本研究內容為以電腦視覺為基礎之室內監控系統。
監控系統的首要任務便是隨時都能正確定位出監視畫面上目標物的位置。一開始我們引入單台攝影機監控環境下的單人追蹤。此部分使用Particle Filter配合顏色統計作為追蹤目標的特徵。接著增加監控人數。由於室內環境的限制,勢必經常出現多人間互相遮蔽的現象。而此現象將使追蹤目標的特徵在監控視野中暫時消失，導致追蹤失敗。因此我們發展一套多攝影機為基礎之追蹤演算法解決此問題。
我們將這些追蹤目標在三度空間上的狀態投影到每個影像平面上，並估測在二維投影平面上屬於每個追蹤目標物的狀態空間。在此我們視複雜的多人互動過程為貝氏網路中的隱變數。這些隱變數可以從每個時刻下所有追蹤目標的狀態聯合推算。藉由推斷這些隱變數可以幫助我們在下一時刻有效分配追蹤演算法的運算資源。我們因此可以在較受信賴(較可能取得有意義資訊)的視野上使用更多的計算量。此外，這些隱變數也幫助我們整合在每台攝影機擷取之影像上得到的追蹤結果，使我們推斷出目標更精確的位置。這個修正過後的追蹤結果可以經由轉換矩陣傳遞到發生遮蔽的視野上，以幫助該視野在物體脫離遮蔽後繼續追蹤。在本論文中我們使用DLT演算法找出攝影機監控畫面之間的空間對應關係，以此對應矩陣連結我們估算出的隱變數。

Multiple human tracking is an important research topic in computer vision. In a crowded environment, occlusions occur frequently that makes the tracking problem even more challenging. This thesis presents a multi-view-based cooperative tracking of multiple human objects. Based on the homographic relation between two views, we proposed a so-called cooperative tracking which consists of particle filter tracking for the objects in different views. The multiple view tracking is modeled as different sequences of hidden process and observation. In addition, based on the interaction among the targets, a hidden variable is added in to reveal the reliability of the tracking result in that specific view. With this hidden variable, the cooperative tracking allocates computational resources for tracking the objects in different views. Experimental results show the efficiency of the proposed method.

[1] M. Isard and A. Blake, “CONDENSATION-Conditional Density Propagation for Visual Tracking”, IJCV, 1998, pp. 5 – 28.
[2] P. Pérez, C. Hue, J. Vermaak, and M. Gangnet, “Color-Based Probabilistic Tracking”, In European Conference on Computer Vision, 2002, number 2350 in Lecture Notes in Computer Science, pp. 661–675.
[3] K. Nummiaro, E. Koller-Meier, and L.Van Gool, “An Adaptive Color-Based Particle Filter”, IVC, Vol. 2(1), 2003, pp.99-110.
[4] Y. Wu, T. Yu, and G. Hua, “Tracking Appearances With Occlusions”, Proceedings IEEE CVPR 2003, Volume 1, June 2003, pp. 789-795.
[5] M. Hu, W. Hu, and T. Tan, “Tracking People Through Occlusions”, ICPR Aug. 2004, pp.724 – 727.
[6] T. Matshuyama and N. Ukaita, “Real-Time Multitaget Tracking by a Cooperative Distributed Vision System”, Proceedings of the IEEE, Vol. 90, No. 7, July 2002, pp. 1136 -1150.
[7] A.Utsumi, H. Mori, J. Ohya, and M. Yachida, “Multiple-View-Based Tracking of Multiple Humans”, Proceedings of ICPR Aug. 1998, pp. 597 – 601.
[8] K. Otsuka and N. Mukawa, “A Particle Filter For Tracking Densely Populated Objects Based on Explicit Multiview Occlusion Analysis”, Proceedings of ICPR, 23-26 Aug. 2004, pp. 745 – 750.
[9] S.Kang, B.-W. Hwang and S.-W. Lee, “Multiple People Tracking based on Temporal Color Feature”, International Journal of Pattern Recognition and Artificial Intelligence, Vol. 17, No. 6, 2003, pp. 931 – 949.
[10] R. Hartley and A. Zisserman, “Multiple View Geometry in Computer Vision”, Cambridge University press, 2003.
[11] D. Fox, “KLD-Sampling: Adaptive Particle Filters”, Advances in Neural Information Processing Systems (NIPS), 2001.