透過您的圖書館登入
IP:3.235.46.191
  • 學位論文

利用Kinect V2在監控環境下量測人體肩膀高度

Human Shoulder Height Measurement based on Kinect V2 Skeleton under Surveillance Conditions

指導教授 : 陳永耀

摘要


近年來,監控式攝影機在全球已經越來越普及,致力於將監控系統自動化以辨識景中人物的研究愈加受重視。然而,一般傳統式生物特徵,利如臉、指紋或是虹膜等,都會需要使用者的配合才能取得足夠的資料進行分析;若要將其應用在監控環境中,考量到人物可能會朝向任何方向,所處的距離範圍也可能差異非常大,取得之資料品質會受到嚴重影響。為了解決這一問題,一種稱為軟式特徵的生物特徵正開始發展,例如髮色、種族或是身高等。相較於傳統生物特徵,軟式特徵較不會受距離遠近影響,而且也無需使用者特別的合作就能取得,是一種特別適合在監控環境中使用的生物特徵。而在眾多軟式特徵中,身高是一個較具分辨力且容易被看見的特徵,所以,本論文提出一套基於Kinect V2所提供之人體骨架計算肩膀高度的方法。在人體骨架資訊輸入後,我們會先過濾掉不正確的骨架,以確保後續計算的準確性;在得到初步估算的肩膀高度後,一個經由統計而得的補償方程式可將估算之肩高修正至我們定義之真實肩膀高度。最後,實驗結果顯示出,靜態下的量測平均誤差是16mm,在動態下的量測平均誤差和標準差分別為16mm和19mm。另外,在多人環境下測試的結果指出,即使是在遮蔽頻繁的狀況下,我們的方法仍舊能保持單人環境時的表現。

並列摘要


Interest in the security of individuals has increased in recent years. This trend led to much wider deployment of surveillance cameras both indoor and outdoor. Consequently, more approaches are focusing on improving the mediocre performance of classical biometrics, such as face or iris, under uncontrolled conditions such as illumination and facing directions. To address the problems, soft biometrics, including hair color, ethnicity or height, were proposed. These type of biometrics can be obtained at a distance without subject cooperation, making them ideal for surveillance applications. Among many soft biometric traits, the height trait is one of the most visible and distinctive traits in unconstrained environment. Therefore, an approach to measure human shoulder height based on Kinect V2 skeleton is proposed to assist in human identification. By analyzing the data provided by Kinect V2 skeleton, the preliminary shoulder height can be estimated. Then, a compensation is applied to make the estimated shoulder height match with human true shoulder height. The results show that the accuracy error in static pose is 16mm and the mean absolute error and fluctuation deviation in dynamic pose are 16mm and 19mm, respectively. In addition, the test results in multi-person scenarios show that the proposed approach is more resistant to occlusions, which indicates more applicable in real scenes.

參考文獻


[37] H.-W. Lee, C.-H. Liu, K.-T. Chu, Y.-C. Mai, P.-C. Hsieh, K.-C. Hsu, et al., "Kinect Who's Coming—Applying Kinect to Human Body Height Measurement to Improve Character Recognition Performance," Smart Science, vol. 3, pp. 117-121, 2015.
[14] A. Jain, A. A. Ross, and K. Nandakumar, Introduction to biometrics: Springer Science & Business Media, 2011.
[33] C. BenAbdelkader, R. Cutler, and L. Davis, "View-invariant estimation of height and stride for gait recognition," in International Workshop on Biometric Authentication, 2002, pp. 155-167.
[2] A. Dantcheva, C. Velardo, A. D’Angelo, and J.-L. Dugelay, "Bag of soft biometrics for person identification," Multimedia Tools and Applications, vol. 51, pp. 739-777, 2010.
[3] J. A. Unar, W. C. Seng, and A. Abbasi, "A review of biometric technology along with trends and prospects," Pattern Recognition, vol. 47, pp. 2673-2688, 2014.

延伸閱讀