利用紅外線影像之人臉追蹤

Abstract

在無照明的環境下，紅外線影像可以捕捉物體本身主動散發的輻射熱，其有別於一般可見光影像的特性，能夠發展出不同於可見光影像分析的應用。例如使用紅外線熱影像在黑暗環境中做安全監控時，其影像分析功能之一是進行人臉追蹤的即時處理，在本論文中，是所要探討的課題。我們首先實作影像子空間模型的方法，而不同於一般追蹤所使用的模型，如利用機率或類神經的模型來做人臉追蹤。我們再將一些可能改進的方法加入到影像子空間模型中。我們所加入方法包含運動速度的預測、目標物形狀估算的改進，以及設定臨界值來判定新的影像是否該用於系統更新。實驗結果顯示，這些方法的加入可以使得人臉追蹤達到更精確的效果。

In the dark environment, infrared imaging offers a promising alternative to visible light image in various applications owing to its capability of sensing object heat emissions. For example, we can use infrared image to track human face for security surveillances in the dark in real time. In this thesis, we first implement the presented method in [17]. that uses an image subspace model. It is different from general tracking models, such as probabilistic or neural network model. In this paper, some adjustments are incorporated into the proposed method, including target motion prediction, bounding box refinement, and outlier detection. Experimental results show that the above approach can indeed improve tracking accuracy.