost characteristics. In this work, a new pallet detection based on adaptive structure feature (AS) and direction weighted overlapping ratio (DWO) to aid forklifts in picking up the pallet process model is proposed from a single-lens camera erected at the forklift. Combining AS and DWO for pallet detection, the proposed method can remove most of the non-stationary (dynamic) background and significantly increase the processing efficiency. In our approach, Haar like-based Adaboost scheme with AS of pallets algorithm to detect pallets is presented. It can detect the pallet at luminance less environment. Finally, by calculating the DWO between the detected pallets and tracking record, it can avoid those error responses in object tracking. Therefore, this work improves the pallet detection to solve the problem with an effective design. As results, the hybrid algorithms proposed in this work can increase the average pallet detection rate by 95%.

中文摘要 .......................................................................... I
英文摘要 ......................................................................... II
目錄 ................................................................................. III
圖目錄 ............................................................................. V
表目錄 ........................................................................... VII
第一章	緒論 ............................................................................... 1
1.1	研究動機 ............................................................................... 2
1.2	論文架構 ............................................................................... 4
第二章	文獻回顧 ..................................................................... 5
2.1	基於影像的棧板檢測 ....................................................... 5
2.2	基於感測器的棧板檢測 .................................................. 7
2.3	物件追蹤 ............................................................................... 9
第三章	棧板檢測 ................................................................... 11
3.1	設定感興趣區域 (Region of Interest, ROI) ............... 13
3.2	降低影像解析度 ............................................................... 14
3.3	哈爾特徵 (Haar-like feature) ..........................................17
3.3.1	Haar-like特徵 ........................................................... 17
3.3.2	積分影像 .................................................................... 18
3.4	Adaptive Boost Algorithm (Adaboost Algorithm) ..... 21
3.5	階層式分類器 (Cascade Classifier) ............................. 24
3.6	變異數 .................................................................................. 26
3.7	自適應結構特徵 ............................................................... 31
第四章	追蹤系統 .................................................................... 37
4.1	Direction Weighted Overlapping Ratio (DWO) .......... 38
4.2	候選區重疊情況比較 ..................................................... 44
第五章	實驗結果 .................................................................... 48
5.1	實驗環境與樣本 ............................................................... 48
5.2	實驗結果與分析 ............................................................... 50
第六章	結論 .............................................................................. 56
參考文獻 ....................................................................................... 57

圖目錄
圖 1.1 工業發展的演進.............................................................. 1
圖 1.2 實驗環境 (a) Cui et al. [1] (b) Chen et al. [2] ........ 3
圖 2.1尋找下半邊界的處理過程[1] ...................................... 6
圖 2.2棧板中心點求法[2] ......................................................... 7
圖 2.3 RFID ceiling tag environment[6] ................................... 8
圖 3.1 系統流程圖 .................................................................... 12
圖 3.2 攝影機安裝位置 ........................................................... 13
圖 3.3 輸入影像 .......................................................................... 14
圖 3.4  感興趣區域影像 ......................................................... 14
圖 3.5 (a)低低頻帶係數 (b)低高頻帶係數 (c)高低頻帶係數 (d)高高頻帶係數 ............. 16
圖 3.6 (a)原始影像 (b)低低頻帶影像.................................... 17
圖 3.7基本型Haar-like特徵矩形............................................ 18
圖 3.8積分影像算法示意圖...................................................... 20
圖 3.9 哈爾特徵值算法示意圖................................................ 21
圖 3.10 階層式分類器示意圖.................................................. 24
圖 3.11 工廠環境下之 (a)上貨物棧板 (b)貨箱的彩色影像 (c)棧板的灰階影像 ........... 26
圖 3.12 RGB色彩空間[18] ......................................................... 28
圖 3.13 HSV色彩空間[19] ......................................................... 29
圖 3.14 棧板樣本(一) ................................................................. 31
圖 3.15棧板樣本(二) .................................................................. 31
圖 3.16 灰階數值分布(%).......................................................... 32
圖 3.17 二值化結果..................................................................... 33
圖 3.18 滑動統計式意圖........................................................... 34
圖 3.19亮度分布曲線................................................................. 34
圖 3.20 棧板亮度模型................................................................ 35
圖 4.1 重疊區域示意圖.............................................................. 39
圖 4.2 81種重疊情況下傳統重疊率所算出之比率。.... 40
圖 4.3 DWO ratio示意圖............................................................. 40
圖 4.4  81種重疊情況下方向加權重疊率所算出之比率....... 41
圖 4.5重疊情況之比較................................................................ 44
圖 4.6 尋找最佳候選區示意圖 (a)重疊情況 (b)最終結果..... 45
圖 5.1 棧板訓練正樣本............................................................... 48
圖 5.2 用於訓練之負樣本[23] .................................................. 49
圖 5.3 部分辨識結果示意圖...................................................... 50
圖 5.4 實驗場景一 (a) 廠房內未開燈僅有自然光線的情況 (b) 廠房內有開燈的情況..................... 51
圖 5.5 實驗場景二 (a) 廠房內未開燈僅有自然光線的情況 (b) 廠房內有開燈的情況.................... 53
圖 5.6 實驗場景三.......................................................................... 54

