利用足跡分析來加速基於消失點射線取樣之人群定位演算法

Abstract

近年來，以視覺為基礎的人群定位與追蹤越來越受到重視，也不斷發展出新的技術與應用。然而，大部分的方法都需仰賴大量的計算方能處理嚴重遮蔽的問題，且往往需倚賴特殊硬體才能達成即時的定位與追蹤。不同於這些研究，本論文提出一快速且準確的多攝影機人群定位演算法，對前景區域建立以消失點為基礎的二維樣本線段，並將之投影於地平面，利用足跡分析找出線段相交密集處，有效限縮人物立足點在地平面的可能範圍。再透過二維前景影像，對人物立足點可能區域做進一步的篩選與驗證，有效率地估計出人物的位置與高度。本篇論文不需大量分析人物特徵點，有效率地降低系統的計算成本以符合即時運算的需求。經實驗證明，本篇論文演算法相較於先前研究[9]的人物三維重建方法，在多人且嚴重遮蔽的環境中可提升至十倍計算速率，且依然不失偵測正確度與定位精準度，進而達成即時的三維人群定位。

With the popularity of vision-based camera surveillance, the research on people localization appeals to much attention. In this study, we propose an efficient and effective system capable of locating a crowd of dense people in real time, using multiple cameras. For each camera view, line samples, originated from a vanishing point, of foreground objects are projected on the ground plane. Ground regions containing a high density of projected lines are then used to find people locations. Enhanced from previous works, the people localization approach proposed in this study needs not project all foreground pixels of all views to multiple reference planes or compute pairwise intersections of projected sample lines at different heights, resulting in significant improvement in computational efficiency. Furthermore, the people heights can also be estimated. Experimental results on real surveillance scenes show that comparable accuracy in people localization can be achieved with ten times in computing speed compared with our previous approach.