請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/46886
標題: | 使用具空間相互關係的區域特徵模型之部分式夜間行人偵測 Part-based Nighttime Pedestrian Detection Using Spatially Interrelated Local Feature Model |
作者: | Li-An Chuang 莊理安 |
指導教授: | 傅立成 |
關鍵字: | 行人偵測,夜間,近紅外線,幾何資訊,空間關係,基於片段, pedestrian detection,nighttime,near-infrared,geometric information,spatial relationship,part-based,poselet,HOG, |
出版年 : | 2011 |
學位: | 碩士 |
摘要: | 在智慧型運輸系統中,行人偵測是一個重要的議題,因為行人並非光源體,在夜間黑暗的環境下,駕駛者更不容易注意到行人,使得夜間行人偵測較日間行人偵測更為重要。本論文提出一個在移動式車輛上使用之基於近紅外線影片的夜間行人偵測方法,在近紅外線影像中會顯示輻射近紅外線以及反射經由我們的近紅外線投射燈投出的紅外線的區域。但在某些情況下,行人身上衣物的材質會吸收大部分投射在身上的紅外線,造成這個行人穿著此材質衣物的區域無法在紅外線影像中顯示,這種情況會使基於整體的行人偵測方法失敗,為了解決這類的議題,我們提出了一種基於片段的行人偵測方法。在過去傳統的方法中,若使要解決此類遮蔽的問題需要大量的遮蔽樣本,為了解決這個問題,我們將會學習任兩個片段間的空間關係,因此我們提出了在空間上具有相互關係的區域特徵,對於每個片段擷取此特徵以訓練分類器以及空間關係,藉由空間關係模型,我們不需要蒐集所有的遮蔽樣本,學習了各個片段間的空間關係後,即使遇偵測的行人部分被遮蔽,我們還是可以強化對偵測到的行人片段的信心。 Pedestrian detection is an important issue in the intelligent transportation system field. Since the pedestrian is not an apparent object at nighttime, it is more critical for the vision system to help detecting the pedestrian for assisting driving. Accordingly, this thesis proposes a nighttime pedestrian detection method based on near-infrared video stream on a moving vehicle. The objects in the nighttime environment will radiate infrared or reflect the infrared projected by our emitted spotlight. But in some cases, the clothes on the pedestrian will absorb most of the infrared and makes the pedestrian partially invisible in that part. To deal with this, we propose a part-based pedestrian detection method which divided pedestrian into many parts according to the feature points marked on them. Those human body parts with specific feature point configuration named poselets. For each poselet, a proposed feature named spatially interrelated local feature (SIL feature) is extract for construct the poselet classifier. For the purpose of obtaining an effective detector, the tradition methods need to collect large number of samples with occlusion and thus complicates the training process. To alleviate this problem, the spatial relationship between each two parts is automatically learned by using SIL feature. Additionally, the learned spatial relationship can enhance the confidence of the detected parts even if some parts are occluded. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/46886 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊網路與多媒體研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-100-1.pdf 目前未授權公開取用 | 2.34 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。