為了提升工廠巡檢效率，智慧化指針式儀錶辨識是重要的研究議題之一，因此，本論文提出一套儀錶影像數值辨識技術。本論文提出一套基於視覺的指針式儀錶數值識別系統。提出的方法可分為三個部分：ROI偵測、數值辨識和時間分析。為了減少背景的影響並提高效率，本論文提出先透過深度學習為基礎，用以偵測來源影像中的儀錶影像中的ROI(regions of interest）區域。在ROI偵測之後，本論文提出一種基於參考影像的數值識別算法。為了校正輸入影像和參考影像，透過ORB和RANSAC算法找到正確的對應點後，進行轉換參數估算。透過幾何空間變換後，透過Hough Transform定位指針位置。透過指針位置，計算指針的角度與儀錶數值之間的比例關係，並將其角度轉換為儀錶數值結果。在時間分析中，本論文透過參考連續性儀錶影像的數值，獲得最終的結果為了評估所提出的方法，本論文收集數組連續性儀錶影像用於測試。實驗結果表明，本論文不僅可以檢測ROI，也可以有效地辨識指針式儀錶影像的數值。

In order to improve the efficiency of factory inspection, intelligent numerical value reading from pointer gauge images is one important issue. In this thesis, we proposed a vision-based numerical value reading scheme for pointer gauge images. The proposed scheme is composed of three parts: ROI detection, numerical value recognition, and temporal analysis. To reduce the impact of background and increase the efficiency of the proposed scheme, the regions of interest (ROIs) are first detected from the captured pointer gauge images based on deep learning. After ROI detection, a numerical value recognition algorithm was proposed. The numerical value recognition algorithm is a reference-based approach. To align the input image and the reference image, the ORB and RANSAC algorithms are used to find correct corresponding point pairs and then estimate the transform parameters. After geometrical transform, Hough transform is used to find the pointer position. After analyzing the orientation of the pointer, the corresponding numerical value can be obtained. In temporal analysis, we consider the numerical values measured from several images to obtain the final result.To evaluate the proposed scheme, some image sequences are captured for testing. The experimental results show that the proposed scheme cannot only detect ROIs but also recognize numerical values from pointer gauge images well.

中文摘要 iii
ABSTARCT iv
誌謝 v
目錄 vi
圖目錄 viii
表目錄 x

第一章 緒論 1
1.1研究動機與目的 1
第二章 文獻回顧 3
2.1 深度學習 3
2.2 物件偵測 7
2.2.1 R-CNN 7
2.2.2 Fast R-CNN 8
2.2.3 Faster R-CNN 10
2.2.4 SSD 10
2.3 儀錶數值辨識 12
2.4 霍夫變換 13
2.5 ORB 14
第三章 研究方法 17
3.1ROI偵測 18
3.1.1訓練階段 19
3.1.2測試階段 20
3.1.3數值辨識 21
3.1.4直線檢測 23
3.1.5特徵點匹配 24
3.1.6仿射變換 26
3.1.7數值估算 27
3.2時間分析 30
第四章 實驗結果與分析 32
4.1 實驗環境 32
4.2 評估因子 34
4.2.1 ROI分類評估因子 34
4.2.2 ROI回歸評估因子 35
4.2.3 數值檢測 36
4.3 效能評估 36
4.3.1 ROI偵測分類評估 37
4.3.2 ROI偵測回歸評估 47
4.3.3數值辨識評估 48
4.4 效能比較 52
第五章 結論 55
參考文獻 56

[1] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” arXiv:1311.2524, 2014.
[2] R. Girshick. “Fast R-CNN.” in Proc. IEEE International Conference on Computer Vision, pp. 1440-1448, 2015.
[3] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards realtime object detection with region proposal networks,” in NIPS, pp.91-99, 2015.
[4]F. C. Alegria and A. C. Serra, “Computer vision applied to the automatic calibration of measuring instruments,” in Measurement, vol. 28, Iss. 3, pp. 185-195, October 2000.
[5]W. Qi, T. Xiling, D. Cheng, H. Yao, and F. Yanjun, “Automatic alignment system based on center point recognition of analog measuring instruments dial,” in Proceedings of the 39th Annual Conference of the IEEE Industrial Electronics Society (IECON 2013), pp. 5532–5536, November 2013.
[6]P.A. Belan, S.A. Araujo and A.F.H. Librantz, “Segmentation-free approaches of computer vision for automatic calibration of digital and analog instruments,” in Measurement, vol. 46, Iss. 1, pp. 177-184, January 2013.
[7]Jiqiang Song and Michael R. Lyu. “A Hough transform based line recognition method utilizing both parameter space and image space,” in Pattern Recognition, vol. 38, Iss. 4, pp.539-552, April 2005.
[8]J. Matas, C. Galambos and J. Kittler, “Robust Detection of Lines Using the Progressive Probabilistic Hough Transform,” in Computer Vision and Image Understanding, vol. 78, Iss. 1, pp. 119-137, April 2000.
[9]E. Rublee, V. Rabaud, K. Konolige and G. Bradski, “ORB: An efficient alternative to SIFT or SURF,” in Proc. IEEE Int. Conf. Computer Vision, pp. 2564-2571, November 2011.
[10] E. Rosten and T. Drummond, “Fusing points and lines for high performance tracking,” in IEEE International Conference on Computer Vision (ICCV 2005), vol. 2, pp. 1508–1511, October 2005.
[11]M. Calonder, V. Lepetit, C. Strecha, and P. Fua. “BRIEF: Binary Robust Independent Elementary Features,” in Proceedings of the European Conference on Computer Vision (ECCV), 2010.
[12] Jiqiang Song and Michael R. Lyu. “A Hough transform based line recognition method utilizing both parameter space and image space,” in Pattern Recognition, vol. 38, Iss. 4, pp.539-552, April 2005.
[13] J. Matas, C. Galambos and J. Kittler, “Robust Detection of Lines Using the Progressive Probabilistic Hough Transform,” in Computer Vision and Image Understanding, vol. 78, Iss. 1, pp. 119-137, April 2000.
[14] Analog Gauge Reader。取自：https://www.codeproject.com/Articles/1221181/Analog-Gauge-Reader，2019年6月。
[15] Wei Liu , Dragomir Anguelov , Dumitru Erhan , Christian Szegedy , Scott Reed , Cheng-Yang Fu , Alexander C. Berg, SSD: Single Shot MultiBox Detector, ECCV Computer Vision and Pattern Recognition, 2016.