表目錄
表 3.1階層式分類器各層之特性比較..................................... 25
表 4.1傳統重疊率與DWO Ratio比較表................................ 43
表 5.1實驗結果(一) ........................................................................ 52
表 5.2實驗結果(二) ........................................................................ 53
表 5.3實驗結果(三) ........................................................................ 54

參考文獻
[1]	G. Z. Cui, L. S. Lu, Z. D. He, L. N. Yao, C. X. Yang, B. Y. Huang, and Z. H. Hu, “A robust autonomous mobile forklift pallet recognition,” International Asia Conference on Informatics in Control, Automation and Robotics, pp. 286-290, 2010.
[2]	G. Chen, R. Peng, Z. Wang, and W. Zhao, “Pallet Recognition and Localization Method for Vision Guided Forklift,” International Conference on Wireless Communications, Networking and Mobile Computing, pp. 1-4, 2012.
[3]	G. Garibotto, S. Masciangelo, M. Ilic, and P. Bassino, “ROBOLIFT: A vision guided autonomous fork-lift for pallet handling,” IEEE/ RSJ Int. Conf. on Robots and Systems, pp. 656-663, 1996.
[4]	M. Seelinger, and J. D. Yoder, “Automatic Pallet Engagment by a Vision Guided Forklift,” IEEE International Conference on Robotics and Automation, pp. 4068-4073, 2005.
[5]	J. Pagès, X. Armangué, J. Salvi, J. Freixenet, and J. Martí, “A computer vision system for autonomous forklift vehicles in industrial environments,” Mediterranean Conf. on Control and Automation, pp.379-384 , 2001.
[6]	S. Jeon, M. Choi, G. Kim, and B. Hong, “Localization of Pallets Based on Passive RFID Tags,” International Conference on Information Technology: New Generations, pp. 834-839, 2010.
[7]	Z. He, X. Wang, J. Liu, J. Sun, and G. Cui, “Feature-to-Feature Based Laser Scan Matching for Pallet Recognition,” International Conference on Measuring Technology and Mechatronics Automation, pp. 260-263, 2010.
[8]	D. Lecking, O. Wulf, and B. Wagner, “Variable Pallet Pick-Up for Automatic Guided Vehicles in Industrial Environments,” IEEE Conference on Emerging Technologies and Factory Automation, pp. 1169-1174, 2006.
[9]	M. Fogel, N. Burkhart, H. Ren, J. Schiff, M. Meng, and K. Goldberg, “Automated Tracking of Pallets in Warehouses: Beacon Layout and Asymmetric Ultrasound Observation Models,” IEEE International Conference on Automation Science and Engineering, pp. 678-685, 2007.
[10]	Z. Kalal, K. Mikolajczyk, and J. Matas, “Tracking-Learning-Detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 7, pp. 1409-1422, 2012.
[11]	X. Liu, D. Tao, M. Song, L. Zhang, J. Bu, and C. Chen, “Learning to Track Multiple Targets,” IEEE Transactions on Neural Networks and Learning Systems, vol. 26, no. 5, pp. 1060-1073, 2015.
[12]	R. Cucchiara, M. Piccardi, and A. Prati, “Focus based feature extraction for pallets recognition,” British Machine Vision Conf., pp. 1-10, 2000
[13]	S. Byun, and M. Kim, “Real-Time Positioning and Orienting of Pallets Based on Monocular Vision,” IEEE International Conference on Tools with Artificial Intelligence, pp. 505-508, 2008.
[14]	C. H. Hsia, J. M. Guo, and J. S. Chiang, “Improved Low-Complexity Algorithm for 2-D Integer Lifting-Based Discrete Wavelet Transform Using Symmetric Mask-Based Scheme,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 19, no. 8, pp. 1202-1208, 2009.
[15]	P. Viola, and M. Jones, “Rapid object detection using a boosted cascade of simple features,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol.1, pp. I-511-I-518, 2001.
[16]	R. Lienhart, and J. Maydt, “An extended set of Haar-like features for rapid object detection,” International Conference on Image Processing, vol.1, pp. I-900-I-903 2002.
[17]	G. Zhao, S. Wang, T. Wang, and J. Chen, “HSV Color Space and Face Detection Based Objectionable Image Detecting,” International Conference on Future Generation Communication and Networking Symposia, pp. 107-110, 2008.
[18]	http://www.infocellar.com/Graphics/color-theory.htm
[19]	https://zh.wikipedia.org/wiki/HSL%E5%92%8CHSV%E8%89%B2%E5%BD%A9%E7%A9%BA%E9%97%B4
[20]	Y. R. Horng, Y. C. Tseng, and T. S. Chang, “Stereoscopic images generation with directional Gaussian filter,” IEEE International Symposium on Circuits and Systems, pp. 2650-2653, 2010.
[21]	L. Zhang, and W. J. Tam, “Stereoscopic image generation based on depth images for 3D TV,” IEEE Transactions on Broadcasting, vol. 51, pp. 191-199, 2005.
[22]	RealTouchApp Corp. Ltd. Factory Warehouse.
[23]	http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives/