基於機率佔有率圖人物定位方法的強化

Abstract

在以視覺為基礎的人物定位研究中，人物遮掩是一個重要且具挑戰性的研究課題。隨着視覺監視系統的大量發展，人物定位方法的精準度和效率有明顯的提升。基於機率占有率圖(probabilistic occupancy map，POM)人物定位方法[1]因爲能在嚴重的遮蔽情況下得到精準的人物定位結果，所以成爲人物定位方法的主流之一。方法[1]假設影片是在頭或眼睛高度拍攝，所以人的模型會是矩形的，可以使用積分影像(integral image)加快運算速度。 本論文提出一系列基於垂直線消失點 (vanishing point of vertical lines)的影像轉換，以提升基於機率占有率圖(probabilistic occupancy map，POM)人物定位方法[1]定位效果。本計畫利用垂直線的延伸線會交於消失點的特性和影像/地面的座標系統產生轉換影像。轉換影像的主要特性爲畫中的人物會是直立的。如此，因爲攝影機架設方式的限制，而導致方法[1]的人物定位結果不佳現象可以被減緩。實驗結果證明，在一般攝影機的架設方式下，本計畫提出的影像轉換能有效的改善基於機率占有率圖人物定位方法的結果。 在一般攝影機的架設方式下，雖然基於機率占有率圖人物定位方法的定位精準度可以透過提出的影像轉換提升，但此人物定位方法的計算複雜度非常高。因此，一個線段取樣的方法於此論文提出。我們不分析整張影像，我們只對影像中的線段取樣部份進行分析。此外，因爲是線段，所以可以同時解決攝影機架設的限制。實驗結果顯示，提出的方法能有效的提升基於機率占有率圖人物定位方法的速度，且和影像的解析度無關。同時，當影片是一般攝影機架設方式下得到的，人物定位的精準度也能有效的提升。爲了更進一步增加基於機率占有率圖人物定位方法的處理速度，我們提出兩個方法：(i)快速找出人物可能在的位置，和(ii)時間化機率佔有率迴圈停止條件。實驗結果顯示提出的方法比基於機率占有率圖人物定位方法快7.25倍，且擁有相似的人物定位精準度。

With the pervasive deployment of vision-based security surveillance systems, the improvement in accuracy and efficiency of people localization has attracted remarkable research efforts. Recently, employing probabilistic occupancy map (POM) [1] becomes one of the main research trends in people localization due to its great localization accuracy under severe occlusions and adverse lighting conditions. In [1], it is assumed that videos are taken at head or eye level so that rectangular human models, and the computationally efficient integral image, can be utilized. In this thesis, a series of novel image transforms based on the vanishing point of vertical lines are proposed for enhancement of the POM-based people localization scheme. Utilizing the characteristic that the extensions of vertical lines intersect at a vanishing point, the proposed transforms, based on image or ground plane coordinate system, aims at producing transformed images wherein each standing/walking person will have an upright appearance. Thus, the degradation in localization accuracy due to the deviation of camera configuration constraint specified in [1] can be alleviated, while the computation efficiency resulted from the applicability of integral image can be retained. Experimental results show that significant improvement in POM-based people localization for more general camera configurations can indeed be achieved with the proposed image transforms. Although the localization accuracy of the POM-based approach is enhanced by the proposed image transforms for general camera configurations, the computation complexity of the POM-based approach is high. Accordingly, a novel enhancement scheme for the POM-based approach is proposed in this thesis by employing a line sampling scheme. While high computation complexity can be alleviated by dealing with line samples instead of the whole image, and the camera configuration constraint is resolved by properly choosing the origin of these line samples, i.e., the vanishing point of vertical lines in the scene. Experimental results show that the proposed approach, which is essentially independent of image resolution, can indeed improve the efficiency of the POM-based approach, and also increase the localization accuracy for general camera configurations. For further enhancing the efficiency of the POM-based approach, two enhancement schemes are proposed in this thesis: (i) quick screening of potential people locations, and (ii) timely termination of iterations for occupancy probability estimation. Experimental results show that the proposed approach achieves up to 7.25 times speed-up compared to the POM approach, while delivering comparable people localization accuracy.