標題: 以正規轉換為基礎之日夜人物辨識
Canonical Transform Based Day-and-Night Person Identification
作者: 高仲義
Gao, Jhong-Yi
張志永
Chang, Jyh-Yeong
電控工程研究所
關鍵字: 人物辨識;Person Identification
公開日期: 2013
摘要: 人物辨識系統在電腦視覺領域是很熱門的研究與應用目標。在監控系統中,最常見的方式是使用固定式攝影機,對拍攝場景的人物進行人物辨識。 本論文實現一套監控系統,此系統是在日夜環境中,分別使用多角度步態辨識系統及人臉辨識系統。本文研究對於使用兩台近紅外線攝影機進行人物辨識,一台近紅外線攝影機設置在遠處,用於擷取不同方向的步態影像,另一台近紅外線攝影機設置在近處,用於擷取人臉正面影像。 在人臉辨識系統方面,我們利用近紅外線攝影機擷取人臉影像。人臉擷取的方法是使用Haar疊層分類器,這是一種基於特徵運算的演算法,這種演算法比基於逐點的更快速,接著人臉影像經過特徵空間轉換與正規空間轉換後,累積五張上述人臉影像後,藉由多數決的方式,完成人物辨識。 在步態辨識系統方面,我們利用近紅外線攝影機擷取步態影像。為了擷取出完整的人體部分,本文使用背景相減法在灰階空間與HSV色彩空間建立背景模型,並提升消除影像中陰影部分,使得擷取前景影像能夠更完整,接著步態影像經過特徵空間轉換與標準空間轉換後,累積五張上述步態影像後,藉由多數決的方式,完成人物辨識。
Human recognition system is a very popular subject for research and application. Using a camera to recognize human is widely seen in surveillance system. In this thesis, we implement the surveillance system that can recognize multi-angle human gait and human face of a person in the bright and dark environments. We use two near infrared (NIR) cameras for human recognition. One NIR camera, set in remote location, capture the gait images from different angles. And the other NIR camera, set in the vicinity, capture the face images from the person frontal view. In human face recognition system, face region of an image is extracted based on Haar cascade classifier, which is a feature-based algorithm and works much faster than the pixel-based algorithm. Then, the face image is transformed to a new space by eigenspace and canonical space transformation for better efficiency and separability. The recognition is finally done in canonical space. Moreover, we gather five consecutive face images from video, and use majority vote to recognition the human. In human gait recognition system, we build two background models, one in grayscale and one in HSV color space to extract the foreground image correctly. Then we reduce the shadowing effect. The gait image is then transformed to a new space by eigenspace and canonical space transformation for better efficiency and separability. The recognition is done in the canonical space. Finally, we gather five consecutive gait images from video, and use majority vote to recognition the person.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070060063
http://hdl.handle.net/11536/72487
顯示於類別:畢業論文


文件中的檔案:

  1. 006301.pdf