利用紅外線影像之視覺監控

Abstract

由於紅外線影像是捕捉物體本身主動散發的輻射熱，並不受到亮度影響，讓我們在一般可見光影像的應用之外，能夠做出其他不同的應用。例如，在黑暗環境中的安全監控，以及溫度方面的檢測。在本論文中，我們會先研究紅外線影像的特性。接下來，我們將發展能夠即時處理和以區域生長為基礎的人體切割方法，以提供安全監控方面的應用。不同於目前大家所熟悉的背景相減技術，在分割前景物體中，我們所提出的方法是以溫度分佈為基礎，將高溫的區域成長出來，而且我們不需要使用背景的模型。除此之外，在切割出人體之後，我們能夠再將頭部的位置給偵測出來。實驗結果也證明，在絕大多數的情況下，我們能夠將人體給切割出來以及將頭部偵測出來，並且達到即時的處理。

Infrared imagery offers a promising alternative to visible light imagery in various applications owing to its capability of sensing object heat emissions in the distance. It is thus particularly suitable for security surveillances in the dark and for temperature inspections. In this thesis, we first investigate the characteristics of infrared imaging. Then, for visual surveillance applications, we develop region growing-based methods to segment human bodies in real-time. Different from popular background subtraction techniques for separating foreground objects, the proposed methods are based on temperature distributions to grow target regions and have no need to maintain a background model of the view field. Besides, head positions can thereafter be allocated using the segmented shape information. Experimental results show our methods can segment human shapes and allocate their heads correctly in most cases and achieve the real-time performance.