智慧化手勢辨識與影音播放介面之發展

The Development of Media Player Based on Intelligent Gesture Recognition

鄭至絎(Zhi-Hang Cheng)

指導教授 ： 陳冠宇

中原大學/工學院/機械工程研究所/碩士(2019年)

https://doi.org/10.6840/cycu201900563

摘要

有別於傳統影像演算法，深度學習演算法能做到複雜手勢的識別‧傳統影像處理，計算上有太多限制，即使能做到手指數量的計算、手部輪廓的跟蹤，但遇上複雜度較高的手勢、具有爭議的手勢，傳統演算法的表現將無法滿足使用者的需求。因此，本研究希望結合傳統演算法與深度學習，做出複雜性更高的影像識別。並將影像識別的技術，延伸推廣至更實際的應用層面。
本文的研究基於PC-Based的架構及Python編譯，發展一套手勢互動影音播放介面。實驗流程大致二個重點：影像前處理與深度學習模型建立。從攝影機擷取的即時影像為最原始的RGB影像，經過適當影像前處理，使手勢資料更純粹，並蒐集大量經過處理的手勢影像，加入深度學習模型的訓練資料庫。以大量樣本為基底的深度學習模型，做出各種自定義手勢的識別。最後將識別結果轉換為控制指令，取代以往滑鼠點擊的方式，透過不同的手勢，控制音樂播放介面。
最後本文將討論訓練模型的參數、辨識結果、準確度及不足之處的改善與未來展望。

關鍵字

深度學習 ； 影像處理 ； 機器視覺 ； 互動使用者介面 ； Python

並列摘要

Different from traditional image algorithms, deep learning algorithms can recognize complex gestures; traditional image processing has too many calculations, even if it can calculate the number of fingers and track the contours of hands, it is complicated. With higher gestures and controversial gestures, the performance of traditional algorithms will not meet the needs of users. Therefore, this study hopes to combine traditional algorithms and deep learning to make more complex image recognition. Moreover, extend the technology of image recognition to a more practical application level.
The research in this thesis is based on the PC-Based architecture, Python compilation, and develops a set of gesture interactive video playback interface. The experimental process is roughly focused on two aspects: image pre-processing and deep learning models. The real-time image captured from the camera is the original RGB image. After proper image pre-processing, the gesture data is more pure, and a large number of processed gesture images are collected, and the training database of the deep learning model is added. A deep learning model based on a large number of samples to identify various custom gestures. Finally, the recognition result is converted into a control command, which replaces the way the mouse clicks, and controls the music playing interface through different gestures.
At the end of the thesis, the parameters of the training model, the identification results, the accuracy and the improvement of the deficiencies and future prospects will be discussed.

並列關鍵字

deep learning ； image processing ； machine vision ； interactive user interface ； Python.

參考文獻

[1]S. Mitra and T. Acharya, “Gesture recognition: A survey,” IEEE Transactions on Systems, Man, and Cybernetics, Part C, Vol. 37, No. 3, pp. 311-324, 2007

[2]R. C. Gonzalez and R. E. Woods, Digital image processing. Prentice Hall: Upper Saddle River, New Jersey, USA, 1992.

[3]C. Saravanan, “Color image to grayscale image conversion,” The 2nd International Conference on Computer Engineering and Applications, Bali Island, Indonesia, Mar. 26-29, 2010.

[4]L. J. van Vliet, I. T. Young, and P. W. Verbeek, “Recursive Gaussian derivative filters,” Proceedings of the 14th International Conference on Pattern Recognition, Brisbane, Queensland, Australia, Aug. 20-20, 1998, Vol. 1, pp. 509-514.

[5]羅華強, 類神經網路, 清蔚科技股份有限公司: 新竹, 2001.