基於多重平面與軸線分析之即時多人定位系統

Abstract

以視覺為基礎的人物定位與追蹤的研究中，人物遮蔽是一個重要且具挑戰性的研究課題。在此一研究計畫中，我們提出一多攝影機與多重參考平面的人物定位方法。先前被提出的方法是藉由將多個視角影像中的前景資訊以像素為單位（pixel-based）投影至各參考平面來確認空間中不同高度的參考平面上是否有人物存在，因此比起僅使用單一參考平面，將能夠有效地處理遮蔽之問題。然而將前景像素投影至多重參考平面將耗費大量的計算時間。為了解決上述的問題，我們希望減低投影所需的計算量。在計畫的第一年我們基於消失點對影像中前景區域產生2D樣本線段，並僅將這些2D樣本線段投影至多重參考平面，如此一來可以大幅減少計算時間。後續，我們計算這些2D樣本線段於多重平面上的交點並對這些不同平面的交點進行分析、連接後可形成多條3D的軸線。這些軸線經過評估並淘汰掉不合適的之後，可依據分群的演算法分為數群。各群內的軸線經整合後可重建出人物的位置及身高。

Occlusion has been an important and challenging task in vision-based people localization and tracking. In this research project, we propose a people localization method based on multiple cameras and reference planes. Some pixel-based methods have been proposed to check the existence of people at reference planes of different heights by projecting image foreground pixels from multiple views to these planes； such approaches can deal with occlusions better than using only a single reference plane. However, it spends a lot of computation time for projecting foreground pixels onto multiple reference planes. In the first year, we aim to reduce the amount of calculation due to projection. We use vanishing point-based line sampling to generate 2D line samples of foreground regions. Then, only these 2D line samples are projected to reference planes so that the computation time can be greatly reduced. For the subsequent localization process, intersection points of the 2D line samples on reference planes are calculated, analyzed and integrated to form several 3D axes. After evaluating and eliminating inappropriate 3D axes, those retained 3D axes are then clustered and integrated to reconstruct the locations and heights of people in the scene.