透過您的圖書館登入
IP:3.141.35.99
  • 學位論文

多頂照式魚眼攝影機之人物辨識與追蹤

Topped Fisheyes For Pedestrians Recognizing and Tracking

指導教授 : 王才沛

摘要


在電腦視覺中,多物件追蹤是常見的議題,目前也有很多相關技術與研究供後人參考,但大部分的追蹤影像都是投射型攝影機,而在魚眼影像上的相關追蹤較為稀少,魚眼影像的優點是涵蓋的範圍較廣,而在人物與人物之間很靠近,在投射型攝影機會被遮蔽到的情況下,使用頂照式魚眼影像會有較好的視野資訊。本篇論文使用頂照式魚眼影像提及一個新的追蹤方法,期望在即時影像上偵測、追蹤行人。 在追蹤行人的部分,我們使用三維的voxel資訊儲存前景值,在行人的位置附近上找尋一個最大機率的位置上判斷新的行人位置,並且搭配行人的外觀特徵,用顏色直方圖來輔助追蹤系統,為了達到更高的準備度

並列摘要


In the research of computer vision, multiple object tracking is a very common issue. Most of relative techniques and research are referenced in recent years, but the majority of tracking images is projective camera image. Consequently, there is a few tracking techniques with fisheye image. Fisheye image has advantage in covering wide ground than other camera. Projective camera takes an image that is blocked by people when two people are too closed. Topped fisheye camera has a better information view than projective camera. In this paper, we refer to a new tracking method with topped fisheye camera image. In hope that we can detect and track people in real-time. We use 3-D voxels for tracking people. The position that we find a maximum probability in last frame serve as new position. For the higher accuracy, we use appearance feature to assist our tracking system, like color histogram.

參考文獻


[1] Lou et al., “Multiple Object Tracking: A Literature Review”
[2] Liang Yingyi, Li Xin, He Zhenyu and You Xinge, “Multiple Object Tracking by Incorporating a Particle Filter into the Min-cost Flow Model”
[3] Sang-Il Oh, Hang-Bong Kang, “Multiple Object Tracking using Fuzzy Logic for Handling Uncertaint”
[4] Nii Longdon Sowah, Qingbo Wu and Fanman Meng,” A Classification and Clustering Method for Tracking Multiple Objects”
[5] Lars Meinel, Michel Findeisen, Markus Heß , André Apitzsch , Gangolf Hirtz,” Automated Real-Time Surveillance for Ambient Assisted Living Using an Omnidirectional Camera”