Please use this identifier to cite or link to this item:
Player-Independent Tennis-Stroke Recognition Based on the Wavelet Transform
|Contributors: ||NTOU:Department of Electrical Engineering|
gesture recognition;wavelet transform;artificial neural network
|Issue Date: ||2011-06-22T09:02:23Z
|Abstract: ||本論文的研究目標是利用小波轉換與類神經網路來建立一個可以讓 多使用者操作的網球揮拍動作辨識系統。相較於文獻中所提出的網球 揮拍動作辨識是針對單一使用者, 本多使用者系統在應用領域上更為寬 廣。本系統是以加速度訊號與旋轉矩陣訊號作為辨識之訊號源, 避免掉 以視訊訊號作為辨識訊號源時需處理大量冗餘資料的複雜運算。我們所 提出的方法是將接收到的訊號以小波轉換分解後的近似系數作為訊號的 特徵值, 然後以三軸個別的特徵值合併再交由類神經網路來訓練並更新 網路權重值, 以得到各種網球揮拍動作的類神經網路模型。在辨識的階 段則是將不同於訓練組的網球揮拍動作特徵值輸入訓練過的類神經網路 模型得到辨識結果。所提出的多使用者網球揮拍動作辨識系統可辨識13 種網球基本揮拍動作, 並達到最高100%的辨識率。本論文的研究成果可 以提高人機介面的性能和改善電玩遊戲虛擬實境的效果|
The research goal of this thesis is to build a multiuser tennis-stroke recognition system using wavelet transform and artificial neural network. Compared to the previous work on single-user tennis-stroke recognition that can be found in the literature , the proposed multiuser system has broader areas of application. This system utilizes the acceleration and rotation matrix signals instead of the video signal generated by the tennis-stroke motions as the signal source for recognition. This avoids the necessity of processing massive redundant data and complex operations that one may encounter if the signal source were the video signal. The proposed method uses the wavelet transform to decompose the acceleration signals collected from the three axes, and combines the resulting wavelet approximation coefficients obtained from the three axes to form the feature vector. The feature vector is then use by the artificial neural network as the input for training and updating its weights. By doing so, the artificial neural network models for all the tennis strokes can be obtained. In the recognition phase, tennis-stroke signals that have not been used for training the artificial neural network models are fed to the models to test the recognition performance. It turns out that the proposed multiuser tennis-stroke recognition system can recognize 13 basic tennis strokes and achieve a recognition rate of 100%. The result of this paper can be used to improve the performance of man-machine interfaces in virtual-reality video games.
|Appears in Collections:|
Files in This Item:
There are no files associated with this item.
All items in NTOUR are protected by copyright, with all rights reserved.