人體步態辨識之研究

A Study on Gait-Based Human Recognition

林信安(Shin-An Lin)

指導教授 ： 吳明霓

共同指導教授 ： 吳憲珠(Hsien-Chu Wu)

國立臺中科技大學/資訊與流通學院/資訊工程系碩士班/碩士(2012年)

摘要

影像辨識蓬勃發展，人體步態資訊成為必然的趨勢，本論文提出兩個步態辨識之研究方向。其一是步態姿勢辨識，在社會醫療技術發達使得人口老化成為必然的趨勢，智慧型居家照護已是一個重要的科技發展方向。本研究運用步態資訊提供老人姿態判別，尤其是獨居老人的居家安全管理問題更形重要。其二是步態性別辨識，性別對於資訊的適性化服務提供一個重要的線索，它能提高資訊服務的加值效果，故成為一項重要研究議題。
在步態姿勢辨識之研究中提出一個低運算複雜度的演算法，利用三角形態法則，能快速的偵測人形的異常動作，家中裝設一個機器人，利用機器人攝影機拍攝行走步態，就能即時判斷居家老人的異常動作。實驗結果顯示準確率達90%。若再結合現時普遍使用的手機3G系統，則在發生異常動作時，即時傳送簡訊及動作影像至家人的手機中或啟動通報系統，便能達到即時照護的效能。
性別辨識之研究是以人們行走的步態做為性別分類的依據。性別辨識可運用於百貨公司與商店門口，以定點式攝影鏡頭偵測性別，可以提供賣場商品明細，減少顧客搜尋產品的時間。研究使用CASIA gait datasetB資料庫作為資料來源 ，前置處理要將影像轉為二值化影像，首先將影像轉成灰階，給予一個門檻值，大於門檻值為前景部分，小於門檻值為背景部分，這樣就可擷取前景二值化影像，得到二值化影像後再將影像正規化到一樣大小，再經由GEI(gait energy image)與DEI(denoised energy image)方法做為影像前置處理，並據此擷取特徵。將這些特徵經由SVM(support vector machine)分類器進行訓練與測試。實驗結果顯示，行進人員在固定角度下行走，可達100%之性別辨識率，在容許的角度誤差範圍內行進，亦可達到8成以上的性別辨識率。

關鍵字

物特徵識別 ； Gait Energy Image(GEI) ； Denoised。 Energy Image(DEI) ； Support Vector Machine(SVM)

並列摘要

The image recognition vigorous development of human gait information is one of the most inevitable trends. This paper proposes two gait recognition research topics. First, the aging recognition benefits the development of social and medical technology development. Therefore, smart home care becomes a focus of attention. This research provides the gesture recognition by using the gait information of the elderly living alone, it’s important of the management of their home safety. Second, the gender recognition provides some important clues for the appropriate service information. The gender recognition can improve the effect of value-added information services, and becomes an important research topic.
A low computational complexity of algorithms has been proposed in this research of home gait recognition cares. Using the rule of triangular shape, the abnormal movements of human body can be detected quickly. The abnormal movement of the elderly living alone can be detected and recognized immediately by a video camera. The accuracy of the successful detection and recognition is 90% in the experiment.
In the aging recognition research, we present a preliminary design and experimental results of gender recognition from walking movements that utilizes gait energy image (GEI) with denoised energy image (DEI) pre-processing as support vector machine (SVM) classifier to training and extract the characteristics. The result shows that the proposed method would adopt the few characteristic value but the accuracy can reach to 100% on the same shot angles and more than 80% on the tolerance of 18 degree shot angles.

並列關鍵字

Biometric ； Gait Energy Image(GEI) ； Denoised Energy Image(DEI) ； Support Vector Machine(SVM)。

參考文獻

[3]C. Bauckhage , J.K. Tsotsos and F.E. Bunn , Automatic detection of abnormal gait, Image and Vision Computing archie , Vol. 27 , pp. 108-115 , January 2009.

[4]A .F. Bobick , A .Y. Johnson , Gait Recognition Using Static , Activity-Specific Parameters, Computer Vision and Pattern Recognition , Vol. 1 , pp. 423-430, 2001.

[5]G. Diraco , A. Leone and P. Siciliano , An Active Vision System for Fall Detection and Posture Recognition in Elderly Healthcare, Design, Automation & Test in Europe Conference & Exhibition , pp. 1536-1541, March 2010.

[7]F. Jean ,R. Bergevin and A.B. Albu , Body Tracking in Human Walk from Monocular Video Sequnces, Computer and Robot Vision , pp. 144 – 151 , May 2005.

[9]A. Jain and J. Huang, Integrating independent components and linear discriminant analysis for gender classification,in Proc. 6th IEEE Int. Conf. Automatic Face and Gesture Recognition, 2004, pp. 159–163.