在電腦視覺不斷進步的今日，基於二維影像的人體骨架偵測技術日漸成熟，因此，越來越多的基於人體骨架偵測的應用陸續被開發出來。然而，當輸入影像中的人體被大面積遮擋或是遮蔽物件與人體顏色類似時，皆會對於人體骨架估測結果造成重大的影響。因此本論文希望提出一個基於生成對抗網路 （Generative Adversarial Network） 的演算法，來降低上述的兩大干擾因素，能自動生成人體被遮蔽影響的區塊，使得二維影像的骨架偵測效果能夠被大幅改善。
本論文以居家環境為主要應用情境，在此應用情境中，我們關心的日常生活中常見的動作姿態共有八種，以此為後續分析的目標。由於居家環境中，身體常常容易被各類家具所遮蔽，導致人體骨架估測結果變差。所以，本論文訓練一個生成對抗網路，使得生成對抗網路可以自動生成擬真的圖像，補全原先被遮蔽而可能造成誤判的區塊。藉此進一步改善骨架偵測演算法的準確性。
在不同人的推廣性測試與不同背景下的測試上，本論文提出方法相較於原先直接使用骨架偵測演算法，改善了八成的誤判，證明本系統在遮擋情況下，能有效地提供穩定的填補圖像，改善二維圖像的骨架偵測效果。

Nowadays, with the continuous advancement of computer vision, human body skeleton detection technology based on two-dimensional images is becoming more and more mature. Therefore, more and more applications based on human skeleton detection have been developed. However, when the human body in the input image is blocked by a large object or the object’s color is similar to the human body, it will result in a significant impact on the estimation of the human skeleton. Therefore, this thesis tries to propose an algorithm based on the Generative Adversarial Network to reduce the above two major interference factors. The proposed algorithm can automatically generate the corresponding blocks that are blocked, so that the 2-D skeleton detection effect can be greatly improved.
This thesis takes the home environment as the main application scenario. In this application scenario, there are total of eight common postures in daily life that we care about and these eight postures will be the goal of subsequent analysis. Because of the home environment, the body is often easily occluded by various types of furniture, resulting in poor estimation of the human skeleton. Therefore, this thesis tries to train a generative adversarial network, so that the network can automatically generate the corresponding body image to complement the area that was originally blocked by a furniture. Via this kind of amendment, the accuracy of the skeleton detection algorithm can be further improved.
Based on the generalization performance comparisons of different people and different backgrounds, the proposed method improves the 80% misjudgment compared with the original skeleton detection algorithm. These simulation results demonstrate that the proposed algorithm can effectively solve the occlusion problem and provide a stable recovery image so as to improve the performance of the original 2-D skeleton detection algorithm.

[1] I. Goodfellow et al., "Generative adversarial nets," in Advances in neural information processing systems, pp. 2672-2680, 2014.
[2] Z. Cao, G. Hidalgo, T. Simon, S.-E. Wei, and Y. Sheikh, "OpenPose: realtime multi-person 2D pose estimation using Part Affinity Fields," arXiv preprint arXiv:1812.08008, 2018.
[3] Kinect, Wikipedia. [Online]. Available: https://en.wikipedia.org/wiki/Kinect [Accessed: 20-Jun-2019]
[4] H. T. Kam, "Random Decision Forest," in Proceedings of 3rd International Conference on Document Analysis and Recognition, vol. 17, no. 8, pp. 790-799, 1995.
[5] Y. Cheng, "Mean shift, mode seeking, and clustering," IEEE transactions on pattern analysis and machine intelligence, vol. 17, no. 8, pp. 790-799, 1995.
[6] Z. Cao, T. Simon, S.-E. Wei, and Y. Sheikh, "Realtime multi-person 2d pose estimation using part affinity fields," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 7291-7299, 2017.
[7] M. Andriluka, L. Pishchulin, P. Gehler, and B. Schiele, "2d human pose estimation: New benchmark and state of the art analysis," in Proceedings of the IEEE Conference on computer Vision and Pattern Recognition, pp. 3686-3693, 2014.
[8] Wikipedia perceptron. [Online]. Available: https://zh.wikipedia.org/wiki/感知機 [Accessed: 21-Jun-2019]
[9] 蘇木春、張孝德, 機器學習：類神經網路、模糊系統以及基因演算法則，第二版. 全華科技圖書, 民國一百零一年.
[10] "IBM Deep learning architectures." [Online]. Available: https://www.ibm.com/developerworks/library/cc-machine-learning-deep-learning-architectures/index.html.
[11] "Wikipedia Convolution Neural Network." [Online]. Available: https://en.wikipedia.org/wiki/Convolutional_neural_network.
[12] "github generative adversarial networks." [Online]. Available: https://github.com/jonbruner/generative-adversarial-networks/blob/master/gan-notebook.ipynb.
[13] "GAN Introduction." [Online]. Available: https://ithelp.ithome.com.tw/articles/10196828.
[14] Y. Yu, Z. Gong, P. Zhong, and J. Shan, "Unsupervised Representation Learning with Deep Convolutional Neural Network for Remote Sensing Images," in International Conference on Image and Graphics: Springer, pp. 97-108, 2017.
[15] P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Image-to-image translation with conditional adversarial networks," in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1125-1134, 2017.
[16] J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, "Unpaired image-to-image translation using cycle-consistent adversarial networks," in Proceedings of the IEEE international conference on computer vision, pp. 2223-2232, 2017.
[17] J. Yu, Z. Lin, J. Yang, X. Shen, X. Lu, and T. S. Huang, "Generative image inpainting with contextual attention," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5505-5514, 2018.
[18] P. KaewTraKulPong and R. Bowden, "An improved adaptive background mixture model for real-time tracking with shadow detection," in Video-based surveillance systems: Springer, pp. 135-144, 2002.