臨床上要檢測一個病人是否有扁平足，通常醫師會站在病人後方以肉眼觀察其走路的狀態並給出診斷，但這個診斷通常只會有像是正常、輕度、或是中度之類的評語，不會有一個很精準的量化。要量化扁平足有很多種指標，在此我們選擇一種臨床上常用的指標－後足角度(Rear Foot Angle)，其定義為小腿中線與後腳跟中線的夾角。
本研究係利用SiamRPN++追蹤演算法追蹤貼在病人腳上多個標記點的影像座標，並使用自組Stereo相機系統搭配PSMNet演算法來取得取得其3D資訊，模擬常見動作捕捉系統，藉此來量化並計算每一偵中左腳與右腳之後足角度。除了原相機影像平面所量測之後足角度之外，為了解決固定位置相機可能產生之視角誤差，本研究還提出一種基於虛擬相機跟拍視角之評估方式，將原於相機座標的標記點之3D座標投影至此虛擬相機影像平面後計算後足角度之量測方法，使用此方法量測RFA在步態於Swing Phase時能有較穩定的結果。

To detect whether a patient has a pronated foot, the doctor will usually stand behind the patient and directly observe their walking state and then give a diagnosis. However, this diagnosis usually only has comments such as normal, mild, or moderate. There will not be a very precise quantification. There are many indicators to quantify the flat feet level. Here we choose a commonly used indicator in a clinical practice called Rear Foot Angle, RFA, which is defined as the angle between the midline of the lower leg and the midline of the heel.
This research uses the SiamRPN++ tracking algorithm to track multiple marker dots attached to the patient’s foot in the image and uses the self-assembled stereo cameras with the PSMNet algorithm to obtain its 3D position to simulate common motion capture systems. Then estimating the RFA of the left and right leg in each frame to do the quantification. In addition to the RFA measured in the original camera image plane, in order to solve the viewing angle error caused by using a fixed-position camera, this research also proposes a virtual camera follow-up based estimation by perspective projecting 3D dots position from camera coordinates to the virtual camera one, then estimating RFA in the virtual image plane. Using this method to measure RFA can have a more stable result in the swing phase.

[1] Adrienne E Hunt, Alexander J Fahey and Richard M Smith, "Static measures of calcaneal deviation and arch angle as predictors of rearfoot motion during walking", Australian Journal of Physiotherapy (2000).
[2] Thomas G. McPoil and Mark W. Cornwall, "Relationship Between Three Static Angles of the Rearfoot and the Pattern of Rearfoot Motion During Walking", Journal of Orthopaedic & Sports Physical Therapy (1996).
[3] Z. Zhang, "A Flexible New Technique for Camera Calibration", IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol.22, no.11, pp.1330-1334, 2000.
[4] Jia-Ren Chang and Yong-Sheng Chen, "Pyramid stereo matching network", In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.
[5] Rıza Alp Guler, Natalia Neverova and Iasonas Kokkinos, "DensePose: Dense Human Pose Estimation In The Wild", arXiv:1802.00434. (2018)
[6] Bo Li, Junjie Yan, Wei Wu, Zheng Zhu, Xiaolin Hu, "High Performance Visual Tracking with Siamese Region Proposal Network", CVPR2018.
[7] Yihong Xu, Aljosa Osep, Yutong Ban, Radu Horaud, Laura Leal-Taixe and Xavier Alameda-Pineda, "How To Train Your Deep Multi-Object Tracker", arXiv:1906.06618
[8] Wenhan Luo, Junliang Xing, Anton Milan, Xiaoqin Zhang, Wei Liu, Xiaowei Zhao and Tae-Kyun Kim, "Multiple Object Tracking: A Literature Review", CVPR2017.
[9] Luka Cehovin Member, Ales Leonardis and Matej Kristan, "Visual object tracking performance measures revisited", IEEE Transactions on Image Processing (2016).
[10] L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy and A. L. Yuille, "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs", arXiv:1606.00915, 2016.
[11] Pierre Merriaux, Yohan Dupuis, Rémi Boutteau, Pascal Vasseur and Xavier Savatier, "A Study of Vicon System Positioning Performance", MDPI 2017.
[12] R.Girshick, "FastR-CNN", In Proceedings of the IEEE international conference on computer vision, pages 1440–1448, 2015. 5.
[13] Bo Li, Wei Wu, Qiang Wang, Fangyi, Junliang, Junjie "SiamRPN++: Evolution of Siamese Visual Tracking with Very Deep Networks", arXiv:1812.11703.
[14] Qiang Wang, Li Zhang, Luca Bertinetto, Weiming Hu, Philip H.S. Torr "Fast Online Object Tracking and Segmentation: A Unifying Approach", arXiv:1812.05050v2.
[15] Luca Bertinetto, Jack Valmadre, Jo˜ao F. Henriques, Andrea Vedaldi, Philip H. S. Torr, "Fully-Convolutional Siamese Networks for Object Tracking", arXiv:1606.09549v2.
[16] Chao Ma, Jia-Bin Huang, Xiaokang Yang, Ming-Hsuan Yang, "Hierarchical Convolutional Features for Visual Tracking", ICCV 2015.
[17] Hyeonseob Nam, Bohyung Han, "Learning Multi-Domain Convolutional Neural Networks for Visual Tracking", CVPR 2016.
[18] J. Henriques, R. Caseiro, P. Martins and J. Batista, "Highspeed tracking with kernelized correlation filters", TPAMI, 2015.
[19] Ben Langley, Mary Cramp and Stewart C. Morrison, "Selected static foot assessments do not predict medial longitudinal arch motion during running", Journal of Foot and Ankle Research 2015.
[20] Vivienne H Chuter, "Relationships between foot type and dynamic rearfoot frontal plane motion", Journal of Foot and Ankle Research 2010.
[21] T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollar and C. L. Zitnick, "Microsoft coco: Common objects in context", In ECCV, pages 740–755. Springer, 2014.
[22] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg and L. Fei-Fei, "ImageNet Large Scale Visual Recognition Challenge", IJCV, 2015.
[23] "LearnOpenGL-Camera", https://learnopengl.com/Getting-started/Camera.
[24] Graham Sellers, Richard S Wright Jr., Nicholas Haemel, "OpenGL SuperBible: Comprehensive Tutorial and Reference", Addison Wesley
[25] Kaiming He, Xiangyu Zhang, Shaoqing Ren and Jian Sun, "Deep Residual Learning for Image Recognition", arXiv:1512.03385v1.