熱影像行人偵測與人臉辨識演算法及其於夜間陌生人偵測之應用

Abstract

近年在防盜系統等等應用的需求下，行人偵測演算法與人臉辨識演算法已成為關鍵的研究領域。而由於熱影像能清楚地反應人體皮膚表面溫度的輻射，因此熱影像已經被使用來避免由於光源的照度變化對系統所產生的影響。然而傳統的熱影像行人偵測與人臉辨識系統的偵測效能仍然不足，所以本論文提出新的熱影像行人偵測演算法與人臉辨識演算法來提升系統的效能。本論文提出的熱影像行人偵測演算法，設計一組新的行人輪廓特徵，不同於傳統行人特徵，該組特徵由行人候選區域的選取技術所發展而來，本論文提出的熱影像行人偵測演算法，可以改善傳統熱影像行人偵測系統的偵測率。另外一方面，本論文提出的熱影像人臉辨識演算法，不同於傳統辨識系統只使用溫度值統計資訊做為特徵，也使用較高溫區域的分布位置資訊，可以改善傳統熱影像人臉辨識演算法的效能。本論文將提出的熱影像行人偵測演算法與人臉辨識演算法，搭配可見光人臉偵測演算法，應用於夜間陌生人偵測。實驗結果顯示，可有效偵測夜間的陌生人。

Pedestrian detection and face recognition are critical research fields given the requirements of applications such as burglarproof systems. Thermal imagery has been proposed to eliminate variations in illumination because it clearly identifies the heat radiating from the human skin. However, the performance of traditional thermal pedestrian detection and face recognition systems remains insufficient. Thus, this dissertation presents pioneering thermal imagery algorithms for pedestrian detection and face recognition to enhance the performance of systems. The proposed pedestrian detection algorithm of thermal image involves pioneering outline features developed from the ROI generation method of pedestrians, which are different from traditional features. The proposed pedestrian detection algorithm of thermal image improves performance of the traditional thermal pedestrian detection system. The proposed thermal face recognition algorithm uses not only statistical temperature information but also the position information of areas with higher temperature. The proposed face recognition algorithm of thermal image improves performance of the traditional thermal face recognition system. The proposed pedestrian detection and face recognition algorithms of thermal image interact with face detection algorithm of visible images, and are applied to stranger detection at night. Thus, the proposed system can detect strangers at night in the presence of pedestrians.