一個基於多節梯形的無軀幹人像活動表示方法

Abstract

由於人類的平均年齡逐漸上升，不管在安養中心或是醫院，都需要大量的看護與醫療資源。因此我們提出一個基於電腦視覺的監控系統，幫助醫療人員或家屬能夠透過遠端即時的觀察被觀測者的活動情況。醫療監控系統不同於一般的安全監控系統，由於被觀測者通常不希望自己的真實影像被記錄和檢視，因此須注重保護被觀測者的隱私。由於被觀測者發生的情況經常需要緊急的處理，醫療監控系統同時也需要提供醫療人員即時性的資料。因此建立一個注重隱私，且具有較少資料量以及運算量的人體模型，是本篇論文的研究主題。在本篇論文中，我們提出一個快速且簡單的演算法，藉由監視攝影機，獲得人類影像中的肢體部位。最終提供一個基於多節梯形的無軀幹模型，並利用此無軀幹的肢體模型達成人像活動分析的目的。

Because of the increase of average human age over the years, plenty of medical and nursing resources are needed, either in a nursing center or a hospital. To that end, we propose a surveillance system based on computer vision to help medical staffs or family members to monitor patient activities via remote access. Different from the surveillance for security purposes, it is usually desirable that original videos of patients’ daily life are not available to the public. Therefore, the main feature of the proposed system is that the personal privacy is always ensured by using only a synthetic yet realistic human pattern to represent a patient. Since there are urgent situations in which instant care of patients is needed, a simple torso-less human pattern of multi-joint trapezoids is proposed in this paper, which can be generated near real-time through a computationally efficient analysis of a surveillance video. Experimental results show that one can indeed comprehend human activities reasonably well by examining the proposed torso-less human pattern.