效能導向與硬體導向之時頻分析步態凍結偵測演算法

Performance-Oriented and Hardware-Oriented Freezing of Gait Detection Algorithms Using Time-Frequency Analysis

胡豪(Hao Hu)

指導教授 ： 丁建均

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2014年)

https://doi.org/10.6342/NTU.2014.00771

摘要

帕金森氏症患者在日常生活中對行走會造成許多不便，稱為凝態步伐。這會使他們深受其影響，影響到他們的生活。
我們提出了兩種可以偵測凝態步伐的演算法。第一種演算法是效能導向的凝態步伐演算法。我們使用時頻分析還有在時預方向上的訊號來偵測凝態步伐。我們的模擬結果準確率可以達到82.83%。因此，我們的演算法可以準確地找出病人們在凝態步伐的時間。很重要的一點是，演算法的延遲時間總共只有0.95秒。跟其他現有的方法比較，是非常少的，且我們方法的效能也比現有方法來的好。
第二部分是硬體導向的凝態步伐演算法，我們使用整數轉換和使用IIR濾波器來取代平滑化濾波器來降低複雜度。我們使用遞迴方程式來取代原本的傅立葉轉換以降低複雜度，還有，我們將原本所使用的非對稱窗格使用遞迴式的非對稱窗格來取代，也就是使用0階、一階、以及四階的遞迴多項式。階數越高、效能越好，但同時複雜度也會升高。使用了以上的調整，快速演算法的敏感度能達到77.47%、特異度能達77.61%、準確度能達到78.02%，平均每一點輸出的運算量只需要14個乘法。運用效能導向的演算法平均每一點輸出的運算量需要19.16個乘法。因此，與效能導向的演算法相比，因為較低的複雜度以及其具有競爭力的效能，硬體導向的演算法更加適合在硬體上實現。

關鍵字

時頻分析 ； 整數轉換 ； 起始偵測 ； 帕金森氏症 ； 凝態步伐

並列摘要

Parkinson’s disease (PD) patients have difficulty in walking. They usually suffer from the Freezing of gait (FOG) problem, which interferes PD patients’ life.
We develop two algorithms which can detect FOG of PD patients. The first algorithm is performance-oriented. In this part, we develop an algorithm which uses fast time-frequency analysis methods and onset detection to detect FOG in real time. Simulation results show that the specificity can achieve 81.83%, and the sensitivity and accuracy are 82.66% and 82.83%, respectively. The sensitivity and the accuracy are higher than those of other algorithms. Our algorithm can help PD patients overcome the difficulty of walking and help them live better if the algorithm can be realized on a healthcare devise. More importantly, the time delay of the proposed algorithm is only 0.95 second, which is less than that of other algorithms.
The second algorithm is hardware-oriented. In this part, we use the integer transform and replace the smooth filter by the IIR filter to reduce complexity. We use a recursive formula of Fourier Transform to implement the output instead of Fourier transform in order to decrease the complexity. Moreover, the original asymmetric windows are replaced by recursive asymmetric rectangular window, triangular window, and polynomial window, which is zero order, first order, and forth order polynomial formula of recursive form, respectively. The more the order is, the better the performance will be, meanwhile, the complexity will be higher. The fast method with forth order polynomial form with asymmetric polynomial window is able to detect the FOG with sensitivity of 77.47%, specificity of 77.61%, and accuracy of 78.02%. The average amount of multiples of each point of the proposed hardware-oriented FOG algorithm with rectangular asymmetric window needs 11. On the other hand, the average number of multiples of each point for the original performance-oriented method is 19.16. Therefore, compare to the performance-oriented FOG detection algorithm, the hardware-oriented FOG detection algorithm is even more suitable for implementing on a hardware-device because of its lower complexity and competitive performance.

並列關鍵字

Time-frequency analysis ； integer transform ； onset detection ； Parkinson’s disease ； freezing of gait

參考文獻

[4] M. J. Bastiaans, “Gabor’s expansion of a signal into Gaussian elementary signals,” Proc. IEEE, vol. 68, pp. 594-598, 1980.

[5] S. C. Pei and J. J. Ding, “Relations between Gabor transforms and fractional Fourier transforms and their applications for signal processing,” IEEE Trans. Signal Processing, vol. 55, no. 10, pp. 4839-4850, Oct. 2007.

[6] E. P. Wigner, “On the quantum correlation for thermodynamic equilibrium,” Phys. Rev., vol. 40, pp. 749-759, 1932.

[8] F. Hlawatsch and G. F. Boudreaux-Bartels, “Linear and quadratic time-frequency signal representation,” IEEE Signal Processing Magazine, pp. 21-67, Apr. 1992.

[9] L. J. Stankovic, S. Stankovic, and E. Fakultet, “An analysis of instantaneous frequency representation using time frequency distributions-generalized Wigner distribution,” IEEE Trans. on Signal Processing, pp. 549-552, vol. 43, no. 2, Feb. 1995.