深度學習應用於動作辨識中的key pose分類

Key Pose Classification for Gesture Recognition using Deep Neural Networks

李文富(Wen-Fu Lee)

指導教授 ： 蔡志忠

國立中正大學/工學院/資訊工程學系/碩士(2015年)

摘要

深度神經網路近年來是一個熱門的方法在機器學習領域。他能夠從原始資料學習特徵且不需要太多預備知識，以及些許的前處理。
學者們對深度神經網路的興趣不只來自於上述優點，還有它們在機器學習應用上良好的結果，像是在動作辨識、物件辨識、語音辨識等。

手勢辨識基於key pose，是其中一種有效辨識手勢的方法，一個手勢被視為由一連串的pose所組成，辨識出正確的手勢非常依賴於辨識key poses的準確性，
因此我們聚焦在key pose的分類、辨識上。本篇論文分別使用兩種深度神經網路: 多層感知機與卷積神經網路，
在之前的研究，使用機器學習的方法像支持向量機來分類、辨識key pose。我們將我們的兩種方法實驗在14個key poses上，
實驗結果顯示我們的兩種方法比先前的相關論文還要好。

並列摘要

Deep neural network is a popular method in machine learning field in recent year. It can learn features from raw data without using any prior knowledge and little pre-processing. Reason why researchers are interested in deep neural network which is not only we mentioned above but also its successful results in machine learning tasks, such as action recognition, object recognition and speech recognition, etc.

Gesture recognition based on key pose is one of the most effective methods to recognize a gesture, which is considered as a sequence of poses. Precise gesture recognition depends heavily on the accuracy of key poses recognition. Hence, we focus on the problem of key poses classification. In this thesis, we use two deep neural networks methods respectively: Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN). In previous research, machine learning techniques such as SVM are used to classify key poses. We evaluated our two methods on total fourteen key poses and the experimental results show that both MLP and CNN models give comparable result with related work.

並列關鍵字

Convolutional neural network ； Kinect ； deep neural network ； Key pose classification ； Gesture recognition

參考文獻

Kinect Applications” National Chung Cheng University,2014

[14] Hsien-I Lin†, Ming-Hsiang Hsu, and Wei-Kai Chen, ” Human Hand Gesture

Vieira, Mario F. M. Campos, “Real-time gesture recognition from depth data

[2] Shih-Hai Lin, ” Real-time Gesture Recognition using Key Pose Classification for

Cairo, Egypt.