低解析度紅外線影像系統之家庭保全應用

Abstract

本論文探討低解析度檢知器與適當光學組成之熱像系統設計，以為低成本之家庭保全應用。檢知器為免凍式之熱阻型感測焦面陣列，其設想以微機電技術製作，並與標準CMOS單晶片製程匹配；同時考慮元件相關之材料結構所表現的熱導、熱容、電阻溫度係數與雜訊，以得最佳化之設計。在熱像系統的光學方面，分析以傳統的折射式透鏡、反射鏡，以及塑膠材的折射式傅涅耳透鏡等的可行性；又為增加檢知器受光的填充因子以提高靈敏度，本論文亦以單一特殊圖樣光罩設計多步階繞射式微透鏡。在熱像系統工程中，進一步分析待測目標輻射特性與各種典型天候下的大氣對紅外光譜的衰減影響，確定熱型感測系統用8-14□響應譜帶較3-5□為佳。利用上述結果，設計室內保全用途之低解析度熱像系統，由系統需求估算辨識室內目標物時必需的最小特徵尺寸，以及檢知器面陣列的相關尺寸與解析度。研究中，充分利用紅外輻射特性做影像前置處理，並以目標物的形狀特徵建構判斷法，進行人或寵物的影像辨識模擬。結果顯示用類神經網路方法可得極為可靠的辨識機率，且目標物的最小尺寸僅8□8即可。另有關火焰之分析，則利用其高溫及尺寸擴增率等特徵以獲致可靠之早期預警效果。本研究之結論為，低解析度之單晶片免凍式熱阻型焦面元件足以做為室內紅外線感測的保全應用，具有高可靠度與經濟性之雙重優點。

A thermal imager, using a low-resolution detector and appropriate optics for low-cost home security applications, is discussed in this thesis. The detector is envisaged on an uncooled microbolometer IRFPA, fabricated with MEMS technology, which is compatible with standard CMOS processes. An optimal design for this device, considering its material and structure are related to thermal conductance, thermal mass, TCR, and noise is proposed. The feasibility of a conventional refractive lens, reflective mirror, plastic refractive Fresnel lens was analyzed for the optical system. The article also presents a diffractive micro lens achieved using a single mask with a special pattern to increase the detector fill-factor for improving the sensitivity. In thermal imager system engineering, the system spectral band is determined by analyzing the target signatures under typical atmospheric conditions, concluding that 8-14□ is better than 3-5□. According to the above results, a low-resolution thermal imager for home security applications was designed. The sensor size and resolution were evaluated to meet the optical constraints of the system and make the minimum target feature recognizable. Features of human beings and family pets were extracted using a special infrared image pre-processing and recognized with target-shape distinction methods. This demonstrates that highly reliable detection of human beings or pets, with a minimal target feature of 8□8 pixels, can be obtained using the neural-network method. In addition, a fire can be detected early using the higher temperature and size enlargement rate.